National Academies Press: OpenBook
« Previous: Front Matter
Page 1
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 1
Page 2
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 2
Page 3
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 3
Page 4
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 4
Page 5
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 5
Page 6
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 6
Page 7
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 7
Page 8
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 8
Page 9
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 9
Page 10
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 10
Page 11
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 11
Page 12
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 12
Page 13
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 13
Page 14
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 14
Page 15
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 15
Page 16
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 16
Page 17
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 17
Page 18
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 18
Page 19
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 19
Page 20
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 20
Page 21
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 21
Page 22
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 22
Page 23
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 23
Page 24
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 24
Page 25
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 25
Page 26
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 26
Page 27
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 27
Page 28
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 28
Page 29
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 29
Page 30
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 30
Page 31
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 31
Page 32
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 32
Page 33
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 33
Page 34
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 34
Page 35
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 35
Page 36
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 36
Page 37
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 37
Page 38
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 38
Page 39
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 39
Page 40
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 40
Page 41
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 41
Page 42
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 42
Page 43
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 43
Page 44
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 44
Page 45
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 45
Page 46
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 46
Page 47
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 47
Page 48
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 48
Page 49
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 49
Page 50
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 50
Page 51
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 51
Page 52
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 52
Page 53
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 53
Page 54
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 54
Page 55
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 55
Page 56
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 56
Page 57
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 57
Page 58
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 58
Page 59
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 59
Page 60
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 60
Page 61
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 61
Page 62
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 62
Page 63
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 63
Page 64
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 64
Page 65
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 65
Page 66
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 66
Page 67
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 67
Page 68
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 68
Page 69
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 69
Page 70
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 70
Page 71
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 71
Page 72
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 72
Page 73
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 73
Page 74
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 74
Page 75
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 75
Page 76
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 76
Page 77
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 77
Page 78
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 78
Page 79
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 79
Page 80
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 80
Page 81
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 81
Page 82
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 82
Page 83
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 83
Page 84
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 84
Page 85
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 85
Page 86
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 86
Page 87
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 87
Page 88
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 88
Page 89
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 89
Page 90
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 90
Page 91
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 91
Page 92
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 92
Page 93
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 93
Page 94
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 94
Page 95
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 95
Page 96
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 96
Page 97
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 97
Page 98
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 98
Page 99
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 99
Page 100
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 100
Page 101
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 101
Page 102
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 102
Page 103
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 103
Page 104
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 104
Page 105
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 105
Page 106
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 106
Page 107
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 107
Page 108
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 108
Page 109
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 109
Page 110
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 110
Page 111
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 111
Page 112
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 112
Page 113
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 113
Page 114
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 114
Page 115
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 115
Page 116
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 116
Page 117
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 117
Page 118
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 118
Page 119
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 119
Page 120
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 120
Page 121
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 121
Page 122
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 122
Page 123
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 123
Page 124
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 124
Page 125
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 125
Page 126
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 126
Page 127
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 127
Page 128
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 128
Page 129
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 129
Page 130
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 130
Page 131
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 131
Page 132
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 132
Page 133
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 133
Page 134
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 134
Page 135
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 135
Page 136
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 136
Page 137
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 137
Page 138
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 138
Page 139
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 139
Page 140
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 140
Page 141
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 141
Page 142
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 142
Page 143
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 143
Page 144
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 144
Page 145
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 145
Page 146
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 146
Page 147
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 147
Page 148
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 148
Page 149
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 149
Page 150
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 150
Page 151
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 151
Page 152
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 152
Page 153
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 153
Page 154
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 154
Page 155
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 155
Page 156
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 156
Page 157
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 157
Page 158
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 158
Page 159
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 159
Page 160
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 160
Page 161
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 161
Page 162
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 162
Page 163
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 163
Page 164
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 164
Page 165
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 165
Page 166
Suggested Citation:"Final Report." National Academies of Sciences, Engineering, and Medicine. 2005. Predicting Air Quality Effects of Traffic-Flow Improvements: Final Report and User's Guide. Washington, DC: The National Academies Press. doi: 10.17226/13797.
×
Page 166

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Final Report

The NCHRP 25-21 project identified and investigated most significant impacts of traffic-flow improvements on travel behavior and air quality suspected or known at this point in time. The impacts of traffic-flow improvements on household trip mak- ing, destination choice, time-of-day choice, mode choice, and route choice have been considered and included in a recommended comprehensive methodology for pre- dicting the air quality impacts of traffic-flow improvements. The long-term impacts on the redistribution of future economic activity from less accessible areas of the region to more accessible areas have also been considered and incorporated into the methodology. Only two identified impacts of traffic-flow improvements on air quality have been intentionally excluded from the methodology: the potential direct impact on the over- all growth of a metropolitan region and the potential indirect impact of traffic-flow improvements on actual or perceived accessibility (via nonmotorized modes) for tran- sit, pedestrian, and bicycle modes. Both impacts were excluded because of the lack of necessary data and limitations of project resources. The NCHRP 25-21 methodology was applied to a series of case studies, and the results were compared with more general results reported in the literature. The facility- specific results showed travel time and volume changes that were consistent with the- ory and expectation. However, it was harder to validate the methodology’s predictions for system-level (i.e., regionwide) performance. Some of the results fell within the broad range of results that have been reported in the literature. Other results fell outside the range of results reported in the literature. Indeed, application of the methodology to the same traffic improvement at different locations in the region showed a wide range in pre- dicted systemwide impacts. The same type of project (adding an HOV lane, for exam- ple) resulted in net benefits or disbenefits to regional emissions, depending on its location. The NCHRP 25-21 methodology was applied to 10 case studies. The impacts of individual traffic-flow improvement projects on regional daily vehicle-miles traveled (VMT) were on the order of a few hundredths of 1 percent. A 30-year improvement program impacted regional VMT by less than 1 percent. The impacts varied from a SUMMARY PREDICTING AIR QUALITY EFFECTS OF TRAFFIC-FLOW IMPROVEMENTS FIN A L R EPO RT

reduction in emissions to an increase in emissions, depending upon the specifics of each case study. The case study results suggest that more applications of each traffic-flow improvement type in different facilities, in different area types, and at dif- ferent congestion levels are needed to better understand the conditions under which traffic-flow improvements contribute to an overall net increase or decrease in vehicle emissions. 2 FI N A L R EP O RT

FIN A L R EPO RT 3 CHAPTER 1 INTRODUCTION This report presents the results of the research for the National Cooperative Highway Research Program (NCHRP) Project 25-21 to develop a methodology to predict the long- and short-term mobile source emission impacts of traffic- flow improvement projects. 1.1 ORGANIZATION OF THIS REPORT This first chapter outlines the structure of this paper and reviews the research project objectives. The second chapter provides an overview of the theory and evidence for the impact of traffic-flow improvement projects on mobile source emissions. Chapter 3 presents a review of the state of the practice at typical planning agencies. That chapter outlines the methodologies used to forecast the emission impacts of transportation projects used by some of the more advanced metropolitan planning organizations (MPOs) in the United States. A lengthy critique of the shortfalls of current practice is included in that chapter. Chapter 4 presents an overview of the available methodolo- gies in the literature for improving current practice. Chapter 5 describes current methodologies in the literature for conduct- ing sketch-planning analysis, such as the Highway Economic Requirements System (HERS), the Surface Transportation Efficiency Analysis Model (STEAM), and the Spreadsheet Model for Induced Travel Estimation (SMITE). Chapter 6 discusses the state of the art in land-use forecasting. It delves into current land-use modeling practice, its shortfalls, and two of the more promising models currently available (the High- way Land Use Forecasting Model [HLFM] and UrbanSim). The next three chapters consider potential state-of-the-art improvements to conventional travel models. Chapter 7 presents some of the more promising advances in the field of travel demand forecasting (the Portland Tour-Based Model, the Transportation Analysis Simulation System [TRANSIMS], and the Short-Range Transportation Evalu- ation Program [STEP]). Chapter 8 presents vehicle opera- tions models (The Highway Capacity Manual and Corridor Simulation [CORSIM] are highlighted) and describes tech- niques for linking travel demand models to these vehicle models. Chapter 9 presents mobile source emission models: Mobile 6, Georgia Tech’s Mobile Emission Assessment System for Urban and Regional Evaluation [MEASURE], and the NCHRP 25-11 modal emission model. Chapter 10 evaluates various strategic approaches to the development of a methodology to predict the emission impacts of traffic-flow improvement projects. Chapter 11 presents the recommended methodology. The next five chapters (Chapters 12–16) present the deriva- tion and testing of the individual modules of the methodology. Chapter 17 presents the results of various investigations into the validity of the methodology. The final chapter, Chapter 18, summarizes the results of the research and presents a recommended program for dis- seminating the results to the professional community. 1.2 SUMMARY OF PROBLEM BEING RESEARCHED With the passage of the Intermodal Surface Transportation Efficiency Act of 1991 (ISTEA) and its reauthorization, the Transportation Equity Act for the 21st Century (TEA-21), increased emphasis has been placed on informed decision making regarding the full range of environmental, system performance, financial, and other implications of statewide and metropolitan transportation plans and programs. A major component of providing accurate impact assessments centers on effective data collection and analytic methods to support decision makers. The total air quality effects of transportation projects, espe- cially those designed to improve traffic flow, are not fully understood. Projects may result in beneficial or detrimental impacts over the short or long term. For example, traffic-flow improvement projects may have a short-term air quality ben- efit by reducing congestion and increasing speed yet have a negative effect by facilitating additional travel. Also, trans- portation actions such as high-occupancy vehicle (HOV) proj- ects, tolling strategies, and reduction in parking availability may have long-term air quality benefits by reducing trips and VMT, yet might make air quality worse in the short term by increasing congestion and queuing. Research is needed to improve the information available to support decision mak- ing in project evaluation, selection, and priority program- ming. Further, more accurate and objective information is needed by transportation decision makers regarding the full range of effects and impacts associated with traffic-flow improvement projects over the life of those projects.

1.3 OBJECTIVES OF THE RESEARCH The objective of this research has been to develop and demonstrate, in case study applications, a methodology to predict the short-term and long-term effects of corridor-level, traffic-flow improvement projects on carbon monoxide (CO), volatile organic compounds (VOCs), oxides of nitrogen (NOX), and particulate matter (PM). The methodology should eval- uate the magnitude, scale (such as regionwide, corridor, or local), and duration of the effects for a variety of representa- tive urbanized areas. This project will result in analytical methods for assessing long- and short-term air quality and other effects; however, it is hoped that a visionary approach can be applied to the broadest range of issues and options. The research will focus on analytical methods that can be implemented in a broad range of existing software used for travel demand modeling. The potential audience for this research will be broad, including both technical and nontechnical interests. The final product will become a tool for effective decision making in investing transportation resources and should provide both qualitative policy direction as well as a “state of the practice” methodology for analyzing emission impacts. There is no expectation for the research to predict pollu- tant concentrations or ozone formation resulting from traffic- flow improvement projects. Rather, the research is expected to use the best available emission factors and vehicle opera- tions and activity data to estimate net changes in emissions of ozone precursors, particulates, and carbon monoxide. 1.4 OVERVIEW OF THE WORK PLAN The work plan consists of seven tasks: • Task 1. Literature Review: The objective of this task was to conduct a review of transportation and air qual- ity literature to determine previous and current research studies that will support the objectives of and provide tools for the research. • Task 2. Devise Methodology: The objective of this task was to devise a methodology to predict short-term (less than 5 years) and long-term (more than 10 years) air pol- lutant emission effects (and consequently the air quality 4 effects) of completed traffic-flow improvement projects. The methodology should evaluate those effects at the local, corridor, and regional scales. Projects include, for example, added freeway lanes, arterial widenings, inter- section channelization, access management, HOV lanes, signal coordination, transit improvements, ramp meter- ing, and park-and-ride lots. The methodology should include consideration of secondary effects of traffic- flow improvements, including possible changes in emis- sions resulting from project impacts on (1) safety and accessibility for pedestrians, bicyclists, and transit users and (2) land use. To the extent possible, the methodol- ogy should be designed to use data sources commonly available in the transportation planning process. • Task 3. Develop Case Study Criteria and Candidate Projects: The objectives of this task were to (1) develop, for panel review and comment, criteria to select case studies and (2) identify a variety of project types and urbanized areas (e.g., high- and low-growth areas and areas with heavy and light congestion) for which there are available appropriate data. This task presented poten- tial case study project types and urbanized areas that meet the criteria developed in this task. • Task 4. Interim Report: The objective of this task was to prepare an interim report covering the work performed and the findings from Tasks 1 through 3. The interim report recommended any necessary work plan modifi- cations for panel review and approval. • Task 5. Test and Validate Methodology: This task involved the selection of a variety of traffic-flow improvement projects and testing and validation of the methodology. This task identified deficiencies in the analytical approach. A brief technical memorandum was developed summarizing the findings and recom- mending improvements in the methodology for review by the panel. • Task 6. Refine Methodology: This task refined the methodology based on the comments of the panel and the validation results from the previous task. • Task 7. Final Report: This task documented in the final report the results of the research, including both the methodology and the results of the case studies. A user’s guide was produced describing the recommended methodology and providing example problems (i.e., case studies) illustrating the application of the methodology. FI N A L R EP O RT

FIN A L R EPO RT 5 CHAPTER 2 THE IMPACTS OF TRAFFIC IMPROVEMENTS ON EMISSIONS This chapter presents a selective review of the literature and the general knowledge of how traffic-flow improvements affect mobile source emissions. This chapter focuses on a basic understanding of the subject. Later chapters will look at the state of the practice and will review potential method- ologies in the literature for improving the ability to predict the impacts of traffic-flow improvements on emissions. 2.1 PROBLEM STATEMENT Within the past decade, transportation professionals have reluctantly accepted that many of the transportation projects that are implemented affect the level of travel demand. Most importantly, following a landmark court case in the San Fran- cisco Bay Area,1 the existence of induced demand for travel has been recognized and must be dealt with in planning trans- portation facilities. Induced demand, however, is not the only effect that changes to transportation facilities will have. The context within which induced demand has surfaced is that of the addition of capacity, usually through widening an exist- ing, congested roadway. However, also of importance is the effect of a myriad of transportation demand and supply man- agement and low-cost investments, many of which are aimed at achieving the opposite of induced demand, namely the reduction of vehicular travel. One of the primary driving forces in looking at demand and supply strategies that might reduce vehicular travel is that of conforming to air quality standards in metropolitan areas. Although technology has made major strides in reduc- ing per-vehicle pollutants, increases in vehicular travel may eventually outpace the ability of technology to reduce pollu- tants. The result is that many metropolitan areas have found it difficult to achieve reductions in emissions from vehicles. During the late 1980s, and throughout the 1990s, many agen- cies have invested time and effort into a variety of strategies aimed at reducing vehicular travel, increasing occupancy of private vehicles, and shifting travel into public transit or non- polluting modes like bicycle and walking. The major question that needs to be addressed with respect to these strategies, and which is the focus of the current research project, is that of how much these various strategies are able to impact the level of travel. The question is asked as to which strategies are effective and by how much can they change vehicular traffic, in both the short run and the long run. Included in the types of projects and strategies to be con- sidered in this respect are added freeway lanes, arterial widen- ing, intersection channelization, access management, HOV lanes, signal coordination, transit improvements, ramp meter- ing, and park-and-ride lots. It is also desired that the method- ology developed to assess the impacts of such projects be capable of assessing not only the primary, direct effects, but also the secondary effects, such as changes in safety and accessibility for pedestrians, bicyclists, and transit users, and also for land-use changes. 2.1.1 Theories of Change in Travel Demand Economic theory is clear that if all other factors are equal, a change in the price of a commodity will result in a change in the quantity demanded by consumers. The application of such economic theory to travel demand has long been established as appropriate (Oi and Shuldiner,2 Wohl and Martin,3 and oth- ers). Of potentially more concern is what constitutes “price” in the context of travel demand. A number of studies in the late 1960s and early 1970s, aimed at putting a value on travel time, established fairly clearly that time and monetary costs make up price (Quarmby,4 Lisco,5 Haney,6 Groneau,7 Hensher,8 and Watson9). In addition, another researcher, de Donnea,10 established that the circumstances under which the time was spent, i.e., the comfort, convenience, and other attributes, also affected the value of the travel time and therefore the price. A further contribution in this debate, and one on which much of the later work relied, was that of Lancaster,11 who postulated that economic price could contain a number of attributes of a commodity that contribute to the satisfac- tion or enjoyment of the consumer. In spite of this work, and the recognition of a downward- sloping demand curve for travel and the activity to which a person traveled, the transportation planning profession was slow to accept the notion that a change in the transportation facilities, such as the addition of capacity under conditions of congestion, might actually result in a net increase in travel. Such a conclusion was pointed out by P. R. Stopher and A. H. Meyburg (Urban Transportation Modeling and Plan- ning, Lexington Books, D.C. Health and Co., Lexington, Mass- achusetts, 1975, pp. 220–221), but largely ignored by the pro-

fession, because of the complexities and expense of account- ing for such changes in demand. It is now, however, much more widely accepted in the profession that changes in trans- portation supply will, all other factors being equal, result in changes in the level of demand. Thus, adding new capacity to an existing congested roadway in the hopes of reducing pollutants from vehicles caught in stop-and-go driving con- ditions could actually have the reverse effect if the level of induced traffic is such as to more than offset the pollution reductions obtained from speeding up the traffic. In some cases, the amount of additional traffic generated or diverted may be such as to exceed in pollutants the pre-capacity addi- tion situation, thereby leading to a worsening in air quality. As Stopher13 pointed out, there are several possible reactions that transportation users may have to a change in transportation facilities. Stopher’s arguments applied specifically to capacity increases, but also apply equally well to a wide range of other transportation system changes. The principal reactions are • Change route of travel; • Change time-of-day of travel; • Change mode of travel; • Change destination of travel; • In the longer term, change work place location or home location, i.e., make substantial and significant change to repeated origins and destinations; and • Change amount of travel. Only the last one of these reactions is actually a change in the level of demand (induced traffic if the change is an increase in trips, and suppression of travel if the change is a reduction). The other reactions result in diverted trips. In theory, then, when a change is made to transportation facilities, a range of possible results can occur, in which traffic is diverted away from a facility that loses capacity and to a facility that increases capacity. A combination of diversion and either induced or suppressed travel is what will give rise to changed volumes on the facility that has experienced the change. 2.1.2 Estimation of Demand Changes The next issue is to consider if it is possible to estimate or forecast the extent of such changes. To a large extent, diver- sions of trips can be forecast or estimated provided that the travel-forecasting models are sensitive to both travel time and congestion. Probably, a model that includes only travel time will underestimate diversion because it does not take into account the circumstances under which the time is spent. Congested travel is considered more onerous than uncon- gested travel of the same duration. If capacity is increased for a transportation facility, all other factors being equal, con- gestion will be reduced and travel diversion will take place as a result of both the reduced travel time afforded by the capacity increase and the improved circumstances under which the travel takes place (i.e., a reduction in the level of 6 congestion). Conversely, if capacity is reduced or price of travel increased for some alternatives, then travel diversion will take place, all other factors being equal, as a result of increased travel times or prices and potentially increased congestion. The increase in congestion will likely cause a larger diversion of travel away from the facility that has been changed than would be estimated by travel time effects alone. Estimating induced or suppressed travel and changes result- ing from a change in home or work location are more difficult to estimate or forecast. In the case of changes in residence or work place, it may be possible to gain some idea of the mag- nitude of such long-term changes through a transportation- sensitive land-use model. However, most land-use models use transportation accessibility measures that are rather aggre- gate and insensitive. It can be expected that the true extent of such locational changes is underestimated by current land- use models, with the possible exception of UrbanSim.14 Induced or suppressed travel is largely missing from the capa- bilities of most current models, although some recent devel- opments have provided some means to estimate these changes. The real problem here is that traditional trip generation mod- els, which are the ones that should provide such estimates, have limited sensitivity to the transportation system and there- fore will not produce a change in the estimated amount of travel. The only source in conventional models for such a change would have to come through feeding new travel times into the land-use models and finding some change in the dis- tribution of land use that resulted in a change in trip genera- tion. Such changes are generally quite small. The method that is used most commonly at this time is to apply an estimated elasticity to the total level of trip making.15 According to Noland and Cowart, elasticities of VMT (a mea- sure of the total level of demand) and travel cost (including time) are of the order of −0.5 to −1.0, while the elasticity of VMT with respect to lane-miles of roadway appear to be in the range of 0.2 to 0.5 in the short term, and 0.7 to 1.0 in the long run. Such elasticities have to be applied outside the conven- tional model system, because there is nowhere in the models for this to operate to increase or decrease total levels of demand. 2.1.3 Transportation Facility and Demand Changes Much of the literature has concentrated on the issue of increased highway capacity (Noland and Cowart,15 Fulton et al.,16 Noland,17 Litman,18 Marshall,19 Chu,20 Stopher13). Adding or removing a lane of a multilane facility, or adding lanes to a two-way, two-lane facility represent major trans- portation changes in the corridor of the affected roadway and can be expected to produce quite substantial demand changes. However, these are by no means the only transportation facility changes of interest. Of considerable importance are the introduction of HOV lanes, changes in traffic signaliza- tion and channelization, and improvements to transit ser- vices, among others. Often, the changes in travel times and FI N A L R EP O RT

costs produced by these types of facility changes are small. The ability of the models to reflect the demand changes, particularly given that the models probably tend to under- estimate the changes resulting from travel time changes, is probably quite small to nonexistent. This situation has been troublesome for those who champion the cause of such changes as HOV lanes, transit improvements, and traffic sig- nal improvements. Although theory is clear that these changes in facilities will create some level of demand change and diversion, the models are not sufficiently sensitive to estimate the magni- tude of the changes. In addition, as is discussed later, the non- transportation changes occurring in the metropolitan areas where these transportation changes are implemented are so large that they may overshadow the transportation changes. The effects of the transportation changes may be noticeable only in the very short term. Most transportation networks are sensitive principally to changes on the order of not less than 1 minute in travel times. A typical signalization improve- ment may result in only a number of seconds of travel time change. Of course, changes will again take place in the lev- els of congestion perceived by the user, but these changes are not taken into account in any models. 2.1.4 THE TROUBLE OF ASSUMING “ALL OTHER THINGS BEING EQUAL” Rarely are all other things equal, and they certainly do not remain so for very long if they are equal at all. The fabric of the country is in continual change. Equilibrium is a useful construct in theory but does not happen for any significant period of time in reality. The earlier statements about diver- sion and induced or suppressed demand are all conditioned on no other change taking place in the system. To see the problems that this condition raises, it is necessary only to consider a few limited examples. Consider the situation in Baton Rouge, Louisiana, between 1997 and 1999. A section of Interstate 10 and Interstate 12 were to be widened, with Interstate 10 being widened from three lanes in each direction to four and Interstate 12 being widened from two lanes to three in each direction from their point of confluence/divergence, and for a distance of a few miles on each side of that point. The project began in mid- 1997 and was completed at the end of 1999, thus taking a period of 2.5 years to complete. One would like to measure the effects of this widening on travel on the Interstates and parallel surface streets. While the project was being com- pleted, two intersections of surface streets in the close vicin- ity of the Interstates were modified to add new turning lanes. During the 2.5-year period, the population of Baton Rouge grew by about 5 percent, with most of the growth concen- trated in areas that are likely to be served at least in part by the widened freeways. Gasoline prices declined to an all-time low price in Baton Rouge (in inflation-adjusted dollars), with 7 the price drop being on the order of 25 percent over the con- struction period. Unemployment declined by about 2 per- centage points over the period. The local economy grew sig- nificantly. Under any definition, this does not describe the conditions implied by all other factors being equal. On the contrary, each of the changes mentioned is likely to affect travel on the widened freeways. Consider now the situation that exists almost 1 year after completion of the project. Gasoline prices have risen by almost 40 percent, population growth has continued at an annual 2 percent, and the economy continues strong with very low unemployment. Surface road projects in the vicin- ity of the widened freeways continue to be undertaken. Again, the situation clearly does not meet the concept of all other factors being equal. Under these conditions, how does one determine the effects of each of these different changes on the levels of traffic on the widened facilities? One could possibly argue for building some type of linear model, but such a model would assume that all of these factors were linearly additive in their effects on the amount of travel on the facilities, an assumption that seems rather improbable. There is little theory and no empir- ical evidence available to explain how all of these factors might combine to change traffic flows, particularly because none of the listed factors except population increase are taken into account in conventional travel demand modeling. Yet, it is clear that each of these factors affects levels of travel in a corridor. The situation described here is by no means unique, but rather probably describes the type of situation that will arise in many highway widening projects. Of course, in some instances, the changes will be in the opposite direction, which will lead to the suspicion, if these changes are not accounted for, that widening roadways does not lead to any significant change in demand levels. Potentially, the situation is worsened further because the changes in population, employment, gasoline prices, the econ- omy, etc., are described in global terms (at least on a metro- politan area level), while the effects to measure are corridor specific. Not all of the changes discussed take place, and cer- tainly those that take place do not do so at the same levels, within what one might define as the influence area of the widened roadways. This brings into view the question of the geographic limits of the factors to be examined. For exam- ple, in the case of the Baton Rouge freeways, does the widen- ing of Interstate 10 affect traffic that is passing through the area, possibly diverting traffic from Interstate 20 (200 miles to the north of Interstate 10)? Because diversion of travel from other routes and destinations is one of the posited effects of the widening, from how far away may those diversions occur? How much diversion may occur from a parallel sur- face street that is itself undergoing reconstruction or widen- ing, but that is experiencing exacerbated delays due to con- struction, after the opening of the widened freeways? How far away must one look at surface streets to capture the majority FIN A L R EPO RT

of the effects and to determine how much VMT is diverted and how much actually increased? These issues are clearly difficult to resolve, but must be resolved to meet the goals of this project and be able to mea- sure the extent of the effects of a variety of factors on con- gestion, VMT, and vehicular emissions. Some attempts have been made to put some sort of an estimate together of some of these effects, and these are reviewed next. 2.1.5 Attempts to Estimate Demand Changes Since the landmark San Francisco Metropolitan Trans- portation Commission (MTC) case in 1989 and the passage of the Clean Air Act Amendments of 1990 and ISTEA in 1991, there has been an increasing level of interest and effort in trying to determine if certain projects, such as road widen- ing, affect demand for travel. Most of the attempts that have been made to date are based on existing data, not on new data collected for the specific purpose of estimating the effect of transportation facility changes on travel demand. Most of the extant and recent studies on the topic have focused on (1) using VMT as a measure of demand and (2) analyzing aggregate data for metropolitan areas across the country. 2.1.5.1 VMT as a Measure of Demand At first glance, VMT appears to be a reasonable measure of demand. It focuses first on vehicles, which are certainly the elements of concern in estimating emission and conges- tion levels. Also, it is a reasonably available and easy-to-use measure, as is noted by Noland and Cowart.15 However, it is in other respects not a particularly good measure. It is not the principal measure of concern in estimating vehicle emissions. Although emission factors are multiplied by VMT, the fac- tors themselves are highly dependent on vehicle speed, which is the major determinant of both emission levels and conges- tion. Slow-moving vehicles and vehicles in stop-and-go con- ditions generate substantially higher emissions per vehicle- mile than do vehicles traveling at cruising speeds of 50 to 70 mph. Also, it is known that travelers in the urban context are largely insensitive to distance. They are, however, highly sensitive to travel time, which is a derivative of speed and VMT. In general, however, people budget time, not distance, in deciding what travel to undertake. Thus, travel time is almost certainly the measure that captures congestion and to which people respond when capacity is increased. There is a further problem in the use of VMT as a measure of demand. In early discussions among researchers, induced travel was defined very specifically to be the new travel that was not undertaken before the improvement. By deciding to focus on VMT as a measure of demand, researchers have, in effect, redefined induced travel to be any increase in travel distance, whether arising from changes in route, shifts in mode, changes in destination, or changes in the number of 8 trips made. At the same time, the revised definition excludes time-of-day shifts. The decision to use VMT as a measure has required the definition of induced travel to change to meet what VMT is capable of measuring. However, if one is to be complete in estimating the air quality effects of trans- portation facility changes, then all changes, including time- of-day shifts, become significant. If one is simply interested in estimating how many new trips take place, then the num- ber of trips—not VMT, travel time, or speed—should be the measure of interest. In summary, VMT confounds diverted and induced demand, but does not completely measure diverted demand. Some elements of route diversion and all aspects of time-of-day diversion are ignored by VMT. It is also possible that some destination diversion might not be captured if the new desti- nations do not add significantly to VMT. In this regard, time- of-day shifts may be more important than some of the other aspects of induced and diverted demand because of the poten- tial impacts on speeds of shifts between peak and off-peak times. VMT is not the major influencing variable on emis- sions, which are more subject to change because of speed than because of VMT. Also, VMT is not what people are intrinsically demanding. They demand travel time (or, per- haps, reductions in travel time). Elasticities of VMT with respect to capacity change are not helpful to modelers. VMT is only an incidental output of the modeling process and does not feature as a significant input or output of any step of the modeling procedures. Elasticities of VMT are not helpful to the modeler in establishing amounts of induced travel. Thus, equating induced travel with VMT is confusing and unhelp- ful. Travel time and speed changes are far more germane to the issue, as is also the number of trips. The largest problem by far in the modeling process is that trip generation is unaf- fected by price of travel in even its most general sense, so that it remains invariant with increases or decreases in transporta- tion capacity. This is the problem that needs to be addressed from the modeling standpoint. Any changes in quantity of trips forecast in trip generation will then be correctly picked up in trip distribution, mode choice, and assignment to reflect diversions among destinations, modes, and routes, and the new speeds of travel can be output readily from the assign- ment process, thereby leading to estimation of emissions consequences. 2.1.5.2 VMT and Travel Time Budgets Zahavi21 was probably one of the first researchers to propose that people have a travel time budget. His work was largely ignored because of problems perceived in his examination of very aggregate data. More recently, his work has begun to be accepted as having some considerable merit as modern activ- ity analysis seems to bear out his contention of travel time bud- gets (Stopher and Metcalf,22 Gordon and Richardson23). In essence, the notion of a travel time budget proposes that out of each 24 hours, people have a limited amount of time that FI N A L R EP O RT

they are willing to spend on travel. In most investigations, this seems to amount to somewhere in the region of 1.25 to 1.5 hours, although there may be some significant varia- tion around these mean figures. The notion of a travel time budget has significant implications for the analysis of the response of demand to transportation facility changes. In the event that a transportation facility were improved so that users experienced a travel time saving, then much of that travel time saving would be used elsewhere in travel. If the user did not use the travel time saving elsewhere, then the user’s total travel time expenditures would decrease. Consider now the tracking of VMT. If VMT were to increase inelasti- cally with capacity increases, there could be net travel time decreases to users because, while users would travel farther, they may take less time overall. If this were to happen, then the travel time expenditure would decrease progressively below the person’s travel time budget, as each new capacity increase occurred. Eventually, this trend would seem to lead to the absurd notion of no travel time being expended. It seems then that the elasticity of VMT with respect to capac- ity increases should be fairly close to unity in order for peo- ple to continue to use their travel time budgets. However, the ability to continue using travel time budgets also somewhat depends on the proportionate travel time changes. It seems, then, that the notion of a travel time budget is fur- ther evidence that VMT is not the correct measure. It also seems that interpretation of elasticities of VMT with respect to capacity may have relatively little meaning. 2.1.6 Empirical Measurement The Travel Model Improvement Program (TMIP) confer- ence on “The Effects of Added Transportation Capacity” in 1991 addressed the issues of empirical measurement in some detail. Much of the discussion in the conference proceed- ings24 on measurement appears to be germane to the present situation and has not been superseded by any significant expe- rience or change in design issues. This conference identified three primary categories of experimental approach: the case study, attitudinal and preferential surveys, and longitudinal panel surveys. It was noted that these options are not mutu- ally exclusive, but may offer opportunities for composite designs and studies that would take advantage of the partic- ular merits of each approach. 2.1.6.1 Case Studies Case studies are seen as perhaps the most obvious method to track the effects of capacity changes on travel demand. The TMIP conference proceedings suggested that the case study is the most useful starting point to consider the issues in an empirical design. The suggestion from the conference proceedings appears to necessitate a series of surveys, with a 9 before survey taking place prior to any construction or other changes in the subject corridor so as to measure the baseline. The conference proceedings recommended a series of after surveys taking place from 1 year following completion of the project to up to 10 years after, to measure the long-term effects. Even 10 years is likely to be insufficient to measure some of the long-term land-use changes that may result. The proceedings also recommended that the surveys include res- idents, employers, and developers, at a minimum. The problems of case studies are numerous, however: • How to sample—identifying both the users and poten- tial users, as well as those whose travel on other facili- ties may be affected by the capacity change in question. Of particular difficulty here is measuring in the before survey those who might decide to travel on the facility after the capacity expansion has taken place. Without measuring these individuals before the capacity change, there is no valid way to determine the induced travel of such people in the after surveys only. • The time frame—there are many other changes that will take place during the more than 10 years that the surveys need to span. These changes will impact the amount of travel. Controlling for these impacts or disentangling their effects poses enormous problems. • Measuring change—there are also problems in measur- ing change because of the baseline of the change of interest. Ideally, the change in travel demand to be mea- sured after the capacity change should be compared with the change in travel demand that would have resulted had the capacity change not taken place. Such a comparison is measurable only if one can find a parallel situation to use as a control. In ensuing discussion at the TMIP conference, several com- ments were made. The first was that there is a tendency to focus on a single case study, where in fact valid results are obtainable only from multiple case studies. Analyzing mul- tiple case studies would require a systematic design for the case studies that could be applied to multiple case studies. Conference members further suggested that researchers should not simply choose five cases at random, but rather should use cluster analysis and the development of topologies of capac- ity addition problems. Other problems noted included the dif- ficulty of identifying parallel routes, conceptualizing the data collection scheme, and maintaining the data over the lifetime of the project. A final problem that was noted is that of antic- ipatory development, or development that is spurred by the expectation of a capacity addition. Anticipatory development may occur many years in advance of the commencement of construction, thus making it even more difficult to define when the before study should be done. An example was quoted of the Bay Area Rapid Transit (BART) system, which had discussions for 20 years before construction began, but dur- ing which time anticipatory developments took place. FIN A L R EPO RT

2.1.6.2 Attitudinal and Preferential Studies Some discussion at the TMIP conference focused on what is now generally described as stated preference (SP) tech- niques. Conference members suggested that preferential studies would be preceded by setting up focus groups to help design the survey. The focus groups should consider the hypothetical situations of both a capacity addition and no change in capacity but worsening congestion. Attitude surveys of developers and employers may also be a necessary component of the SP survey to determine the air quality impacts of highway capacity improvements. The atti- tude surveys would elicit information about stated intentions if capacity were to be added or if congestion were to worsen. The TMIP conference proceedings note several advan- tages and disadvantages to the SP approach. The attitude and stated preference surveys do not need a specific project set- ting in which to be conducted, nor do they necessarily involve complex sampling. Nevertheless, the SP surveys should be undertaken with travelers who are presently experiencing some level of congestion on an identifiable highway so that situations described in hypothetical questions can be realis- tic enough to be meaningful. For example, it would not be useful to ask a person living in Baton Rouge about conges- tion increasing above levels that are only commonly experi- enced in, say, Los Angeles, nor would it be sensible to talk about major capacity additions in a corridor where there is no space for such additions. The conference proceedings also note that the focus groups can help substantially to design the survey, and other questions can be included that will control for externalities that complicate the before-and-after survey. There are also some disadvantages to the SP approach. First, in the absence of revealed preference data, it is difficult to scale the coefficients in a derived model from SP data, so that the actual elasticities with respect to capacity addition are probably not derivable. Second, there are still some sig- nificant issues with respect to the accuracy of behavioral intent measurement. Third, some of the items that may need to be included may not be easily quantified or described for inclusion in the survey. It was observed in the conference proceedings that the long-term reliability of SP results has not been ascertained. This observation is still true today. At the TMIP conference, there was considerable discus- sion of the attitudinal and preference approach, also. First, it was suggested that the changes ensuing from a capacity increase were sufficiently complex that it would be impossi- ble to remove the attitudinal and preference modeling from computer modeling and simulation. It was also pointed out that it is difficult for people to visualize a future condition, especially if this condition will affect their behavior or their attitudes. If the change presented is too simple, it may be mean- ingless, while if too complex, it may not elicit the responses desired. It was suggested that video technology could be used to overcome part of this problem. A second problem that was raised is that people are willing to provide a response as long 10 as it does not cost them anything. It was suggested that cost changes could invalidate results. 2.1.6.3 Longitudinal Panels A longitudinal panel is selected at the beginning of the process and is re-interrogated at intervals after the comple- tion of the project to measure changes in behavior. Survey methodology is clear on the advantages that accrue from pan- els as a means to measure change. However, a panel must be established in one or more control areas to determine how much change will likely take place in the absence of a behavior-changing project. Measuring system performance is also necessary and would also be required of a case study, using cross-sectional samples. This requirement poses some problems because of the lack of good performance measure- ment of the highway system. It was also noted in the TMIP conference that the panel approach applies more to residents than to employers and developers. The idea of longitudinal panels engendered more discus- sion than either case studies or SP studies. The benefits of panels were mentioned a number of times, both in terms of measuring change and also in terms of maintaining updated data generally. It was noted that the make-up of the panel is critical. New households need to be added, and households that move should be retained in the panel population. The ability to get information about households that move to new locations, it was felt, would provide useful insights on the changes caused by the capacity change itself. Keeping track of households, however, also can make panels much more expensive to maintain. The length of time that the panel would need to remain active was also discussed. It was noted that the panel would need to be in place for some years, which also causes condi- tioning effects and fatigue. It was also noted that the type of information obtained from a panel may not be sufficient to provide satisfaction to a legal court. 2.1.6.4 Conclusions From the TMIP conference, no clear method of measuring the effects of added capacity emerged. In fact, a review of the discussion suggests that there remain very substantial prob- lems, not least of which are the need for a control area and the issue of controlling for externalities that change travel demand along with a capacity change. Defining the extent of the study area also poses problems unless one is willing to (1) set up panels or other measurement techniques in various locations where capacity additions are not currently planned and (2) hope to intercept some capacity increases in the future. This approach, however, appears to be much too expensive and is unlikely to be considered feasible. Issues of how long measurements must continue to capture longer-term changes, FI N A L R EP O RT

such as home and job relocation, and new development that may follow are also unanswered. 2.2 RELATIONSHIP BETWEEN TRAFFIC-FLOW IMPROVEMENTS AND EMISSIONS A complex chain of effects connects traffic-flow improve- ments to mobile source emissions (see Figure 1). Traffic-flow improvements, by definition, improve overall vehicle operating speeds and reduce congestion. Reduced congestion means fewer and less extreme vehicle acceleration and deceleration events for the facility. These first-order effects (see Box 1 in Figure 1) usually mean a change in the vehicle emission rates for the facility. Fewer acceleration and deceleration events will result in lower emission rates. Higher speeds may increase or decrease the vehicle emission rates. However, there are second-order effects as well. The higher speeds mean lower travel times. Lower travel times may encourage vehicle drivers to make more trips, make longer trips, and change their mode, route, and time of day for mak- ing their trips. These second-order effects usually occur fairly soon (within a year) of the facility improvement. Traffic-flow improvements for one mode may also adversely affect accessibility and travel times for other modes. For exam- ple, a street widening may improve auto speeds, but will increase pedestrian crossing times. Thus, an improvement may adversely affect one mode at the same time as it bene- fits another. 11 Longer-term, third-order effects take many years to occur (see Box 3 in Figure 1). These effects involve individuals and businesses relocating to take advantage of the better travel times. The second- and third-order effects increase the demand for the facility and reduce its first-order travel time savings and emission reduction benefits. These effects will feed back until a theoretical equilibrium is reached and there is a final estimate of the mobile source emission impacts of the traffic- flow improvement. Conventional analyses of the impacts of traffic-flow improvements usually focus on only the first-order (opera- tional improvement) effects while neglecting the very real second- and third-order effects. Some advanced analyses, using a modeling process called feedback or equilibration, have been able to take into account some but not all of the second-order effects. They take into account the geographic distribution, mode choice, and route choice effects of the improvements, but often fail to take into account increased trip making. A few very advanced analyses have been able to take into account the longer-term, third-order effects, but only at the cost of very large investments in the analysis process, thus discouraging the application of these analyses for all but the most extensive traffic-flow improvement projects. The remainder of this chapter reviews the current state of knowledge regarding the various ways that traffic-flow improvements affect demand and ultimately emissions. The reader will note that there is a great deal of literature on the impacts of traffic-flow improvement on demand, but the vast majority of the literature stops short of evaluating the impacts on emissions. 2.3 EMPIRICAL STUDIES OF THE IMPACT OF HIGHWAY IMPROVEMENTS ON TRAVEL DEMAND Empirical studies of the impact of highway improvements on travel demand have tended to be macroscopic statistical studies of how regional vehicle activity (almost always mea- sured in terms of daily VMT) has correlated with changes in highway capacity (often measured in terms of lane-miles added). Statistical studies of the impact of highway improvements on travel demand are in essence “uncontrolled” experiments. The observer looks at changes in aggregate behavior but is unable to interrogate the users as to why they made their change in behavior. Thus, while the observer can identify the correlation, the observer cannot determine how much of the correlation is due to increased capacity and not due to other changes. When reviewing the results, one must take into account the precise definition of induced demand and the model forms used in each study before extrapolating the results to broad conclusions. Most of the studies have taken a great deal of care to control for extraneous factors and effects, but almost all FIN A L R EPO RT Traffic-Flow Improvements 1. Operational Improvements - Higher Speeds - Fewer Acceleration Events 2. Short-Term Demand Changes - More Trips & Farther Trips - Change Mode, Route, Schedule 3. Long-Term Changes - Relocate Home/Business Lower/Higher Emission Rates More VMT More Trips Mobile Source Emissions Figure 1. Chain of effects tying flow improvements to emissions.

suffer from being limited to a subset of the entire urban area transportation system. Thus, their elasticity results include the broad scale shifting of traffic from the local street system to the state highway system that would be expected to occur when the state highway system is improved. The measured increases are subsystem but not regional changes in VMT. Still, the statistical studies serve a valuable purpose. They point investigators to the conclusion that highway capacity and demand are closely correlated and that demand-modeling practices need to take this correlation into account. 2.3.1 U.S. Conferences and Committees The U.S. Department of Transportation has sponsored several conferences and committee sessions to review the impacts of highway capacity improvements on travel demand. The most recent federal effort was a special session at the 1998 Transportation Research Board (TRB) Annual Meeting sponsored by the Federal Highway Administration (FHWA).25 Four papers on the subject representing different viewpoints were presented at this session. They include a summary paper by Kevin Heanue of FHWA on this topic. All of the papers are syntheses of past work. The session participants identi- fied several high-priority areas for future research on inducted traffic: • The development of simplified procedures to account for induced traffic in benefit-cost analyses of highway improvements; • More basic research on travel behavior oriented toward understanding the role of changes in travel times and costs on the amount of travel by households and businesses; • Retrospective studies, which compare observed volumes in highway corridors with forecasts; and • Before-and-after studies of major improvements in high- way capacity. A special committee was appointed by TRB to study the impacts of highway capacity improvements on air quality and energy consumption in 1995. The committee’s report covers contributions of motor vehicle transportation to air pollution and energy consumption, traffic-flow characteris- tics, travel demand, and land-use and urban form. Some of the key conclusions are that major highway capacity addi- tions vary over time, and the effects of highway capacity additions on emissions highly depend on the state of vehicle design, automotive and motor fuel technology, and emission controls. The report notes that initially, adding highway capac- ity under heavily congested traffic conditions tends to reduce emissions and energy use by smoothing traffic flows, all else being equal. However, the travel time savings from conges- tion relief can stimulate travel demand and, over the long term, set the stage for development and travel growth if other conditions are present. The greatest probability of large devel- opment and travel impacts occurs where major highway 12 capacity additions provide access to developable land in out- lying suburban areas. TRB Special Report 24526 suggests that the range of dis- agreement between highway proponents and opponents on the subject of induced travel has narrowed considerably. The report also notes that there are widely differing elasticities of travel demand with respect to capacity reported in the litera- ture summaries. A review of empirical studies contained in an appendix to the TRB committee report summarized the elasticities of VMT found in the literature for various trans- portation supply measures (see Table 1). The U.S. Department of Transportation27 sponsored a spe- cial conference on the subject in Bethesda, Maryland, in 1991. The topics covered at the conference included effects of added capacity on system performance, travel, and devel- opment; institutional and financial context; environmental effects; forecasting models; and experimental design. Some noteworthy conclusions include the following: • Longitudinal panels can provide information on changes in income, behavior, and facility use that would be valu- able in assessing congestion and capacity impacts. • The effect of added capacity on freight movement needs to be part of the research agenda. • The effects on nonwork travel and off-peak travel need to be considered. • “Backcasting” could provide information on how effective land-use and transportation modeling efforts have been. • It is important to consider redistributive impacts of trans- portation facilities on land development (as opposed to increased total growth). One of the papers at the U.S. Department of Transporta- tion conference notes that the most appropriate way to fore- cast a derived demand is to forecast the demand for the final good or activity at the trip end. Progress28 published by a consortium of pro-environment, anti-road advocacy groups, examines the emerging evi- dence that building roads generates traffic and the corollary that a more balanced set of transportation choices can reduce congestion and improve the local economy. Several short Progress articles, synthesizing the reports of others, provide selected information on effects of increasing or decreasing capacity. The Bureau of Transportation Statistics (BTS) assembled a series of articles on the traffic impacts of the Los Angeles FI N A L R EP O RT Transportation Supply Measure Elasticities Average Highway Speed +0.58 to +1.76 Total lane-miles of highway +0.13 to +0.15 Seat-miles of transit service -0.0098 Source: TRB Special Report 245: Expanding Metropolitan Highways, Transportation Research Board, 1995. TABLE 1 Estimated elasticities of VMT

Northridge earthquake in 1994. This collection of articles pro- vided an example of the effects of catastrophic capacity reduc- tions in the United States. A special issue of the BTS Journal of Transportation Statistics29 includes articles on transit, high- way, goods movement, and transportation-related economic losses due to the temporary closure of the Interstate 10 Santa Monica freeway and other road closures caused by the Janu- ary 1994 Northridge earthquake. The earthquake provided a unique opportunity to examine travel behavior responses in an emergency. An important limitation of this work was that it focused on the short-term responses, but it provided sup- port for the contentions found in other short-term studies. The key conclusions were as follows: • Change in trip scheduling was the largest single impact on travel of the loss of capacity: in the Interstate 5 corri- dor, almost 30 percent of commuters said they left from home earlier or later because of the earthquake. Work schedule changes (such as 4/40 [4 days, 40 hours] and 9/80 [9 days, 80 hours]) were reported by significant numbers of people: 7 to 8 percent depending on the cor- ridor. • Route changes in affected corridors were quite high: 31 percent. • Modal shift effects were more modest, but the most fre- quently found modal shift was from drive alone to car- pool/vanpool; depending on the corridor, between 4 and 6 percent of surveyed commuters indicated this response. It was countered by a shift of some motorists from car- pool to driving alone, perhaps due to disruptions of schedules. • Shifts from drive alone to transit were very small: less than a fraction of a percent in all corridors. 2.3.2 The Standing Advisory Committee on Trunk Road Assessment Study and Related U.K. Research The U.K. Standing Advisory Committee on Trunk Road Assessment (SACTRA) prepared a series of reports about statistical correlation between capacity and demand. They were prepared by an independent advisory committee to the U.K. Department of Transport (DoT). The purpose of the initial report30 was to inform the DoT about evidence of “the circumstances, nature and magnitude of traffic redis- tribution, mode choice and generation [resulting from new road schemes], especially on inter-urban roads and trunk roads close to conurbations; and to recommend whether and how the Department’s methods should be amended, and what if any research or studies should be undertaken.” Trunk roads in Britain are intended to serve long-distance through travel, but may include two-lane roads as well as freeways (i.e., motorways). The principal conclusions of the SACTRA report were the following: 13 • Induced traffic can and does occur, probably quite exten- sively, though its size and significance is likely to vary widely in different circumstances. • The economic value of a scheme (i.e., plan) can be over- estimated by the omission of even a small amount of induced traffic. • Induced traffic is of greatest importance when the net- work is operating close to capacity (or will in the future), where traveler response to changes in travel times or costs is high, and where an improvement causes large changes in travel costs. The SACTRA report makes a number of recommenda- tions, including the use of variable-demand methods (rather than fixed trip tables), improved monitoring, and project appraisal that includes induced traffic in environmental and economic analyses. The SACTRA report includes case stud- ies (of both traffic volume and land-use changes) based on the openings of new highway projects in Britain. The SACTRA report, done by highly credible consultants and academics, has been widely cited by the environmental community as supporting the notion that induced demand is significant, but is being ignored by highway advocates. How- ever, critics have questioned the applicability of the report to the United States. Britain, despite being an industrialized country, has a relatively poor system of high-performance roads. Other differences with the United States include a greater degree of traffic congestion in urban areas, an exten- sive urban and intercity railway network, a less developed airway network, and a high density of development (i.e., there is little open land legally or physically suitable for devel- opment). The critics question if these factors—reminiscent of patterns in the United States at the dawn of the freeway- building era—are really applicable to the kind of marginal changes in highway network improvements being proposed in most U.S. cities today. National spending by Britain on major roads is relatively low by U.S. standards: about $55 per year per capita; FHWA’s current budget is approximately three times this, not to mention considerable spending by state governments. Major new freeway investments, such as the M25 motorway circling London, have made dramatic improvements in highway accessibility in the affected corri- dor. Projects of this magnitude are not likely to occur in many U.S. cities. The DoT staff prepared its own brief (21-page) response to the SACTRA report.31 The response states the following: • Much of the added traffic growth that SACTRA has attributed to induced travel is more properly attributed to increased economic growth rates (particularly in the 1980s). • The negative effect on road benefits is smaller than claimed by SACTRA. Further, SACTRA did not con- sider some of the benefits that occur to nonproject motorists, in terms of redistribution and retiming. FIN A L R EPO RT

• DoT will develop methods to allow, where necessary, variable-trip matrix analysis to be carried out as a mat- ter of course. Coombe and his co-authors32 looked at the opposite effect generated by capacity reductions. Their article looks at empirical evidence of trip reductions due to reductions in highway capacity. The study included the theoretical frame- work for the analysis of travel behavior in the face of reduced road space and practical ways of estimating the traffic impact of selective withdrawal of highway capacity. Evidence from over 100 locations was collected, with more than 60 provid- ing primary case study material. Available evidence showed a wide range of results. Coombe et al. note some general caveats and problems of interpretation. They also conclude that reductions in traffic only occur given certain network conditions, that behavioral responses partly depend on nat- ural variability in behavior, and that behavioral changes vary over time. The authors make points about considerations and model practices that should yield the most accurate results, although the authors do not provide examples of applications to real data. Where generalized cost changes as a result of a capacity reduction are significant, the authors argue for an “elastic” assignment that allows for changes in the genera- tion rates and distribution of traffic. Goodwin33 and his co-authors also looked at capacity decreases. They prepared a paper that examines the effect on traffic flows of 100 cases of capacity reductions in Britain that were caused by re-allocation of existing capacity to buses or pedestrians, maintenance, or natural disasters. The cases fell into three broad groups: cases where, on closer examina- tion, there was no actual capacity decrease; cases where there was a real capacity reduction on the treated route but there was spare capacity on alternate routes; and cases where the capacity reduction actually happened and there was no spare capacity on alternate routes. The latter cases were the only ones where traffic was found to decrease on the routes being studied. 2.3.3 Research Authored and Co-Authored by Noland Noland, with the use of his carefully crafted lagged effect statistical models, has contributed greatly to refining the def- inition and measurement of induced demand. The following studies of his have meticulously separated demand induce- ment from other effects in the available data. Noland and Lem34 reviewed the literature, theory, and definition of induced travel and attempted to clarify much of the confusion in the literature by defining induced travel to be an increase in VMT for the entire region that is “attribut- able to any transportation infrastructure project that increases capacity.” Changes in number of trips are excluded from the definition of induced demand. Noland and Lem cited various VMT elasticities with respect to travel time decreases and 14 with respect to lane-mile increases from several studies in the United States and Great Britain. Short-run elasticities range from 0.3 to 0.6 for VMT with respect to lane-miles, while long-run elasticities range from 0.7 to 1.00 for VMT with respect to lane-miles. The authors concluded that the theory of induced travel—namely, that increased capacity con- tributes to substantial increases in demand—is confirmed by the various studies and suggest that much of the benefits of highway projects may come from redirection of urban growth rather than from congestion reduction. Noland and Cowart35 fitted various cross-sectional time- series models to 14 years of daily VMT and lane-mile data for arterials and freeways located in 70 metropolitan areas in the United States. Fixed effects were included across urbanized areas and across time. The authors also applied a “two-stage, least-squares” approach to address the issue of causality bias (do planners build capacity in response to demand increases, or do increases in capacity cause increases in demand?). The models that Noland and Cowart constructed employed variables to isolate the effects of population growth, income growth, fuel costs, and population density changes from the effects of capacity increases. The authors found short-run capacity elasticities of around 0.7 for demand. However, the elasticities varied from 0.3 to 0.7 depending upon the vari- ables included in each model. Two-stage least squares gen- erally improved the fit of the models to the data, but also yielded elasticities that ranged from 0.3 to 0.8. The authors went on to identify the proportion of metropolitan VMT increases that could be attributed to capacity increases and found that the proportions varied widely by region, ranging from a low of 7 percent to a high of 34 percent. Noland and Cowart initially attempted to extend their analysis to the entire road system within each metropolitan area but found conflicts between sources and gaps in avail- able data. Thereafter, they focused on arterials and freeways where the data were more reliable. Unfortunately, this approach means that an unknown portion of the VMT increases that the authors detected may simply be traffic shifted between the local roads and the arterial or freeway system in each region. Thus, the actual metropolitan area VMT increase is probably lower than what the authors measured. The authors noted that the arterial and freeway systems included in their study accounted for an average of 64 percent of the daily VMT and 28 percent of the lane-miles in each metropolitan area. Noland36 estimated the correlation between statewide VMT and lane-miles. He computed elasticities based on several econometric specifications. Lane-miles were found to gener- ally have a statistically significant relationship with VMT. He found elasticities of between 0.3 and 0.6 in the short run and between 0.7 and 1.0 in the long run. Elasticities are larger for models with more specific road types. This is one of the few studies that claims (from empirical data) to find unit elas- ticities. The author believes that about 25 percent of VMT growth can be attributed to lane-mile additions, assuming historical rates of growth in road capacity. The principal short- FI N A L R EP O RT

coming of the paper (as far as drawing conclusions useful to urban areas) is that it is highly aggregative. Statewide data are used, so California and Rhode Island are considered comparable units of analysis. 2.3.4 Research by Others Several researchers have also analyzed the available field data in the United States. Some have even used the same data as Noland and have come to moderately different conclu- sions on the magnitude of the effect. Marshall37 used the same 70 metropolitan area Texas Transportation Institute (TTI) data set as Noland and Cowart to evaluate induced demand. However, Marshall focused on only a single year’s data (1996) and performed a cross- sectional analysis of it for daily VMT per capita and lane- mile per capita changes solely on freeways and arterials. The only non–lane-mile factor included in Marshall’s model was area encompassed by each metropolitan area. Marshall’s elasticities of 0.8 to 0.9 without long-term effects are high compared with other research results. Fulton et al.38 evaluated induced demand at the county level for the states of Maryland, Virginia, North Carolina, and Washington, D.C., using Highway Performance Moni- toring System (HPMS) data that dated back to 1969. The lane-mile and daily VMT data were limited to the state- maintained highways in each county. The county data for Virginia excluded any data for incorporated cities within each county for that state. Induced demand was defined in this study as an increase in each county’s daily VMT on the state- maintained highways that was caused by an increase in the lane-miles of state-maintained highways within that county. Fulton et al. fitted a series of models to the data. A fixed- effects model was found to have demand elasticities of between 0.3 and 0.6 for each county in each state. A “first differences model” was found to have a slightly wider range of demand elasticities (between 0.15 and 0.61). A distributed lag model was found to result in short-run elasticities of between 0.1 and 0.4 and long-run elasticities of 0.5 to 0.8. Fulton et al. found that population growth had an equal or greater effect on VMT than capacity had (because Fulton et al. were predicting total VMT, not VMT per capita, like Noland). Income had a comparatively minor effect on VMT. Fuel cost effects were not directly evaluated. A Granger test of precedence found that lane-mile growth precedes VMT growth (thus indirectly addressing the ques- tion of causality). A follow-up test of the hypothesis that con- gested areas are more sensitive to capacity increases than uncongested areas are yielded inconclusive results. Note that because Fulton et al. limited themselves to same- county impacts of lane-mile increases, the mitigating effects of traffic decreases from other counties are missed. The result is the potential for an unknown amount of overestimation of the elasticities. 15 Chu39 performed a cross-sectional study of 391 urbanized areas in the United States. He fitted a static equilibrium demand model to data obtained from the HPMS and the FHWA’s 1997 Highway Statistics Report. He measured demand inducement in terms of the change in traffic density (daily VMT per lane-mile) on all nonlocal roads in each region. Local roads were excluded because of data reliability problems. He tested several models for each facility type and concluded that congestion/capacity elasticities ranges from 0.03 to 0.37, depending on the facility type and the amount of lane-miles of each facility type already present in the urbanized area. The absolute value of the elasticities for free- ways and minor arterials increased significantly with increas- ing lane-miles already in place. Chu studied a more comprehensive set of facility types (only local roads were excluded) than Noland and Fulton did and obtained similar or lower cross-sectional elastici- ties than Noland or Fulton did. Chu, however, was unable to estimate long-term effects with his cross-sectional approach. Hansen40 prepared an article that summarizes previous work in the field. He noted that, “conventional wisdom aside, we simply don’t know whether new highway capacity affects travel behavior and, hence, traffic volumes.” He provided a lucid review of problems in measuring this effect, espe- cially the direction of causality (“do roads generate traffic or does traffic generate roads?”). He noted that although cross-sectional estimates of VMT with respect to highway capacity have been in the 0.13–0.7 range, the studies yield- ing estimates in the lower end of this range have controlled for more variables. In order to avoid problems associated with causality (i.e., simultaneity), Hansen used a distributed lag model to esti- mate VMT growth at the county and metropolitan levels in 30 California urban areas. He concluded that a 1-percent increase in lane-miles soon induces an immediate 0.2-percent increase in traffic, building to a 0.6-percent increase within 2 years after the lane-miles are added. At the metro level, he noted that the elasticity could be 0.9 percent after as little as 4 years. The major limitations of this work are that they apply to VMT on state highways only and do not separate diverted traffic (i.e., traffic from route shifting). Hansen’s study also makes the assumption of constant elasticity, which seldom holds for most other economic goods except in the case of small changes in the explanatory variable. Distributed lag models have been criticized by Brian Field in his book, Forecasting Techniques for Urban and Regional Planning (ULC Press, 1992): The level of complexity of these [distributed lag] models masks an underlying theoretical inadequacy. The causal struc- ture is poor and the independent variables [VMT] may conceal numerous specific causal factors linked in different ways to the dependent variable [lane-miles of capacity]. It also seems unlikely that the parameters will remain constant—as required FIN A L R EPO RT

for forecasting. There are also problems which arise from the statistical requirements for using the technique—the inde- pendent variables must be normally distributed and indepen- dent (p. 35). Brodahl41 looked at the effect of a major new freeway opening in the Los Angeles area. His study reports on traffic volume and travel time data collected on the Interstate 105 Glenn Anderson Freeway (formerly known as the Century Freeway) in Los Angeles. The freeway was a major new facility opened in October 1993. Traffic volume data are pre- sented not only for the new freeway, but also for parallel and feeding major surface streets. The largest impact noted by Brodahl was a reduction in traffic volumes on the Route 91 freeway, located about 4 miles south and parallel to Interstate 105. Parallel surface streets generally showed decreases in volumes, with the largest decreases occurring nearest the new freeway; the numbers vary widely from one location to another. Many of the cross (perpendicular) streets showed either large increases or sub- stantial decreases in traffic volumes. Streets having freeway access (i.e., interchanges) showed the largest increase. A con- siderable amount of backup statistical information was pre- sented with this report. Downs42 provides an early insight into the subject. His paper is noteworthy primarily for its early date of publication (1962), just as Interstate/high-performance highway construc- tion was getting into full swing. Downs posits that “on urban commuter expressways, peak-hour traffic congestion rises to meet maximum capacity.” Because of its early publication date, the paper does not distinguish between such important concepts as generated versus diverted traffic. Although the arguments put forward in favor of the hypothesis are persua- sive and compelling, they are based primarily on what hap- pened with the opening of early expressways and freeways and two hypothetical case studies. Also, the work was com- pleted before the planned high-performance highway net- work was completed (and in fact, the planned network has not been completed in most U.S. urban areas today). 2.4 BEHAVIORAL STUDIES OF THE IMPACT OF TRAVEL TIME ON TRAVEL DEMAND Behavioral studies look at travel behavior at the disaggre- gate household or individual traveler level in order to identify the response of the individual traveler to capacity increases or differences in accessibility. Many of these studies relate to the development of a new class of travel demand models called “activity-based models.” Activity-based models seek to predict travel behavior as a derived demand from the scheduling of daily activities both within and without the home. These studies strive for a better understanding of how people use their time and how they trade off time spent on various activities each day or week. 16 2.4.1 Study of Travel Time Elasticity Barr43 performed a cross-sectional study of 27,000 house- holds surveyed in the 1995 National Personal Transportation Survey. He defined induced demand as a change in annual household VMT due to a change in the trip travel time. He found that the elasticity of VMT with respect to trip time ranged from 0.3 to 0.5. The advantage of his approach is that it includes all VMT generated by the household, regardless of the facility type, so all substitution effects are accounted for. His approach also accounts directly for the expected effect of highway capacity on demand, through reductions in travel time. Capacity changes that do not affect travel time will not affect demand as well. His results, however, cannot be directly compared with other lane-mile–based elasticities, because Barr’s elasticities are for travel time, not lane-miles. Also, Barr was unable to determine long-term effects because of his use of cross-sectional rather than longitudinal data. 2.4.2 Econometric Model of Travel Behavior Kockelman44 fitted a system of utility theory consistent demand equations to the 1990 San Francisco Bay Area Travel Survey of 10,000 households. She tested two model specifica- tions and found that the elasticity of discretionary travel time was generally less than 1 (ranging from 0.4 to 1.0, depending on the model specification and trip length). She also used the models to test various hypotheses on the sensitivity of travel demand to travel time and cost: • Hypothesis 1: Total time spent traveling by a house- hold is independent of trip time. This hypothesis was rejected based on the calibrated model results. The result- ing effects of trip time changes were mixed, though. Kockelman noted that total time spent traveling increased as trip time to closer activities increased, indicating the inelastic nature of demand for these activities. However, as trip times to distant locations increased, total time spent traveling by households decreased as closer activ- ities were substituted for more distant activities. • Hypothesis 2: The total number of household trips is independent of trip time. This hypothesis was also rejected based on the calibrated model results. How- ever, Kockelman found that the relative magnitude of the change in trips with respect to trip time changes was comparatively negligible. Higher trip speeds seemed to imply longer-distance trips more so than more trips, although both effects were observed. 2.4.3 Studies of the Use of Time and Travel Time Savings Use of time studies attempt to get at the heart of the expected relationship between traffic-flow improvements and travel demand. Traffic-flow improvements affect demand FI N A L R EP O RT

by changing travel times. If researchers can understand bet- ter how people respond to travel time changes, researchers can better understand how these travel time changes affect travel demand. Robinson and Godbey45 produced landmark research on how people in the United States have changed their use of time over the years. Their study shatters many preconceived notions on the use of time in the United States. They observe that, contrary to popular wisdom, Americans have gained 1 hour of free time each day of the week since 1965, while at the same time feeling more harried for time than ever before. Robert Putman, in his forward to the book, observes that the extra free time has come in small packets and has therefore been spent on television (where small packets of time can be most easily spent) rather than on more satisfying leisure activities. Robinson and Godfrey found that the amount of time each week that women spend traveling has increased by almost 2 hours per week while it has held almost constant over 30 years for men. They attribute this to a large increase in the participation of women in the labor force between 1965 and 1995. Robinson and Godfrey’s research found that people do not automatically invest increased free time (whether it comes from labor-saving devices, shorter work hours, or better roads) into more travel to more out-of-home activities. The vast majority of extra free time comes in packets too small to be used in new activities, and it is therefore used to extend exist- ing activities in the home or out of the home. Dowling and Colman46 conducted an SP survey of 676 adults in California to identify how travelers would respond to various increases or decreases in trip times. Participants 17 were questioned in detail on their prior day’s activities. Each prior day’s trip was then reviewed with the respondent and they were asked what they would have done differently if the trip time had been increased or decreased by a randomly selected amount of time (the maximum amount of change was capped at 50 percent of the total trip time to preserve realism). Dowling and Colman found that the traveler’s willingness to change travel behavior varied according to the amount of trip time savings or increase offered (see Figure 2). With travel time changes of plus or minus 5 minutes, more than 90 percent of the respondents indicated they would make no change at all in their previous day’s trip. The predominant response from participants was that they would change their trip start time to compensate for any changes (increase or decrease) in trip travel time. The percentage of respondents indicating they might make an extra stop did not rise to 5 per- cent until the time savings approached 20 minutes. The large number of “other” responses shown in the chart for trip time increases was respondents who indicated they would try to find some way to avoid the trip time increase. The results of Dowling and Colman are interesting in that most respondents did not consider the travel time savings that they were offered to be significant enough to warrant changes in their previous day’s travel patterns. They might have responded differently to larger time savings, but the experi- mental design intentionally limited the amount of time sav- ings offered to participants to an amount that bore some real- istic relationship to their current trip times. This restriction in the design of the survey was required because realism is the key to a reliable SP survey. SP sur- veys allow the experimenter to structure the experiment very FIN A L R EPO RT 0% 5% 10% 15% 20% 25% 30% 35% 40% 45% -20 -15 -10 -5 0 5 10 15 20 Minutes Difference in Trip Time Pe rc en t R es po nd en ts C ha ng in g Tr av el B eh av io r Other Make Extra Stop Change Start Time Figure 2. Traveler response to trip time changes.

precisely and thus better understand the decision-making process (and therefore hopefully the behavior) of travelers. However, SP surveys may have problems. As aptly put by Ben-Akiva and Lerman,47 “People often do not actually do what they say they would do under hypothetical circum- stances. These biases can be reduced through a careful exper- imental design that maximizes the realism of the questions.” Ortuzar and Willumsen48 grant the many experimental advan- tages of SP data when the experiment is carefully designed. However, they recommend that revealed preference data be used in combination with SP data (rather than relying exclu- sively on SP data) for model development. Fujiwara49 identi- fies bias problems occurring with longitudinal panel SP sur- veys and recommends means for dealing with them. A San Jose State University study50 analyzed household survey data collected by Caltrans in 1991 as part of the Cali- fornia Statewide Travel Survey. Only urban households, encompassing about 6,200 dwelling units, and 64,000 person trips, were used for analysis. After geocoding household loca- tions, the authors tested the hypothesis that freeway accessi- bility, as measured by the proxy of the number of freeway interchanges within a 3-mile radius, would be positively cor- related with the number of private vehicle trips made by a household after controlling for demographic variables. In other words, all other things being equal, a household living closer to a freeway ramp would be more likely to make pri- vate vehicle trips as a result of the increased mobility afforded by close access to a freeway (or freeways). The San Jose researchers concluded that freeway access, at least as mea- sured by the proxy variable of interchanges within 3 miles, does not play a very strong role in determining how many vehicle trips the household makes. Households in low-density areas (under 2,500 persons per square mile) with no freeway access within 3 miles averaged nearly the same private vehi- cle trip generation rate as those with freeway access. Because freeway spacing is often closer in high-density areas, the researchers note a problem with the possible multicolinear- ity of survey data. 2.5 STUDIES OF THE URBAN FORM IMPACTS OF TRANSPORTATION IMPROVEMENTS There is a great deal of literature in the field of economics on how transportation infrastructure in theory affects urban development. Although it is recognized that the spatial loca- tion of transport facilities is a significant predictor of devel- opment patterns in a growing area, the effects of highway capacity improvements within an already developed urban area are less clear. This review focuses on three studies that attempted to measure actual impacts on urban form of trans- portation improvements in a built metropolitan area. Pernot’s51 retrospective study of urban growth attributed urban sprawl primarily to increased financial well-being, which causes people to buy more land. Highways had a lesser effect. In the case studies of the Interstate 294 and Interstate 18 88 freeways in the Chicago area, Pernot found that major pop- ulation gains occurred 10 years before the asphalt was poured. An American Association of Bay Area Governments (ABAG) study52 of the sensitivity of land development pat- terns to differing highway investment strategies in the San Francisco Bay Area found mixed results. ABAG undertook an analysis of the land-use implications of the 1989 Trans- portation Improvement Program (TIP) in the Bay Area using the Projective Optimization Land Use Information System (POLIS) model. Two tests were performed: The first test used existing land-use policies and ABAG’s Projections ‘90 land uses as the basis for the analysis of the travel time net- work on growth distribution. The second test looked at the impacts of the transportation network scenarios on growth distribution, relaxing the constraint of local development policies. Various corridors where major transportation improve- ments were planned in the TIP were analyzed under “build” and “no build” conditions. The model results suggest that the effects of capacity increases may be highly location specific. In most cases, the relative magnitude of shifts was not great, but in certain less developed areas (e.g., Half Moon Bay), the unconstrained land-use test and highway improvements led to substantial growth. However, in Marin County, highway improvements tended to keep growth from spreading to the less congested northern part of the county (Novato). A simi- lar conclusion was reached in Sonoma County in the North Bay: “. . . the existing transportation network—independent of the build–no build scenarios—will probably facilitate already existing development pressure to further decentral- ize jobs into the northern [less developed] portion of the [101] corridor.” The central Interstate 80 corridor (Vallejo- Vacaville-Fairfield) was the area found to be most sensitive to highway improvements. In the build scenario, this area would have 50,000 jobs in the year 2010, while under the no- build it would have 46,000 jobs. Still, this difference is only about 10 percent. The general conclusion of the ABAG study is that high- way improvements in the core area have little effect on devel- opment in the core area, but highway improvements on the fringes attract more development to the fringes. An impor- tant qualification to this study was the inability of the POLIS model at that time to distribute land development between counties in the nine-county Bay Area (this distribution had to be done manually). POLIS only distributes development to smaller areas within a county based on county control totals. TCRP Report 16: Transit and Urban Form53 observes that there is “little evidence of land-use impacts from the construction and operation of busways in California, Wash- ington State, and Washington D.C.”; however, “busways that provide service comparable to rail systems can influence urban form.” This report also notes, “In most urban areas where transit operates, its comparative advantage in the reduc- tion of individual trip times is only felt on selected trips. Thus models that forecast land use on the basis of travel imped- FI N A L R EP O RT

ance are not sufficiently sensitive to transit’s contribution to increasing accessibility, especially in downtown areas.” The TCRP report identified the following four mecha- nisms by which rail transit influences urban form: • Rail transit influences the value of adjacent land and its improvements. • Rail transit influences the intensity of development (especially for nonresidential development). • Rail transit influences urban structure (i.e., urban versus suburban development). • Rail transit influences the timing of development. The TCRP report cites several case studies for each mech- anism (Philadelphia, San Francisco, Washington D.C., Atlanta, San Diego, Boston, Los Angeles, Miami, Sacramento, and San Jose) as well as a few cases where urban form impacts were not observed (these cases were usually slower, lower- capacity rail systems). 2.6 EXAMPLES OF THE IMPACTS OF TRAFFIC- FLOW IMPROVEMENTS ON EMISSIONS Prior sections have addressed pieces of the entire chain of events connecting traffic-flow improvements to emissions. This section attempts to look at the entire picture and define better the size of the target for the current research. This sec- tion borrows results from some selected conventional and advanced studies to illustrate the magnitude of the emission impacts that might be expected from traffic-flow improve- ments and shows the required degree of precision for the cur- rent research effort. 2.6.1 A Typical Conventional Analysis: The San Francisco Metropolitan Transportation Commission’s Long-Range Transportation Plan MPOs routinely apply conventional travel demand models to estimate the impacts of their long-range transportation plans on mobile source emissions. In fact, it is so routine that few examples make it into the research literature. The fol- 19 lowing example exemplifies the hundreds of analyses rou- tinely conducted each year in the United States. The San Francisco Metropolitan Transportation Commis- sion (MTC) evaluated the impacts of $17 billion of trans- portation infrastructure improvements contained in its 20- year updated 1998 regional transportation plan (RTP).54 The improvements would add 706 lane-miles of capacity to the regional highway network (an increase of 4 percent) and would increase transit system capacity in the region by 346,028 peak-period transit seat-miles per hour (an increase of 11 percent). MTC applied a relatively advanced travel demand model (see Chapter 3 for a description) to the analysis of the impacts. However, the trip generation component of the model is not directly sensitive to travel costs, and the emission factors (EMFAC7g) were applied to average speeds, not mode of vehicle operation. The analysis concluded that the RTP would cause very minor reductions in VMT in the region and sig- nificantly greater reductions in emissions (see Table 2). The increase in highway capacity was apparently compen- sated by the transit capacity improvements, thus resulting in a net reduction in vehicle activity. This result shows how even large-scale highway and transit improvement plans for a region often represent a very small change for the region as a whole (considering the magnitude of the transportation system already in place) and (at least by conventional forecasting techniques) result in miniscule changes in vehicle activity. Interestingly, though, the miniscule vehicle activity changes were magnified into larger-scale mobile source emission reductions, which suggests the need for extreme precision in predicting vehicle activity changes. 2.6.2 NCHRP 8-33 Impacts of Transportation Control Measures on Emissions Recent research for NCHRP Project 8-33 provides an illustration of the impacts of transportation system improve- ments on mobile source emissions that can be expected when a more advanced analytical tool is applied to the analysis. The NCHRP 8-33 investigators55 developed an advanced tour- based modeling approach for the Portland urbanized area (described in a later chapter) and used this model to test the effectiveness of various transportation control measures FIN A L R EPO RT Performance Measure Impact % Change Number of Vehicles in Use No Change 0% Daily VMT - 154,000 per day < 1/10 of 1% Engine Starts - 4,000 per day ~1/100 of 1% Daily Vehicle Trips - 3,000 trips ~ 2/100 of 1% Reactive Organic Gasses - 3.1 tons/day 8% reduction Carbon Monoxide -76.2 tons/day 11% reduction Nitrous Oxides -8.5 tons/day 6% reduction PM 10 No change 0% reduction Impacts of 4-percent increase in highway lane-miles and 11-percent increase in peak-period transit capacity. TABLE 2 Impacts of MTC long-range transportation plan on emissions

(TCMs) for reducing regional VMT. The impacts of TCMs on emissions are of interest because TCMs are almost the inverse of traffic-flow improvements (in that they usually either make highway travel more difficult, or they make alternative non- vehicle options more attractive) and thus give us an idea of how emissions respond to traffic-flow changes in general. The TCM policies evaluated by NCHRP Project 8-33 using its advanced tour-based model include the following: • Pricing auto travel: Double the long-term parking cost in downtown area and have an SOV toll of one dollar for all peak-period travel within the metropolitan area (excludes urban growth area). • Telecommuting incentives: Double the percentage of workers working at home. • Transit improvements: Within the metropolitan area (excludes growth areas), cut transit fare in half for all times of day and double bus service frequency at all times of day. Table 3 shows the forecasted impacts of the TCMs on emissions. The combined impact of all of these TCM poli- cies was to reduce travel and mobile emissions by less than 5 percent, as shown in the table. The NCHRP Project 8-33 investigators also looked at TCM policies that influence residential location. They used the results of Portland’s household SP survey to predict loca- tional responses to policies (improved shopping opportuni- ties, transit service, better safety, better schools in the city center, etc.). Based on the SP survey, the researchers pre- dicted that the TCM policies would increase the forecasted number of households locating in the urban center by 16 per- cent. Four percent of the households in the region would move from the outer suburbs to within the urban growth area. The effect of these relocations of households was found to be a 2-percent increase in auto trips and a 1-percent increase in daily VMT. The effect of moving suburban families to down- town was to increase the number of trips generated per down- town household. 2.6.3 The Sacramento Area Council of Governments Integrated Land-Use Transportation Model Study Rodier et al.56 demonstrate the effect of using an integrated land-use and transportation model to predict the emission 20 impacts of various transportation improvement scenarios in the Sacramento metropolitan area. The Marcial Echenique Plan (MEPLAN), an integrated land-use and travel forecast- ing model, was used to predict the travel behavior and emis- sion impacts of four transportation improvement scenarios over a 15-year time frame. These results were then compared with the results obtained when using an advanced conven- tional travel forecast model, the Sacramento Metropolitan Travel Demand Model (SACMET), without a land-use mod- eling component. Table 4 shows the results for the HOV sce- nario (i.e., expansion of the existing 26 lane-miles of HOV lanes to 179 lane-miles). Forecasted land-use changes caused by the increased mobility provided by the expanded HOV system resulted in a significant change in daily VMT, how- ever; the decrease in the number of trips resulted in no dif- ference between the models in predicted increases of total organic gasses and CO and modest reductions in the predicted increases of NOX and PM. Putman et al.57 describe a parallel study of integrating a land-use model with the Sacramento SACMET model but do not report numerical results. A series of maps are presented illustrating the land-use impacts (over a 30-year period) of the transportation improvement measures studied in the report. According to the authors, “The results of these analyses show small but significant differences in the outcomes of the sev- eral scenarios examined.” 2.6.4 Implications for Current Research Effort None of the above studies take into account all of the known short-term and long-term effects of traffic-flow improvements on demand and therefore emissions, but they illustrate a key point for the current research. The impacts of typical traffic-flow improvement projects on travel activity are miniscule (less than 10 percent when compared with regionwide travel activity), but these miniscule impacts are magnified several-fold when translated into mobile source emissions. Changes in peak-period activity have significant effects on mobile source emissions. The challenge for the cur- rent research effort will be to predict vehicle activity changes resulting from minor changes to a built network accurately enough to reliably predict their effect on emissions. FI N A L R EP O RT Type of Impact Forecasted Impact Impact on Daily Trips - 1 % Impact on Daily VMT - 2 % Impact on AM Peak VMT - 4 % Impact on AM Peak VOC - 3 % Impact on AM Peak NOX - 3 % Impact on AM Peak CO - 3 % Type of Change SACMET (no land effects) MEPLAN Daily Trips +0.2 -0.6% Daily VMT +1.9% +6.3% Mean Trip Speed +2.0% +2.1% Total Organic Gasses +1.3% +1.3% CO +1.6% +1.6% NOX +3.0% +2.3% PM +1.3% +0.3% TABLE 3 NCHRP Project 8-33 forecasted impacts of TCMs on emissions TABLE 4 Projected changes in emissions over baseline for HOV system expansion

2.7 CONCLUSION This chapter reviewed some of the attempts that have been made to measure induced travel and ideas for measuring induced travel in the future. First, induced and diverted traf- fic occur as a result of transportation system facility changes. There appears to be no dispute in the profession at this point on this issue; rather, it seems to be widely accepted that such changes occur and need to be estimated. Second, there have been a number of recent attempts not only to establish that induced travel takes place as a result of capacity changes, but also to estimate the elasticity of this demand. However, these attempts have largely concentrated on using VMT as a mea- sure of the induced travel demand and have estimated the elasticities of VMT with respect to capacity changes or changes in lane-miles. The notion of using VMT as a measure of induced demand has been called into question. First, VMT confounds ele- ments of induced and diverted traffic without completely measuring the latter. Second, VMT is not the most important component of demand that impacts emissions. (This compo- nent is vehicle speed.) In addition, it is argued that people do not demand VMT, and so measuring an elasticity of VMT with respect to capacity changes is a rather barren concept. Rather, it is suggested that people have a travel time budget. If a capacity increase is implemented that increases speeds on a facility, then the amount of a person’s travel time bud- get that must be consumed in existing travel that used that facility or that can be diverted to that facility is reduced. This reduction leaves spare travel time within the individual’s budget that can be used for additional travel (i.e., more trips), changing an existing destination to a further location, chang- ing time of day of travel, changing mode of travel, or chang- ing the route of some travel. In the longer run, a reduction in the amount of travel time required from a person’s budget may lead to either a change in residence or a change in job location to take advantage of the lower expenditures required and to live or work in an area that is considered more desirable. The emissions implications are greatly varied. If people divert from a congested route to a less congested route, the increase in speed will most likely lead to a reduction in emis- sions. Similarly, some people may divert from a destination that requires driving on congested roadways to a further desti- nation that requires less congested driving, although a longer distance may be traveled. If the speed is sufficiently increased, the longer distance may be offset in its impacts on emissions by the improved speeds. Diversion from transit modes, or nonmotorized modes, to automobiles will necessarily result in increased emissions. Changes in time of day of travel may add to emissions, decrease emissions, or result in no change. Added trips will almost certainly result in increased emis- sions, no matter when or where these trips take place. This chapter contains an extensive review of statistical studies of highway capacity increases or decreases. These statistical studies suffer from the common weakness of all 21 uncontrolled experiments. Correlations are found, but it is difficult to go beyond the statistical conclusion of correlation to a causal mechanism. It is impossible to isolate the specific effects of the traffic-flow improvement from the effects of other changes in the environment. However, the effects all illustrate a point. Traffic-flow improvements impact travel demand, and the impacts are on the order of a 10-percent increase in daily VMT for every 10-percent increase in lane- miles of new capacity. Table 5 summarizes the conclusions of the most recent research of this type. The statistical studies suggest that every 10-percent increase in capacity is absorbed by a 10-percent increase in demand. However, this trend applies to only the higher-speed subsys- tem of the entire regional transportation network. The behavioral studies indicate that the vast majority of time savings that Americans have received over the last 30 years (whether from new freeways or from labor-saving devices) has gone to nontravel activities. Nevertheless, while these behavioral studies may contradict the magnitude of impacts suggested by the statistical studies, they confirm the basic conclusions of the statistical studies that travel time savings (and therefore traffic-flow improvements) result in increases in travel demand. Using VMT changes to estimate the induced and diverted traffic is clearly deficient. Such use does not capture the complexity of the changes outlined here, nor does it relate directly to the estimation of emissions effects of the trans- portation facility changes. One of the thorniest problems to be resolved in measuring the effects of changes in the transportation system is the con- dition of “all other factors being equal.” This condition, made throughout the economics of demand and supply and very much of importance in considering both induced and diverted demand, assumes that everything else remains the same dur- ing the period of interest in which the transportation facility change is being made. Unfortunately, it is almost guaranteed that nothing remains unchanged while the capacity change or other facility change is implemented. This reality is particu- larly true when the change being implemented will have long-term consequences or when the change requires a sig- nificant time to implement. In such circumstances, there are likely to be changes in population, the economy, the supply of jobs, participation in the work force, fuel prices, the exis- tence of destination opportunities in new locations, etc. Thus, to be able to measure the effects on existing travel demand of a transportation facility change requires the analyst to be able to control for, or estimate the separate effects of, all the other changes that take place. This requirement poses a very difficult problem to solve, one that does not appear to have been solved to date. This chapter then considered issues of the empirical mea- surement of induced travel. Three methods have been pro- posed for empirical measurement. Two of these are closely related and are case studies and longitudinal panels. Case studies usually involve a series of cross-sectional surveys of FIN A L R EPO RT

the affected population, and longitudinal panels repeat mea- surements with panel members over some period of time. The chapter identified numerous problems that arise with either of these two methods, including the difficulty of defin- ing the area in which affected residents, employers, and devel- opers may be found; identifying the affected persons in the before period; and obtaining a control sample that is similar to the main sample, unaffected by the transportation facility change (or any transportation facility change), but equally affected by the population, demographic, fuel price, and other changes that affect the main sample. A number of other prob- lems and issues were also identified for each of these meth- ods. No effective solution was identified for these problems. The third empirical method that was discussed is attitude and preference surveys. These surveys have the potential to get around some of the difficulties of case studies and pan- 22 els, in that they allow for inclusion of, and control for, some of the externalities. In addition, these surveys do not neces- sarily require that a specific project is contemplated in order to obtain some measurements. However, in the absence of revealed preference data, the actual magnitude of elasticities cannot be determined. Acquiring the revealed preference data and controlling for the externalities raise the same prob- lems as for the case studies and panels. The chapter also noted that any type of empirical mea- surement, especially if it is intended to determine the long- term impacts of transportation facility changes, would need to be conducted over a substantial period of time. This period may need to last at least 10 years after the change has been fully implemented, which would lead, in many cases, to a total period of 13 to 15 years at least. The difficulty of main- taining consistent data collection over such a period is con- FI N A L R EP O RT Source Definition of Induced Demand Data Source Model Type Results Noland, Robert B., and William A. Cowart, “Analysis of Metropolitan Highway Capacity and the Growth in Vehicle Miles of Travel,” Transportation, Vol. 27, No. 4 (2000), 363-390. Increase in daily VMT per capita on freeways and arterials in a metropolitan area due to increase in lane-miles per capita on those facilities. TTI congestion report data on 70 U.S. metropolitan areas from 1982 to 1996. Panel (longitudinal) data. Distributed lag, with fixed effects. Short-run elasticity = 0.3. Long-run elasticity = 0.9. Marshall, Norman, “Evidence of Induced Demand in the Texas Transportation Institute’s Urban Roadway Congestion Study Data Set,” Pre-Print CD-ROM, Transportation Research Board Annual Meeting, Washington, D.C., 2000. Increase in daily VMT per capita on arterials and freeways in a metropolitan area due to increase in lane- miles per capita on those facilities. TTI congestion report data on 70 U.S. metropolitan areas for 1996 only. Cross- sectional data. Regression model (only noncapacity factor included is area). Elasticities of 0.9 for freeways and 0.8 for arterials. Noland, Robert B., “Relationships between Highway Capacity and Induced Vehicle Travel,” Transportation Research A, Vol. 35, No. 1 (2001), 47-72. Increase in statewide VMT on nonlocal roads due to increase in statewide lane-miles on those facilities. FHWA Highway Statistics 1984-1996 for 50 states. Distributed lag, with fixed effects. Short-run elasticities = 0.3–0.6. Long-run elasticities = 0.7–1.0. Fulton, Lewis M., Robert B. Noland, Daniel J. Meszler, and John V. Thomas, “A Statistical Analysis of Induced Travel Effects in the U.S. Mid- Atlantic Region,” Journal of Transportation and Statistics, Vol. 3, No. 1 (2000), 1-14. Increase in daily VMT on state-maintained highways within a county that is due to an increase in lane-miles on those facilities within that same county. HPMS systems for Virginia, Maryland, North Carolina, and Washington, D.C. Panel data for 220 counties, 1969 to present. Fixed effects: first difference: distributed lag: 0.3–0.6 elasticity. 0.15–0.6 elasticity. Short run = 0.1–0.4. Long run = 0.5–0.8. Chu, Xuehao, “Highway Capacity and Areawide Congestion,” Pre-Print CD- ROM, Transportation Research Board Annual Meeting, Washington, D.C., 2000. Increase in urbanized area traffic density (daily VMT per lane-mile) due to increase in lane-miles on nonlocal roads. FHWA Highway Statistics and HPMS. Cross-sectional data for 391 urbanized areas for 1997. Static equilibrium model. 0.03–0.4 elasticities depending on facility type and extent of lane-miles already present in each urbanized area. Barr, Lawrence, “Testing the Significance of Induced Highway Travel Demand in Metropolitan Areas,” Pre-Print CD-ROM, Transportation Research Board Annual Meeting, Washington, D.C., 2000. Increase in annual household VMT due to reduction in travel time. (Not limited by facility type, only includes capacity improvements that affect travel time.) Cross-sectional study of 27,000 households surveyed in 1995 NPTS. Regression model. Elasticities of 0.3–0.5 (with respect to time, not lane-miles). Note: most elasticities shown in this table are with respect to lane-miles of capacity or some variation of that measure (see “definition of induced demand” column for indication). HPMS = Highway Performance Monitoring System. TABLE 5 Comparison of recent induced-demand study results

siderable, as is clearly evidenced by the difficulties that arise for most MPOs with budgeting for a single cross-sectional survey and by the almost complete lack of panels for trans- portation measurement. The studies of the impacts of trans- portation improvements on urban form in already built urban areas suggest that the long-term impacts will be hard to dis- tinguish from other factors. 23 The examples of past conventional and advanced efforts to study the emission impacts of traffic-flow improvements suggest that the magnitude of the impacts will be quite small on a percentage basis compared with basinwide activity. The current research will require an exceptionally precise tool to isolate the emission impacts associated with traffic- flow improvements. FIN A L R EPO RT

24 FI N A L R EP O RT CHAPTER 3 STATE OF THE PRACTICE This chapter reviews the travel demand and emission- forecasting procedures used by MPOs and other practition- ers to evaluate the impacts of traffic-flow improvements in the United States. 3.1 REVIEW OF CONVENTIONAL PRACTICE This section describes the current demand-modeling and emission-estimating procedures used by seven leading MPOs in the United States. The procedures illustrate intermediate to relatively advanced practices and indicate the resources that might be available for an advanced methodology for predict- ing the emission impacts of traffic-flow improvements. This section details the phases, or “steps,” of the procedures. 3.1.1 Portland, Oregon This section describes the currently operational travel demand model developed by Metro for the Portland, Oregon, metropolitan area. (Note that a later chapter describes the experimental activity-based model currently being tested in Portland. The experimental model, however, is not currently used for production work by the MPO.) The Portland metro- politan area has a population of 1.8 million people and covers a land area of 6,954 square miles (18,080 square kilometers). The population estimate was taken from the 1990 Official Census count, based on 1992 definitions of the consolidated metropolitan statistical area (CMSA). The definition, and thus the population figure, may differ from that actually included in the regional model area. The Portland Metro model was calibrated against a 1994/ 1995 household activity and a behavior survey of 4,500 households. The input data for this model are as follows: • Socioeconomic and land-use data – Households cross-classified by four income cate- gories, four age-of-household-head categories, and four household size (persons per household) cate- gories (a total of 64 cells) – Employment (retail and other) – Land use (residential acres, industrial acres, and other acres) • Access measurement data – Degree of mixed land uses in zone – Retail and other employment within 1 mile of zone – Density of local intersections in zone – Total employment within 30 minutes via transit from zone • Special generators data – Shopping center floor area – Hospital staff – College students and staff – Weekday zoo attendance – Weekday attendance at the Oregon Museum of Sci- ence and Industry (OMSI) • Other data – Average weekday traffic volumes at external stations – Household and transit coverage factors (percent within zone that are within 1/4 mile of bus line or 1/2 mile of light rail line) – Zones with park-and-ride lots The steps of this model are as follows: • Pre-Generation. This step of the model consists of three independent multinomial logit models: a worker model, a children model, and an auto ownership model. The worker model estimates the proportion of house- holds in each zone that have 0, 1, 2, or 3+ workers. The children model estimates the proportion of households with 0, 1, 2, or 3+ school age children. The auto owner- ship model estimates the proportion of households in the zone that own 0, 1, 2, or 3+ autos. These models are sen- sitive to the household size and the age of the head of household. The worker and auto ownership models are sensitive to the household income. The auto ownership model is sensitive to the density of local street inter- sections in the zone, the degree of mixed uses in the zone, and the transit accessibility to employment of the zone (i.e., the number of jobs accessible within 30 min- utes via transit). • Trip Generation. Trip generation is estimated for six purposes (home-based work, home-based school, home- based college, home-based other, non–home-based work,

25 FIN A L R EPO RT and non–home-based other). A combination of cross- classified tables of trip generation rates and linear regres- sion equations are used to predict daily person trip pro- ductions and attractions. The trip generation rates and regression equations are sensitive to household size, workers per household, autos per worker, retail employ- ment, and other employment. The school trip generation estimates are sensitive to household size (persons per household) and children per household. The college trip production estimates are sensitive to household size (persons per household) and the age of the head of the household. The college trip attraction estimates are sensitive to special generator data gathered for each college. A separate modeling process is used to predict trips to the Portland International Airport. • Trip Distribution. A multinomial logit model is used to distribute the trips. The model is sensitive only to the number of attractions in the destination zone and the travel time between zones (the same as a standard grav- ity model). Special district-level (a geographic grouping of zones) adjustment factors (K) are applied to certain trip interchanges to better match the household survey trip distribution. These factors vary by district pairs and are constant. • Mode Choice. Mode choice is performed in two steps. First, the bicycle and walk trips are separated out. Then, the remaining trips are split between vehicle modes. The proportions of trips using walk mode and bike mode are computed using multinomial logit equations that vary by trip purpose and by mode. The mode split equations are sensitive to trip distance, cars per worker, local street intersection density, and the mix of land uses. School trip mode split is not computed using the logit equations; instead, the mode split is obtained from a table. There is a set of mode splits for each of four major areas within the metropolitan area. The walk and bike mode computations are con- strained by the following maximum allowed distances for these modes: Maximum Maximum Walk Bike Distance Distance Trip Purpose (miles) (miles) Home-Based Work 5 15 Home-Based Other 4 6 All Other 3 5 The motorized mode person trips are then split among the vehicular modes according to multinomial logit equations. The one exception is trips generated by 0-car households. These trips are split between transit and car passenger modes based on fixed percentages. Home-based work trips and non–home-based work trips are split among drive alone, shared ride, walk to transit, and auto to transit modes. All other trip purposes are split between auto and transit modes. The vehicle mode splits are sensitive to access time, in-vehicle time, cost, workers per household, cars per worker, trip distance, residential density, and employ- ment density. A central business district dummy vari- able is employed to account for the special transit usage characteristics of the downtown. A special adjustment process is used to shift some bus trips to light rail to account for the observed light rail ridership. Fixed auto occupancy rates by trip purpose are used to convert auto trips and shared-ride trips to equivalent vehicle trips. • Time of Day. Fixed percentages by trip purpose, direc- tion of trip (production to attraction, or attraction to pro- duction), and peak period are used to predict the num- ber of trips made during the AM peak 2 hours, the AM peak hour, the PM peak 2 hours, and the PM peak hour. • Traffic Assignment. Portland has a 1,244-zone network. The auto and truck vehicle trip tables are assigned to the highway network using a multiclass equilibrium assign- ment. Trucks are assigned in terms of their passenger car equivalents to account for their greater consumption of capacity. The truck table is developed through a separate process that is independent of the development of the auto trip table. Transit trips are assigned using a multipath assign- ment. Transit speeds are a function of the auto volumes on each link. • Feedback and Equilibration. No formal procedure was documented in the user’s guide or model description. 3.1.2 San Francisco, California This section describes the travel demand model developed by the Metropolitan Transportation Commission (MTC) in 1997 for the San Francisco Bay Area.59 The San Francisco Bay Area has a population of 6.3 million and a land area of 7,368 square miles (19,150 square kilometers). The model was calibrated against a 1990 household travel survey of 9,359 households. Another 1,479 households were surveyed for multiday (three weekdays) travel patterns. The steps of this model are as follows: • Pre-Generation. Demographic and socioeconomic fore- casts for the region are based upon national, state, and local trends. The POLIS model is used to spatially allo- cate the forecasts. These forecasts are performed by ABAG, which is separate from the MTC. A nested logit model is used by the MTC to predict the distribution of workers per household and vehicles per household based on the socioeconomic forecasts provided by ABAG. The top level of the nested logit model predicts the proportion

26 FI N A L R EP O RT of households with 0, 1, or 2+ workers per household. The second level of the model predicts the conditional proportion of households with 0, 1, or 2+ cars per house- hold given the number of workers per household. • Trip Generation. Daily person trip generation is esti- mated for the following trip purposes: home-based work, home-based social or recreation, home-based school, home-based other, and non–home based. Trip generation is estimated using linear regression equations that are sensitive to workers per household, household income, employment density, retail employment, service employ- ment, vehicles per household, and household size. School trips are divided into grade school, high school, and college subpurposes. The number of trips produced is a function of the school age population. School attrac- tions are a function of enrollment. • Trip Distribution. Trips are distributed using a gravity model based on a blend of peak and off-peak travel times. For each trip purpose, the peak and off-peak travel times are weighted according to the percentage of trips of that purpose that occur during peak and off-peak periods. The result is a table of weighted mean zone-to-zone travel times for each trip purpose. Home-based work trips are stratified by household income quartile. Each income quartile is distributed with its own friction factor curve. Fixed adjustment (K) factors are applied to specific trip interchanges to account for variations in trip mak- ing not adequately explained by the gravity model. • Mode Choice. A set of nested logit models is used to forecast mode choice by trip purpose. For the home- based work trip purpose, the top level of the model sep- arates trips by bicycle, walk, and motorized modes. The next level divides the motorized trips by drive-alone auto, two-person shared-ride auto, three-person shared- ride auto, and transit. Then the transit trips are further divided at the third level into auto access trips and walk access trips. The other trip purposes employ less exten- sive nesting and fewer modes. • Time of Day. A binomial logit model is used to predict the proportion of home-to-work auto person trips that are made during the 2-hour morning peak period. The model is sensitive to delay and household income. All other trip purposes and modes of travel are assigned to the peak period using fixed percentages. • Traffic Assignment. The MTC uses a 1,099-zone sys- tem plus 21 external gateways. The highway network has about 31,000 one-way links. The transit network has 25 transit operators and over 700 transit lines. Static user optimal equilibrium is used to assign vehicle trips to the highway network. • Feedback and Equilibration. A feedback procedure has been used by the MTC for years. Depending on the model run (e.g., existing versus future), between three and eight iterations are required for closure. Closure is based primarily on professional judgment. The direct method (rather than averaging previous runs) is used. 3.1.3 Dallas–Fort Worth, Texas This section describes the travel demand model process currently being used in the Dallas–Fort Worth area by the North Central Texas Council of Governments (NCTCOG).60 The Dallas–Forth Worth area covers 9,105 square miles (23,670 square kilometers) and has a population of more than 4 million. This model was calibrated against a 1984 home interview survey. The NCTCOG maintains land-use and socioeconomic data in a 5,000+ traffic survey zone system; however, this zone system is aggregated to 960 regional analysis areas and 61 external gates when used with the travel demand model. The steps of this model are as follows: • Pre-Generation. The Disaggregate Residential Alloca- tion Model (DRAM) and Employment Allocation Model (EMPAL) are used to predict land use in 5-year incre- ments for 191 super districts. The super-district fore- casts are disaggregated to the 5,000+ zone system. Nev- ertheless, the travel model cannot operate on such a large number of zones, so before trip generation is com- puted the necessary socioeconomic data are first aggre- gated from the 5,000+ traffic survey zone system to the 960 regional analysis area system. Household income distribution curves, which are derived from 1980 Cen- sus data, are used to compute the proportion of house- holds within each zone that fall into each income quar- tile. The curves relate the proportion of households in each quartile to the zonal median income. A similar process, which uses distribution curves from the Cen- sus, computes the proportion of households by house- hold size as a function of the mean household size for each zone. • Trip Generation. Four two-dimensional, cross- classification tables of trip rates (one for each trip pur- pose) are used to compute daily person trip production for home-based work, home-based other, non–home based, and other (external, truck, and taxi vehicle trips) trip pur- poses. The “other” trip purpose rates per employee vary by employment type and area type. The home-based and non–home-based trip rates per household vary by income quartile and household size. The home-based work trip productions are divided into four household income quartiles. Home-based work attractions are computed separately for each income quartile. A series of cross- classification tables are used to compute the zonal attrac- tions as a function of the employment in the zone (basic, retail, and service). Trip generation for regional malls, colleges, hospitals, airports, and regional recreational facilities are estimated separately as special generators.

27 FIN A L R EPO RT • Trip Distribution. A gravity model is used to distrib- ute trips. A Bessel function is used for the friction fac- tors. The home-based work trips for each income quar- tile are distributed separately, each with its own Bessel function. Intrazonal travel times are computed by divid- ing each zone into 13 concentric squares and by com- puting the average distance from the zone centroid to the perimeter of each square. A table of speeds by area type and time of day is used to compute the mean intra- zonal travel time from the average intrazonal trip dis- tance. K factors are used to account for trip behavior not adequately modeled by the gravity model. External- external vehicle trips are added to the “other” trip pur- pose trip table. • Mode Choice. Multinomial logit models are used to predict mode choice by trip purpose (home-based work, home-based other, and non–home based). Home-based work trips, which are stratified by income quartile, are split between drive alone, two-person shared ride, three- plus–person shared ride, transit with walk access, and transit with auto access. Similar modes are used for the other two trip purposes, with the exception that the two shared-ride modes are collapsed into a single two-plus– person shared-ride mode. • Time of Day. A fixed set of time-of-day factors by trip purpose is used to estimate peak-hour volumes from daily trips. • Traffic Assignment. An incremental capacity restrained assignment process is used. Link impedances used in the assignment process are a function of not only the link travel time, but also the link length and the link travel cost. Exponential functions are used to predict the impact of traffic volumes on link travel times. The expo- nential functions are capped so that the link speed never drops below 1 mile per hour. Different functions are used for the daily assignment and for the AM and PM peak-hour assignments. The highway network is coded using eight link types and centroid connectors. The link free-flow travel time is estimated by dividing the posted speed limit into the link length and adding the estimated control delay due to stop signs and signals on the link. Between 4 and 12 seconds of control delay is added depending upon the area type and the functional class of the link. Link capacities vary by number of lanes, median type (divided or not), area type, and functional class. • Feedback and Equilibration. The mainframe-based regional model uses final link speeds that are fed back to trip distribution and mode split until the change in the VMT-weighted highway assignment speed difference is less than 5 percent for each facility type. A direct method is used, i.e., the results from the previous iteration(s) are not combined with the current iteration to obtain a new overall solution. Rarely have more than one iteration been required. 3.1.4 Philadelphia, Pennsylvania This section describes the travel demand process used by the Delaware Valley Regional Planning Commission (DVRPC) for the Philadelphia metropolitan area.61 The region encom- passes 5.2 million people in two states. The model was calibrated against a 1987–88 survey of 2,500 households. The DVRPC maintains socioeconomic data for 1,395 zones. There are 114 external gates. The steps of this model are as follows: • Pre-Generation. Pre–trip-generation models are not employed. • Trip Generation. Daily person trip generation is fore- casted for home-based work, home-based nonwork, and non–home-based trip purposes. Fixed-trip rates by area type and by vehicle ownership category are used to esti- mate trip generation. Vehicle trips made by external trips, trucks, and taxis are estimated using separate rates. • Trip Distribution. Gravity models are used to distrib- ute the person trips and the truck, taxi, and external vehi- cle trips. • Mode Choice. The mode choice is predicted using a binary probit model that splits the person trips into auto and transit modes. An auto-occupancy model is used to predict drive-alone and shared-ride trips. The transit trips are assigned to submodes (commuter rail, subway/ elevated, and surface bus) during the assignment process according to the shortest transit path. The person trip table is stratified into 18 tables accord- ing to the trip purpose (home-based work, home-based nonwork, and non–home based), the transit submode that is used by the transit shortest path (commuter rail, subway/elevated, and surface bus/trolley), and the house- hold auto-ownership type (zero cars and one or more vehicles). Binary-mode choice (transit or auto) is com- puted for each of the stratified trip tables. The auto-occupancy model predicts the mean number of persons per vehicle for home-based work trips and for home-based other trips. It consists of a pair of linear equations that are a function of only the highway travel time. The linear equations are subject to allowable max- imum and minimum auto-occupancy values. • Traffic Assignment. Traffic assignment is performed using static user equilibrium. A standard Bureau of Pub- lic Roads equation is used to predict the impact of traf- fic on travel speeds. Transit trips are assigned to the sin- gle shortest path. • Feedback and Equilibration. No formal procedure is included in the user’s guide or model documentation. 3.1.5 Chicago, Illinois This section describes the travel demand model devel- oped by the Chicago Area Transportation Study (CATS) for

28 FI N A L R EP O RT the Chicago area.62 The CATS region includes the Illinois counties of Lake, McHenry, Cook DuPage, Kane, Kendall, Grundy, and Will and the Indiana county of Lake. The CMSA includes 8.2 million people in 6,931 square miles (18,000 square kilometers). The model was calibrated against a series of household surveys, the latest of which occurred in 1990 and consisted of 19,000 households. The land-use and socioeconomic data are tabulated for a system of 1,640 traffic analysis zones. The steps of this model are as follows: • Pre-Generation. The households in each traffic analy- sis zone are stratified into 21 categories according to the estimated number of workers per household and the estimated number of persons per household. Survey- developed distribution curves, which plot the percent- age of households in each category as a function of the mean persons per household or the mean workers per household, are used to estimate the percentage and num- ber of households falling in each category for each zone. • Trip Generation. Daily home-based person trip gener- ation is computed using trip rates in a cross-classification table stratified by workers per household in one dimen- sion and by persons per household in the other dimension. The number of trips generated by workers in the house- hold is added to the estimated number of trips generated by the remaining persons in the household to obtain the total trips generated by each category of households. Trips are generated for three purposes: home-based work, home-based other, and non–home based. Linear regression equations, which are sensitive to seven employment categories and the total number of households in a zone, are used to predict trip attractions and trips produced outside of the homes in each zone. Special floor space–based trip generation rates are used to predict trip generation for the central Chicago area. A separate model is used to predict truck trip generation. • Trip Distribution. Chicago uses an “intervening oppor- tunities” model (similar to a gravity model) to distribute trips. Separate friction factor curves are used for each trip purpose and for each of 15 different districts within the region. Vehicle trip tables for through trips, visitor trips, school trips, truck trips, and taxi vehicle trips are estimated separately. • Mode Choice. A binary logit model is used to split per- son trips into auto and transit modes. A unique Monte- Carlo simulation approach is used to trap the impact of variances in parking costs and the income of the traveler on mode split. The mode split probability is computed for each individual trip between zones, and a random number generator is used to select a parking cost and income for that trip. The results of this simulation are then summed over all trips between the pair of zones to obtain the transit trips going between the zones. • Traffic Assignment. The transit network has 642 tran- sit lines coded. • Feedback and Equilibration. No formal feedback or equilibration procedure is included in the user’s guide or model documentation. 3.1.6 Washington, D.C. This section describes the travel demand model currently being used by the Metropolitan Washington Council of Gov- ernments (MWCOG) for the Washington, D.C., area. The socioeconomic data are stored in a system of 1,972 traf- fic analysis zones that are aggregated to 333 districts for the trip generation and distribution steps. Then the district-level trip table is proportionally disaggregated to the zone level for the mode split step. The steps of this model are as follows: • Pre-Generation. A household vehicle ownership model is applied at the district level to estimate the number of homes in each district owning 0, 1, or 2+ vehicles. • Trip Generation. Motorized person trips are generated for home-based work trips only. Vehicle trips are gen- erated for three noncommercial purposes (home-based shop, home-based other, and non–home based) and two truck purposes (medium weight and heavy weight). Lin- ear equations, which are sensitive to households by auto ownership category and by five categories of employ- ment, are used to forecast trip productions and attrac- tions for all six purposes. • Trip Distribution. Trip distribution is computed using the gravity model. A preliminary assignment of vehicle trips is made to obtain “first cut” congested travel times for use in distributing home-based work trips. • Mode Choice. The mode choice model is applied only to the home-based work trips. Home-based work trips are split between low-occupancy vehicle, high-occupancy vehicle, and transit modes. • Traffic Assignment. Traffic assignment is performed using an incremental capacity restraint algorithm. • Feedback and Equilibration. Congested highway travel times are fed back to the trip distribution model only for home-based work trips. This cycle is repeated twice. Other trip purposes are assumed to be unaffected by traffic congestion. 3.1.7 Seattle, Washington This section describes the travel demand model that is currently being used by the Puget Sound Regional Council (PSRC) for the Seattle area.63 The Washington CMSA popu- lation is over 3 million with a land area of 7,224 square miles (18,780 square kilometers). The PSRC has been gathering a longitudinal panel household travel behavior data set for

29 FIN A L R EPO RT 1,700 households since 1989. The panel has been surveyed eight times since 1989. Approximately 50 percent of the original households are still in the panel. New households have been recruited to replace those leaving the region so that the current panel remains at about 1,700 households. The steps of this model are as follows: • Pre-Generation. The PSRC uses a linear regression model to predict the regional control totals for house- holds and employment. The regression model is not sen- sitive to changes in accessibility. The DRAM/EMPAL models then are used to allocate the regional totals to 219 districts, which are then further disaggregated to 832 traffic analysis zones. • Trip Generation. Daily motorized person trip produc- tions and attractions are estimated for home-based work, home-based other, and non–home-based trip pur- poses, plus college student trips, school trips, and com- mercial vehicle trips. Productions are estimated using cross-classification tables that, for home-based work, home-based other, and non–home-based trips, are sen- sitive to household size and the number of workers in the household. College and school productions are sen- sitive to the number of college age students and school age students per household. Commercial vehicle trips are factored from the non–home-based trips. Linear regression equations are used to predict attractions. Trip generation is not sensitive to travel time, access time, or auto ownership. The PSRC currently is testing an update of its model that includes nonmotorized modes in the trip generation step. The new model also splits out home-based shop trips from the home-based other category and groups home-based work trip productions and attractions by income quartile. • Trip Distribution. Trip distribution is done using a grav- ity model with K factors to correct for underestimates or overestimates by the gravity model. • Mode Choice. A logit mode choice model is used to predict the percentage of home-based work, home-based other, and non–home-based trips that are transit with walk access, transit with auto access, and automobile. Home-based work auto trips are further split into car- pool and single-occupancy (noncarpool) person trips. AM peak-period travel times are used to estimate home- based work mode choice. Midday travel times are used for predicting the mode choice for the other trip pur- poses. The auto operating costs are included in the mode choice analysis for transit trips with auto access. The PSRC currently is testing a combined mode choice and trip distribution model that includes walk and bike mode choices. • Time of Day. The daily trips by mode are split into trips made during the 3-hour AM peak period and the 3-hour PM peak period using fixed percentages. The remainder of the trips are midday and off-peak trips. • Traffic Assignment. Single-occupancy auto, carpool, and commercial vehicle trips are assigned to the high- way network using multiuser equilibrium assignments. The auto access portions of transit and auto access trips are also assigned to the highway network. Transit trips are assigned using EMME/2’s capacity constrained algorithm. • Feedback and Equilibration. Travel time results are routinely fed back to the trip distribution and mode split steps. Major changes in the transportation system are fed back to the DRAM/EMPAL land-use allocation step. 3.2 CRITIQUE OF CONVENTIONAL PRACTICE Litman64 has criticized conventional travel demand mod- els for underestimating the demand-inducing impacts of highway capacity increases. He is one of many voices to crit- icize the current state of the practice. As noted by Deakin and Harvey65 in their review of the state of the practice, the quality of models in practical use varies significantly. Merely bringing all MPOs up to current standard practice would be quite an improvement. The key shortcomings of current practice that they identified include • Omission of key variables for predicting travel behav- ior (household income, parking and auto operating costs, and number of workers per household); • No trip generation variables beyond auto ownership and income (e.g., household size would be a good predictor); • Inadequate representation of trip attractions; • Omission of transit and walking accessibility in trip dis- tribution models; • Lack of peaking information by trip type and market segment; • Simplistic representation of socioeconomic variables affecting travel behavior; and • Simplistic characterization and modeling of nonwork travel. Deakin and Harvey also note that many MPOs are not gathering the data they need to develop and maintain ade- quate travel models. They recommend regular collection of land use, land-use regulations, travel behavior surveys, net- work, and monitoring data. They also recommend additional staffing to maintain and operate the models. Stopher66 has noted that current travel demand models suf- ficiently predict the impact of travel cost changes on mode choice, but not the impact of cost on overall demand for travel. The typical problems of conventional travel demand models are that • They cannot reflect changes in trip making per household, • They lack feedback (i.e., equilibration of demand with supply),

30 FI N A L R EP O RT • They fail to use land-use models to reflect the impacts of transportation changes on land use, • They have large aggregation errors with large zones, and • They cannot accurately predict real-world travel speeds. Many of the advanced MPOs described in Section 3.1 have already addressed many of the above problems. Stopher’s observations still apply to the vast majority of MPOs that have not yet addressed these issues. Stopher observes that only a subset of TCMs, those that are quantifiable, could be reasonably evaluated using conven- tional travel demand models. The primary quantifiable TCM strategies that can be evaluated with conventional travel demand models are • Price-related TCMs (transit fare subsidies, parking costs, and tolls), • HOV lanes, • Transportation system management (TSM) improve- ments that have measurable impacts on speeds and capacities, • Transit service improvements (exclusive of reliability changes), and • Park-and-ride lots. Examples of nonquantitative TCMs are informational, promotional, and marketing TCMs. Stopher and Fu67 identified a set of improvements that could be made in the short term to improve the accuracy of conventional travel demand models and the accuracy of mobile emissions produced from demand model output. The following improvements are identified: • Pre-Trip Distribution Diurnal Factoring. It is pro- posed that the factoring of daily trips to time of day be moved to immediately after the trip generation step. Thus, the model will run the distribution, mode split, and assignment steps for five time periods using the travel times and costs appropriate for that time of day. • Link-Specific Capacities. It is proposed that more pre- cise capacities be computed on a link-specific basis rather than relying on general capacity values based on the area type and the number of lanes. • More Realistic Speed-Flow Curves. It is recommended that modelers adopt more realistic speed-flow curves that show a much steeper drop in speeds when demand exceeds capacity. • Feedback. Congested travel times should be fed back to the trip distribution step. • Seasonal and Day of Week Factoring. Seasonal and day of week factors should be developed to convert the average weekday volumes produced by models into spe- cific season and day of week data needed for air quality analyses. Feedback, or equilibration of travel times with the assumed travel times, is a major issue for all travel models. A report by the Comsis Corporation68 for the U.S. DOT TMIP identi- fies conditions when feedback should be used, explains how it can be implemented, and describes its effects on model results. Harvey Miller69 has prepared a guide on maintaining internal consistency within travel models that presents the basic theory of transportation system equilibrium and describes the various types of equilibrium (i.e., user opti- mal, dynamic user, and stochastic user). The appendixes provide the formulas and properties for network and market equilibria. Replogle70 recommends additional data collection, includ- ing panel surveys, traffic counts, time and delay studies, supply inventories, pricing data, goods movement data, special gen- erator data, and land development inventories. Replogle also makes the following recommendations for model methodol- ogy improvements: • The trip generation models must predict person trips not vehicles trips, be sensitive to changing demographics and urban structure (i.e., access time), be sensitive to trip chaining, and consider job/housing balance in fore- casting land use and external trip patterns. • Trip distribution models must use travel times internally consistent with later stages of the model, integrate multi- modal factors, and provide for departure time choice. • Mode choice analysis must improve treatment of tran- sit access options, better represent auto access to transit, better represent nonmotorized access modes to transit, become sensitive to variations in pedestrian and bicycle friendliness, and become sensitive to auto ownership. • Networks must have increased zone and network detail, and intersection capacity and delay must be separated from link capacity and delay. • Models must be sensitive to alternative land-use sce- narios. • Models must be able to represent transportation demand management (TDM) programs. From the perspective of statewide travel forecasting, Horowitz71 identifies appropriate methodologies for differing analysis needs. One of the several methodologies he presents is a four-step modeling procedure for forecasting statewide passenger travel: 1. Trip generation, 2. Trip distribution, 3. Mode choice, and 4. Vehicle assignment. Although Horowitz’s original model includes four steps, some applications of the model have elaborated on the basic four steps, adding such steps as pre-generation and time of day. Thus, the applications included in this chapter include

31 FIN A L R EPO RT more than four steps, although they follow Horowitz’s orig- inal four-step model structure. 3.3 MODELING NONMOTORIZED TRAVEL Relatively few models take into account nonmotorized travel. Only some of the more advanced research models have attempted to explicitly model nonmotorized travel. This section is a condensed version of the FHWA 1998 overview titled Guidebook on Methods to Estimate Non-Motorized Travel: Overview of Methods (Publication No. FHWA-RD- 98-165) and also refers to an article by Thomas Rossi of Cambridge Systematics. The term “nonmotorized travel” refers mostly to bicycling and walking, yet also could include in- line skating, scooting, skateboarding, or horseback riding. This review discusses the various methods that are used to predict future demand of nonmotorized travel. Other meth- ods are available that support demand forecasting such as the usage of land-use and population data, before-and-after stud- ies, preference surveys, facility and environment characteris- tics, and geographic information systems (GIS). This review only covers demand estimation methods of discrete choice and regional travel models because these approaches reflect the state of the art of nonmotorized demand estimation techniques. Thomas Rossi72 prepared a paper on methods for incorpo- rating nonmotorized vehicle modes (bike and pedestrian) into travel behavior models. He identifies four reasons for incor- porating nonmotorized travel in models: better modeling of mode choice, analysis of transportation demand management measures, analysis of alternative land-use patterns, and pre- diction of transit access. Rossi describes three examples of nonmotorized models: the Central Artery/Tunnel project, the Land Use Transporta- tion Air Quality (LUTRAQ) project, and Philadelphia. The Central Artery/Tunnel model is a submodel of the regional model system and is focused on downtown Boston. A special pedestrian trip generation, distribution, and assignment model was developed along with a pedestrian network. The Port- land LUTRAQ model extended the preexisting pedestrian/ bicycle modeling capabilities of the Portland model. Pedes- trian environment variables and data were added to the Port- land model, which enabled more sophisticated auto ownership and mode choice forecasts. The Delaware Valley Regional Planning Commission (DVRPC of Philadelphia) added non- motorized trips to the trip generation model and then sepa- rated them out using a binary mode choice model prior to trip distribution. Rossi notes several limitations with pedestrian environ- ment variables: They are, of necessity, limited to zones, since that is the smallest analysis unit within travel behavior mod- els. A significant amount of time is required to develop and update environmental variables for each zone. Many of the components of the pedestrian environment variables are sub- jective, making it difficult to ensure consistency between model operators and between calibration and forecasting. Rossi concludes that environmental variables have been suc- cessfully applied, but do require a great deal of care. 3.3.1 Discrete Choice Models Discrete choice models predict individuals’ choice of mode or route as a function of variables such as parking availability or the traffic level. The model is used to estimate how travel- ers would respond to a specific policy change or facility improvement like increased bicycle parking at transit sta- tions. The underlying data set of the model consists of the following characteristics: individual attributes such as age and income, alternative route or mode choices, geographical location, and individual trip decisions. These data sets are developed from revealed and SP surveys. Revealed prefer- ence surveys quantify actual behavior, whereas SP surveys illustrate the choices that travelers would make given differ- ent scenarios. The results are limited by the questions asked in the SP survey. The possible outputs include the probabil- ity of an individual to choose bicycling or walking for each scenario, elasticities that show the percent change of bicy- cling or walking when one variable changes, and the total number of travelers who are expected to change for each sce- nario. The results could be incorporated into regional travel models as bicycle and pedestrian submodels. 3.3.1.1 Work Trip Mode Choice In the late 1970s and early 1980s, the Wisconsin Depart- ment of Transportation (WisDOT) developed a series of work trip mode choice models to determine the impacts of trans- portation policy in urban areas throughout the state. The bicy- cle variables included the presence of bike lanes, street surface quality, and street traffic. The pedestrian variables included the presence of sidewalks, season, and distance to work. The pres- ence of bicycle lanes on all the applicable streets caused a 39-percent increase in summer bicycle trips. A deterioration of pavement quality from smooth to rough caused a 42-percent reduction of summer bicycle trips. These results remained fairly constant for the four urban areas that the model covers. The calibration process showed a reasonable correspondence between the model estimates and actual travel behavior.73 3.3.1.2 Transit Access Mode Choice in Chicago The Chicago Transit Authority (CTA) developed discrete choice models to predict the travel impacts of bicycle and pedestrian improvements at rail stations. CTA used two nested logit models: one to measure the access mode to the com- muter rail and the other for the rapid rail. These models used the following variables to estimate changes in mode split:

32 FI N A L R EP O RT travel time, parking availability for autos, parking costs, other costs, number of buses, and bicycle and pedestrian improve- ments. The bicycle variables that had high statistical signifi- cance include debris, bicycle parking, curb lane width, and slow traffic. The presence of bicycle facilities was not sta- tistically significant. The pedestrian variables included side- walks, recreation paths, slow traffic, no turn on red, cross- walks, pedestrian lights, and walk islands. The model was used to prioritize stations, select case study locations, identify design improvements, and estimate the cost-effectiveness of improvements.74 3.3.2 Regional Travel Models Regional travel models use existing and future land-use data, transportation networks, and human behavior to predict future travel patterns. The traditional four-step approach of trip generation, trip distribution, mode choice, and network assignment typically has been oriented toward autos and tran- sit. More sophisticated models predict nonmotorized mode splits and route choice. Models specifically for bicycling and walking also exist. The primary factors that are assumed to influence bicy- cling and walking are trip distance or time, trip purpose, indi- vidual characteristics, and environment factors. The different environment factors cover the following: sidewalk availabil- ity, terrain, land-use mix, building setbacks, transit stops, street crossings, and bicycle infrastructure. The model out- puts include the nonmotorized trip generation of each traffic analysis zone and the trip distribution between the zones. The advantage of regional travel models is that they exist in every major metropolitan area in the United States. With sufficient nonmotorized infrastructure and demand data, these models represent the state-of-the-art method for nonmotorized travel demand estimation. The disadvantages of regional travel models include insufficient data on nonmotorized travel pat- terns, inadequate knowledge about the nonmotorized network characteristics, inability to consider recreation trips, and a level of detail that is too coarse to analyze shorter trips. 3.3.2.1 Edmonton Transport Analysis Model The Edmonton Transport Analysis Model includes bicy- cling and walking as separate modes and uses bicycle network characteristics to predict nonmotorized travel. A link was coded as a facility with a bicycle path, a bicycle lane, or mixed traffic. Time-equivalent penalties were given for each facility type. SP surveys showed that bicyclists would ride 1 minute on mixed-use facilities, 2.8 minutes on bike paths, or 4.1 min- utes on bike lanes. Feedback loops made it possible for the model to show the affects of bicycle network improvements on trip generation, trip distribution, and mode choice.75 3.3.2.2 Pedestrian-Bicycle Environment Factor Models Pedestrian-bicycle environment factors such as sidewalk and bikeway availability help predict nonmotorized trips. These factors describe the attractiveness of an area to bicy- clists and pedestrians. The Portland regional model used the following pedestrian environment factors: sidewalk avail- ability, ease of street crossing, connectivity of street and side- walk system, and terrain. Each zone was ranked according to its quality of pedestrian environment. The mode choice step included a motorized versus nonmotorized option, which was a function of the pedestrian environment, travel distance, ratio of cars to workers in households, and employment within 1 mile of the zone. The Maryland–National Capital Park and Planning Commission developed a nested logit mode choice model that included bicycle and pedestrian variables in its environment factor. The walk/bike mode was used to deter- mine transit access.76 3.4 MODELING TRUCK TRAFFIC 3.4.1 Literature Review: Freight Flow Models The freight flow model literature review is a condensed version of the summary provided by Cambridge Systemat- ics titled Review of Current Freight Flow Models (Draft), which was prepared for the Florida Department of Trans- portation Freight Model Development Project in September 2000. The review provides a discussion of vehicle-based and commodity-based models and includes descriptions of existing truck models that use the respective techniques. Modeling truck movements separately from passenger cars is a relatively new phenomenon. Most truck models have focused on the statewide level. A few models have been devel- oped for urban areas (e.g., Portland, Phoenix, and Sacra- mento), but they generally operate independently with limited interfaces to the conventional travel demand models already present in the region. Models of truck movements in an urban area approach the problem in either of two ways: they model the truck traffic flows directly or they model the flows indirectly by model- ing the movements of commodities and then deriving the truck flows necessary to carry the commodities. 77 Commod- ity flow models are sensitive to many economic variables that affect the amount of goods that must be moved; however, they often exclude the various truck trips not associated with the movement of goods (such as service trips). Truck flow models can only indirectly predict the amount of truck traf- fic through land-use proxies, but have the advantage of mod- eling all truck trips. The ideal regional model would use both the vehicle-based and the commodity-based model approaches. The advantage of the vehicle-based model is that it includes service-oriented

33 FIN A L R EPO RT trucks such as distribution and air express delivery move- ments. Commodity-flow models are better at estimating long- haul truck movements. The Southern California Association of Governments (SCAG) model, which is described under the commodity-based model section, provides the best exam- ple of this combined approach. 3.4.2 Vehicle-Based Models 3.4.2.1 Background Vehicle-based models are based on three of the four steps used in the traditional person transportation modeling process: trip generation, trip distribution, mode choice, and vehicle assignment. The mode split step is not necessary since the model only focuses on one mode. Trip generation applies trip rates to traffic analysis zone employment or household data. Special generators such as seaports, airports, and intermodal rail yards are also considered. External stations include the effect of long-distance truck trips. The trip distribution step uses the gravity model and considers truck trip lengths. The vehicle assignment step focuses on a subset of roads that con- sists of the highway network and other truck routes. Other modifications include revisions to the network’s speed, capac- ity, and toll rates. Two sample vehicle-based models include the New Jersey and Phoenix truck models shown below. 3.4.2.2 New Jersey Statewide Truck Model The New Jersey statewide truck model was developed to establish more accurate truck trip tables. The previous truck model used a commodity flow approach that did not ade- quately reflect service-oriented truck trips. Truck trips are estimated based on information from a variety of sources like cordon-line and facility-specific surveys as well as state bor- der crossing count data. Separate gravity models exist for medium- and heavy-truck types and for trip end types such as internal/internal (I/I) and external/internal/external (E/I/E). Truck trips are assigned to the truck routes along with auto trips. This model is used to estimate the impacts of toll changes, road construction, and major new developments. 3.4.2.3 Phoenix Truck Model The Phoenix truck model also uses three of the four steps of the person transportation modeling process. Different trip gen- eration and distribution models exist for the following truck types: less than 8,000 pounds, 8,000–28,000 pounds, and more than 28,000 pounds. The truck generation models are based on a survey of daily truck trips. The trip distribution step uses weight class-specific gravity models, which are calibrated to observed trip length distributions and their averages. 3.4.3 Commodity-Based Models 3.4.3.1 Background Commodity-based models forecast freight flows usually for statewide transportation networks using commodity flow data for at least two modes: truck and rail. Commodity data come from the nonproprietary U.S. Census’s Commodity Flow Survey or from the proprietary sources of Reebie and Asso- ciates or TRANSEARCH. Trip generation rates derived from population and employment data are used to better under- stand annual or daily travel flows. Future changes in the high- way network are also considered. The zone system is at the county level and typically includes fewer than 200 zones. The trip distribution step consists of gravity models that are based on five or six commodity groups. Using tons per truck or railcar for each commodity type, a conversion from commodity flows to vehicle flows is possi- ble. After the trip distribution step, conversion to trucks by size and weight occurs. The vehicle assignment method is an all-on-nothing procedure since trucks have limited route alternatives. The following five models reveal the different variations that are possible using the commodity-based mod- eling approach. 3.4.3.2 Indiana Statewide Freight Model The Indiana statewide freight model uses both nonpropri- etary and proprietary data to forecast the freight flows of trucks and rail within the state of Indiana. The freight flow data come from the latest U.S. Census’s Commodity Flow Survey. The existing and future county-level population and employment data come from the databases of Woods and Poole Economics, Inc. (www.woodsandpoole.com). The lat- est edition of the Interstate Commerce Commission’s (ICC’s) Rail Waybill Sample is used to convert commodity flows to vehicle flows for railcars. The conversion for heavy trucks is assumed to be 40 percent of railcars. 3.4.3.3 Kansas Statewide Agricultural Commodity Model The Kansas statewide agricultural commodity model fore- casts the flows of five major agricultural commodities by truck, rail, and barge. The model uses nonproprietary data and new data collected specifically for the model effort, such as mail surveys, telephone reports, base year traffic counts, and field interviews. The highway network acts as the start- ing point with revisions for link grades and toll rates. The data are tabulated in mode- and commodity-specific trip tables. Gravity models are based on origin/destination data. The model is able to test changes to the existing transportation system, yet has a limited forecasting capability.

34 FI N A L R EP O RT 3.4.3.4 Michigan Statewide Truck Model The Michigan statewide truck model uses a variety of national and international data to develop base year truck trip tables. To obtain projections, the model uses data from the Bureau of Economic Affairs (BEA) and from a proprietary source, Regional Economic Models, Inc. Interindustry Fore- casting at the University of Maryland (Inforum) forecasts truck movements between the United States, Canada, and Mexico. The model is highly compatible with the person transportation model, yet has limited forecasting abilities except for route choice. 3.4.3.5 Portland Commodity Flow Model The Portland commodity flow model has two major com- ponents: the strategic model database (SMD) and the tactical model system. The SMD analyzes the existing and future freight flows. It covers eight modes of travel, including pri- vate truck, less-than-truckload, truckload, intermodal, rail, barge, sea, and air. Seventeen commodity groups are included, as well as five origin and destination areas such as northern and southern United States. The SMD is based on both pro- prietary and nonproprietary data. The tactical model system is a more in-depth analysis of heavy-truck trips. It uses data provided in the SMD and relies on the Portland regional per- son travel model to do the following: summarize the heavy- truck flows, allocate their origins and destinations, simulate reloading and terminal usage, convert commodity flows to heavy-truck trips, add empty trailer and tractor trips, add through truck trips, and assign heavy-vehicle trips to the highway network. The model forecasts changes in the trans- portation infrastructure, yet lacks sensitivity to changes in transportation policy and private-sector costs. 3.4.3.6 Southern California Association of Governments Regional Heavy-Duty Truck Model The Southern California Association of Governments (SCAG) regional heavy-duty truck model consists of three submodels to forecast transportation and air quality in the Los Angeles region: • The intraregional model is not commodity based due to the lack of commodity data for the 1,300 internal zones. Instead, population and employment data are used to esti- mate existing and future light-heavy, medium-heavy, and heavy-heavy truck trips. Gravity models are developed for the three truck classes. • The external-internal/external-external model com- bines multiple commodity flow data sources at the county level. • The special generator model predicts truck trips as opposed to commodity flows at shipping ports and air- ports within the SCAG region. The submodel results are combined and converted from daily to hourly trip tables and then are assigned to the high- way network. Six trip tables are provided: three truck-type tables for external-internal truck trips and three for internal truck trips. The models focus on truck types, time period, and trip-end type allowing for congestion and truck flow analyses. 3.5 NCHRP PROJECT 8-33 RECOMMENDATIONS FOR IMPROVED PROCEDURES NCHRP Project 8-33, “Quantifying Air Quality and Other Benefits and Costs of Transportation Control Measures,” is a research project to develop and test an improved framework for analyzing the air quality impacts of TCMs. This project was completed December 1999 and as such is a predecessor to the current research effort. NCHRP Project 8-33 investigated the current state of the art for analyzing TCMs and concluded that while the con- ventional modeling approach of linking models works, “seri- ous reservations exist concerning the accuracy of these results, the robustness of the underlying data, and whether the cor- rect set of variables are captured in the model systems repre- senting current practice.”78 The Project 8-33 researchers recommended that a new modeling framework consisting of the following modules be developed for the purpose of evaluating TCMs: • Disaggregate and activity-based demand. These newly emerging modeling approaches focus on the individual, the household, the vehicle, and the trip, rather than the aggregate groups of households used in more traditional approaches. A daily activity plan by hour of day (including trip making) is developed for each person or household in response to the char- acteristics of the person, the household, and that per- son’s environment (accessibility to jobs and other activities, etc.). Nonmotorized trips are included explicitly in the daily activity plan. • Household sample enumeration. Rather than aggre- gating households by traffic analysis zone within the region and then predicting the mean trip patterns for that group, individual persons or households are randomly selected, and their individual travel patterns are predicted. These individual trips are then expanded and summed to represent total travel of all households in the region. Individual trip-making patterns and linkages are pre- served with the sample enumeration method. • Incremental analysis. Incremental analysis involves comparing the “changes” produced by specific strate-

35 FIN A L R EPO RT gies rather than the absolute magnitude. The philosophy behind this approach is that behavior models tend to be more accurate at predicting changes in travel behavior than at predicting the total magnitude of travel. The mod- els that predict “changes” or “deltas” are then added to the “well-calibrated” base-year trip table (obtained from some other source than the travel demand model) to obtain the future trip table. • Traffic microsimulation. Traffic microsimulation is needed to obtain accurate modeling of congestion effects (i.e., speed, delay, queuing, and volume) and to output vehicle operating mode predictions (i.e., acceleration, cruise, deceleration, and idle). • Household travel survey data with SP data to sup- port policy analyses. Extensions include information on characteristics of vehicles used (model, year, type, and odometer reading), seasonal variation of travel patterns, time of day, weekend trips, SP responses to potential TCMs, and panel surveys to track longer-term responses and to monitor the reliability of SP responses. The Project 8-33 researchers also recommended various improvements to current emission models, including • Update the Federal Test Procedure, • Update speed correction factor test cycles, • Improve the speed correction factor methodology, and • Develop link-specific emission rates. Specifically, the researchers noted that the Federal Test Procedure and the speed correction factor test cycles need to be reviewed and updated in light of new information on how people actually drive their vehicles. The current speed cor- rection factor methodology considers only the mean speed of the trip. It does not consider the underlying distribution of speeds and acceleration that vary by facility type and degree of congestion. Trip-based emissions are not appropriate for estimating emissions for improvements to individual segments of the roadway system. Recognizing that all of these improvements will take time to research and implement, NCHRP Project 8-33 researchers have also developed a series of improvements that can be implemented in the mean time (see Table 6). 3.6 ENVIRONMENTAL PROTECTION AGENCY ANALYSIS The Environmental Protection Agency (EPA) has issued reports on appropriate travel activity methodologies for the analysis of intelligent transportation systems (ITS)79 and pric- ing measures to reduce transportation emissions.80 The EPA’s ITS analysis report identifies the need for two major method- ological improvements to evaluate the emission impacts of ITS measures: • Modal emission models that can predict second-by- second tailpipe emissions under a variety of conditions and • Travel demand models linked to microsimulation mod- els with feedback from simulation back to trip genera- tion, distribution, and mode choice. The EPO proposes that a range of modeling approaches and analytical processes would be required to analyze ITS compo- nents since the different components affect different aspects of the transportation system at differing levels of spatial aggre- gation. A series of candidate analytical tools are considered Area of Improvement Recommendation Feedback Linkage Incorporate a feedback linkage that equilibrates congested travel time predictions with assumed travel times used in trip distribution and mode choice Land-Use Model Adopt a formal land-use model that provides capability to assess changes in location in response to transportation system changes Vehicle Ownership Model Provide a policy-sensitive model of vehicle ownership choice Trip Generation/Distribution Add nonmotorized modes Mode Choice Provide a nested logit, add a nonmotorized mode, and access modes Time of Day Provide a time-of-day choice model or a peak spreading model sensitive to predicted congestion levels Route Choice Incorporate the effects of tolls Household Surveys Add TCM-related SP questions Household Sample Enumeration Aggregate individual household trip patterns, rather than forecasting aggregate trip making for aggregates of households Traffic Microsimulation Link the travel model output to microsimulation models Network Coding Increase the coverage and precision of the network Emissions Analysis Configure the EPA vehicle emission factor model (MOBILE) for operating mode corrections (cold start, hot start, or stabilized), develops trip-based emission estimates (separate start emissions from running exhaust emissions), and links mode of travel (auto, transit) to vehicle class (cars, light-duty trucks, heavy-duty trucks, motorcycles, etc.) TABLE 6 Short-term improvements identified by NCHRP Project 8-33 researchers

36 • The accuracy of the pricing relationships that are included in models is uncertain because of the lack of in-place pricing policies for validating the relationships; • The income effects of traveler response to pricing rarely are included in models; • The use of zonal averages in the models restricts the pricing detail that can be considered in the models; and • Models are not set up to predict the impact of pricing on through trips or commercial trips. The recommended improvements revolve around the inclu- sion of cost (or its equivalent in time impedance units) in the computation of vehicle ownership, vehicle trip generation, trip distribution, mode choice, and route choice. The STEP model is presented in the appendix of the EPA report as an example of an advanced modeling approach. FI N A L R EP O RT in the report: INTEGRATION, Traffic Network Simulation (TRAF-NETSIM), Air Quality (AirQ), MOBILE6, CMEM, MEASURE, Transportation Analysis Simulation System (TRANSIMS), Mitretek’s travel demand modeling process, and Short-Range Transportation Evaluation Program (STEP). The EPA’s pricing measures report focuses on recommen- dations regarding the appropriate analysis methodologies for evaluating the emission reduction potential of the following: parking pricing, modal subsidies, at-the-pump charges, emis- sion fees, and roadway pricing. The pricing report notes that the conventional four-step travel forecasting process was not designed for testing pricing policies. The conventional process has shortfalls for testing pricing policies: • Pricing is not rigorously included in the model’s struc- ture (usually pricing is only considered in mode choice);

FIN A L R EPO RT 37 CHAPTER 4 AVAILABLE METHODOLOGIES This chapter presents an overview of the leading available methodologies for estimating the emission impacts of traffic- flow improvement projects. 4.1 TYPOLOGY It is useful to categorize the various methods and models in the literature according to the portion (or portions) of the total air quality estimation process that each method is designed to address. As shown in Figure 3, the entire process is divided into five major analytical steps: land-use forecasting, travel demand estimation, transport system operations analysis, emis- sions estimation, and air quality forecasting. This research will focus only on the first four steps and will leave air qual- ity forecasting to others. Each analytical step in the process can be considered a “link” in the analysis chain. When all the links are completed and connected, they form a comprehensive and complete procedure for analyzing the emission impacts of any policy or investment option for the air basin. Each link is defined as follows: • The land-use step forecasts growth in population and demographic and socioeconomic changes and spatially allocates people, households, and commercial activity within the air basin. • The travel demand step converts the locational data gen- erated by the land-use model into estimates of travel activity. • The systems operation step estimates the impacts of the forecasted travel activity on the operation of the region’s transportation system. This step predicts travel time, speed, delay, and vehicle modal activity. The travel times predicted in this step influence the prior steps, land use and travel demand. • The pollutant emissions step uses the vehicle modal activity data to make emission predictions, which are in turn fed into an analysis of the basin’s air quality. It is useful to add a second dimension to the typology by establishing the level of aggregation for which each method is designed to be applied. The research objective and the nature of the air quality analysis problem is such that the abil- ity to evaluate policy and investment options at a micro- scopic level will be a valuable attribute of any methodology considered in this research project. The levels of aggregation are as follows: • The areawide level of aggregation is typical of sketch- planning models. These models and the methodologies behind them are designed to require and produce only basinwide averages of VMT, delay, and emissions. • The traffic analysis zones level of aggregation is an intermediate level of aggregation typical of most trans- portation forecasting models in the United States. House- holds and commercial activity are aggregated into geo- graphic units, or zones. The real-world transportation system is represented by a subset of key facilities and coded as “links.” The models work with and produce results that reflect averages for each zone and link. An air basin is typically split into no more than 1,500 geo- graphic analysis zones that are often aggregates of cen- sus tracts (a few regions with GIS capabilities store their socioeconomic data at a smaller level of disaggregation, but travel models rarely can employ that full level of detail). • The individual or household level of aggression is the lowest level of analysis. Each household or each person within the household is evaluated separately, and the results are summed to obtain estimates of aggregate behavior. In most cases, the household-level behavior forecasts are made for only a random sample of house- holds in the region and the results are expanded by a fac- toring process to represent all households in the region. 4.2 OVERVIEW OF AVAILABLE METHODS Figure 4 presents a graphic overview of the available methodologies most relevant to this research project. They are classified according to the analytical processes they con- tain and their target level of aggregation for application. The best methodologies for accomplishing the NCHRP 25-21 objectives will tend to lie at the lower (i.e., individual) level of the chart and will have the broadest horizontal coverage. The following sections provide a brief summary of each methodology.

4.2.1 Areawide Analysis Tools At this level of analysis, the available methodologies pre- dict the changes in areawide VMT caused by transportation improvement projects or transportation control measures. These methodologies do not generally consider the implica- tions of shifting traffic between routes or between types of facilities. The standard error of the vehicular activity esti- mates produced by these models is generally greater than the predicted benefits of any individual transportation improve- ment project. TCM Tools is a sketch-planning methodology designed to estimate the change in vehicle activity and emissions resulting from any one of a couple dozen TCMs. The methodology works at the regional or areawide level of aggregation. It does not track link- or location-specific impacts. It does not deal with long-term land-use impacts. The software user must identify the percentage of travelers likely to respond to or participate in the TCM. The method- ology does not deal with the synergistic effects of multiple TCM projects. Highway Economic Requirements System (HERS) is a highway program investment tool that computes the likely areawide benefits of different program investments by com- puting the highway operation and air pollution impacts of 38 improvements made to a sample of highway links (the HPMS system) within urban areas. HERS has the same modeling capabilities (demand, supply, and emissions) as TCM Tools. HERS uses elasticities to estimate the likely magnitude of latent demand for each improvement project. SMITE and Sketch-Planning Analysis Spreadsheet Model (SPASM) are various economic and latent demand estima- tion models developed by Patrick deCorla Souza of FHWA. These models are primarily sketch-planning models designed to yield estimates of areawide changes in VMT due to trans- portation improvement projects. The models generally employ demand elasticities of approximately one-half of those used in HERS. Their capabilities are similar to those of TCM Tools. STEAM is primarily an economic benefit assessment model designed to function as a postprocessor at the end of a traditional four-step demand model process. STEAM is designed to improve on the system operations analysis meth- ods contained in conventional travel demand models. 4.2.2 Land-Use Forecasting Tools This section discusses the available land-use analysis tools that function at the zonal and disaggregate level. Land-use FI N A L R EP O RT Land Use Forecasting Travel Demand and Estimation Transport Systems Operations Analysis Emissions Estimation Air Quality Forecasting Figure 3. Land-use/transportation/air quality chain. Le v el o f A gg re ga tio n Analysis Stage Land Use Travel Demand Systems Operations Pollutant Emissions DRAM/ EMPAL Traditional 4-Step Model Mobile5/6 HLFM/QRS In di vi du al STEP CORSIM NCHRP 25-11 NCHRP 25-14 A re aw id e TCM Tools Zo ne s UrbanSim Tran sims SMITE, SPASM, HERS, STEAM HCM Figure 4. Analytical breadth and level of aggregation of available methodologies.

models translate demographic, natural resource, and infra- structure data into forecasts of land use. DRAM/EMPAL is the most widely applied set of land- use models in the United States. DRAM allocates households within the region, and EMPAL allocates employment. Both models are Lowry-type models that allocate households and jobs according to accessibility. Land availability is taken into account, but land prices are not. The models do not forecast growth; they merely spatially allocate it. DRAM/EMPAL must be manually interfaced with a travel demand model to obtain zonal accessibility data. These models generally per- form better (i.e., produce fewer irrational results requiring manual intervention) when applied to large aggregations of analysis zones (typically no more than 100 land-use alloca- tion districts for the air basin). HLFM is a simplified Lowry type allocation model that has been integrated with a travel demand model process called the Quick Response System (QRS). The combined HLFM/QRS model is one of the few if not the only model to provide full equilibration between route choice, mode split, trip distribution, and land-use location. However, it does not have the capability of predicting time of day or new gener- ated trips effects of different land-use patterns. UrbanSim is a dynamic metropolitan area land-use fore- casting model. It was developed in 1996-1998 as part of Ore- gon’s Transportation and Land-use Model Integration Project (TLUMIP). Unlike more traditional approaches to land-use modeling, UrbanSim does not seek a cross-sectional equilib- rium between the demand for and supply of land. UrbanSim models land-use changes as a dynamic process where people and businesses have certain price and accessibility demand functions but are not perfectly mobile so as to take full advan- tage of available land supply opportunities. UrbanSim shows how changes in land-use policies and transportation supply affect the movement of households and businesses on a year- by-year basis. Moving costs and other constraints delay the response of the actors to changes in land supply, price, and accessibility. 4.2.3 Travel Demand Estimation This section describes travel demand methodologies at two levels of aggregation, the traffic analysis zone level and the disaggregate household level. Conventional four-step travel demand models function at the link and zone level of aggregation. TRANPLAN, EMME/2, MINUTP, TRANSCAD, and TP+ are examples of software that implement analytical methodologies at this level of detail. The vehicle activity results are considered to be more accurate than can be obtained from sketch-planning approaches; however, the results are still averages and over- look much of the temporal and individual trip variation pres- ent in a typical urban area. 39 Conventional four-step travel demand models divide peo- ple’s complex travel behavior decision-making process into four sequential steps for the sake of computational conve- nience: trip generation, destination choice, mode choice, and route choice. A fifth step is often added: time-of-day choice, although this step frequently consists of “hard wired” per- centages that are not sensitive to changes in traffic conges- tion (see Figure 5). Disaggregate methodologies analyze travel behavior and vehicle activity at the individual traveler or household level. This level of detail allows the greatest ability to trap all pos- sible effects of transportation improvements, but often comes at the cost of excessive data requirements and computation requirements. STEP is a comprehensive disaggregate demand-forecasting tool, identified by the EPA as a promising tool for evaluating the air quality impacts of ITS projects. STEP operates at the household level directly off of household survey databases. These databases are usually collected by MPOs as part of the MPOs’ calibration every 10 years of the MPOs’ traditional four-step travel demand models. STEP contains trip fre- quency, trip destination, mode choice, and vehicle ownership models that are sensitive to the travel time and cost changes caused by traffic-flow improvement projects. The Portland Tour-Based Model is a disaggregate model, like STEP, that is applied at the household level. However, unlike STEP, which employs a variation of the traditional four-step procedure, the Portland model uses a tour-based approach to predict travel activity. A nested multinomial logit model is used to predict each person’s decisions about daily activities, whether the activities are performed inside or outside of the home, whether the person stops along the way to the destination, the time of day the person will make the trip, and the person’s choice of destination and mode of travel. This model is still undergoing development and is not yet used in Portland’s day-to-day planning process. TRANSIMS is a multiyear project of the FHWA and the University of California Los Alamos National Laboratory (LANL) to develop a travel demand model that retains the identity of individual “synthetic” travelers throughout the entire travel demand forecasting and traffic operations analy- sis process of the model. TRANSIMS incorporates tour-based FIN A L R EPO RT Trip Generation Destination Choice Mode Choice Time of Day Choice Route Choice Figure 5. Conventional travel demand model.

travel demand, intermodal trip planning, traffic microsimula- tion, and air quality analysis, all with a single unified pro- gram architecture. TRANSIMS simulates the demand patterns of individual travelers (rather than the households used in the STEP model). However, in order to microsimulate traffic operations for a large region, TRANSIMS has adopted a traffic microsimu- lation approach (called cellular automata) that is slightly more aggregate than Corridor Simulation (CORSIM) (thus, TRANSIMS is “tilted” in Figure 4). TRANSIMS currently requires an order of magnitude increase in the data collection resources and computer processing capabilities of MPOs, although this requirement will change as TRANSIMS is fur- ther refined for the commercial market. 4.2.4 Transportation System Operation Simulation Travel demand models always include a very crude trans- portation system operation component so that the interrela- tionship between demand and route choice can be modeled. This crude transportation system operation component is often a speed-flow curve, which is not a very good predictor of mean vehicle speeds, but is sufficient to produce reason- able demand estimates for each link of the transportation sys- tem. Speed-flow curves, many of which are based on a Bureau of Public Roads (BPR) equation, generally underestimate the impact of congestion on travel speeds and thus contribute to underestimates of vehicular emissions. In particular, the use of link-based estimates of speed such as the BPR equation miss the impacts of cross-street demand on signalized inter- section delays. The Highway Capacity Manual (HCM) provides one basis for improving vehicle operation forecasts. The manual con- tains a series of procedures for predicting the steady-state traffic conditions at a macroscopic level. Traffic performance in terms of mean delay, mean travel speed, and mean density are predicted for the peak 15-minute period within the peak hour. Dynamic effects such as the build-up of traffic queues over several time periods and the impact of one time period on the following time period are not explicitly considered (although a few of the procedures allow users to manually account for these effects). Modal activity (acceleration, decel- eration, idle, and cruise) is not predicted by the HCM proce- dures. QRS is one travel demand–modeling software pack- age that has implemented the HCM procedures within the demand-modeling context. CORSIM is an example of disaggregate transportation supply simulation model. It requires that overall demand be fixed within each subperiod of analysis and has no capabili- ties for revising the demands in response to simulated traffic congestion. It has the capability to show the build-up and dis- sipation of congestion over the analysis period and generates emission estimates using Mobile 5 data. Microsimulation 40 software such as CORSIM tends to be so data-intensive as to be unworkable for basinwide applications. It is generally lim- ited to segments of facilities no more than 10 miles long. 4.2.5 Mobile Source Emission and Air Quality Models A variety of air quality–modeling approaches can be used to assess the effects of transportation projects. Air quality models can be characterized according to the methods they use for the “source term” and the “dispersion term.” Spatial and temporal variability in emissions, as well as release con- ditions (temperature, stack heights, and velocities) are all potentially important elements of the source term. The dis- persion term may include both pollutant transport and dis- persion based on meteorological conditions, the effects of terrain and street canyons on dispersion, and chemical trans- formation and removal processes. MOBILE5 and MOBILE6 are emission factor models developed by the U.S. EPA. MOBILE6 was released in 2002. Emission rates are produced for different vehicle classes and age distributions for specified calendar years. MOBILE5 pro- duces a single set of speed-dependent running emission rates (in grams per mile), whereas MOBILE6 produces different speed-dependent emissions for arterials and freeways, along with non–speed-dependent rates for ramps and local road- ways. Emissions associated with “trip-ends” (i.e., excess emis- sions during starts and evaporative emissions from hot soak, diurnal, and resting loss) can be obtained from these models to assess the effects of changes in the number of trips. At the time of the NCHRP Project 25-21 research, the Hybrid Roadway Intersection Model (HYROAD)81 was undergoing final revisions under NCHRP Project 25-6. HYROAD is a disaggregate emission model that models the geographic dispersion of CO emissions in the vicinity of an intersection. The vehicle demands are given to the model, which then disaggregates the activity data by vehicle type, modal activity, and distance from the intersection. At the time of the NCHRP 25-21 research, NCHRP Proj- ect 25-11 was producing a modal emission model capable of responding to second-by-second operating conditions for light-duty vehicles. The model can be used to investigate the effects of congestion on emissions in ways not treated by the cycle-based approaches used in MOBILE. In particular, it is sensitive to power demand, including the increased likeli- hood of vehicles going into power enrichment with mild accel- eration under high-speed, low-congestion conditions. At the time of the NCHRP 25-21 research, NCHRP Proj- ect 25-14 was producing analytical tools for predicting the effects of various transportation planning policies on heavy- duty vehicle activities and the associated emissions. The first phase of this research involves the inventorying of heavy- duty vehicle usage patterns. This project was still active as of April 2003. FI N A L R EP O RT

FIN A L R EPO RT 41 CHAPTER 5 SKETCH-PLANNING APPROACHES This chapter reviews macroscopic sketch-planning approaches for predicting the impacts of highway capacity improvements on traffic volumes. Some investigators have proposed that the elasticities used in sketch-planning models could provide a quick solution to conventional travel models. The elasticity factors would be used to factor-up the forecasts produced by conventional four-step models. This chapter pre- sents some of the sketch-planning approaches that employ the use of an elasticity to account for the demand-inducing effects of new highway capacity. 5.1 HERS The FHWA has developed the Highway Economic Requirements System (HERS),82 which uses data from the Highway Performance Monitoring System (HPMS) segments to estimate the investment requirements for the urban areas of the United States. HERS was recently updated to include a procedure for estimating induced demand on a link-by-link basis and for including that effect in the computation of the consumer benefits. Douglass Lee et al.83 developed a method- ological framework that splits induced demand into short- and long-term components (see Figure 6). Lee et al. deal with some of the important measurement issues with calculating elasticity, which are often given scant treatment in other papers. For example, capacity may be a poor explanatory variable because it is actually changes in travel speed (or travel time) that drive increases in travel demand (widening a free-flowing facility presumably would do little to directly stimulate demand). Lee et al. conclude that long-run elastic- ities (over 5 years or more) are probably in the range of −1.0 to −2.0, about double the short-run elasticities, which are in the range of −0.5 to −1.0. The elasticities have the same mag- nitude as those identified by R. Noland.84 Although HERS looks at specific links, it looks at only a sample of links and thus cannot distinguish between new travel in the region and diverted (i.e., reassigned) travel between links or facilities. Its results are reliable for areawide analy- ses of broad program investment decisions, but not for eval- uating the impacts of specific improvement projects. 5.2 TRAVELER RESPONSE TO TRANSPORTATION SYSTEM CHANGES INTERIM HANDBOOK, TCRP PROJECT B-12 The Traveler Response to Transportation System Changes Interim Handbook85 provides a comprehensive, interpretive documentation of how travel demand and the usage of trans- portation facilities and services are affected by various trans- portation system changes. An interim handbook was released in March 2000 (via the web) that makes available seven topic areas that were completed under TCRP Project B-12. They include HOV facilities, vanpools/buspools, demand- responsive services, transit scheduling/frequency, bus routing changes, transit pricing/fares, and parking pricing/fares. Although TCRP Project B-12 provides a possible model to emulate in terms of report format and interpretation, the material presented has little overlap with that in NCHRP Proj- ect 25-21. TCRP Project B-12 is intended for use by trans- portation planners as an aid in the development and screen- ing of alternatives and quick preliminary assessments. The report emphasizes nonhighway modes (e.g., a transit planner might want to know, for a specific transit market, what the likely patronage impact is of changing the headways on a bus route from 30 minutes to 20 minutes). TCRP Project B-12 pro- vides excellent guidance on this topic. TCRP Project B-12 is not intended to cover general-purpose highway capacity increases, nor is it intended to provide guidance on entirely new facilities (as opposed to changes in existing facilities). The closest that TCRP Project B-12 comes to general-purpose highway capacity increases is in the HOV facilities section. Throughout most of the report, considerable use is made of elasticities; thus, baseline (“before change”) demand levels must be a known value. In some cases, before-and-after market shares and percent changes are used. The transportation system changes dealt with in this handbook are principally not single-occupancy vehicle (SOV) modes; the final version of the report will include park-and-ride facilities, busways, light rail transit, commuter rail, transit information, road value pricing, land use and site design, pedestrian/bicycle facilities, parking management, and TDM.

5.3 SPASM, SMITE, AND OTHER SKETCH-PLANNING TOOLS DeCorla-Souza and Cohen86 and Cambridge Systematics87 have developed a series of sketch-planning models that treat demand inducement using elasticities. Their elasticities tend to be about half those of HERS, the rationale being that the latter model is predicting route shifts from non-HPMS seg- ments, while the former models deal with entire urban net- works and predict only net increases in demand, not shifts in routes. SPASM88 is a spreadsheet-based, sketch-planning model appropriate for “screening analysis” of alternatives. It pro- vides first-cut estimates of public capital and operating costs, employer costs, system user costs and benefits, air quality and energy impacts, performance, and induced demand. Induced demand is based on an elasticity of demand (i.e., trips) with respect to travel time of −0.5. This elasticity was chosen because of its use in TRB Special Report 245, but this value can also be changed by the user. Another sketch-planning model is SMITE, which was developed for estimating travel increases caused by capacity increases. The model predicts the change in VMT as a func- tion of travel time change. It is particularly useful where a traditional four-step travel forecasting model is not available, or where comparing alternatives would be too resource inten- sive in the four-step environment. A paper by DeCorla- Souza and Cohen89 demonstrates the application of SMITE. In their hypothetical case study, a 50-percent increase in capacity (widening a freeway from four lanes to six lanes) induces an increase in VMT of 5–8 percent, depending on the initial level of congestion. Even with a unit elasticity (−1.0), VMT increases range between 9 and 11 percent corridorwide due to the project. Denvil Coombe et al.90 point out several problems with using simple elasticities. COBACHECK is a model similar to HERS for evaluating the economic value of transportation improvement decisions. The paper by Coombe et al. cri- 42 tiques the Metropolitan Transport Research Unit’s COBA- CHECK method for computing economic benefits of road improvements and counting as a “disbenefit” all induced traf- fic. COBACHECK assumes a long-run elasticity that is twice the value of the short-run elasticity. The article concludes that COBACHECK’s method of reducing benefits due to induced demand is inaccurate and that only a model using trip matrices (rather than aggregate demand) can accurately be used to evaluate project benefits. Austin et al.91 developed a workbook of sketch-planning techniques for estimating the emission reduction and travel activity impacts of TCMs. The method estimates direct reduc- tions in trip generation, indirect increase in trip generation, temporal shifts, trip length shifts, and speed shifts. The California Air Resources Board and Caltrans92 have developed a manual of sketch-planning methods for evaluat- ing the cost-effectiveness of new bus service, vanpools, shut- tles, signal coordination, bicycle facilities, telecommuting programs, and ridesharing/pedestrian facilities. The manual does not deal with secondary demand effects of the projects. The user must input the primary mode choice effects of the projects. TCM Tools is a sketch-planning methodology designed to estimate the change in vehicle activity and emissions result- ing from any one of a dozen transportation control measures. It works at the regional or areawide level of aggregation. It does not track link- or location-specific impacts, nor does it deal with long-term land-use impacts. The software user must identify the percentage of travelers likely to respond to or participate in the TCM. The methodology does not deal with the synergistic effects of multiple TCM projects. Crawford and Krammes93 provide more information about this and other sketch-planning tools. Commuter is a pivot point logit model methodology for estimating the emission reductions from (1) voluntary mobile source emission reduction and commuter choice incentive programs contained in the state implementation plan and (2) employer-based commuter choice programs.94 Commuter predicts the impacts of these programs on the commute mode split and the percentage of commute trips shifted to off-peak periods. However, Commuter is not designed for transportation control measures that are mas- sive enough to affect travel speeds and does not predict the impacts of the measures on trip lengths (except what occurs when shifting modes). 5.4 SKETCH-PLANNING POSTPROCESSORS Sketch-planning postprocessors are specialized analysis tools designed to work with the trip tables and transportation networks produced by a typical regional planning model. The base case trip table and network produced by the planning model are modified by the postprocessor to reflect the impacts of street improvements or ITS improvements. The post- FI N A L R EP O RT Pr ice ($ / V M T) Quantity (VMT) Short Run (Present) Short Run (Future) Long-Run Demand P1 P2 V1 V2V1s Figure 6. Short- and long-term demand effects.

processors have the advantage of working with a richer data- base than is available to typical sketch-planning tools, and they save on the expense of adapting and rerunning the tra- ditional regional models for the purposes of evaluating the impacts of specialized projects. STEAM is a corridor sketch-planning tool, sponsored by FHWA, that uses output from conventional four-step travel forecasting models to generate more accurate forecasts of speeds, volumes, and benefits for highway capacity improve- ments.95 STEAM recomputes the link speeds given the fore- casted volumes and capacities contained in the base model network. It performs a risk analysis and outputs systemwide results of net present worth and benefit/cost analyses. The ITS Deployment Analysis System (IDAS) performs a similar postprocessing function as STEAM, but is designed for analyzing the costs and benefits of ITS projects.96 IDAS contains routines for recomputing vehicle speeds and vehicle demand based upon estimated changes in travel time and costs resulting from various ITS project strategies. IDAS com- putes mode shifts, temporal shifts, route shifts, and induced 43 or foregone demand. IDAS also computes fuel consumption and air pollution impacts. 5.5 ASSESSMENT The sketch-planning approaches are appealing because of their simplicity. However, with the exception of STEAM and IDAS, sketch-planning models do not produce the level of detail required for mode of operation emission estimates without interfacing them with more detailed travel models. A further criticism is that the sketch-planning models incorporate fixed elasticities instead of behavioral models, so it is difficult to know if one is not double-counting demand effects when mixing a sketch-planning model with a con- ventional travel demand model that also accounts for some demand effects. This problem is a concern even when apply- ing STEAM and IDAS. For these reasons, sketch-planning models will not be considered further in the development of the NCHRP 25-21 methodology. FIN A L R EPO RT

44 FI N A L R EP O RT CHAPTER 6 LAND-USE MODELS There are a variety of methods for examining the impacts of the transportation system on land use, including informal methods such as simple judgment, expert judgment and Del- phi panels, formal modeling approaches such as simple regres- sion models, economic frameworks, and complex transport/ land-use models. The decision to use one method instead of another for assisting planning policy depends on a number of criteria, including the relevance of the indicators, the valid- ity of the method, the plausibility of the results, and the con- tribution of the results to the needs of the regional planning process. Still et al.,97 for example, list four criteria for any forecasting model of land-use changes in response to a trans- port investment: • The model must be intuitive and internally consistent, with clear and supportable assumptions and recognition of key sensitivities. • The model must be able to yield forecasts of house- holds, populations (including workers), and employment indicators. • The zoning disaggregation must be fine enough for plan- ning detail, but capable of aggregation up to planning units and political jurisdictions. • The method and results must be transparent to be as accessible as possible to a wide audience and to increase credibility. The complex transport and land-use models are able to satisfy the last three criteria, but do not fully satisfy the first criterion. What seems to be missing from this list is the capac- ity of the model for policy analysis over the range of alter- natives facing planners. Policy instruments link to policy objectives, so models should be capable of assessing the abil- ity of these policy instruments to achieve the policy objec- tives. The objectives stem from current transport policy issues, including congestion, energy use, safety, environmental degra- dation, accessibility for disadvantaged and challenged persons, social inequality, fiscal restraint and privatization. These issues give rise to one or more of the following policy instruments: zoning and traffic restrictions, gas taxes, transit subsidies, infrastructure investment, transit investments, transportation system and demand management options, road pricing, priva- tization, and deregulation. There will be ranges of weak and strong linkages between these policy objectives and the list of instruments just cited. The selection of a methodology to evaluate the outcomes of changes in transportation management options, invest- ments, and policies in a variety of network settings requires a balance between detail and flexibility. The principal pur- pose in developing this framework is for broad applicability, yet at the same time the need for accuracy leads one to adopt a finer grid of spatial detail. In the end, it may not be possi- ble to develop a methodology that fits all users, but the level of detail should reflect a balance of benefits and costs of portability across issues, options, and sizes of urban areas. Before undertaking a detailed assessment of the alternative models, the question arises, Are the land-use effects large enough or predictable enough to be a necessary part of the modeling effort? Hunt et al.,98 for example, state, “The exis- tence of a relationship between land use and transportation is axiomatic. The need to consider land use and transportation as important determinants of air quality is also well estab- lished, both practically and legally.” The authors go on to claim that air quality depends not only on travel activity but also on urban form and the distribution of population and employment. However, they point out that the nature of the relationship is not well understood. Pickrell,99 however, argues that the past influence of transportation on urban land is unlikely to be replicated in the future. In the past few years, innovations and investments in transportation facilities have had a diminishing effect on land-use changes. Also, he argues, the influence of land-use patterns themselves on travel behav- ior is modest. The statistical relationship between travel and land-use measures is weak and unreliable. Thus, Pickrell claims, “the lack of compelling evidence of these relation- ships means that the changes in land-use patterns likely to be fostered by metropolitan planning cannot be relied upon to alter the volume or geographic pattern of urban travel in pre- dictable ways, despite planners’ frequent assertions to the contrary.” Pickrell is concerned that land-use planning will be used as a substitute for rational investment levels and pric- ing policies. The decision to include or not include land-use effects should be based on an assessment of the role the land-use model plays in the modeling framework. The assessment should examine what benefits the model yields and what costs it imposes. Several models are being evaluated on a number

45 FIN A L R EPO RT of different dimensions, including Integrated Transportation and Land Use Package (ITLUP, also referred to as DRAM/ EMPAL), Marcial Echenique Plan (MEPLAN), New York Metropolitan Transportation Council Land Use Model (NYMTC-LUM), and UrbanSim. The evaluation focuses on the current state of practice using these models. The models are examined in terms of their treatment of time, land, space, transportation networks and services, and the economic agents, as well as in terms of their behavioral relationships, how they undertake the spatial allocation process, what poli- cies can be modeled using the models, and the performance of the models. 6.1 INTEGRATED LAND-USE AND TRANSPORTATION MODELS NCHRP Report 423A: Land-Use Impacts of Transporta- tion: A Guidebook provides a comprehensive review of the current state of the art in land-use models.100 This is the first volume of a two-volume report on the development of the UrbanSim model (the second volume will be a user’s guide for UrbanSim). The first volume reviews current land-use forecasting practice and available tools and recommends improvements. The following formal land-use models are reviewed in the NCHRP report: • DRAM/EMPAL; • MEPLAN; • TRANUS (an integrated land-use and transportation model developed by Dr. Tomas de la Barra (formerly known as “Transporte y Uso del Suelo,” or “Transporta- tion and Land Use”); • MetroSim; • The Highway Land Use Forecasting Model (HLFM II+); • The Land Use Transportation Interaction Model (LUTRIM); and • The California Urban Futures (CUF). Among these models, DRAM/EMPAL has seen widest use, particularly in the United States. TRANUS and MEPLAN have not seen application in the United States (although TRANUS is currently being tested in Sacramento, Califor- nia; in Baltimore, Maryland; and in the state of Oregon). DRAM/EMPAL consists of three components: the dis- aggregate residential allocation model, the employment allo- cation model, and a set of travel demand models (MPOs usu- ally substitute their own travel models). DRAM/EMPAL has seen widespread use in the United States. DRAM/EMPAL is based on the Lowry gravity model, which assumes that acces- sibility is the prime explanatory variable of locational choice. The user provides forecasted employment and population con- trol totals for the region, plus vacant “developable” land (the user must decide what is “developable”). DRAM/EMPAL then allocates the growth to districts within the region. NCHRP Report 423A notes that the DRAM/EMPAL mod- els “do not perform well with very disaggregate zonal sys- tems or where there is sparse activity within certain zones.” Development has ceased on DRAM/EMPAL. A new model- ing system, the Metropolitan Integrated Land Use System (METROPILUS), is currently under development. HLFM II+ is also based on a Lowry gravity model and (like DRAM/EMPAL) assumes that accessibility and land availability are the key explanatory variables of locational choice. HLFM (like all Lowry models) is a full equilibrium model. It does not predict land use for a given year but rather identifies where the region should be given the land supply and what the accessibility situation is like. A more detailed description is provided later in this chapter. MEPLAN and TRANUS are two closely related model sys- tems that have seen little practical application in the United States. Both models use discrete choice logit models to pre- dict choices of household and business location and trip mak- ing (i.e., trip generation, distribution, and mode choice). The models use the concept of markets for (1) land (floor space and housing); (2) transport; and (3) labor (and other economic factors of production). Input/output modeling is used to repre- sent interactions between economic sectors, households, and land markets. Demand and supply are balanced in each mar- ket by adjusting prices. Accessibility and prices lag demand. These models are therefore temporally “dynamic” rather than “equilibrium.” A noteworthy feature of TRANUS is its use of an expo- nential model to predict the variation of travel demand as a function of travel cost. TRANUS presumes that the number of trips between zones will decrease from a maximum (if travel cost is zero) to a minimum as travel cost goes to infin- ity. The theory is that there is always a minimum number of trips that must be made regardless of cost, and there is also a finite limit on demand regardless of how cheap it is to travel. Stated in formula form, Equation 1 Where: Tij = the number of trips between zone i and zone j (a sep- arate computation is performed for each trip purpose, trip purpose index not shown); Qij = a measure of the potential travel demand (for exam- ple, the number of jobs in zone j filled by workers liv- ing in zone i); a = the minimum number of trips per unit of Qij (for example, trips per worker) that must occur; b = the number of trips per unit of Qij that would be affected by trip cost; B = the calibration parameter (set to zero for work trips, which always must be made, regardless of cost. Non- zero for other purposes.); and cij = the generalized cost to go from zone i to zone j. T Q a b B cij ij ij= ∗ + ∗ − ∗( )[ ]exp

46 FI N A L R EP O RT The maximum number of trips between zones i and j would be Qij ∗ (a + b). The minimum number of trips between zones i and j is Qij ∗ a. MetroSim, LUTRIM, and CUF are models that were under development at the time of this report and have seen little practical application anywhere. MetroSim was being devel- oped by Alex Anas at the State University of New York at Buffalo. LUTRIM was being developed by William Mann. CUF was being developed by John Landis at the University of California. NCHRP Report 423A notes that users of all of these land- use models generally complain about the difficulty of apply- ing any of these models, specifically the high staff time costs, the extensive data requirements, the need to hire the devel- oper of the model to calibrate it, the inaccuracy of results, the lack of integration with transportation models, and the insuf- ficient documentation. The NCHRP report notes that the models based on Lowry gravity models assume that accessibility is the key explana- tory variable of locational choice. The models generally do not adequately represent nonaccessibility factors that influ- ence household and firm location choice. The report provides a conceptual description of the general framework and mechanisms of the land market and comes to the following conclusions regarding needed improvements to current land-use models: • The models must have measures of accessibility that reflect the complex decision making of households and firms (i.e., consider access to services, recreational oppor- tunities, school, etc., as well as traditional work location). • The models must recognize that affordability and other factors may be just as or more important than accessibility. • The models must recognize the limitation of public poli- cies for shaping the pattern of development. • The model must recognize that in-fill and redevelop- ment can accommodate a significant share of growth (i.e., the model must not overestimate the demand for vacant land). Both the NCHRP report and Mark Harvey101 note that one problem with all of these models is obtaining longitudinal data (i.e., panel data gathered over a long period combining time-series and cross-sectional data) for calibration. Most models have been calibrated using cross-sectional data rather than longitudinal data. This implies that they might be valid for shorter periods consistent with the cross-sectional data used to calibrate them. The NCHRP report also notes that the models tend to per- form poorly at the disaggregate level. Even the TRANUS/ MEPLAN disaggregate models perform poorly if the zone system is too detailed. Other noteworthy reviews of land-use models are those by Rosenbaum and Koenig for the EPA, Southworth for the Oak Ridge National Laboratory, Berechman and Small of the University of California, and Oryani and Harris for the Delaware Valley Regional Planning Commission. Rosenbaum and Koenig102 assess the ability of currently available land-use models and integrated land-use and trans- portation models (DRAM/EMPAL, MEPLAN, and TRANUS) to evaluate the impact of land-use policies and strategies designed to reduce travel demand. The authors looked at the ability of the models to predict the impacts of zoning and land regulation incentives on travel patterns. They noted problems with the minimum size (i.e., aggregation) of zones required for these models. DRAM/EMPAL does not reflect land-use zoning impacts as well as MEPLAN and TRANUS do. DRAM/EMPAL is also insensitive to monetary incen- tives to encourage mixed use or higher densities. Rosenbaum and Koenig recommend the following improvements to standard land-use and transportation mod- eling tools so as to facilitate their use in evaluating the impact of land-use strategies and policies: • Development of data and procedures to allow land-use analysis at fine spatial resolutions, such as census tracts; • Development of data to determine the relationship between special land-use features of interest (e.g., pedestrian-friendly environments and mixed land-use development) and neighborhood attractiveness; • Development of data and procedures to allow incorpo- ration of pedestrian and bicycle modes, as well as pub- lic transit, into travel demand models; • Development of data to determine the relationship between mixed use development and travel mode selection; • Development of data and procedures to allow incorpo- ration of trip chaining into travel demand models; and • Development of data and procedures to allow incorpo- ration of temporal choice into travel demand models. Southworth103 reviews the state of the art in operational urban land-use and transportation simulation models. Mod- els reviewed include DRAM/EMPAL, ITLUP, Projective Optimization Land-Use Information System (POLIS), MEPLAN, Kim’s Chicago model, MASTER model, and the Dortmund model. Southworth provides mathematical descrip- tions of the models and identifies several practical issues related to the applications of these models (their analytic complexity, their significant data requirements, and their sig- nificant demands on computational resources). Berechman and Small104 note that modelers must choose between tractability and suitability. Most tractable models exclude agglomeration economies and lack a dynamic struc- ture suitable for handling rapid disequilibrium growth. Oryani and Harris105 reviewed three candidate land-use models for the DVRPC: DRAM/EMPAL, MEPLAN, and MetroSim. The authors interviewed other MPOs’ experience with using DRAM/EMPAL and evaluated two case studies (Orlando and Tampa) of the application of DRAM/EMPAL.

They recommended that DVRPC implement the DRAM/ EMPAL model and provided cost and data collection needs for the implementation. Although POLIS is part of a three-tiered modeling sys- tem, it is the only tier that is directly sensitive to accessi- bility and congestion. The MTC/ABAG POLIS model uses a sophisticated mathematical programming process that allocates land uses to zones based on cost minimization (i.e., microeconomic theory). The model includes a travel time matrix provided from the regional travel model. The objective function of the model is to develop a “solution” of job, household, and labor distribution that maximizes “locational surplus” associated with a specific location, subject to the policy and economic constraints associated with each time period. Population, new housing units, employment (five sectors), number of work trips (by mode and zone pair), and shopping trips (by mode and zone pair) are distributed to 107 zones in the MTC region. The model was calibrated using Census Bureau household and busi- ness data between 1964 and 1980. The following sections provide more detailed technical information on two models: HLFM and UrbanSim. HLFM is an example of an equilibrium Lowry land-use model similar to the widely applied DRAM/EMPAL model, but somewhat simplified. HLFM is notable for its tight integration with travel model software. UrbanSim is an example of an advanced prac- tice dynamic microscopic land-use model currently being tested in Oregon and Washington that bears a resemblance to the MEPLAN and TRANUS models in their use of dis- crete choice models and economic markets. UrbanSim is designed to interface with external travel demand software. These two models bracket the range of available land-use model applications. 6.2 THE HIGHWAY LAND-USE FORECASTING MODEL The Highway Land-Use Forecasting Model (HLFM) was developed by Dr. Alan Horowitz. It is based on the Lowry- Garin land-use model first described in 1964–66. HLFM uses information on the highway system, land uses (existing and proposed), demographic data, and socioeconomic data to pre- dict the amount of employment and population likely to locate in each zone within an urban area. HLFM extends the Lowry- Garin model by taking into account the availability of land by activity type. HLFM is a very long-term “equilibrium” land-use model. The model predicts the land-use demand and supply equi- librium toward which the urban area is heading. HLFM is intended to give a good indication of the global trends in urban development, not detailed land-use information at the zonal level. HLFM will not predict land use for any given year or for the base year (if the urban area is currently sub- stantially out of equilibrium). 47 The data requirements of HLFM are substantially less than those of DRAM/EMPAL. Since HLFM integrates with QRS, it is sensitive to the impacts of both highway and transit on land use, as well as the effects of traffic controls. HLFM is designed to be integrated with a travel demand/ supply model. The travel times and costs are computed for all pairs of districts (i.e., aggregations of traffic analysis zones) in the urban area. The Lowry-Garin model is then used to allocate population and employment to the districts. The population and employment allocations are fed into a travel demand model that predicts the resultant travel demand, traffic volumes, and travel times. The new travel times are fed back into the Lowry-Garin model to compute new popu- lation and employment allocations. The process is repeated until the travel patterns are “unchanged.” 6.2.1 Allocation Process To start, the Lowry-Garin model requires the location and amount of basic industry employment in the region. Basic industries are industries that choose their location primarily according to their proximity to needed natural resources and urban infrastructure. Once the model has this initial infor- mation, the allocation process is then begun. First, the conditional probabilities are computed for worker resident locations and for service employment locations. The model then computes the total employment in each district and the population residing in each district. The model revises the relative attractiveness of each district based upon the results of the current iteration. This process is iterated in the HLFM model until the user-specified maximum number of iterations has been reached. 6.2.2 Conditional Probabilities Equation Three conditional probability matrices are computed: the probability of a person working in district j residing in dis- trict i (matrix A); the probability that an employee working in district j is served by another employee working in district i (matrix H); and the probability that an individual living in district j is served by an employee working in district i (matrix B). The conditional probabilities are computed using singly constrained trip distribution equations with an exponential deterrence function (these equations are mathematically iden- tical to logit models). Equation 2 Where: aij = the conditional probability that an individual working in district j will live in district i; a w t w tij i ij i ij i = ∗ − ∗( ) ∗ − ∗( )∑ exp exp β β FIN A L R EPO RT

48 The F, G, and Q matrices contain the information typically associated with base multipliers for a region. 6.2.4 Adjustments to Attractiveness The attractiveness weights (w) that will be used in the next model iteration to compute the conditional probability matri- ces are based upon the remaining net developable area result- ing from the current model iteration. 6.3 THE URBANSIM MODEL The UrbanSim model is a dynamic metropolitan area land- use forecasting model (see Figure 7). It was developed in 1996-1998 as part of Oregon’s Transportation and Land-Use Model Integration Project (TLUMIP).106 Unlike more traditional approaches to land-use modeling, UrbanSim does not seek a cross-sectional equilibrium between the demand for, and supply of, land. UrbanSim models land- use changes as a dynamic process where people and busi- nesses have certain price and accessibility demand functions but are not perfectly mobile so as to take full advantage of available land supply opportunities. UrbanSim shows how changes in land-use policies and transportation supply affect the movement of households and businesses on a year-by- year basis. Moving costs and other constraints delay the response of the actors to changes in land supply, price, and accessibility. UrbanSim does not have a transportation model integrated within it to compute accessibility. UrbanSim obtains acces- sibility statistics from a separate transportation model, thus allowing users to interface UrbanSim with most any metro- politan transportation planning model and software. A more integrated version of UrbanSim is under development for Honolulu. FI N A L R EP O RT wi = the attractiveness of district i; tij = the travel cost, time, or disutility of travel between districts i and j (from travel model); and β = a calibration parameter. The attractiveness (w) of a district is specified in terms of the net developable area for the computation of residential location probabilities (matrix A). Net developable area for service industries is used for computing the other two loca- tion probability matrices (B and H). 6.2.3 Employment Location Equation The key equation for finding employment in each district is E = EB + ER + EW Equation 3 Where: E = the vector of total employment in each district; EB = the vector containing the amount of basic employ- ment in each district; ER = the vector of employment that service residences (people); and EW = the vector of employment that serves workers (or other businesses). Since ER and EW are functions of total employment, it is necessary to solve this vector equation for total employment E to eliminate EB from the right side. The following equation is used to simultaneously compute the total number of jobs in every individual district as a function of basic employment in all districts individually: E = (I − GBQA − HF)−1 EB Equation 4 Where: I = the identity matrix; a diagonal matrix of 1’s, all other values of the matrix being 0; A = the matrix of aij; the conditional probability that an individual working in district j will live in district i; B = the matrix of bij; the conditional probability that an individual who lives in district j will obtain services from a job located in district i; H = the matrix of hij; the conditional probability that an employee working in district j is served by another employee located in district i; F = a diagonal matrix containing the ratio of service employment to all employment for each employment district, usually set uniformly to the regional average; G = a diagonal matrix containing the ratio of total employ- ment to population for each residential district, usu- ally set uniformly to the regional average; and Q = a diagonal matrix containing the ratio of population to employment in each residential district. Regional Forecast Demographic Transition Economic Transition Base Year Land Use Household Relocation Business Relocation Market Clearing Future Land Use Transportation Model Accessibility Figure 7. UrbanSim flow chart.

49 FIN A L R EPO RT The basic philosophy behind UrbanSim is that urban devel- opment over time and space is the outcome of the choices and actions of four sets of actors: households, businesses, devel- opers, and government. The different actors make their loca- tion decisions within different time frames. Household and business location decisions are assumed to occur within 1 year of a change in conditions. Building construction decisions by developers take longer, and infrastructure decisions by gov- ernment take the longest. UrbanSim requires the following input data: • Base year land use, • Population, • Employment, • Regional economic forecasts, • Transportation system plans, • Land-use plans, and • Land development policies (such as density constraints, environmental constraints, and development fees). The “Demographic Transition Module” and “Economic Transition Module” predict temporal changes in the distribu- tion of household and business types (e.g., age, income, and industry type) for the region. These predicted changes are based upon the user input regional population and employ- ment forecasts. The model then predicts the location of businesses, house- holds, new construction, and the price of land/buildings. The Household Mobility and Location Module simulates house- hold relocation decisions (stay or move; if move, then to where and what housing type). The Business Mobility and Location Module simulates business relocation decisions (stay, move, building type, and location). The characteristics of the house- hold and the businesses influence the choices taken. The land development component simulates developer choices to convert vacant or developed land to urban uses (including type of improvement and density). The model takes into account profit expectations and governmental con- straints (e.g., zoning and infrastructure). The market clearing is simulated by adjusting the land prices in response to competing demands. The ratio of demand to supply is used to proportionally adjust the land price. The new land prices affect the demand for the following year. The model produces information on land uses, prices, den- sity, and the distribution of population and employment that can be input to a transportation model for any desired fore- cast year. 6.3.1 Household and Economic Transition Models For businesses, the economic transition model is Equation 5B B R Rnst nst nsi nso s = −( ) − ∀ ∑ 1 Where: Bnst = the regional total number of businesses of industry n, size s, at time t or t − 1; Rnsi = the rate of business formation in the region and immi- gration to region for type n, size s businesses; and Rnso = the rate of business closure in the region and exo- dus from region for type n, size s businesses. The rate of business formation is computed as follows: Equation 6 Where: R = the rate of business formation, B = the number of businesses in the base year, and E = the number of business events (formations in this case) forecasted to occur over the n-year time period. The rate of business closure is computed similarly. The rates are adjusted to achieve specific target values for busi- nesses for target years. For households, the demographic transition model is Equation 7 Where: Hht = the regional total number of households of type h at time t (t is in units of 1 year), Hht − 1 = the regional total number of households of type h at time t − 1, Rhi = the rate of household formation in the region and immigration to region for type h households, and Rho = the rate of household death or dissolution in the region and exodus from the region for type h households. The household rates are computed similarly as for busi- nesses. 6.3.2 Household and Business Relocation Models The following models first determine how many households or businesses in each zone are likely to move that year and then determine which buildings and zones they will move to. 6.3.2.1 Mobility Submodel The following logit choice model is used to predict the prob- ability that a household or business will move in a given year: H H R Rht ht hi ho h = −( ) − ∀ ∑ 1 R B E B n = +[ ] −1 1/

50 To calculate the logsum of the conditional choice of loca- tion l, use the following equation: Equation 12 Where: Vl = the utility of choosing the location l. The conditional probability of choosing a particular loca- tion (or zone) is given by the following equation. If the avail- able data do not support a nested logit form (e.g., µ = 0), then a multinomial logit specification is used that simultaneously combines the location and building type choice. Equation 13 Where: Equation 14 And where: Vl = the utility of choosing the zone l, µ = the calibration scaling factor for utility, Chl = the bid function for consumer h for lots in zone l, pl = the price of lots within the zone l, and Sl = the size of the choice of lots within the zone l. The bid function is a linear function of density, number of available housing units, income, travel time to the central business district, general accessibility to employment and retail opportunities, and other factors. Similar factors with some variations appropriate to businesses are used to predict business location choice. Accessibility, Accessi, is measured with the following accessibility index: Equation 15 Where: Aj = the amount of activity (e.g., jobs) in zone j, Lij = the travel time, cost, or composite utility for travel between zone i and zone j, and β = the utility scaling parameter. There is some concern about whether the use of a scaling parameter for utility violates the theoretical basis for logit discrete choice models. 6.3.3 Market Clearing Model The market clearing model sets the land price. Each busi- ness type has its own bid function which is a function of Access A Li j ij j J = ( )∑ exp β V C p Sl hl l l= − +( )µ ln P l b V V l l l ( ) = ( )( ) ′ ′ ∑ exp exp ′ = ( )∑V Vb l l ln exp FI N A L R EP O RT Equation 8 Where: p(m) = the probability of a business or household moving in a given year and Vm = the utility of moving. For businesses the utility of moving is Vm = β1 + β2i + β3 s Equation 9 Where: βn = the calibration parameter vectors, i = the industry type, and s = the size of business. For households, the utility of moving is Vm = β1 + β2a + β3c + β4i+ β5 s Equation 10 Where: βn = the vectors of calibration parameters, a = the age group of the head of the household, c = the dummy variable (0,1) for presence of children in the household, i = the household income level, and s = the size of household. 6.3.2.2 Location/Building Choice Submodel The number of moving businesses is added to the net increase in businesses in the region to obtain the total num- ber looking for a new location. A nested logit model is used to predict the probability of a particular building type and location (i.e., zone) being selected by the businesses looking for a new location. The zone choice is the lowest level of the choice model. The zone selection is conditional upon the building type selection. Buildings are grouped into four types: industrial, wholesale, retail, and office. To compute the mar- ginal probability of choosing building type b, use the fol- lowing equation: Equation 11 Where: P(b) = the marginal probability of choosing building type b, Vb = the utility of building type b, Vb′ = the logsum of the conditional choice of location l, and µ = the logsum calibration coefficient. P b V V V V b b b b b ( ) exp exp = + ′( ) + ′( )′ ′ ′ ∑ µ µ p m Vm ( ) exp= + ( ) 1 1

zonal characteristics (lot density, employment density for that industry, population, income, presence of freeway access, etc.) and building types (available building stock, age of build- ings, etc.). Household bid functions are stratified by income and the presence of children in the household. Variables included in the household bid function are accessibility, net building density, age of housing, and housing type (single family, quadplex, or multifamily). Land prices are adjusted in response to differences between the current vacancy rate and the normal vacancy rate for each building type: Equation 16 Where: Plbt = land price at location l for building type b and year t, Plbt–1 = land price at location l for building type b and year t − 1, αb = the normal vacancy rate for building type b, Vlbt = the current vacancy rate (for location l, building type b, at year t), Vbt = the current mean vacancy rate for the region (for building type b at year t), λ = the weighting parameter, and β = the scaling parameter. A Land Development/Redevelopment Module interfaces with the market clearing model to determine how the supply of buildings responds to the changes in land prices. Projects are assumed to be constructed until the user-specified “nor- mal” vacancy rate for each building type is reached. Devel- opers are assumed to construct first the projects that yield the greatest profit. The expected profit of a project is computed by subtracting land cost, hard construction cost, and soft con- struction cost (impact fees, permit costs, demolition costs, service extension costs, etc.) from the expected sales price of the building project. The new supply of building space is assumed to become available in the following year. 6.4 THE IDEAL MODEL The ideal model would include markets for land develop- ment, residential housing, commercial floor space, and labor in which the demand and supply functions are well defined and equilibrium is established through price signals. The explicit modeling captures evolving behavior over time. The demographics of the model should be endogenous to ensure that the population characteristics are representative over time and are at a level of detail to complement the behavioral relationships in the housing and transportation models. The impact of regional economies should be endogenously mod- eled so urban consumption and production both influence P P V Vlbt lbt b lbt b bt= + −( ) + + −( ) +    −1 1 1 1 α λ λ β α 51 and are influenced by land and transport market outcomes. The travel demand component should be activity based to provide a level of disaggregation to ensure that policy instru- ments can be adequately molded. Finally, the auto ownership decision should be explicitly modeled and linked to the travel demand component. Such an ideal model is represented in Figure 8, where the core behavioral components are shown in the shaded area. A key idea here is that location choice and land development are distinguished, as is the supply side of the land market. The four major drivers of the urban system are demo- graphics, the regional economic makeup and level of activity, government policy (including zoning, taxation, regulation, and macro variables such as interest rates), and the transport system (which proxies the supply side of the network). Each component of such an ideal model involves a com- plex series of submodels. Even if the market-based demand and supply relationships are at a more aggregate level, the key issue is to correctly model the interactions of the eco- nomic agents. Failure to properly model the demand-supply interactions may mean that the dynamic evolution of the urban system will not be properly captured. Certainly, in all the models reviewed below and others not included (e.g. Australian and Japanese models), a key setoff assumption is the strong separability between transportation demand and the demand for other parts of the consumption bundle. In an ideal model, the full range of consumption activities would be modeled. Two broad categories of transport and land-use models have been developed. The first category is simulation mod- els, which attempt to replicate the land-use patterns by sim- ulating the process of urban development and transportation investment that produces the land-use patterns. Figure 9 illustrates a typical structure of a simulation model.107 The second category of model is an optimization model, which represents an equilibrium between an urban area’s transporta- tion system and the land market. These models are used to FIN A L R EPO RT Land Use Location Choice Auto Ownership Activity/Travel & Goods Movement Demographics Regional Economy Government Flows/trips External Impacts Land Supply Transport System Figure 8. Idealized transport land-use model.

establish the optimal distribution of land uses in an urban area or flows over the transportation network. Interestingly, simulation models have been used much more in practice to investigate the interaction between transportation changes and land use and to support plans of urban areas while opti- mization models have been used for research purposes. The Lowry-Mills type represents land-use simulation mod- els. In the group reviewed here, ITLUP represents the Lowry- Mills type. Optimization models generally use some mathe- matical programming or assignment techniques to establish the equilibrium of different land uses within the urban area. 6.5 MODEL REVIEW The review in this section draws heavily from other work.108 The models considered here fall into two categories. ITLUP is an operational package with a relatively long history of application in the United States and elsewhere. MEPLAN, NYMTC-LUM and UrbanSim, however, have a shorter his- 52 tory of application, but they have a current set of operational applications and detailed representations of land markets in which price provides the equilibrating mechanism. 6.5.1 ITLUP ITLUP was developed at the University of Pennsylvania, has a 25-year history, and is the most widely used spatial allocation framework used in the United States. It includes a number of submodels, including DRAM (Disaggregate Res- idential Allocation Model) and EMPAL (Employment Allo- cation Model). It is a Lowry-type model with four population income levels, four types of employment, and travel patterns for public and private transport. A multinomial logit model is used to determine mode split while trip generation and distribution are developed within DRAM. Household location is established concurrently with trip generation and distribution. A considerable amount of detail can be added to this model, since DRAM and EMPAL FI N A L R EP O RT Initial Conditions Employment in zone i in period t Eit Population characteristics in zone i, period t N it Full costs of travel between zones i and j F ijt Employment change in zone i in period t + 1 E it+1 Employm ent in zone 1 in period t + 1 E it+1 Population in zone 1 in period t + 1 N it+1 Interzonal travel flows in period t + 1 T ijt+1 Transport system capacity in period t K t Transport system capacity in period t + 1 K t+1 Capacity changes in period t ∆K Employment in zone 1 in period t + 1 Eit+1 Population in zone 1 in period t + 1 N it+1 Full costs of travel between zones i and j in period t + 1 F ijt+1 Figure 9. Land-use simulation model.

can be used separately in conjunction with other travel demand forecasting models such as Tranplan; Urban Transportation Planning System (UTPS); and Equilibre Multimodal, Multi- modal Equilibrium (EMME/2). A significant benefit of this model is that the data required are not large in comparison with other models and the data are generally readily and easily available. The data relate specifically to population, household, and employment data. A major weakness of this model is that it does not account for the land market clearing process.109 The program originated in Formula Translation (FOR- TRAN); however, it has been maintained, running on a per- sonal computer (the personal computer version is called Metropolis) under an Arcview shell, providing linkages to Arcview GIS.110 6.5.2 MEPLAN MEPLAN is contained in proprietary software developed in the United Kingdom. It has developed over a 25-year span, and the principal has significant expertise in the area. It has been applied in a number of regions of the world, includ- ing the United States (Sacramento and Oregon) and Canada (Edmonton). MEPLAN is highly flexible and has an aggregate perspec- tive. Space is divided into zones with quantities of house- holds and economic activities allocated to the zones. Flows of interactions among the factors in the different zones give rise to travel demand. The distinguishing feature of this model is the use of an input-output matrix, which is spatially disaggregated. This matrix provides a means of overcoming (to some degree) the assumption of strong separability between transportation and other components of the consumption bundle contained in other models. This social accounting matrix includes variable technical coefficients, labor sectors, and space sectors. All economic activities, including house- hold activities, are treated as producing and consuming activ- ities. The spatial disaggregation is achieved by having further production arise to satisfy consumption allocated among the zones according to a discreet choice model, which responds to price levels for such production. Travel demand gives rise as a result of this interaction. Temporal change is modeled by considering sequential points in time. Space in each zone is fixed at each point in time. Space, both land and commercial floor space, cannot be moved between zones. Space must therefore be consumed in the zone in which it is produced. In order to reach equilibrium, the tech- nical coefficients for the consumption of space (i.e., demand for housing) are elastic with respect to price. Prices equate demand with supply and are endogenously determined at each point in time. Prices for outputs in other sectors are also determined endogenously by the consumption-production relationship reflected in the social matrix. Travel demand, which arises from the interaction in the land, goods, and housing market, is determined for each point in time. This 53 travel demand is allocated to modes and the network on the bases of nested logit models for mode and route choice. This allocation accounts for congestion in full costs. Any disutil- ity in the transport sector feeds back into the next time period as response lags. Exogenous demand provides the initial impetus for eco- nomic activity and changes in the study area demands. It also provides the amount of space in each zone from one time period to the next, thus driving economic change. These changes are allocated to the different zones under equilibrium conditions. MEPLAN is personal computer based, but it is proprietary. 6.5.3 NYMTC-LUM NYMTC-LUM reflects the work of Anas111 over the past two decades. It is a simplified version of Metropolis, which is the PC version of ITLUP. Like Anas’s earlier work, NYMTC- LUM is anchored in microeconomic theory where demand and supply interactions determine the equilibrium price. The model simultaneously shows the interactions between resi- dential housing, commercial floor space, labor, and nonwork travel markets with explicit representations of demand and supply in each market. Housing prices, floor rental rates, and wages are all endogenously determined in the model and serve as the arbiter between demand and supply in their respective markets. Finding prices and wages that balance demand and supply in the markets of interest generates static equilibrium in the forecast year. This final equilibrium is not path depen- dent and therefore does not require solutions in a sequence of years. The model uses traffic zones as the geo-statistical unit of observation and therefore provides a fine level of detail for policy analysis. For example, the application in New York has 3,500 zones. However, at present the model does not have significant disaggregation in households, employment, and buildings. This fact could be changed with added data and some level of complexity. The land-use component is not integrated with the travel demand model. Rather, it is connected to the existing MTC travel demand model in that it receives model utilities as inputs from the mode choice model. This connection is sim- ilar to the case of UrbanSim and ITLUP models. NYMTC-LUM’s adaptability is illustrated in the applica- tion in New York, where the primary interest is in evaluating transit programs, investments, and strategies. Features of the model, which facilitates its application in the New York con- text, include small traffic zones, integration of the detailed transit network, and use of the mode choice models. The microeconomic foundations provide a range of economic evaluation measures, including property values and consumer and producer surplus. NYMTC-LUM works on a PC platform and is written in FORTRAN. It is commercially available through Anas. FIN A L R EPO RT

6.5.4 UrbanSim UrbanSim is an operational model that was originally developed for Hawaii, Oregon, and Utah, but is currently being further developed at the University of Washington. It is an operational model of urban land and floor space and is integrated with a traditional four-step model. The unique fea- ture of the model is that it has been placed in the public domain and is accessible via the Internet site for both soft- ware and documentation. The most notable features of UrbanSim are the level of detail for both spatial disaggregation and disaggregation across households, employment, and land use. The model is grounded in microeconomic theory and emphasizes theoret- ical consistency and rigor. The model operates as a disequi- librium model in which stock supply and demand are built incrementally over time. Demand for building stock (com- mercial floor space in other models) is based on willingness to pay (WTP) or bids (observed prices, since WTP is diffi- cult to observe). Buyers are utility maximizers who attempt to maximize their consumer surplus (i.e., the residual between WTP and price paid), whereas sellers maximize price paid per unit. Suppliers of building stocks are profit maximizers given observed demand. Markets are assumed to be compet- itive. Building stock prices are determined within a market clearing process, which occurs at the submarket level of the traffic analysis zone and property type. UrbanSim works as a path-dependent model, requiring a solution in each year, and operates in dynamic disequilib- rium in each year. The profit-maximizing supplier provides parcels for development based on expected profit. This profit uses expected revenue from the previous year prices, and new construction choices are not supplied to the market (for occupancy) for the subsequent year. Demand is based on lagged prices, and current supply and prices adjust to the bal- ance of demand and supply in each submarket of each year. The demand side of the market uses traffic analysis zones as the spatial unit of analysis. The level is highly disaggre- gated in all current applications, offering a fine level of detail. On the supply side, the model uses individual land parcels as the unit of land development. This feature is unique to this model. All other models treat a more aggregate level for the land supply function. The level of disaggregation carries over to the economic agents. For example, in some applications it has used 111 household types. The model is based on policy scenarios of varying levels of detail that include comprehensive land-use plans and growth management regulations, such as mixed densities, green areas, and environmental restrictions. The model has also been developed to assess pricing policies and instruments associ- ated with a range of infrastructure and transportation policies. 6.6 ASSESSMENT Tables 7 though 17 provide a comprehensive examination and comparison, across a number of dimensions, among the 54 four models. The models differ widely in some respects and are quite similar in others. The extent of their operating expe- rience varies dramatically. This variation in some ways reflects the evolution of the models. The models are all PC based and are an outcome of a consultancy and are therefore proprietary, except UrbanSim. Three of the four are static equilibrium models that leap directly to the year-end equilibrium or are moved to it in 5-year steps. The exception, again, is Urban- Sim, which operates on a 1-year time step and does not assume equilibrium. All the models are zone based, and the older models have a coarse zone; NYMTC-LUM uses traffic zones and is fairly detailed. UrbanSim has two levels of detail: the traffic zone for the demand side and the parcel for the land supply side. The transportation system in all models is some form of a multimodal network model. MEPLAN has integrated net- work capacity. NYMTC-LUM and UrbanSim are connected to stand-alone, four-step modeling systems, while ITLUP can operate either in conjunction with the four-step model- ing systems or on its own. In all cases, the information passed from the transportation model to the land-use model are based on random utility models. In nested models, the log- sum is used to transfer utility. In MEPLAN, the composite utilities from the mode choice model are used as inputs into the land-use model. The land-use model uses these utilities in simulating spatial economic flows in determining the trip origin-destination (OD) table. The remaining models use the composite utility derived from destination choice models in the land-use model and simulate the OD tables within the transportation model. These differences affect the capability of the models to analyze transportation policy. The integration of transportation and land use does not affect this capability as much as the quality of the four-step transportation model. The range of transportation policy or management options depends primarily on the travel demand model. All models can handle transit issues, albeit at a coarse level. Goods movement is handled only by MEPLAN. Other issues such as HOV, carpooling, and other ITS applications are not handled in the models. The interaction and integration of the various economic agents (persons, households, firms, and public authorities) differs slightly among the models. The models are all house- hold based, and developers and carriers are explicitly iden- tified because of the role they play in the interaction between transportation and land use. Interestingly, none of the mod- els treat the “person” except the individual trip generated in the transportation model. This feature also shows up on the production side, where, for example, the models (except UrbanSim) deal directly with the location of employment but not with firms. All models except ITLUP have an explicit representation of the building supply development process. All the models have a strong basis in economic theory; however, the ways in which the eight potential markets are handled differ among the models. The housing markets in all FI N A L R EP O RT

55 distributed simultaneously with the spatial allocation of households. In all other models, the conventional four-step model travel demand model is used. Auto ownership and transportation infrastructure markets are treated exoge- nously. The demand for infrastructure is implicit in the travel demand model. The objective functions for the different markets are con- ventional utility functions or profit maximization functions. The travel demand models and housing markets have logit or nested logit models. Supply functions are generated by profit-maximizing pursuits. The equilibrium is determined FIN A L R EPO RT models (except ITLUP) include demand and supply. Also, the models (except ITLUP) treat the commercial floor space mar- ket with well-defined demand and supply functions. The labor markets differ across the models. In UrbanSim, the travel demand work trip distribution model determines worker-job linkages, while in MEPLAN the labor market is explicitly modeled. In ITLUP, little distinction is made between the housing and labor market. The personal transportation travel demand predicted by MEPLAN arises out of the spatial consumption-production process. In ITLUP, three trip purposes are generated and Model Time Land Developed Space ITLUP Equilibrium established at each step, 5 years generally. There are information lags from the previous step. Transport costs at time t are the basis of employment allocation at time t. Household allocation of time is based on transport costs and employment costs. Zone based with generally larger units. Developable land is exogenous. No micro scale is represented. No explicit representation of buildings or floor space. MEPLAN Equilibrium established at each time step. Information lags exist for previous time steps. Transport disutilities at time t are the basis of time allocation in t + ∆t where ∆ t is set exogenously, generally 5 years. Zone based, typically large zones. Technical coefficients for production and consumption become unrepresentative for smaller zones. Land categorization is needed in the model. Developable land is exogenous. Model is flexible enough to include the causal chain running from land→buildings→activity→representation of floor space category. Includes floor space, prices, and density. NYMTC- LUM Direct step to equilibrium for horizon year but can be used with multiple time steps with equilibrium calculated at each time step. Small zone based. Land area categorized by housing type, basic industry (exogenous), nonbasic industry (endogenous) and vacant land. Housing by category – number of units, floor space, by zone, basic and nonbasic floor space by zone. Number of categories limited by data only. Prices are explicitly calculated. UrbanSim Dynamic disequilibrium, 1- year time step with lagged responses to price signals. Demand side uses traffic zone and property types. Supply side uses parcel for land development or redevelopment. Developable land is exogenously specified. Land-use plans, regulations and environmental Explicit representation by housing type, nonresidential floor space by type, density, price, and age of development. constraints are integrated at the parcel level. TABLE 7 Time, land, and developed space

56 FI N A L R EP O RT TABLE 8 Transportation network and services Model Transportation Network Transit Representation Goods Movement Other Transport Services ITLUP Uses road network for assignment of travel costs. Accessibility is specified exogenously by DRAM/EMPAL (it can be endogenous or via link to exogenous travel demand forecasting model). Since it is a composite model, iterations between transportation and land-use models require data transfer between independent sub- structure models. Can link to exogenous demand- forecasting models. Population/employment distribution independent of work trip distribution developed in travel demand model. Inherent in general accessibility term. Depends on travel demand model used. Commonly auto- only access is used. Not present. Inherent in general accessibility term. Not explicitly considered in the endogenous travel demand model. MEPLAN Multimodal networks used. Integrated interactions between land use, modal split/assignment. Assignment is static not dynamic. Course network an issue. Can interface with external travel demand models but different zone systems an issue. Nested logit used for mode- split. Explicit transit representation exists with submodels for rail and bus. Includes transit capacity representation. Links can carry different modes. Course network restricts level of detail for transit. Explicit goods movement by all relevant modes in considerable detail. Terminal costs explicit, shipping costs included. Taxi included for some applications. Flexible enough to consider other modes and allocate costs to users. NYMTC- LUM Accessibilities imported from separate four-step travel demand model. Depends on travel demand used. Transit effects enter via mode choice model log sum terms. Small zone system gives good transit system Not present. Depends upon modal split model used. Incorporated in accessibility term. sensitivity. UrbanSim Connected with travel demand models – generally activity based. Uses composite utility to develop access measures to activities as part of business and household location models. Workplace choice is predicted within the travel modes. Depends on travel demand model used. Transit effects enter via mode choice log sum term. Small zone system provides good transit system sensitivity. Implicit in use of auto accessibility terms as proxy for congestion costs on shipping and affecting employment location decisions. Does not model flows of goods. Depends on modal split model used. Incorporated in accessibility term.

57 FIN A L R EPO RT TABLE 9 Economic agents Model Persons Households General Establishments Developers Carriers Public Authorities ITLUP Not explicit. Total population is exogenous. Household based. Generally four income bands. Aggregate number of households per zone. Aggregate number of jobs/zone. Typical categorization by four basic industry groups but Standard Industrial Classification (SIC) is possible. Firms not explicitly represented. Not explicit. NA Exogenous policy inputs (transportation system, developable land). No taxation effects considered. No endogenous public-sector responses. MEPLAN Person-trips generated by households. No explicit representation of person or their attributes. Household based. User-specific categorization – income, occupation of household head. Aggregate number of households/type/ zone. Explicit outputs of production processes. Represented by various proxies such as employment. Space is developed or redeveloped as function of prices and availability. Developers are implicitly represented by total space by type to be developed/ redeveloped. Exogenous input. Implicit in multimodal framework representation. Cost structure is explicit. Exogenous policy inputs re: serviced land, zoning, transport network, land tax. No endogenous public-sector response except some transit frequencies change with demand. NYMTC-LUM Not explicit. Can adjust equations for multiworker households. Household based but no categorization by household type. Average No explicit representation of firms. Employment is explicit by zone. Supply of building stock in each zone responds to market values for buildings of type by zone subject to available NA Transport network or services are exogenous inputs. No sensitivity to zoning or other land-use controls. household income in each zone determined by allocation of workers from workplace to residential zone. Lowry concept of basic and nonbasic industry is used. This division is used to keep track of land use and floor space. land. Some land-use policy is sensitive on a scenario basis determined by exogenous inputs of population and nonbasic employment totals. UrbanSim Not explicit at present. Future versions to include workplace choice. Detailed representation of households; 11 household types. Model predicts births, deaths, moves, building type and location choices. Business establishments are explicit. User classified by industry and number of employees. Model predicts births, deaths, moves, and building type and location choice. Major buildings can be excluded from simulation to reflect low mobility and lack of information. Developers explicit as decision-makers. Currently, simulates development/ redevelopment at the land parcel level based on expected profitability. Number of policy inputs used in determining feasibility and cost. Revenue from current market prices. Can handle large multiyear projects. NA Explicit policy inputs include land-use plans, density constraints, growth boundaries, environmental constraints, transportation infrastructure, pricing, and service levels.

58 FI N A L R EP O RT Model Housing Market Floor Space Market Goods and Services Market ITLUP Demand for land explicit, allocates households to zones by type. Supply is implicit but defined by exogenous constraints. There is no price mechanism or signal, static equilibrium. Supply implicit. No price signal or mechanism. NA MEPLAN Supply function, developers allocate from exogenous total. Housing allocation by type to zones based on prices and current capacity. Includes dynamic lagged response. Demand for housing integrates the idea that the amount of space consumer per household is elastic with respect to price in zone. In a time period, prices are adjusted by zone until there are no vacancies: amount of space consumed per household and distributed among zones is in balance with supply. Same as housing market except with production processes producing labor. Explicit based on input- output framework with variable technical coefficients. Elastic with respect to price, logit style substitution. Households maximize utility subject to budget constraint (dual used in model). NYMTC- LUM Supply of housing by type by zone is a function of housing prices, interest rates, and development costs. Demand for housing by type by zone is determined by a logit model of worker joint choice workplace and place of residence. Choice is a function of wage in employment zone, price of housing in residential zone, accessibility terms, and other variables. Housing prices are determined by equilibrium. Similar to housing market except demand for nonbasic floor space is a function of rent of floor space and demand for nonbasic floor space located in a given zone. NA UrbanSim Uses bid rent model but does not impose equilibrium. Demand is based on willingness to pay, or bid function. Consumers are assumed to maximize surplus (bid-price). Households are price takers with Identical to and integrated with housing market. Land parcels developed into most profitable use that regulation will allow. Exogenous levels of employment by sector, endogenous mobility, and location of businesses. price adjusting between 1-year price steps based on aggregate demand and supply within each submarket (traffic zone or property type). Developers produce supply, maximizing profit based on current market conditions. Supply is assumed inelastic within each 1- year time step but is elastic between time steps. The short time step facilitates seeking model equilibrium, but retains rigorous microeconomic foundations. This provides a realistic representation of competition between residential and commercial land uses within the constraints of the land policy or zoning laws. Location choice incorporates access to labor market, localization, and inter- industry linkages. Strength of the locational influence of inter-industry links and localization is determined empirically during estimation of each sector. TABLE 10 Behavioral framework

by an endogenous price, which equates demand and supply. UrbanSim differs in this respect in that it is a dynamic model that moves to equilibrium but does not necessarily achieve it. Demographics are important model attributes that are treated exogenously. UrbanSim is moving in the direction of treating demographics endogenously. 59 All the models can handle a range of pricing- and infrastructure-related prices. This ability is unsurprising, since the models were developed for this purpose. The models have not been flexible in handling emerging policies, however. The models cannot, for example, handle regulatory policies or policies associated with transportation management or FIN A L R EPO RT TABLE 11 Behavioral framework Models Labor Market Personal Transportation Market Goods Movement Market ITLUP Implicitly modeled in that the jobs-housing market is one process. The demand for labor is determined exogenously to the submodels (i.e., employment per zone is known). The spatial distribution of labor (where workers live) is determined by the spatial allocation model. No labor price (wage) mechanism. Endogenous (within ITLUP) or exogenous linkages to external demand models. Practically, there is no feedback to the activity system. NA MEPLAN Labor supplied by households as demanded by production activities. Generally, labor costs paid by employers are the costs faced by households – households must give up leisure time to earn income. The framework is sufficiently flexible to allow the simulation of a market process, where wages are set endogenously (practical application generally treats wages exogenously). OD demands arise from the spatial distribution of flows from consumption to production. This includes all types of trips. Includes all modes. There is feedback if the model is iterated with flows assigned to networks with congestion. Feedback to travel decision occurs in same time period, feedback to activity location lagged one period. NA NYMTC- LUM Demand for labor in each zone is a function of wage level. External travel demand model. If the model is used with instantaneous feedback, the equilibrium between travel and land use occurs. Place of residence-place of work linkages and residence – nonwork activity location linkages are determined within the model. Not currently used in transportation demand management model. NA UrbanSim Location of jobs and workers determined as households and External travel demand model. Instantaneous NA firm location choices, with firm location affected by access to labor market and household location affected by access to jobs. Direct linkage of individual workers to individual jobs is not currently implemented. feedback within the travel model; lagged feedback effect to the household and business location choice model.

ITS options. These limitations reflect weaknesses in the stan- dard four-step, travel demand–modeling system. The prob- lem is not with integrated models, but with improving the travel demand models. 6.7 SUMMARY AND RECOMMENDATIONS: LAND-USE MODELS This chapter’s review of four representative models, which range across the state of the art in integrated land-use and transportation demand models, illustrates that no model is ideal. One model that stands out for a number of reasons is UrbanSim. First and foremost, UrbanSim is the most disag- gregate of the frameworks. It models land development at the parcel level and travel demand at the traffic zone level. Sec- ond, while most frameworks are based on strong equilibrium assumptions, UrbanSim takes a dynamic approach. Third, while most applications are temporarily aggregate and use up to 5-year steps, UrbanSim calculates changes in 1-year steps and thus provides a framework for ongoing assessment of policy instruments. Fourth, while other models are propri- etary, UrbanSim is a public domain model and is accessible by a broad group of practitioners. It will therefore progress in a way that reflects the attempts to find solutions to current shortcomings. This progression will form part of an evolving literature. The other models, because they are proprietary, will progress only to the extent that urban areas provide these questions and resources to the firms owning the software. The current models all have, to different degrees, strengths in a number of areas. They are based on solid microeconomic foundations with well-developed housing and land and floor space markets. They provide logical frameworks for assess- ing interactions between transportation and land use. In all, 60 the transportation network is multimodal, so a number of substitution possibilities can be considered. At the same time, all the current models have shortcomings in terms of excessive spatial aggregation, excessive reliance on strong equilibrium assumptions, aggregate representation of households and firms, and failure to explicitly represent the individual decision maker. These shortcomings mean that aggregation is not flexible across different dimensions. The models all suffer from the lack of endogenous demographic processes, auto-ownership processes, and the heavy reliance on four-step, travel demand–modeling techniques. A signifi- cant deficiency in all the models is the failure to integrate trans- portation and housing choice into the broader consumption bundle. The current set of models implicitly treats housing and transportation choice as strongly separable from the rest of people’s consumption activities. Changes in relative prices in the other areas will have no impact, as the models are presently structured, on the demand for housing or transportation or both. Even the specifications in the transportation-housing choice utility linkages have a maintained hypothesis of weak separability. On balance, however, UrbanSim has the least number of shortcomings, and the model’s remaining shortcomings are evolving in the direction of the ideal model. On the downside, UrbanSim is complex and data hungry. It requires expertise to populate it with data as well as to run it and translate the output to make it transparent to policy makers and transportation managers. But this requirement is not unique to UrbanSim. At this point, one needs to ask how UrbanSim satisfies the research objectives. The key points of the research objectives are to use the case studies to illustrate the methodology and to develop recommendations for the evolution of an analysis framework. In both cases, UrbanSim is a good candidate. First, FI N A L R EP O RT TABLE 12 Behavioral framework Model Transportation Infrastructure Market Auto (Vehicle) Market ITLUP Exogenous supply, from public authorities (or other providers of transportation infrastructure). Demand is implicit in the travel demand forecasting models. Not considered explicitly in ITLUP, although could be reflected in DRAM trip generation rates, modal splits, and vehicle occupancy rates. An exogenous travel demand model could include an explicit auto ownership choice submodel. MEPLAN Exogenous supply. Demand is implicit in the travel demand forecasting models. Can have categorization of households by auto ownership level, with exogenous transition among categories. NYMTC- LUM Exogenous supply. Demand is implicit in the travel demand forecasting model. Exogenous travel demand model could include auto ownership choice submodel. UrbanSim Exogenous supply. Demand is implicit in the travel demand forecasting model. Can categorize households by auto ownership level, with exogenous transition among categories, or can have an auto choice submodel in travel demand model.

61 FIN A L R EPO RT TABLE 13 Spatial allocation processes Models Housing Supply Housing Demand Floor Space Supply Floor Space Demand ITLUP Implicit in the model; allows zone- or sector- specific constraints that correspond to zoning and planning regulations and other land-use policies. Density constraint processes being tested in Metropolis. Logit allocation model for households, given known workplaces. Implicit. Implicit. MEPLAN Total development is exogenous in each time period and is allocated among zones as a function of price in the previous time period and of the availability of space. (Uses log-linear form of utility function.) Households containing workers demanded in zone j are allocated among zones according to logit functions. The utility function includes costs of location in zone i + travel disutility from i to j + the alternative specific constant. Set of explanatory variables can be expanded although rarely done in practice. Household type can include alternative- specific constants. Costs of locating in zone i include costs of consumption to produce in zone i. Same as housing supply. Same as housing demand. NYMTC- LUM Number of housing units per zone is a function of market value (as a function of price, interest rates) and development costs). Logit model is a function of household income, housing price, accessibility to work and goods/services, and other variables. Same as housing supply. Function of demand for nonbasic services and rent. UrbanSim Developers convert vacant parcels or redevelop parcels with existing development on the basis of expected profit using market prices and development Household choices of moving, building type, and location may be modeled as connected choices in a nested logit or may be separated into mobility Same as housing supply. Same formulation as housing demand, with bids a function of access to labor, localization effects, inter- costs. Housing supply is inelastic within 1 year, but elastic from year to year. Price triggers changes in the profitability of development. choice and joint choice of building type and location. Location choice is a function of consumer surplus, and bids are a function of housing type, density, access to jobs and shopping, age of housing stock, housing supply, zonal income distribution, land-use characteristics, and proximity to central business district. industry linkages, building types, density, age, zonal land-use mix, presence of highway, and proximity to central business district.

case studies by their very nature are highly focused and disag- gregate, and these characteristics are features of UrbanSim. Second, the nature of policies to be examined will require an evolution in a direction that UrbanSim is moving in. UrbanSim also compares favorably with most of the more specific NCHRP 25-21 methodology requirements: • UrbanSim is a good candidate for evaluating short-term impact analysis, since it provides outputs on a yearly basis. Although the model is not as good of a candidate for the long term because it is not an equilibrium model, UrbanSim provides a sense of direction. UrbanSim’s 62 inability to provide a final outcome is not a major weak- ness, since a manager has flexibility to respond to changes in a dynamic setting. • UrbanSim operates at the required level of disaggre- gation. • Although UrbanSim does not satisfy the objective that the model not require new data collection activities, this objective is at odds anyway with the other objectives. Meeting the objective of not requiring new data collec- tion activities would have a significant cost in terms of higher levels of aggregation and the consequent inabil- ity to examine projects and policies. FI N A L R EP O RT TABLE 14 Spatial allocation processes Model Goods & Services Supply Goods & Services Demand Labor Supply/Demand Demographic Processes ITLUP Implicit in nonwork person trip allocation model. Implicit in nonwork person trip production model. Treats job and housing market effectively as one and the same without differentiating between the two (i.e., housing location decisions that are conditional on workplace choice decisions and that tend to be dominated by the workplace decision). Exogenously specified total households by income category, which are then allocated by zone, in each time period. MEPLAN Explicit in production- consumption processes as modeled in input- output framework. Same as supply. Same as ITLUP. In practice, little or no demographics. Unemployed/retired households exogenously specified or are implicitly in labor to household ratios. NYMTC- LUM Implicit in nonbasic employment, which is determined by residential demand for these services. Residence- nonwork linkages submodel estimates the number of nonwork trips from each residential zone to each nonbasic employment zone as a function of travel impedances, accessibilities, income, and other socioeconomic attributes. Supply of labor jointly determined from workers household residence by logit model. Model is a function of wage rate by zone, travel impedances, and other variables. Demand for nonbasic labor is a function of wages, rent, and demand for nonbasic goods as determined by the residence- nonwork linkage model. No explicit demographic processes within the model structure. Total population for the forecast year is an exogenous input, as are other demographic attributes. UrbanSim Implicit in nonwork trip attraction model. Implicit on nonwork trip production model. Household and business location processes modeled independently, but with information about the lagged access to jobs or labor market. Links determined by Full range of demographic transitions is envisaged (birth, death, aging, household dissolution, in/out migration). However, current trip distribution model. Model is not workplace driven. model has static transition probabilities.

• Because UrbanSim is not proprietary, contributors from the academic community can develop the model and software in the direction of the ideal model. • Regarding the ability to model heavy vehicles, only MEPLAN handles goods movements. One method of integrating the use of heavy vehicles into UrbanSim is through the consumption of goods and services by households and a submodule that translates this demand into truck emissions. The same submodule can be used to assess different emission reduction policies by chang- ing the relative prices of goods. • UrbanSim is the only model evaluated herein that pro- vides a direct linkage between worker and jobs. In MEPLAN and the other models, there is little distinc- tion between the labor market and housing market. The Integrated Land Use, Transportation, Environment (ILUTE) , being developed as a joint project among four uni- versities in Canada, shows great promise.112 The model is closer to UrbanSim than to the traditional ITLUP model. It is presently in the development stage, but should be monitored closely as a candidate model for environmental assessment. 63 6.8 DEMOGRAPHICS: A BRIEF DISCUSSION Demographics are key components of any microsimula- tion model. Transportation and housing choices are influ- enced to a great extent by age, gender, education, household makeup, and the spatial distribution of the population with its attendant attributes. In discussing the ILUTE model that is being developed for Canadian cities, Miller and Salvini113 note that the first step in any modeling effort is to accurately represent the population of interest in the study area. This step requires detailed demographic attributes on a highly dis- aggregated basis. This step provides the base population from which the simulations are made. Population synthesis procedures generally use some sort of Monte Carlo procedure to draw the representative population from aggregate population data. Miller and Salvini describe a relatively new procedure to create a synthetic baseline population—a two-step iterative proportion-fitting procedure. The procedure estimates the multiway distributions for each census tract in a public use area, such that each distribution satisfies the marginal distribution for the census tract and has the same overall correlation structure as the public-use data multiway distribution. FIN A L R EPO RT Policy Category Specific Policy ITLUP MEPLAN NYMTC-LUM UrbanSim Land Use Pricing Taxation Subsidies Development Charges N N X E E E E E E E E E Infrastructure and Services Public Housing Servicing Land Government Buildings N X N E E E E E E E E E Regulations Zoning Urban Design X N E N E N E E Transportation Pricing Road Tolls Gas Taxes Subsidies Transit Fares Parking Prices I I N X X E E N E E I I N I I I I N I I Infrastructure and Services Build Roads Build Rail/Transit Ways Operate Transit ITS Parking E X X I I E E E N N E E E N N E E E N N Regulations Parking Regulations Traffic Regulations Nonpricing Transportation Demand Management Licensing Inspection/Maintenance N X N X N N E N N N N E N N N N E N N N Other Pricing Auto Tax License Charges Income Redistribution N N N N N E N N E N N E Infrastructure and Services NA Regulations Air Quality Standards Emission Standards Noise Safety Technology Standards N I I I N N E I I N N I I I N N I I I N E = explicit and can normally be done in model, N = no, I = implicit, X = can respond but only with exogenous parameter changes. TABLE 15 Policy capabilities of current models—land use

Related to the development of the above procedure is the process by which the population is “aged” over time. This process is part of the evolution of the urban area. A model is required that will process births, deaths, aging, in-migration, out-migration, marriages, divorces, employment changes, auto ownership, residential mobility, household formation, and household dissolution. This module will vary in complexity 64 and will require some flexibility within the overall modeling framework to reflect both the availability of data and the desired level of aggregation. The model being developed by Miller and Salvini is a composite of the traditional ILUTE model and UrbanSim. Miller and Salvini’s model has a sophisticated evolutionary process engine. FI N A L R EP O RT Models Scope Theoretical Consistency Spatial Precision Temporal Precision Validation Transit Representation ITLUP Partially integrated/partially connected with transport model. Spatially distributed only if totals are exogenously specified. No supply, no prices. Over-reliance on equilibrium; at best quasi- dynamic. Determined by data availability; in practice this means that large zones tend to be used. Five-year step. Smaller increments are possible if supported by data. History of validation and recalibration. Typically, no direct land-use sensitivity to transit, auto- only trips are considered. Transit trip making centers on travel demand model. MEPLAN Integrated with transport model. Full representation of production and consumption with fully endogenous prices. Over-reliance on equilibrium; at best quasi- dynamic. Use of IO framework at such a small spatial scale is questionable. Use of aggregate IO coefficients limits how small the zones can be while still being representative of the economic processes. Five-year step. Smaller increments are possible if supported by data. One-year lag used in land-use model. Some validation. Explicit representation of transit services. Accessibility includes transit effects and can influence land use. Conversion of transit OD times/costs to larger land-use zones an issue. NYMTC- LUM Connected but not integrated with transport model. Over reliance on equilibrium. Microeconomic consistency throughout. Developed at traffic zone level. Current implementation: one step to forecast end- year equilibrium. Could use 5- year steps for full equilibrium at each step. None for current version. Some experience with earlier versions. Transit trip making and land-use sensitivity to transit demands on travel demand model used. UrbanSim Connected but not integrated with transport model. Less stringent equilibrium conditions. Model uses nonequilibrium dynamic framework with one-year time Demand side is implemented at the traffic zone level. Supply side is implemented at the parcel One-year steps. Interaction with travel model may be annual or set at a level to represent significant changes in the transport Process underway. Transit trip making representation and land-use sensitivity to transit depends on travel demand model used. Current increments. level. system. applications have detailed transit network and mode choice model. IO = input-output. TABLE 16 Performance of current models—applicability

65 FIN A L R EPO RT TABLE 17 Performance of current models—feasibility/usability Models Data Requirements/Implementation Technical Requirements Output Presentation ITLUP Simpler model structure means modest data requirements and calibration effort. Expert support is generally required. Arcview shell provides presentation capabilities and has a link to external analytical and graphical displays in Windows. MEPLAN Complex model structure leading to significant data requirements and calibration effort. Set-up requires expert, ongoing support. Full economic evaluation model—prices, consumer surplus, and flows for volumes and times. NYMTC- LUM Since it is a static model, data for one point in time is required for calibration. At a minimum, census journey-to-work and income data and nonwork trip linkages are required for calibrations. Housing floor space data are not necessary for calibration. Unknown. Population and employment distributions by zone. For residence- work and residence- nonwork trip linkages, include outputs of wages, rents, average income by zone, and consumer and producer surplus for economic evaluation. Format for this information is not clear. UrbanSim Requires the following data: parcels with land and improvement values, area, housing units and nonresidential square footage, business establishment inventory, census data, land-use plans, and environmental constraints. Calibration requires use of standard regressions for bid functions and a logit for location models. Not implemented, currently being implemented in three cities. Model outputs include household and business distributions by zone, land use, property values, and housing and nonresidential floor space. Results can be transferred to spreadsheet programs. There is a visualization component.

66 FI N A L R EP O RT CHAPTER 7 TRAVEL DEMAND MODELS This chapter highlights three candidate modeling approaches for the NCHRP 25-21 methodology: TRANSIMS, the Port- land Tour-Based Model, and STEP. 7.1 TRANSIMS Transportation Analysis Simulation System (TRANSIMS) is a multiyear project of the FHWA and the University of Cal- ifornia Los Alamos National Laboratory (LANL) to develop a travel demand model that retains the identity of individual “synthetic” travelers throughout the entire travel demand fore- casting and traffic operations analysis process of the model. TRANSIMS incorporates tour-based travel demand, inter- modal trip planning, traffic microsimulation, and air quality analysis, all within a single unified program architecture.114 TRANSIMS is a rapidly evolving transportation modeling system for which any written description is soon out of date. Interested readers should consult the TRANSIMS homepage at http://transims.tsasa.lanl.gov to obtain the most current information on this system of programs. 7.2 PORTLAND TOUR-BASED MODEL This section describes an advanced model form that employs tour-based analysis and sample enumeration and that is cur- rently being developed and tested for Portland, Oregon.115, 116 Currently under development, this model is not fully opera- tional in that it has not been used yet by the MPO for its plan- ning tasks (the currently operational model in Portland is described in Chapter 3). This section describes the model as it existed near the end of 1998. Figure 10 provides a flow chart for the advanced model. The subsections below describe each of the modules (i.e., boxes) in the model. For the Portland model, it was necessary to back off from some of the ideals of tour-based models in order to achieve rea- sonable computer memory requirements and model run times (no more than 6 hours per model run on the fastest available microcomputer). Specifically, the Portland tour-based model • Simplified time of travel to five time periods per day, • Limits the number of stops on tour to one outbound and one inbound, • Allows only one “main” mode for the tour (rather than segment-specific modes), • Does not perform microsimulation of traffic conges- tion, and • Does not provide vehicle class information (by emis- sion type). 7.2.1 Input The Portland Tour-Based Model requires the following input data: • Households by traffic analysis zone (stratified by per- sons per household, income, and age of the head of the household), • Employment by Standard Industrial Classification (SIC) code and traffic analysis zone, and • Modal accessibility measures. 7.2.1.1 Household Data The Portland Tour-Based Model employs a sample enu- meration technique to develop aggregate travel results. The activity patterns are predicted for a sample of households and are then expanded (with expansion factors) to represent the entire population in the region. In the case of Portland, a synthetic sample of 120,000 per- sons (about 50,000 households) was created from the U.S. Census Public Use Microdata Sample (PUMS). This syn- thetic sample was chosen rather than a 4,451-household sur- vey that Portland had available because of the desire to have a larger sample size for generating a statistically reliable trip table for use in traffic assignments. The majority of analyses were conducted using 10 percent of the synthetic population sample (approximately 12,000 persons). Households were stratified into 64 bins or cells consisting of four household size categories, four income categories, and four age-of-head-of-household categories. For each traf- fic analysis zone, households were drawn randomly from the 5-percent PUMS from the relevant Public Use Microdata Area (PUMA).

67 FIN A L R EPO RT A maximum of one household is selected for each of the 64 household type bins, for each individual traffic analysis zone within the region. For Portland, this selection means a maximum sample size of about 80,000 households (about 200,000 persons) (1,244 zones by 64 household categories per zone). When a household is selected, all persons 16 years or older are included in the sample used in the analysis. The existing auto ownership for each sample household is pulled from the PUMS. The future auto ownership is pre- dicted using an auto ownership submodel. 7.2.1.2 Employment and Modal Accessibility Data Existing and forecasted employment by SIC code must be provided for each traffic analysis zone. Network (i.e., zone to zone) travel times by mode, access times by mode, and cost by mode must be provided for each possible pair of origin and destination zones, by time period of day. 7.2.2 Household-Based Tour Module The Household-Based Tour Module is nested logit and predicts household-based tour activity for the weekday (see Figure 11). A household tour is a sequence of trip segments in which the entire tour starts and ends at home. The Nested Logit Choice Module is organized according to the follow- ing hierarchy: 1. Primary Activity Choice (first row [i.e., top row]), 2. Primary Tour-Type Choice (second row), 3. Secondary Tour Choice (third row), 4. Time-of-Day Choice (fourth row), 5. Destination Choice (fifth row), and 6. Mode Choice (sixth row [i.e., bottom row]). SIC = Standard Industrial Classification. TAZ = traffic analysis zone. OD = origin-destination. Input Employment by SIC and TAZ Representative Sample of Households by TAZ Modal Accessibility Measures Household-Based Tour Module Primary Activity Choice, Primary Tour Type Choice, Secondary Tour Choice, Time-of-Day Choice, Destination Choice, Mode Choice Work-Based Subtour Module Intermediate Stop Location Module (for car driver tours only) Tour-to-Trip Decomposition Output OD Trip Matrices by Mode, Purpose, Time of Day, Income Class Network Model (trip assignment by mode and period of day) Figure 10. Portland Tour-Based Model flow chart. 9 Modes 22 Destination Choices 15 Time Period Combinations 6 Secondary Tour Types 8 Tour Types Work On Tour 9 Modes 22 Destination Choices 15 Time Period Combinations 6 Secondary Tour Types No Tour Work At Home 9 Modes 22 Destination Choices 15 Time Period Combinations 6 Secondary Tour Types 4 Tour Types Maintenance On Tour 9 Modes 22 Destination Choices 15 Time Period Combinations 6 Secondary Tour Types No Tour Maintenance At Home 9 Modes 22 Destination Choices 15 Time Period Combinations 6 Secondary Tour Types 4 Tour Types Discretionary On Tour 9 Modes 22 Destination Choices 15 Time Period Combinations 6 Secondary Tour Types No Tour Discretionary At Home Household and Person Data Figure 11. Portland’s Household-Based Tour Module.

68 FI N A L R EP O RT The choice probability at each level is conditional upon the choices in the levels above it. The following paragraphs explain each level of the nested logit model. 7.2.2.1 Primary Activity Choice The top level of the nested logit model selects the primary daily activity for each person in the household. Three activ- ity types are available (work [which includes school], house- hold maintenance, or discretionary activities), and for each primary activity type, the person can choose to do it outside the home (i.e., on tour) or at home. Thus, six choices are available at this level: 1. Work On Tour, 2. Work At Home, 3. Maintenance On Tour, 4. Maintenance At Home, 5. Discretionary On Tour, and 6. Discretionary At Home. Work on tour (with intermediate stops or workplace sub- tours) account for about half of all trips made in the Portland urban area. 7.2.2.2 Primary Tour-Type Choice The three primary activities that take place outside of the home are then divided into four or eight tour types, depend- ing on the number and sequence of intermediate stops. (A maximum of one stop is allowed for each direction of travel [outbound and inbound] to the home except for work tours, which are allowed one extra stop during working hours.) For nonwork (i.e., maintenance or discretionary) tours out- side of the home, the following four tour types are available: 1. Home to destination, return home (no intermediate stops); 2. Home to intermediate stop, continue to tour destination, return home (one intermediate stop); 3. Home to destination, make intermediate stop on way home, return home (one intermediate stop); and 4. Home to intermediate stop, continue to tour destina- tion, make intermediate stop on way home, return home (two intermediate stops). For work tours outside of the home, the above four types of choices are available plus the additional option of running (or not running) an errand during working hours (i.e., when a person leaves and returns to work). Thus, for work tours outside of the home, a total of eight tour types are available (HWH, HOWH, HWOH, HOWOH, HWOWH, HOWOWH, HWOWOH, HOWOWOH, where “H” is home, “W” is work, “O” is other intermediate stop, and the letter sequence gives the sequence of tour stops). At this level, there are 16 tour types possible for the tours that leave the house, plus 3 primary activities that do not leave the home; thus, there are 19 possible tour types (a com- bination of primary activity and the tour type). 7.2.2.3 Secondary Tour Choice For each of the 19 possible tour types, 6 secondary tour choices are available: 1. No secondary tour, 2. A single household maintenance tour, 3. Two or more household maintenance tours, 4. A single discretionary tour, 5. Two or more discretionary tours, and 6. A single household maintenance tour and a single dis- cretionary tour. The model trades off extra stops on the primary tour against additional secondary tours to and from the home. At this level, there are now 19 ∗ 6 (114) possible branches of the choice tree for each person in the household. Note that, unlike the primary tour model, which is a logit choice model, the secondary tour model uses fixed percentages obtained from the Portland survey data. 7.2.2.4 Time-of-Day Choice At the time-of-day choice level of the nested logit model, each of the 114 possible primary and secondary tour types is then assigned a starting and ending period probability. The weekday is divided into five time periods (Early, AM Peak, Midday, PM Peak, and Late). For Portland, the following limits were selected for these time periods: • Early (3 AM to 7 AM), • AM Peak (7 AM to 9:30 AM), • Midday (9:30 AM to 4 PM), • PM Peak (4 PM to 7 PM), and • Late (7 PM to 3 AM). Since there are five starting periods and five ending peri- ods, a total of 25 combinations are possible; however; over- night trips have been ruled out, so the resulting available starting and ending choice combinations decreases to 15. All intermediate stops occurring on a tour are assumed to occur in the same period as the half-tour to which they are assigned. Thus, an intermediate stop on the way to work is assumed to occur in the same time period when the person leaves the home.

7.2.2.5 Destination Choice At this level, the probability of choosing one of 22 possi- ble zone destinations is computed for each alternative. The possible choice list of 1,244 zones in the Portland area was reduced to a “feasible” subset of 22 zones to improve com- putational efficiency of the model. The total 1,244 possible destination zones are stratified according to their distance from the tour origin zone and the employment. The 22 “fea- sible” destination zones for each origin zone are sampled from the 1,244 possible destinations so as to reproduce the actual distribution of chosen destinations. Separate sets of models are used depending on the tour pur- pose (work [including school], maintenance, or discretionary). 7.2.2.6 Mode Choice For each of the 22 feasible destination zones, the model computes the conditional probability of choosing each of 9 modes. Each selected mode will be the main mode for its respective tour (segment-specific mode choice was ruled out by computation constraints). The 9 main tour modes are 1. Drive Alone, 2. Drive with Passenger, 3. Passenger in Car, 4. Light Rail via Auto Access, 5. Light Rail via Walk Access, 6. Bus via Auto Access, 7. Bus via Walk Access, 8. Walk, and 9. Bicycle. There are 22 ∗ 9 (198) possible mode/destination types for each tour. Note that by assigning a single main mode to a tour, the model does not distinguish between a casual carpooler who hitches a ride to work and takes the bus home from a person that takes a bus going both to and from work. Main modes are generally assigned to each tour with the intent that, as much as possible, they reflect (within the available nine modes to choose from) the auto VMT generated by the tour while recognizing that some detail and precision on non–auto use will be lost. 7.2.3 Work-Based Subtour Module For home-based tours that are predicted to have a work subtour (running an errand during working hours), the work- based subtour model predicts the mode and destination of the subtour. Fixed fractions from the household survey are used to identify the time period when the subtours occur. The time-of-day fractions are conditional upon the time of day when the person’s home-to-work tour begins and ends. 69 Time of travel is not taken into account. Subtours are assumed to be completed within the same time period as when they started. 7.2.4 Intermediate Stop Location Module The Intermediate Stop Location Module predicts the loca- tion of the one intermediate stop (the module currently can- not account for multiple stops) visited between the home and the primary tour destination (conditional upon the main mode, location, and timing of the primary tour activity). This mod- ule is applied only to tours predicted to have intermediate stops and only to tours that are made by car. This module is applied at a more aggregate level than other modules are. Time of travel is not taken into account. The intermediate stop is assumed to occur in the same time period as when the tour started (if it is an outbound stop) or when the return trip started (if the stop is made on the return trip home). Two separate modules are used: one for intermediate stops on work tours and another for intermediate stops on non- work tours. 7.2.5 Tour-to-Trip Decomposition The Tour-to-Trip Decomposition Module decomposes tours into trips. The origin, intermediate stop, destination, work subtour stop, destination, intermediate stop, and return to origin of the tour become the beginning and end points of trips. The model outputs person trip 1,244 ∗ 1,244 zone origin- destination tables for two tour purposes (work and nonwork), four time periods, nine modes, and three household income classes of the traveler. 7.2.6 Output Module The Output Module of the Portland Tour-Based Model produces OD trip matrixes by mode of travel, trip purpose, time of day, and income class. 7.2.7 Network Module The Network Module of the Portland Tour-Based Model is responsible for assigning the OD matrixes to the highway network and producing OD travel times by time of day. 7.2.8 Shortfalls of Current Portland Tour-Based Model The NCHRP 8-33 investigators (who were also the devel- opers of the Portland Tour-Based Model) noted that they had to overlook a few theoretical shortfalls in order to keep the com- putation requirements of the model within reason. Specifically, FIN A L R EPO RT

70 FI N A L R EP O RT they noted the following shortfalls as of the November 1998 implementation of this model: • The model uses an aggregate traffic analysis zone sys- tem rather than a 1-acre grid system. • Time of day is limited to five time periods (not hour by hour). • Tours are limited to one stop outbound and one stop on the return trip. • The model predicts primary mode of tour, but not segment-by-segment mode. • Intermediate stops are aggregated. • Work subtours are not fed back to other activity choices. • Time and space constraints on activities were not imple- mented in the model to account for the dependency between activities. 7.3 THE STEP MODEL The Short-Range Transportation Evaluation Program (STEP) was originally developed by Greig Harvey of Deakin- Harvey-Skabardonis. This section summarizes materials that Harvey and others prepared for the California Energy Commission.117 STEP is a package of microscopic household-level travel demand models designed for planning applications and pol- icy analysis. The STEP model can predict the influence of travel time and cost on residential household location, pri- mary work location, vehicle ownership, trip frequencies, trip destinations, mode choice, and time of day. The STEP model is composed of 5 modules (see Figure 12): • Household Location Choice Module, • Auto Ownership Module, • Daily Travel Activity Module, • Time-of-Day Module, and • Transportation System Performance Module. The results of the Transportation System Performance Mod- ule are fed back into the Household Location Choice Module. STEP uses sample enumeration to obtain aggregate fore- casts. STEP is applied to a sample of households within the Household Location Choice Module: Residential Location Primary Workplace Location for Each Worker Auto Ownership Module: Number of Autos Owned Daily Travel Activity Module: Trip Frequency (HBW, HBS, HBO, NHB) Trip Destination (HBS, HBO, NHB) Trip Mode Choice (HBW, HBS, HBO, NHB) Time-of-Day Module: Work Arrival Time Transportation System Performance Module: Highway Corridor Delay Note: HBW - Home-Based Work Trips HBS - Home-Based Shopping Trips HBO - Home-Based Other Trips NHB - Non-Home-Based Trips Figure 12. Flow chart of STEP model.

71 FIN A L R EPO RT region (typically 5,000 households). The results are then expanded to represent total regional travel activity. STEP outputs daily VMT, fuel consumption, and mobile emissions. 7.3.1 Input Data Requirements The required input data for STEP are • Socioeconomic data (the socioeconomic characteristics of a sample of households in the region obtained from a household travel survey or the U.S. Census PUMS), • Land-use data (population, number of households, and employment by category located by zone or district in the region), and • Transportation level-of-service data (travel times and costs). 7.3.2 Household Location Module The Household Location Module uses a multinomial logit model to predict the probability of a household locating in a district. The probability is sensitive to the relative housing costs, ethnic makeup, crime rate, tax rate, quality of schools, and the accessibility to jobs for each location. Household location districts are typically the U.S. Census PUMAs. The probability that a household will choose to live in district i can be calculated as follows: Equation 17 Where: Pi = the probability that a household will choose to live in district i, Ui = the perceived utility of district i as a residence, and n = the number of districts in the region. The perceived utility of district i as a residence (Ui) can be calculated as follows: Equation 18 Where: pricei = the mean monthly price of the household’s current type (rent, own, single, multi) for dis- trict i, f(income) = a nonlinear transformation of the household’s income, U b f b b b b b i i i i i i i =     + ( ) + ( ) + ( ) + ( ) + ( ) 1 2 3 4 5 6 price income ethnicity crime tax school mode ( ) P U U i i ii n= ( ) ( ) = ∑ exp exp1 ethnicityi = the percentage of households in district i with this household’s ethnicity, crimei = the rate of serious and violent crime per 100,000 residents at location i, taxi = the property tax on a home of average value at district i (homeowners only), schooli = the average per-pupil expenditure in location i (households with children only), modei = the sum of the log of the denominator of the mode choice model for work trips from dis- trict i for each worker in the household across all modes, and bl,…, b6 = parameters fitted by estimation. The parameters (b) vary by the number of workers in the household. A land price model is used to reflect the effect of supply constraints on price and therefore demand. 7.3.3 Vehicle Ownership Module Two logit models are used to predict the probability of a household owning 0, 1, or 2+ vehicles. One model is used for households with workers and the other for households with- out workers. Both models take the following form: Equation 19 Where: Pv = the probability of choosing vehicle ownership level v, Uk = the household’s utility for vehicle ownership level k, Uv = the household’s utility for vehicle ownership level v, and k = the set of vehicles ownership levels (0, 1, 2+). The variables and coefficients for the three utility functions (for 0, 1, and 2+ auto households) are shown in Table 18. For example, the utility of owning one auto is Equation 20 The nonworker household vehicle ownership model has the same form as the worker model. The utility specifications and coefficients are shown in Table 19. For example, the util- ity of owning one auto for a zero worker household is Equation 21U x1 0 8695 0 3188= − + ( ). . ln dinchhsize U1 1 1 4 989 0 3935 0 06814 0 7919 = + − − + + ( ) . . sin . . ln fam 0.05419eden 2.689 autoshhsize twork rinc P U U v v k k = ( ) ( ) = +∑ exp exp 1 2

7.3.4 Daily Travel Activity Module The Daily Travel Activity Module predicts trip frequency, distribution, and mode choice as a function of travel time and cost plus other variables using multinomial logit models with varying utility functions. 7.3.4.1 Home-Based Work Trip Distribution Model Once the residence location i and auto ownership have been determined, a logit model is used to predict the proba- bility of working in zone j: Equation 22P w P E U w P E U d d v m vd v i v m vi vi = + ∗ + ∗ ( ) [ ]( )   ( ) [ ]( )   = + = + = ∑ ∑∑ exp ln . exp ln . 1 811 1 811 1 2 1 2 1 nzones 72 Where: Pd = the probability of choosing destination d as the workplace; wd = the total number of workers attracted to (or jobs available in) zone d; wi = the total number of workers attracted to (or jobs available in) zone i; E[Um |v d] = the expected utility of work mode choice to destination d, given auto ownership level v (the expected utility is defined here as the natural logarithm of the sum of the utilities for travel to destination d, via each mode m, given auto ownership level v); E[Um |v i] = the expected utility of work mode choice to destination i, given auto ownership level v (the expected utility is defined here as the natural logarithm of the sum of the utilities for travel to destination i, via each mode m, given auto ownership level v); FI N A L R EP O RT Variables in the Utility Coefficient Value 0 Vehicle 1 Vehicle 2 or More Vehicles Explanation 4.989 const 1 vehicle ownership constant. 5.689 const 2+ vehicle ownership constant. 0.3935 sinfam Constant for single-family detached unit. Constant for single-family detached unit. 1.342 sinfam -0.05419 eden eden Workers per acre in the home zone. -2.689 autos/hhsize autos/hhsize Autos per person in household. The variable “autos” has the value 1 for v = 1 and 2.25 for v = 2+. 0.5608 tshop A measure of the quality of transit service from the home zone for nonwork trips, defined as the sum of transit utilities divided by the sum of auto utilities for the shopping destination/mode choice model. 0.06814 twork0 tworkl twork2+ A measure of the quality of transit service from the home zone for work trips, defined as the household head’s work trip transit utility divided by the sum of work trip drive and work trip shared-ride utilities. 0.7919 ln(rinc0) ln(rincl) ln(rinc2+) Natural log of the remaining income after housing, auto ownership, and commuting expenses subtracted. TABLE 18 Variables and coefficients for worker households with 0, 1, and 2+ autos Variables in the Utility Coefficient Value 0 Vehicles 1 Vehicle 2 or More Vehicles Explanation -0.8695 const 1 vehicle ownership constant. -8.357 const 2+ vehicle ownership constant. -0.0682 popden Population density in home zone (persons per acre). 0.3188 Natural log of the household disposable income per person. 1.227 Natural log of the household disposable income per person. 0.5608 tshop A measure of the quality of transit service from the home zone for nonwork trips, defined as the sum of transit utilities divided by the sum of auto utilities for the shopping destination/mode choice model.  ln   hhsize dinc  ln   hhsize dinc TABLE 19 Variables and coefficients for nonworker households with 0, 1, and 2+ autos

Pv = the probability of choosing household auto ownership level v; and nzones = the number of zones in the region. Greig Harvey, in his description of STEP, equates the expected value of a utility function across several alterna- tives with the natural logarithm of the sum of the utilities for the alternatives. However, the mathematical equivalence could not be confirmed. Nevertheless, Harvey’s use of the term “expected value” for the logsum has been retained in this discussion. Note that this model is not constrained by the number of available jobs in the destination zone, although the number of jobs influences the likelihood of a worker going to that zone. 7.3.4.2 Home-Based Work Mode Choice Model The probability of a mode being selected for a home-based work trip is multinomial logit, with the utility of each mode defined in Table 20. For example, the utility for drive-alone trips is Ua = −2.512 − 0.00000714dinc − 1.067cbd − 0.0244ivtta − 0.077walka − 21.43(costa/inc) + 1.958autos + 0.677head Equation 23 The auto occupancy for shared-ride, home-based work trips is computed according to the following equation, which is constrained to have a value greater than 2: srocc = Max[(2.542 * 0.00004717dinc + 0.01116ivtt(s)), 2.0] Equation 24 Where: srocc = the shared-ride occupancy; dinc = the household disposable income; and ivtts = the shared-ride, in-vehicle time. 73 7.3.4.3 Home-Based Shop Trip Frequency Model The daily frequency of home-based shop trips is determined according to the following nonlinear regression equation: Equation 25 Where: hbshop = the daily frequency of home-based shop trips; Uhbs = −0.34174 (household size) − 0.51512 (income/100) − 0.52681 E(Udm) + 0.1146 ln(1 + employment density); and E(Udm ) = natural logarithm of the denominator of the home- based shop destination and mode split model. 7.3.4.4 Home-Based Shop Trip Destination and Mode Choice Model Home-based shopping trips are distributed and split between modes using a multinomial logit model that defines each pos- sible combination of zone destination and mode choice as a separate alternative. The basic model form is as follows: Equation 26 Where: a = auto mode; t = transit mode; Uji = the traveler utility for the destination j, mode i combination; Pdm = the probability of taking a shop trip to destination d by mode m; and P U U dm dm ji i a t = ( ) ( ) = ∑∑ exp exp ,j=1 nzones hbshop = + ( ) 0 8194 0 0766 . . exp Uhbs FIN A L R EPO RT Variables in the Utility by Mode Coefficient Value Drive alone Shared ride Transit Explanation -0.00000714 dinc dinc Household disposable income. -1.067 cbd Constant for central business district. -0.347 cbd Constant for central business district. 0.327 nwork Number of workers in household. -0.0244 ivtta ivfts ivttt In-vehicle travel time (minutes). -0.077 walka walks walkt Walk time (minutes). -0.045 waitl Transit initial wait (minutes). -0.0428 Xferwait Transit transfer wait (minutes). -21.43 costa/inc costs/inc costt/inc Cost (cents)/household income. 1.958 autos Number of autos in household. 1.763 autos Number of autos in household. 1.389 autoslaac Number of autos for auto access. -1.237 aac Constant for auto access to transit. 0.677 head Constant for head of household. -2.512 const Drive-alone constant. -3.473 const Shared-ride constant. TABLE 20 Variables and coefficients for drive alone, shared ride, and transit

Udm = the traveler utility for the destination d mode m combination. The utility equations are linear with variables and coeffi- cients as defined in Table 21. For example, the utility of the auto mode (Ua) is Equation 27 7.3.4.5 Home-Based Social or Recreational Trip Destination and Mode Choice Model Home-based social or recreational trips are distributed and split between modes using a multinomial logit model that defines each possible combination of zone destination and mode choice as a separate alternative. The basic model form is the same as for shopping trips. The utility functions are lin- ear, with coefficients and variables as defined in the Table 22. For example, the utility of the auto mode (Ua) is Ua a a = − + + ( ) − ∗( ) − + + 0 8631 0 2563 5 053 0 000202 0 2447 0 0005995 . . . . . . ln( cbd autos/hhsize time inc cost rden rjobs) 74 Ua = 1.844 − 0.215cbd + 2.167(autos/hhsize) + 0.3368rautos − 0.0001097(timea ∗ inc) Equation 28 − 0.0256 costa + 0.0609rden + 0.0244popden + 0.6998 ln(pop/rjobs) + ln(rjobs) 7.3.4.6 Home-Based Social or Recreational Trip Frequency Model The daily frequency of social or recreational trips is a func- tion of household characteristics, home zone characteristics, and destination characteristics (as embodied in the expected utility for social or recreational destination/mode choice). The trip frequency model is as follows: Equation 29 hbsr hhsize) (hhsize nwork) inc seden = ∗ ∗[ ] + ∗ − + ∗ ÷ + ∗ [ ] − ∗ + 0 1398 0 4671 0 005055 0 3963 100 0 06785 0 3213 1 . exp . ln( . . ln( ) . . ln( ) E Udm FI N A L R EP O RT Variables in the Utility Coefficient Value Auto Transit Explanation -0.8631 const Auto constant. 0.2563 cbd Constant for central business district. 0.8912 cbd Constant for central business district. 5.053 autos/hhsize Autos per person in household. -0.000202 timea*inc timet*inc Door-to-door travel time (minutes) weighted by income. -0.02447 costa Cost (cents). -0.02299 fare*hhsize Transit fare (cents) weighted by household size. 0.0005995 rden rden Retail density (employees per population serving acre). 1.0 ln(rjobs) ln(rjobs) Natural log of retail workers in zone. Variables in the Utility Coefficient Value Auto Transit Explanation 1.844 const Auto constant. -0.215 cbd Constant for central business district (destination). 1.19 cbd Constant for central business district (destination). 2.167 autos/hhsize Autos per person in household. 0.3368 rautos Autos not used for work trips. -0.0001097 timea*inc timet*inc Door-to-door travel time (minutes) weighted by income. -0.0256 costa Cost (cents). -0.0108 fare*hhsize Transit fare (cents) weighted by household size. 0.0609 rden rden Retail density at destination (employees per acre). 0.0244 popden popden Persons per acre at destination. 0.6998 ln(pop/rjobs) ln(pop/rjobs) Natural log of population per retail job at the destination. 1.0 ln(rjobs) ln(rjobs) Natural log of retail employment in the destination zone. TABLE 21 Home-based shop trip destination and mode choice variables and coefficients for auto and transit TABLE 22 Home-based social or recreational trip destination and mode choice variables and coefficients for auto and transit

Where: hbsr = the number of daily home-based social or recre- ational trips per household; hhsize = the number of persons in the household; nwork = the number of workers in the household; inc = the household income (dollars per year); E[Udm] = the expected utility from the social or recreational destination/mode choice model, defined as the natural log of the denominator of that model’s logit equation; and seden = the service employment density, in workers per gross acre. All other trip purpose frequencies (home-based work, non– home-based trip) are kept constant in STEP. 7.3.5 Time-of-Day Module A nested logit Time-of-Day Module is used to predict the starting time of work trips during the morning peak period. (This module had not been implemented at the time of Harvey’s description. The description provided here is more of a conceptual guideline rather than an actual module.) The top level of the Time-of-Day Module estimates the binary probability that a worker has a regular schedule (e.g., a work start time between 5:30 AM and 10:30 AM) or an irregular schedule. This probability is a function of house- hold income, household size, and the ratio between AM peak and offpeak highway travel time. For regular schedule workers, the conditional probability is then computed that a regular-schedule worker will start work during any one of the five morning hours between 5:30 and 10:30 AM. This conditional probability is a function of household size and the ratio of AM peak to offpeak travel. 7.4 ASSESSMENT TRANSIMS is still in its early stages of development and, as such, is more a philosophical approach to travel demand modeling rather than an actual model. As criticisms are lev- eled at specific steps, the problems are fixed and TRANSIMS becomes a different model. Efforts are currently underway to replace TRANSIMS’s simplistic Travel Pattern Selection 75 Module with a more theoretically sound Travel Behavior Mod- ule based upon the Portland Tour-Based Model. In essence, the basic philosophy of TRANSIMS is employ- ing unlimited computing power for the microscopic simula- tion of the second-by-second movements of individual peo- ple through the region. As such, the resources required to feed and operate such a model is beyond the range of feasi- ble options for NCHRP Project 25-21. Additional refine- ments to TRANSIMS (such as recent work to make much of the required traffic control information endogenous to the model) and further improvements in personal computers will no doubt improve the model’s feasibility, but the model is not currently a viable option for NCHRP Project 25-21. The Portland Tour-Based Model, in contrast to TRAN- SIMS, has already been trimmed back from its idealized activity-based theoretical foundations to ensure feasible appli- cation on current computer facilities. As such, the Portland Tour-Based Model is a possible option for NCHRP Project 25-21, although some travel demand responses will have to be cut from the model to ensure feasible operation. Since there is little experience with this modeling approach outside of the Portland area, there is concern over its application to other regions. For example, demand responses, which occur infrequently in Portland and thus could be replaced with fixed factors, may be more important in other areas. Then again, demand responses might be equally unimportant elsewhere. It is simply not known at this time. The bottom line, however, is that Portland represents a significant improvement in demand-modeling capabilities and thus should be tested in NCHRP Project 25-21. The STEP model was once the leading state-of-the-art approach to modeling the impacts of transportation supply changes on demand. Its groundbreaking focus on modeling household travel behavior and using sample enumeration to extrapolate household effects to regional effects allowed the consideration of more detailed demand effects than possible for more aggregate models. However, the analytical heart of STEP is several years old. The demand models it is based on have been gradually replaced with more up-to-date nested logit models. The concept of modeling household behavior is still as valid as (if not more so than) when Harvey created STEP; however, the analytical engine needs to be updated. The bottom line conclusion is that the STEP concept is very appropriate for testing in NCHRP Project 25-21, with the actual demand model formulas updated to the latest practice. FIN A L R EPO RT

76 FI N A L R EP O RT CHAPTER 8 TRAFFIC OPERATION MODELS This chapter reviews methodologies for predicting vehicle mode of operation activity. The methodologies range from the simplistic link-based Bureau of Public Roads (BPR) equation to the Highway Capacity Manual to dynamic microsimulation. 8.1 THE BPR EQUATION The standard BPR (predecessor to the FHWA) equation was developed in the late 1960s by fitting a polynomial equa- tion to the freeway speed-flow equations contained in the 1965 Highway Capacity Manual. The standard BPR equation is as follows: Equation 30 Where: s = predicted mean speed, sf = free-flow speed, v = volume, c = practical capacity, a = 0.15, and b = 4. Practical capacity is defined in this equation as 80 percent of the capacity. Free-flow speed is defined as 1.15 times the speed at the practical capacity. The parameter a determines the ratio of free-flow speed to the speed at capacity. The parameter b determines how abruptly the curve drops from the free-flow speed. A high value of b causes speed to be insensitive to v/c until the v/c gets close to 1.0; then, the speed drops abruptly. Planners typically use tables based on area type and facil- ity type for assistance in coding free-flow speed and capacity data. These tables allow planners to use simple road maps and aerial photos to code the free-flow speed and capacity information for 5,000 to 10,000 links in a region. A common error of practitioners has been to overlook that “capacity” in the standard BPR equation is actually practical capacity, which is closer to 80 percent of the actual capacity of the facility. s s a v c f b= + ( )1 / Table 23 shows practical capacity and free-flow speed. The table was developed by the FHWA118 for use with the BPR equation. 8.2 HIGHWAY CAPACITY MANUAL The Highway Capacity Manual119 contains a series of pro- cedures for predicting the steady-state traffic conditions at a macroscopic level. Traffic performance in terms of mean delay, mean travel speed, and mean density are predicted for the peak 15-minute period within the peak hour. Dynamic effects such as the build-up of traffic queues over several time periods and the impact of one time period on the fol- lowing time period are not explicitly considered (although a few of the procedures allow users to manually account for these effects). Modal activity (acceleration, deceleration, idle, and cruise) is not predicted by the HCM procedures. The HCM procedures are generally sensitive to the geo- metric design of the facility (width, grade, number of lanes, etc.), the traffic controls (stop sign, signal, signal timing, coordination, etc.), and the demand (vehicles, vehicle mix, peaking, turning movements, etc.). Demand is assumed to be fixed, peaking by a fixed percentage (selected by the user) within the peak hour. A key step of all of the HCM procedures is the computa- tion of facility capacity. This computation is normally sensi- tive to the facility design characteristics. The following equa- tion illustrates the computation of capacity for the approach to a signalized intersection: c = g/C ∗ s0 ∗ N ∗ fW ∗ fHV ∗ fg ∗ fp ∗ fbb ∗ fa ∗ fLU ∗ fLT ∗ fRT Equation 31 Where: c = the capacity of approach (vehicles per hour), g/C = the ratio of signal green time to total signal cycle length, s0 = the base saturation flow rate (vehicles per hour of green per lane), N = the number of lanes on the approach, fW = the lane width adjustment factor, fHV = the heavy vehicle adjustment factor,

77 FIN A L R EPO RT fg = the approach grade adjustment factor, fp = the parking lane adjustment factor, fbb = the local bus adjustment factor, fa = the area type adjustment factor, fLU = the lane-use adjustment factor, fLT = the left-turn adjustment factor, and fRT = the right-turn adjustment factor. The delay computation is typically sensitive to the capac- ity, the demand, and the traffic control characteristics. The following equations illustrate the computation of intersection approach delay for a traffic signal: D = du ∗ DF + di + d3 Equation 32 Equation 33 Equation 34 Where: D = the approach total delay, in sec/veh; du = the approach uniform delay, in sec/veh; di = the approach incremental delay, in sec/veh; d3 = the residual demand delay caused by queued vehi- cles at the start of the analysis period, in seconds; DF = the delay adjustment factor (function of quality of signal coordination); C = the cycle length, in seconds; G = the effective green time for the lane group, in seconds; d T X X k I X T c i = ∗ ∗ − + − + ∗ ∗ ∗ ∗( ) ( ) ( )[ ] 900 1 1 82 d C G C G C Xu = ∗ ∗ − ( )[ ] − ( ) ∗ ( )[ ]( . ) / / , .0 50 1 1 1 0 2 Min X = the volume/capacity ratio for the subject lane group; c = the capacity for the through lane group; T = the length of the analysis period, in hours; k = the actuated signal control factor; and I = the upstream signal factor. 8.3 PLANNING MODEL TO HCM LINK Many MPOs have attempted to introduce HCM techniques into their estimation of facility capacities, vehicle speeds, and delay. The FHWA120 has produced a guidebook on innova- tive techniques for accomplishing this. The guidebook cites postprocessing approaches that have been used to improve speed estimates produced by travel models. NCHRP Report 387121 presents procedures for implement- ing improved link-based speed and delay estimation proce- dures based on the HCM. These procedures use an improved speed-flow equation based on the Akcelik equation, which was used in the HCM to produce estimates of delay. Horowitz122 adapted the 1994 HCM procedures for use in estimating node-based delay for his QRS model. Node delay procedures are generally considered to be more accurate than link-based procedures, since node-based procedures take into account the demands on the conflicting approaches at an inter- section. Node-based delay procedures, however, introduce the possibility of multiple solutions to the user optimum equilib- rium traffic assignment problem. Horowitz indicates that while this can happen, it has not been a problem. The 2000 edition of the HCM123 contains the following link-based procedure for predicting mean vehicle speeds. The mean vehicle speed for the link is computed by dividing the link length by the link traversal time. The link traversal time TABLE 23 Capacities and speeds for BPR equation Practical Capacity Table for BPR Equation (VPH) Area Type Freeway Expressway Two-Way Arterial (Parking) One-Way Arterial (Parking) Centroid Connector Two-Way Arterial (No Park) CBD 1750 800 600 700 10,000 600 Fringe 1750 1000 550 550 10,000 800 Outer CBD 1750 1000 550 650 10,000 800 Rural/ Residential 1750 1100 550 900 10,000 800 Free-Flow Speed Table for BPR Equation (MPH) Area Type Freeway Expressway Two-Way Arterial (Parking) One-Way Arterial (Parking) Centroid Connector Two-Way Arterial (No Park) CBD 48 37 22 22 10 22 Fringe 48 44 25 29 15 25 Outer CBD 58 37 22 24 15 22 Rural/ Residential 67 47 28 32 15 28 CBD = central business district. MPH = miles per hour. VPH = vehicles per hour.

78 FI N A L R EP O RT (R) is computed according to the following modified Akcelik equation: Equation 35 Where: R = the segment traversal time, in hours; R0 = the segment traversal time at free-flow speed, in hours; D0 = the zero-flow control delay at signals (equals zero if no signals), in hours; DL = the segment delay between signals (equals zero if no signals), in hours; N = the number of signals on the segment (if no signals, set N = 1); T = the expected duration of the demand (typically 1 hour), in hours; x = the segment demand/capacity ratio; L = the segment length, in kilometers; and J = the calibration parameter. Note that the zero-flow control delay (D0) and segment delay (DL) terms are required because the HCM defines free- flow speed on signalized arterials to exclude delays due to signals and segment delays due to close signal spacing. The segment traversal time at free-flow speed (R0) is com- puted from the free-flow speed: Equation 36 Where: R0 = the segment traversal time at free-flow speed, in hours; L = length, in kilometers; and S0 = the segment free-flow speed, in mph. R L S0 0= R R D D N T x x J L xN T L= + = + ∗ −( ) + −( ) + ∗ ∗   0 0 2 2 2 2 0 25 1 1 16 . The zero-flow control delay for signalized intersections (if any) (D0) on the segment is computed using the following equation: Equation 37 Where: D0 = the zero-flow control delay at the signal, in hours; N = the number of signals on the segment; 3,600 = the conversion from seconds to hours; g/C = the average effective green time per cycle for sig- nals on segment (default = 0.44); C = the average cycle length for all signals on the seg- ment (default = 120), in seconds; DF = delay factor; = 0.9 for uncoordinated traffic actuated signals; = 1.0 for uncoordinated fixed-time signals; = 1.2 for coordinated signals with unfavorable progression; = 0.90 for coordinated signals with favorable pro- gression; and = 0.60 for coordinated signals with highly favorable progression. The segment delay between signals (DL) is obtained by multiplying the length of the arterial (or segment) for which a speed or travel time estimate is desired by the segment delay per kilometer shown in Table 24. The number of signals (N) on the facility segment is obvi- ous, except for when there are no signals. When there are no signals on the facility, N is still set equal to 1. This is because N is really the number of delay-causing elements on the facil- ity. Each delay-causing element on the facility adds to the overall segment delay when demand starts to approach and/or exceed capacity at that element or point. Since demand in excess of capacity must wait its turn to enter the facility seg- D N C gC0 2 3 600 2 1= ∗ ∗ −( ), DF Arterial Class: I I I II II III III III IV IV IV Free-Flow Speed (km/h) 88 80 72 72 64 56 56 48 56 48 40 Signal Spacing (km) Segment Delay Between Signals (secs/km) 0.08 n/a n/a n/a n/a n/a n/a n/a n/a n/a 66.9 75.6 0.16 n/a n/a n/a n/a n/a n/a 26.3 21.9 38.8 37.5 47.5 0.24 n/a n/a n/a n/a n/a n/a 20.1 13.1 23.2 18.8 22.5 0.32 18.1 18.1 18.1 18.1 15.6 13.8 15.7 8.8 17.0 12.5 13.1 0.40 15.0 15.0 15.0 15.0 12.5 10.1 10.7 4.4 12.0 7.5 5.6 0.48 11.9 11.9 11.9 11.9 7.5 4.5 0.0 0.0 0.0 0.0 0.0 0.64 8.8 8.8 8.8 8.8 3.8 1.3 n/a n/a n/a n/a n/a 0.80 5.0 5.0 5.0 5.0 1.9 0.1 n/a n/a n/a n/a n/a 1.60 0.0 0.0 0.0 0.0 0.0 0.0 n/a n/a n/a n/a n/a Source: 1994 Highway Capacity Manual, Table 11-4, Segment Running Time Per Mile, which is being included in the 2000 HCM, unchanged. The above table was computed by subtracting the running time if traveling at free-flow speed from running time shown in Table 11-4 and then converting the result to Standard International (SI) units. TABLE 24 Segment delay between signals (secs/km)

ment, there is always at least one delay-causing element (the segment itself) on a facility, even when there are no signals. The more signals there are on a facility, the more points there are where traffic is delayed along the way. The duration of demand (T) is usually 1 hour for a peak- hour analysis but can be longer for a peak-period analysis. The total demand for the peak period is divided by the num- ber of hours to arrive at the average hourly demand rate that is used to compute the average demand/capacity ratio (x) for the peak period. The calibration parameter J is selected so that the tra- versal time equation will predict the mean speed of traffic when demand is equal to capacity. The values for J, shown in Table 25, reproduce the mean speed at capacity predicted by the analysis procedures contained in the HCM. The data for two-lane rural highways are tentative. They are taken from recent, as yet unpublished research to update the HCM methodology for these facilities. The following equation was used to generate the J parameter values in Table 25: Equation 38 Where: Sc = the mean speed at capacity (km/h) and Sf = the mean speed when demand is zero (km/h). J S Sc f = −     1 1 2 79 FIN A L R EPO RT 8.4 MICROSIMULATION MODELS There are numerous microsimulation models, many designed for just one type of facility or one type of intersec- tion. Table 26 provides a succinct inventory of the majority of the models classified according to their target facility types and geographic coverage capacity. The following paragraphs provide descriptions of the four italicized models in the table: CORSIM, INTEGRATION, Paramics, and VisSim. 8.4.1 The CORSIM Model The CORSIM model is a dynamic microsimulation model. Vehicle movements are simulated every second, and statistics are gathered on vehicle operating mode. The CORSIM model is composed of two submodels, FREESIM for the freeway and ramps and NETSIM for the surface street system.124 8.4.1.1 FREESIM FREESIM is based upon the proposition that each vehicle will seek to travel at the driver’s desired speed in the absence of other vehicles and geometric constraints (grades, lane drops, ramp merges, and horizontal curves). The desired speed is link and driver dependent. It is determined by the mean speed coded for each link and the driver’s aggressiveness level (ran- domly selected at the time the vehicle first enters the network Facility Type Signals Per km Free-Flow Speed (km/h) Speed (km/h) at Capacity J n/a 120 86 1.05E-05 n/a 112 85 8.20E-06 n/a 104 83 5.78E-06 n/a 96 82 3.38E-06 Freeway n/a 88 80 1.29E-06 n/a 96 88 8.97E-07 n/a 88 82 7.94E-07 n/a 80 75 6.37E-07 Multilane Hwy n/a 72 67 9.84E-07 n/a 110 70 2.70E-05 n/a 100 60 4.44E-05 n/a 90 50 7.90E-05 n/a 80 40 1.56E-04 Two-Lane Hwy n/a 70 30 3.63E-04 0.333 80 53 2.21E-05 1 80 31 1.83E-04 Arterial Class I 2.5 80 15 1.30E-03 0.5 64 40 4.99E-05 1 64 28 1.96E-04 Arterial Class II 2 64 18 7.91E-04 2 56 17 8.74E-04 3 56 13 1.78E-03 Arterial Class III 4 56 10 3.18E-03 4 48 10 3.17E-03 5 48 8 5.37E-03 Arterial Class IV 6 48 7 7.11E-03 TABLE 25 Recommended traversal time J parameters

80 FI N A L R EP O RT from a default or user-specified distribution of driver types). The driver’s aggressiveness determines how much faster or slower the vehicle will travel than the coded mean desired speed for the link. 8.4.1.1.1 Car-Following Equations. The presence of other vehicles and geometric constraints trigger acceleration or deceleration events as the vehicle adjusts its speed in response to the constraints. FREESIM assumes that in the presence of other vehicles, a vehicle will attempt to maintain a constant distance behind a lead vehicle. The desired following distance is a function of the speed of the following vehicle and the difference in speeds of the lead and following vehicles: Equation 39 Where: d = the desired following distance between the back of the lead vehicle and the front of the following vehicle (feet), k = driver sensitivity for the follower vehicle, v = the speed of the follower vehicle (fps), u = the speed of the lead vehicle (fps), and b = a calibration constant. Aggressive and nonaggressive drivers will have different desired following distances according to their “driver sensi- tivity factor” (k). The driver sensitivity factor (k) is one of the main determinants of freeway capacity in FREESIM. Lower k’s result in higher capacities. d kv bk u v= + + −( )10 2 If the two vehicle speeds are different, then the desire to maintain a constant distance will result in acceleration or deceleration of the following vehicle. The vehicle’s acceler- ation (a) is determined according to the following equation (note that deceleration is the same as negative acceleration): Equation 40 Where: a = the acceleration rate (fpss), d = the distance between back of lead vehicle and front of following vehicle (feet), k = driver sensitivity for the follower vehicle, v = the speed of the follower vehicle (fps), u = the speed of the lead vehicle (fps), b = a calibration constant, and T = the duration of scanning interval (secs). The emergency requirement to avoid collisions overrides the acceleration determined by the car-following equation. The following vehicle must be able to stop safely behind the lead vehicle when the lead vehicle decelerates to a stop at the maximum allowable emergency deceleration. Similarly, the vehicle performance characteristics will limit the acceleration predicted by the car-following equation. 8.4.1.1.2 Geometric Constraints. For lane drops and ramp merges, the vehicle acceleration is computed by comparing the acceleration that would be predicted by the car-following a d v k T bk u v T kT= − − + − −( ) + 2 10 2 2 2 ( ) ( ) SIMULATION TYPE OPERATING ENVIRONMENT MODEL MAC/ ISOLATED FREEWAY RURAL MIC D/S INTERSECTIONS ARTERIALS NETWORKS FREEWAYS CORRIDORS HIGHWAYS CONTRAM MAC D DYNASMART MES D CORFLO MAC D/S CORSIM MIC S EVIPAS MIC S FLEXSYT MIC S FREQ11 MAC D INTEGRATION MIC D/S METACOR MAC D PARAMICS MIC S ROADSIM MIC S SATURN MAC D TEXAS MIC S TRAFFICQ MIC S TRARR MIC S TWOPAS MIC S WATSIM MIC S VISSIM MIC S MAC = macroscopic. MIC = microscopic. MES = mesoscopic. D = deterministic. S = stochastic. Source: Skabardonis, “Assessment of Traffic Simulation Models,” Final Report, prepared for Office of Urban Mobility, Washington State Department of Transportation, May 1999. TABLE 26 Inventory of microsimulation models

81 FIN A L R EPO RT equation (assuming that the lead vehicle is located at the lane drop or merge location and has a zero speed) with the acceler- ation rate required to come to a complete stop at the lane drop or merge location. The lower of the two acceleration rates is selected, subject to the vehicle’s performance capabilities. Changes in the user-coded mean desired speed between two sequential links will trigger acceleration events as the vehicles change speed between links. If the downstream link has a horizontal curve with a safe speed lower than the user- coded mean speed, the safe speed will override the user cod- ing. Similarly, if the downstream link has a steep grade that results in a sustainable speed for trucks that is lower than the user-coded desired mean speed, the maximum sustainable speed will override the user coding (within the range of allowable grades and speeds in FREESIM, the sustainable speeds of passenger cars are unaffected by grades). 8.4.1.1.3 Lane-Changing Criteria. Changing lanes will also trigger an acceleration or deceleration event in FREESIM. The lane-changing vehicle must either accelerate or deceler- ate in order to fit between vehicles in the new lane. Vehicles change lanes according to gap acceptance criteria. An available gap in the desired lane is evaluated accord- ing to two acceleration criteria: the required deceleration for the lane changer to safely fall in behind a lead vehicle in the new lane and the required deceleration rate for the following vehicle in the new lane to safely follow the lane changer. A vehicle makes the lane change if the required leader and fol- lower decelerations in the new lane are within the acceptable acceleration range for the driver wishing to change lanes. The acceptable acceleration varies by the type of lane change. There are three types of lane changes considered in FREESIM: mandatory, discretionary, and anticipatory lane changes. Mandatory lane changes are triggered by lane drops, ramp merges, incidents, and the necessity of exiting the free- way. Anticipatory lane changes occur upstream of an on ramp, when vehicles in the right lane of the freeway shift over one lane to avoid merging with the on-ramp traffic. All other lane changes are “discretionary” and occur when a vehicle seeks to pass a slower vehicle in front of it. The acceptable acceleration rate for lane changes is high- est for mandatory lane changes. It varies by speed for dis- cretionary and anticipatory lane changes. 8.4.1.2 NETSIM NETSIM, like FREESIM, is based upon the proposition that each vehicle will seek to travel at the driver’s desired speed in the absence of other vehicles, traffic control devices, and geometric constraints (e.g., lane drops). The desired speed is link and driver dependent. It is determined by the mean speed coded for each link and the driver’s aggressiveness level (randomly selected at the time the vehicle first enters the network from a default or user-specified distribution of driver types). The driver’s aggressiveness determines how much faster or slower the vehicle will travel than the coded mean desired speed for the link. 8.4.1.2.1 Car-Following Equations. Unlike FREESIM, NETSIM determines the acceleration of a vehicle according to a car-following equation that employs the maximum emer- gency deceleration rate for the vehicles and the driver response lag time (two factors that are missing from the FREESIM car-following formulas). In essence, in each second, NET- SIM first moves the lead vehicle to its new position and then moves the following vehicle to the closest position behind the leader that will allow the follower to avoid colliding with the leader if the leader should decide to emergency brake in the following 1-second simulation period.125 The accelera- tion for the following vehicle is determined by the change in position of the following vehicle in a 1-second time period: Equation 41 Where: Equation 42 F2 = ef (2c + 1) +2v Equation 43 Equation 44 Equation 45 Where: a = the acceleration rate (feet per square second, or fpss), ef = the maximum emergency deceleration rate for the fol- lowing vehicle (fpss), el = the maximum emergency deceleration rate for the lead vehicle (fpss), d = the distance between the front of the following vehi- cle and the back of the lead vehicle (feet), c = the driver response time lag to deceleration (seconds), v = the following vehicle speed (fps), u = the lead vehicle speed (fps), and t = the time remaining to change lanes (seconds). Equation 44 should be used only if the vehicle is changing lanes. Otherwise, T1 equals 0. Equation 45 should be used only if the vehicle is changing lanes. Otherwise, T2 equals 0. If the computed acceleration rate is greater than the maxi- mum emergency deceleration, the computed acceleration rate is reduced. T t e c t vf2 1 2 2 1 2= + + + + +[ ]( ) ( ) T e t vf u v e u u v t l 1 2 1 1 1 2= + −+ −    + − +     ( ) ( )( ) F e d c v u e e vf f l 1 2 22 1= − +[ ] + −( ) a F T F T F T t F T = +( ) + +( ) +( ) + +( ) 1 1 2 2 1 1 2 2 2 2

The speed of the following vehicle is determined from the equation of motion: vt = vt − 1 + aT Equation 46 Where: vt = the speed at time t seconds (fps); vt − 1 = the speed at time t − 1 seconds (fps); a = the acceleration rate (fpss); and T = the duration of the simulation time period (sec- onds), always 1 second in NETSIM. 8.4.1.2.2 Lane Changing. NETSIM has two types of lane changing: mandatory and discretionary. Mandatory lane changing is due to lane channelization (e.g., right-turn-only lane), lane drop, lane closure, or the need to reach the appro- priate lane to make a turn. Discretionary lane changing occurs to pass a slower-moving or stopped vehicle or to move to a lane with a shorter queue. There is no anticipatory lane changing in NETSIM to avoid a downstream queue or lane drop on a downstream link (such as in FREESIM). FREESIM vehicles can look ahead three links to line up in the correct lane to exit a freeway, while NETSIM vehicles can react only to conditions on the link on which they are located. NETSIM vehicles will not line up in the right lane for a right turn more than one block (i.e., one link) in advance of a turn. The motivation for a discretionary lane change is com- puted according to the vehicle speed and headway. The speed that would motivate a discretionary lane change and head- way is computed for the vehicle. 8.4.1.2.3 Size Limits of Software. The publicly released version 4.2 of the CORSIM software currently has limita- tions on network size (see Table 27). With only two parallel facilities, these limits preclude the use of CORSIM for any- thing larger than a 5- to 10-mile-long corridor. 8.4.1.2.4 Field Validation of NETSIM Modal Activity Forecasts. Hallmark and Guensler126 compared NETSIM- estimated, second-by-second vehicle speeds and acceleration against field measurements at 30 locations (approach stop bar and midblock) and found that at a signalized intersection, NETSIM predicted much higher fractions of hard accelera- tions (≥6 mph/s [≥9.7 kph/s]) than were measured in the field. Hallmark and Guensler also found that NETSIM under- 82 FI N A L R EP O RT estimated the variance in vehicle speeds midblock between intersections. Chundury and Wolshon127 compared NETSIM car- following equations to car-following data measured in the field and found generally reasonable correspondence between the predicted and actual car-following distances and speeds; however, they also noted that NETSIM’s predicted acceler- ation and deceleration rates were higher than observed in their field tests. 8.4.2 INTEGRATION The strengths of the INTEGRATION model are the explicit modeling of integrated freeway and arterial networks under time-varying demands and the ability to model different vehi- cle classes under various levels of traffic information provi- sion. INTEGRATION appears as the most comprehensive single model for corridor planning and ITS applications. The model includes several options for traffic assignment for sev- eral vehicle classes and incorporates the effects of traffic dynamics (i.e., queue formations) into the traffic assignment. Aggregate OD flows are converted into individual vehicle departures, with each vehicle having a unique origin and des- tination. Vehicle routings are determined through an equilib- rium traffic assignment at user-specified intervals and micro- scopically from the link travel times of earlier departures of simulated vehicles that act as dynamic vehicle probes. The car-following, lane-changing, and gap acceptance algo- rithms permit the explicit modeling of freeway and surface street traffic-flow dynamics, traffic signal control, and ramp metering. The model’s car-following algorithm is designed to satisfy the link’s macroscopic speed-flow-density relation- ships, in contrast to the rest of microscopic simulation mod- els, which use the driver’s target headway and other criteria in the car-following algorithms. INTEGRATION is not a high-end simulator for vehicle movements. It does not pro- vide the detailed modeling of driver or vehicle characteris- tics through a number of parameters found in CORSIM and other microscopic models. Thus, a number of design and con- trol options are handled approximately. Examples include complex interchange designs, detailed roadway layouts, round- abouts, pedestrians, actuated signal control, transit move- ments, and signal preemption. INTEGRATION has been used in several studies in research and practice. Most of the earlier studies involved the assessment of benefits from real-time route information and guidance (e.g., the Travtek experiment in Florida, the National ITS System Architecture Study). Following the conversion of the model into a fully microscopic one, a number of oper- ations studies have been performed concerned with street cir- culation patterns, interchange redesign, freeway operations, signal control on arterials, and impact studies. A number of software utilities exist for importing data from travel demand models (e.g., TRANPLAN and EMME/2) into Characteristic NETSIM FREESIM Nodes 250 350 Links 500 600 Vehicles present on network at any one time 10,000 10,000 Source: Table 3-5, Traffic Software Integrated System (TSIS) On-Line Help, Version 4.02, 1998 TABLE 27 Size limitations of CORSIM networks

INTEGRATION. An interface between EMME/2 and INTE- GRATION was created as part of the Seattle study conducted by Wunderlich et al.12 A utility was written by the city of Portland to import data from the GIS Map-Info database into the link data file of INTEGRATION. The model requires OD matrices per time period (15 to 30 minutes each). Usually, the OD matrix produced by the conventional trip generation and distribution planning process is not accurate enough because INTEGRATION requires OD flows per time period instead of a single peak-period OD matrix (the same is true for all the models that require OD flows as input). Numerous adjustments and iterations are required to obtain a represen- tative OD matrix for further analysis. A separate software package (QueensOD) is available to estimate OD matrices from traffic counts. Input to the model consists of a series of ASCII files and is accomplished through a text editor. There is no graphical user interface available other than the utilities to directly import data from other sources. 8.4.3 Paramics The major strength of Paramics is its software design for high performance and scalability. It provides for a seamless integrated modeling of networks consisting of freeways, arte- rials, and minor roads; various intersection types (i.e., signals, stop signs, and roundabouts) and parking garages with no limit on the network size (i.e., number of links and nodes); and the number of vehicles that can be simulated. The user interface with multiple graphical windows for data input and output provides an excellent visualization tool. A companion soft- ware (Paramics Analyzer) is available for statistical analysis of the outputs from multiple model runs. Paramics limitations include (a) lack of equilibrium traf- fic assignment, (b) limited options in modeling traveler information/guidance (i.e., the model updates the routing instructions at each intersection instead of each path because updating the routing instructions at each intersection may result in myopic travel paths with extensive turns and oscil- lations); (c) inability to explicitly model a number of control options (e.g., the National Electrical Manufacturers Associ- ation [NEMA]/170 controller and bus signal preemption from mixed lanes); and (d) limited user options in modeling inci- dents and workzones (e.g., specification of lanes occupied by the incident and rubbernecking). The latest version of the model reportedly includes several enhancements to overcome the above limitations, including improved routing algorithm, bus preemption options, and simulation of NEMA controllers. There are a number of Paramics applications, mostly in Britain, on freeway operations and impact studies. Several reports describe the model validation for British conditions. The U.S. applications are still limited. Currently, Paramics is being used to model the design and impacts of a freeway inter- change along Interstate 680 in the city of Pleasanton, Califor- nia, and alternative roundabout designs in the city of Petaluma, 83 FIN A L R EPO RT California. An evaluation and application of Paramics at the University of California, Irvine, indicated that the model accu- rately replicated traffic flow on a single freeway link, but fairly high discrepancies were found between observed and pre- dicted link flows during the simulation of the entire Irvine network. Model developers attributed this finding mostly to improper model application. Paramics includes utilities for importing existing data from travel demand models (e.g., TRANPLAN and EMME/2) and CORSIM into the model. Other utilities include importing of U.S. Geological Survey maps, AutoCAD drawings, and networks generated from geographical information systems (GIS) packages. Like INTEGRATION, Paramics requires time-dependent OD matrices. Input to the model is accomplished through a graphical user interface. Alternatively, a text editor is available for data coding. Considerable time and effort is required, even with imported network data, to correctly represent the real-world street layouts into the model. The program includes an appli- cation program interface (API) to externally specify algo- rithms and control options. This API improves the model’s flexibility, but the user has to design the control logic through “IF-THEN-ELSE” statements, which may not be straight- forward for many traffic operations staff. The model allows for changing model parameters while the simulation is run- ning and for immediately observing the changes through ani- mation, thus cutting the time required during the calibration. 8.4.4 VisSim The model’s primary area of application is detailed mod- eling of traffic flow on urban networks under different vehi- cle types, intersection geometries, and control options. The model can be used for freeway operations studies to simulate interchange configurations, merging, weaving movements, and ramp metering (including HOV bypass). VisSim is not suitable for corridor capacity improvements at the regional level or for evaluation of networkwide effects of traveler information/guidance systems in combined freeway and arte- rial networks. There are no software limits on the size of the network to be modeled, but the practical limit is networks with 60 signalized intersections. VisSim’s particular strength is to explicitly model tran- sit priority (i.e., bus preemption), signal control, pedestrian movements, stop/yield sign control, and roundabouts. Because VisSim’s coding scheme is based on links and connectors, the network physical geometry can be explicitly coded (i.e., scaled of imported AutoCAD drawings and aerial photos). Thus, vehicle paths can be explicitly traced (analogous to “railroad tracks”). This ability provides a realistic simula- tion of vehicle movements, and this realistic simulation is useful in roundabouts, other complex intersection designs, and access control designs. The model can simulate fixed- time, traffic-actuated, and adaptive real-time signal control

strategies through the interface of its signal generator pro- gram. Recent and emerging enhancements to the model include a dynamic traffic assignment algorithm and sensitiv- ity to grades so it can better model truck performance on grade-separated interchanges. VisSim provides several performance measures for autos and transit for impacts assessment. Users can define points in the network to (a) collect travel time data from the simulated vehicles or (b) set up queue counters. The model produces time-space and speed-distance diagrams along a route. Its interface with the Traffic Engineering Application Package (TEAPAC) software relates model predictions with HCM measures and level of service. Animation of vehicle move- ments (especially with background AutoCAD or aerial photos) greatly facilitates the understanding of the impacts of alterna- tive scenarios. The generated outputs on detector calls and sig- nal status would be valuable to signal operations staff working on developing and debugging logic for signal controllers. In Europe, there are several applications of VisSim primar- ily on traffic signal control and transit priority. VisSim (or its predecessor model, MISSION) has been used in Germany to study the effects of speeds limits and incidents on freeways. King County Metro is currently using VisSim on transit sig- nal priority studies in Seattle. Several studies in the United States applied VisSim on intersection and interchange design and operations, mostly through consultant projects. The results from these studies are unpublished, so detailed information on the model’s features and accuracy in replicating real- world conditions are not readily available. VisSim requires a fairly significant amount of time to code the input data. Most of the effort stems from the requirement of the link/connector scheme to represent in detail the inter- section layouts. Also, the interface and coding of detector/ signal logic for signal control (other than fixed-time plans) may require significant effort. 8.5 LINKAGES BETWEEN THE PLANNING MODEL AND THE MICROSIMULATION There are three recent examples of linkages between plan- ning model output and microsimulation input. Two of these, 84 Skabardonis and Dion et al., involve various strategies for decomposing the planning model output into modal activity data. The third, Fellendorf and Vortisch, is an actual software linkage between planning and simulation models. Skabardonis128 developed a travel demand postprocessor for predicting the percentage of vehicle-hours spent in each of four operating modes (cruise, acceleration, deceleration and idle) as a function of the facility type, the physical char- acteristics of the specific segments of the facility, and the travel demand model predicted volume/capacity (v/c) ratio for each segment. The procedure consists of a series of tables that convert v/c ranges into predictions of mean speed and operating mode fraction. Skabardonis used a series of runs of the NETSIM and Inte- grated Traffic Simulator (INTRAS) microsimulation models to develop a set of 33 tables. The microsimulation model runs were performed for 12 real-world arterial street networks (with 104 traffic signals and 334 links) and one real-world freeway (a 9.6-km section of the Interstate 880 freeway). In all cases, the results produced by each simulation model data set had been previously validated for each real-world net- work. The demand levels were then varied on each network to obtain results for a wide range of v/c ratios. The tables are stratified into four different facility types. Each facility type was then further subdivided according to the geometric and traffic control characteristics. Table 28 shows the 33 link types identified by Skabardonis. The 1985 HCM129 defines arterial classes approximately as follows: • Class I = suburban high-design facilities with multi- lane approaches, exclusive left-turn lanes, protected left- turn phasing, and free-flow speeds in the range of 64 to 72 km/h (40 to 45 mph). • Class II = urban/suburban facilities with two to three lanes per approach, some intersection with no exclusive turn lanes (i.e., pockets), and free-flow speeds of 48 to 56 km/h (30 to 35 mph). • Class III = urban streets with no exclusive turn lanes, permitted left-turn phasing, short signal spacings, and free-flow speeds of 40 to 48 km/h (25 to 30 mph). FI N A L R EP O RT Facility Types Classification Criteria Class Values Number of Link Types Section Type Basic, Merge, Weaving Number of Lanes 6, 8, 10 Freeways Design Speed 60, 70 mph 12 1985 HCM Arterial Class I, II, III Arterials Progression Quality Poor Progression, Uncoordinated, Good Progression 9 Number of Lanes 1, 2 Configuration On Ramp, Off Ramp Ramps Metering Signal Yes, No 8 Number of Lanes 1, 2 Collectors Traffic Control Signal, Stop Sign 4 TABLE 28 Link type categories

Table 29 shows an example of one of the 33 tables. Potential problems with the use of tables would be the applicability of tables created for specific conditions to other situations. Little is known about the robustness of these tables for wide application. Dion et al.130 propose a method to determine modal emis- sions from average speeds using regression equations and assumptions. The model determines the number of stops per given average speed and then the time spent in acceleration and deceleration. The regression models were developed using the Oak Ridge National Laboratory (ORNL) database. According to the authors, The fuel consumption and emission rates estimated by the model were compared against the rates estimated by MOBILE5 and the microscopic model used to develop the mesoscopic models. Specifically, fuel consumption and emis- sion estimates were compared for scenarios considering the EPA’s standard urban and highway driving cycle, as well as for a series of real-world urban arterial driving cycles. The results of these evaluations indicate that the mesoscopic model estimates fuel consumption and emission rates that are consis- tent with those produced by the underlying microscopic model in scenarios considering both EPA driving cycles, and those estimated with MOBILE5. The only exception was for the CO estimates, which were significantly lower with MOBILE5. However, it was also found that the MOBILE5 estimates fell between the minimum and maximum emission rates esti- mated by the mesoscopic model. Finally, the test performed with the real-world driving cycles indicate that the meso- scopic model could significantly overestimate fuel consump- tion and vehicle emissions in scenarios including a signifi- cant number of partial stops as a result of inaccuracies in converting partial stops into a single number of equivalent full stops. Fellendorf and Vortisch131 developed a software suite to apply a disaggregate activity-based travel demand model, a dynamic route choice, a traffic microsimulation model, and a vehicle modal emission model, using the VisSim software. According to the authors, Four separate models are integrated in one software suite to cover traffic demand, route choice, traffic flow and pollutant emissions. The traffic demand model follows a behavior- oriented, disaggregated approach. It computes the set of trip chains performed during one day in the analysis area. The dynamic route choice is calculated by an iterated simulation of the entire day. Each individual vehicle travels through the 85 FIN A L R EPO RT road network using the microscopic traffic-flow model of VISSIM. Fuel consumption and exhaust gas emissions of all vehicles in the network are determined based on dynamic engine maps. In addition, the model is capable of consider- ing additional emissions during the warm-up phase of the engine as well as evaporation emissions during parking. It is unclear whether the model includes feedback of traffic- flow results to the activity model. The authors do not recom- mend that the model be applied to large areas because of the calibration effort required, data required, and the computa- tional burden. 8.6 ASSESSMENT OF METHODS FOR ESTIMATING MODAL ACTIVITY The purpose of including a traffic operations model in the recommended NCHRP 25-21 methodology is to predict the VHT by mode of operation (i.e., cruise, idle, acceleration, and deceleration) and by speed and acceleration category. The estimates of vehicle activity are then used with modal emission factors (e.g., University of California, Riverside/ NCHRP 25-1) to produce the emissions estimates. The following sections describe and critique the possible approaches for estimating vehicle activity. The approaches can be classified into two major categories: direct modeling approaches and postprocessing techniques. 8.6.1 Direct Modeling Approach A direct modeling approach involves the simulation of the entire study area using a modeling tool that would predict the vehicle activity, as well as other performance measures of interest. This simulation can be done at either the microscopic or the mesoscopic level of detail. Microscopic models predict vehicle activity by processing individual vehicles’ trajectories. This information is obtained from the output of microscopic simulation models (e.g., COR- SIM, INTEGRATION, Paramics, and VisSim). The process involves the simulation of the entire study area using a microsimulator. This approach provides directly the required vehicle activity data. However, there are a number of issues related to accuracy, data requirements, computational aspects, and implementation into a methodology for use by MPOs. v/c Ratio Range Cruise Acceleration Deceleration Idle Mean Speed 0–0.50 55.90% 22.70% 21.20% 0.20% 57.40 mph 0.51–0.75 54.60% 23.60% 22.40% 0.40% 55.70 mph 0.76–0.90 53.20% 23.70% 22.70% 0.50% 54.30 mph 0.91 and greater 34.60% 31.00% 26.50% 7.70% 32.30 mph Note: These entries are for a design speed of 60 mph. Percent entries are percent of total vehicle-hours traveled (VHT) on link that are spent in specific operating mode. TABLE 29 Mode of operation fractions and mean speed for basic freeway sections

The state-of-the-art microsimulation tools model the move- ment and interaction of individual vehicles based on car- following, lane-changing, and queue discharge algorithms. These algorithms are based on the “fail-safe” principle; that is, they attempt to maintain a minimum safe distance head- way between successive vehicles. Often, the calculated vehi- cle speed changes are higher than the observed field condi- tions, and as a result these models tend to overestimate the magnitude and frequency of accelerations and decelerations. Microscopic simulation models are best suited for opera- tional studies for which the OD patterns or turning move- ments have been determined from other sources. They are not designed to estimate the amount or mode of travel gen- erated and distributed in the study area. Thus, this approach requires the linkage of a four-step planning model with a microscopic simulation model. The four-step planning mod- els provide the input traffic volumes and turning movements to microscopic network models, which in turn simulate the characteristics of individual vehicles and their trajectories in the network. A mesoscopic model simulates individual vehicles, but it assumes that all vehicles travel at the same average speed; that is, the model simulates traffic based on macroscopic speed- flow-density relationships. An example of such a model is DYNASMART-P, recently released by the University of Texas at Austin. The model could perform microsimulation of individual trip-maker decisions (route, departure time, and mode); traffic interactions are modeled using macroscopic speed-flow-density relationships. The advantage of using such a hybrid model is that queuing is explicitly taken into consideration in the traffic assignment process, which leads to improved estimates of traffic volumes and average speeds at reasonable computer costs. The dis- advantage is that the model cannot directly produce vehicle activity data by speed-acceleration category. The micro- or mesoscopic model must be linked in some manner to the planning model to produces the demand forecasts. One possible approach is to sequentially link the four- step planning model with a microscopic model. This process requires detailed operational data and recoding of the net- work in sufficient detail for the microscopic models. For example, a series of street segments could be coded as a single link in the planning model. Microscopic models, however, require coding at the approach/intersection level, as well as specification of the type and characteristics of traffic control. This approach is best suited for subarea analysis because at present it is computationally infeasible to simulate microscop- ically traffic conditions in large areas such as urban counties. Another major drawback of the hybrid approach is that the assigned volumes and turning movements from the planning model are often unrealistic because planning models do not consider queuing in the traffic assignment. Thus, the simula- tion results can be inaccurate. An improvement of the hybrid approach is to feed the travel times from the simulation models back into the four-step 86 assignment algorithm. This iterative process improves the accuracy of the planning model’s volume and speed outputs. The process involves challenging software development and has the same shortcomings as the sequential linkage approach regarding the data collection requirements, the network cod- ing requirements, and the application to large networks. 8.6.2 Postprocessing Techniques Postprocessing techniques involve the development and linkage of an Analysis Module that predicts vehicle-activity to the planning model (or methodology) that produces fore- casts of traffic volumes and speeds. The accuracy of this approach depends on the accuracy of the predicted traffic volumes and speeds. There is no feedback to the other mod- ules of the methodology. Either microscopic or mesoscopic models may be used. A microscopic approach involves the estimation of vehi- cle activity using widely used analytical relationships (e.g., HCM). The approach is called mesoscopic because it involves obtaining microscopic data (i.e., time in cruise, acceleration, deceleration, and idle) using macroscopic relationships. An example procedure illustrates an HCM-based analysis pro- cedure for an arterial link: 1. The researchers have the total link travel time from the volume and speed forecasts from the other modules of the methodology. 2. The researchers use the HCM analysis procedure to calculate the control delay at the signal (as proposed using default values in NCHRP Report 387).132 3. The researchers develop relationships to determine the spatial and temporal extent of the queue and the num- ber of stops. One approach by Erera et al.133 is to predict time spent in the queue from the deterministic queuing diagram used in the HCM. 4. The researchers develop typical vehicle trajectories, assuming typical values of acceleration and decelera- tion rates to determine the time spent in cruise, accel- eration, deceleration, and idle mode. A mesoscopic procedure was developed by Dion et al.134 A set of regression equations was developed to predict the aver- age speed and the number of stops along arterials. A speed- change cycle was calculated using constant rates of accelera- tion and deceleration. The emissions were determined using regression analyses. Evaluations against microscopic models indicated that this mesoscopic approach tended to signifi- cantly overestimate CO emissions and underestimate hydro- carbons (HC) and NOXemissions. The major limitation is that this approach does not adequately account for the speed slow- downs (essentially, the model predicts that most delayed vehi- cles come to a complete stop). A mesoscopic approach can be readily implemented in a postprocessor to planning models. Its implementation would FI N A L R EP O RT

be particularly straightforward by agencies that have incor- porated certain HCM procedures (e.g., node delays) into their planning modeling framework. A number of assumptions have to be made on vehicle acceleration and deceleration rates to determine the time spent in acceleration, deceleration, and idleness from the total delay. These rates depend on the vehicle and roadway characteris- tics. Issues to resolve include the following: • Should different vehicle types be considered, or should average “composite” rates be used? • What rates should be used (e.g., maximum or normal acceleration rates)? What are typical rates for acceler- ation at signalized intersections or ramp meters? For example, an EPA study reports that typical acceleration rates at traffic signals are 50 percent of the maximum acceleration rates. • How should one account for vehicle slowdowns and delays that do not involve complete stops? Should one incorporate a filtering algorithm to estimate total stops based on the amount of delay (similar to the Traffic Net- work Study Tool, Release #7 [TRANSYT-7F] model)? The procedure needs to be sensitive to TCM improve- ments. Therefore, some analytical relationships need to be incorporated between quality of traffic-flow and vehicle activ- ity. For example, the 2000 HCM states that 30 percent of the total delay at the traffic signal is acceleration/deceleration delay and the rest is stopped (idle) delay. However, research findings indicate that the acceleration/deceleration delay is lower than 30 percent of the total delay for arterial links with good progression and much higher than 30 percent on arteri- als with uncoordinated signals.135 The mesoscopic approach does not readily produce vehi- cle activity by speed-acceleration category. Instead, it pro- duces the total amount of time spent in the acceleration or 87 deceleration mode. It can be modified to produce vehicle activ- ity data by assuming a speed-acceleration relationship (e.g., constant or linear) and calculating the time spent in each speed-acceleration cell. A sampling approach using a table was developed as part of a California Air Resources Board (CARB) study that involves the stratification of the network links into distinct link types depending on facility type, design, traffic, and con- trol characteristics. The time spent in each mode is estimated from the link volume and travel time outputs from the plan- ning model and the relationships between the link types and vehicle activity. These relationships were developed through simulation in small-scale networks with the selected link types. The relationships in a form of tables are defined by the link type, v/c ratio, and free-flow speed. These relationships account for the variation of vehicle activity between facility types, undersaturated versus oversaturated conditions, and characteristics within a link type. The first step in the CARB study involved the selection of link types. The researchers recognized that it is practically impossible to capture all the variations in the characteristics of the different highway facilities into separate categories. The determination of link types considered the accuracy of the rela- tionships, time and computational resources to develop the relationships, data collection and coding requirements to implement this approach in the planning model, and the link classification schemes commonly employed in regional mod- els. Thirty-three link types were defined (see Table 30). The test sites used in the simulation experiments consisted of 12 surface street data sets (8 arterials and 4 grid networks), and 2 freeway corridors. The data were coded into the TRAF- NETSIM and INTRAS microscopic models, and several ini- tial runs were performed to verify the accuracy of the coding and the stability of the results. Next, base simulation runs were performed in each site and the outputs were processed through FIN A L R EPO RT Facility Type Classification Criteria Range of Values # Link Types Freeways Section Type Number of Lanes Design Speed (mph) Simple Section/ Merging, Weaving 6, 8, 10 60, 70 12 Arterials Arterial Class* Progression Quality I, II, III Poor, Uncoordinated, Good 9 Ramps Number of Lanes Configuration Metering/Signal 1, 2 On, Off Yes, No 8 Collectors Number of Lanes Traffic Control 1, 2 Stop Sign/Signal 4 *HCM-85 classification. 1 mile = 1.609 km. TABLE 30 Proposed link types for determining vehicle activity relationships

the software to determine vehicle activity. The process was repeated on each site by changing the input volumes to obtain performance estimates for a range of volume-to-capacity ratios. Additional simulations were performed to determine vehicle activity for scenarios not sufficiently represented in the test sites (e.g., different signalization conditions on surface streets and alternative designs on freeway segments). In addition, the trajectories of instrumented vehicles from actual floating car runs on the Interstate 880 freeway were analyzed to com- pare the measured time spent in each driving mode with the predictions of the INTRAS model. The simulation results were analyzed on each site separately for each link, for each portion of the network (e.g., arterials versus cross-streets), and for the entire network. These relationships were then incorporated in a specially written postprocessor to the MINUTP planning model. This postprocessor could be easily implemented for other four-step regional models. The postprocessor produces the following outputs: • Tables with time spent in each speed-acceleration cate- gory for speeds 0–65 mph (at 5-mph intervals) and accel- erations from −7 to 7 mph/sec (at 1-mph/sec intervals), for each link, for each facility type, and for the total network (see Figure 13). This information can be used directly to estimate emissions using modal emission factors. • A summary of the vehicle activity and traffic perfor- mance for each link, for each facility/area type, and for the total network (including VMT, delay, average speed, travel time, and the total time spent in idle, acceleration, cruise, and idle mode). This information can be used to estimate emissions based on simplified modal emission 88 factors (e.g., idle, cruise, stop-to, and stop-from) and speed-based emission rates. The postprocessor approach can be easily interfaced with a typical four-step planning model used by most MPOs to pro- duce regionwide estimates of vehicle activity data. Recoding of the network is not required except for coding additional fields in the link data file to designate the link types. The existing tables are not well suited for evaluating a num- ber of TCM strategies (e.g., ramp metering and related ITS measures), because they are based on v/c ratio and because of basic link characteristics. The tables account for improved signal timing because they include the quality of signal pro- gression on arterials as link type characteristic. Researchers can overcome the above limitation by per- forming additional simulation experiments to generate vehi- cle activity data and to develop relationships between traffic conditions (v/c), link characteristics, and control/management scenarios. 8.6.3 Subarea Microsimulation Subarea microsimulation involves using a microscopic sim- ulation to model a selected sample of links in the region and to expand, through sample enumeration, the predicted modal activity for the sample to the region as a whole. This approach requires a substantial amount of software development to ensure compatibility of the entire study area and the sample region to be microsimulated. This requirement means that por- tions of the network would be designated as “buffers” modeled macroscopically, and at the same time the volumes should be FI N A L R EP O RT Figure 13. Predicted vehicle activity.

consistent with the subarea to be simulated. The subarea net- work needs to include all the representative network conditions to permit extrapolation of the sample to the entire study area. 8.7 CONCLUSIONS The desirability of converting demand model outputs into vehicle modal activity appears clear. As will be discussed in Chapter 9 under mobile emission models, the ability to take into account modal activity effects quadruples the estimated emission benefits of signal coordination. There is some evidence in the literature, though, that cur- rent microsimulation models are not designed to produce real- 89 istic acceleration and deceleration behavior. As described in Section 8.4.1.2.4, Hallmark and Guensler, and Chundury and Wolshon found that NETSIM overpredicts hard accelera- tions and braking. A review of the car-following equations in NETSIM indicates that there is no provision for less than “emergency braking” in NETSIM. The mesoscopic model by Dion et al. or a more sophisti- cated extension of Skabardonis’s work may be worthwhile modules for the NCHRP 25-21 methodology. Directly linking a planning model to a microscopic simulation model does not currently appear advisable, given the tendency of microscopic simulation models to overpredict hard accelerations and decel- erations. Some type of filtering process will be required to reduce the more extreme modal data. FIN A L R EPO RT

90 FI N A L R EP O RT CHAPTER 9 MOBILE EMISSION MODELS This chapter discusses the candidate mobile source emis- sion models for consideration in NCHRP 25-21. 9.1 BACKGROUND—VEHICLE EMISSION PROCESSES Vehicle emissions are a function of vehicle type (light duty, heavy duty, etc.), emission controls (Type 1, Type 2, etc.), the mode of operation of the vehicle (acceleration, decel- eration, idle, cruise), the vehicle’s operating state (cold start, etc.), the amount of vehicle activity (VMT), and the simple presence of liquid fuel–powered vehicles in the air basin (diurnal and evaporative emissions). Historically, emissions from on-road vehicles have been calculated and discussed in terms of the grams of emissions per VMT, or simply grams per mile. This approach dates to the first versions of the EPA’s MOBILE emission factor model, but necessitates the incorporation of emissions into the composite grams-per-mile rates that are not strictly based on VMT. Reasonable emission inventories could be devel- oped based only on total estimated areawide VMT using this method, provided that the assumed average rates for non– VMT-dependent processes were appropriate for the area of interest. The nature of the changes in vehicle activity caused by traffic-flow improvements (including VMT, speeds and accelerations, number of trips, and time between trips) directly affect emission processes. This effect is not directly propor- tional to VMT. There are two major classes of emissions from on-road vehicles: exhaust emissions from fuel combustion and evap- orative emissions. It is perhaps simplest to describe the vari- ous types of vehicle emissions sequentially, starting with the beginning of the first trip of the day. 9.1.1 Start Emissions After being parked overnight, a vehicle is started for its first trip of the day, with engine coolant, oil, and catalytic converter all at ambient temperature. The emission control system does not reach full efficiency until the catalyst has reached operating temperature, and other engine systems are operating at nominal conditions for “hot-stabilized” opera- tion. (Note: for definitions of emission terms, see the EPA MOBILE6 User Guide, publication #EPA420-R-03-010, August 2003.) The emission rates during this period are higher for CO, VOCs, and PM. NOX emissions may be higher or lower than normal. The EPA MOBILE model assumes that a fraction of VMT occurs during “cold-start” conditions and averages the excess emissions during starts into the composite grams-per-mile emission rates. MOBILE6 incorporates a method that is used in California’s Emission Factor (EMFAC) model: separate calculation of excess emissions on a grams- per-start basis. In this approach, the magnitude of the excess start emissions is based on the time passed since the end of the last trip, referred to as the “soak time.” 9.1.2 Running Exhaust Emissions When a vehicle has reached “hot-stabilized” operating conditions, exhaust emissions are generally constant over time for given vehicle operations (i.e., following a specific second-by-second speed profile). Because excess start emis- sions can be measured only by comparing a vehicle’s emis- sions after a cold start with those occurring for the same speed profile under hot-stabilized conditions, running exhaust emissions can be assumed to begin immediately upon the beginning of a trip. A number of factors can influence run- ning exhaust emission rates, including engine load (speed, acceleration, gear, road grade, and air conditioner use, both instantaneous and time history), ambient conditions (tem- perature and humidity), and fuel formulation. Traffic-flow improvements will primarily affect speed and acceleration along specific roadway segments, but may also influence route choice (i.e., traffic volume on specific roadway segments). To the extent that these changes reduce travel time for trips, there may also be changes in the number and timing of trips, including both new trips and the addition of new interme- diate destinations to existing trips (trip chaining). Accurate treatment of the effects of these changes on running exhaust emissions will require explicit treatment of changes in the speed and acceleration profiles, expressed in terms of vehicle- seconds or vehicle-miles accumulated at different speeds and accelerations. Several emission factor models and analytical efforts, including the Comprehensive Modal Emission Model (CMEM), developed under NCHRP Project 25-11, can pro- vide running exhaust emission rates in this form.

91 FIN A L R EPO RT Heavy-duty vehicles present additional problems in esti- mating modal emissions. Emissions testing of heavy-duty vehicles is quite expensive, and databases of existing test results are much smaller than those of light-duty vehicles. NCHRP Project 25-14 is addressing known uncertainties in heavy-duty vehicle emission inventories, and some informa- tion has been developed regarding emission sensitivity to speeds and acceleration for NOX. PM and CO emissions were found to be much more sensitive to transients (e.g., hard accelerations). This work is ongoing and will be reviewed for its applicability to the evaluation of the effects of traffic-flow improvement projects. Current understanding of running exhaust particulate emis- sions is less detailed than that of other exhaust emissions because of difficulties in obtaining accurate second-by-second measurements of PM emission rates. This is particularly true for heavy-duty vehicles (especially diesels), whose PM emis- sions are high relative to the emissions of other vehicle types. CMEM (which is specific to light-duty vehicles) does not address PM emissions. EMFAC2000 (Version 2.02) produces separate start and (speed-dependent) running exhaust emis- sions, but does not directly address acceleration. The EPA PART5 model and EMFAC2000 (Version 2.02) also include on-road “fugitive” emissions of road dust and tire and brake wear. Various formulations of rate equations have been developed, with dependence on factors including vehicle weight, roadway silt loading, traffic volume, and in some cases speed. The rates from these models are consid- ered to be quite uncertain. As a result, an ongoing NCHRP project (25-18) includes an empirical investigation of exhaust and fugitive PM emission rates. Although this project should provide some improvement in the accuracy of average rates, it is unlikely to provide any significant advances in under- standing of emission rate sensitivity to vehicle operations (i.e., speed and acceleration) on specific roadways. At best, available models and data sets are expected to provide emis- sions estimates based on VMT and average speed by road- way functional class. 9.1.3 Running Evaporative Emissions While in operation, gasoline vehicles undergo various changes that influence evaporative losses of fuel. Under- hood temperatures increase, resulting in increased permeabil- ity of fuel hoses. This is potentially aggravated by the inter- nal fuel pressure of portions of the fuel delivery system in fuel-injected vehicles. Evaporative emissions occurring dur- ing vehicle operation are known as running losses and are related most closely to the total time elapsed for a trip. Quan- titative studies of running evaporative emissions (i.e., run- ning evaporative emissions) are much more limited than those of exhaust emissions, but running-loss VOC emission rates are comparable in magnitude to those of exhaust VOC. MOBILE composite grams-per-mile rates include running evaporative emissions, and EMFAC (both EMFAC7G and EMFAC2000, Version 2.02) produce grams-per-hour rates as separate emis- sion rates. These rates are known to change with the duration of trips (increasing with time since vehicle start). Vehicle speed and engine load are not considered to be important fac- tors in estimating running evaporative rates. 9.1.4 “Off-Cycle” Emissions For reasons having to do with the design and operation of current computer-controlled fuel delivery and emission con- trol systems, some vehicle operations can cause significant short-term changes in emission rates. “Power enrichment” events can occur during sustained hard accelerations or dur- ing mild accelerations on positive grades. These events are characterized by brief increases in fuel/air ratios and signifi- cant increases in CO and VOC emissions. Rapid throttle changes can also cause “enleanment” events and associated increases in NOXemissions. These effects are a consideration that may or may not be important in the estimation of run- ning exhaust emissions effects of traffic-flow improvements. Urban intersection and arterial projects are unlikely to cause such effects, but the use of ramp metering on freeways can. 9.1.5 Hot-Soak, Diurnal, and Resting Loss Emissions At the end of a trip when a vehicle is parked and switched off, all exhaust emissions cease, but evaporative emissions continue. During the first hour, evaporative emissions are referred to as “hot soak” emissions. Hot soak emissions from carburetors can be much larger than those of current fuel- injected vehicles. Subsequent to the hot soak, evaporative emissions continue from small seeps at fuel system joints and permeation through fuel lines and seals. In addition, thermal expansion of air and fuel vapors in the gas tank can cause emissions from the carbon canister. Emissions following a hot soak are referred to as either diurnal or resting evapora- tive emissions, depending on whether the ambient tempera- ture is rising or not. Diurnal emission rates (associated with rising temperatures) are higher than resting loss rates because they include the effects of expanding air and fuel vapors forced out of the fuel tank. However, on a grams-per-hour basis, both are substantially lower than running exhaust or evaporative rates. Carbon canisters can become saturated with fuel vapors, resulting in significant “breakthrough” of VOCs, but this effect is primarily associated with “multiday diurnals” from vehicles that remain parked for more than 24 hours. 9.1.6 Refueling and Carbon Dioxide Emissions Gasoline vehicle refueling causes VOC emissions because of both displacement of VOC-laden air in fuel tanks and spillage. The delivery of gasoline to gas stations also causes emissions during both tanker loading and unloading. Stage I

92 FI N A L R EP O RT (gasoline distribution) and Stage II (vehicle refueling) vapor recovery systems are in place in many areas, and these emis- sions are generally small relative to running emission rates. The rates are effectively proportional to the amount of gaso- line sold, so changes in VMT or fuel economy that are the result of traffic-flow improvements will also affect these emis- sions. Carbon dioxide emissions are also effectively propor- tional to fuel consumption and are of interest for global climate change. 9.2 MOBILE6 MOBILE6136 is the update to the MOBILE5 emission fac- tor model being developed by the U.S. EPA. MOBILE6 includes updated basic emission rates, off-cycle driving pat- terns and emissions, separation of start and running emis- sions, improved speed correction factors, and updated fleet information. Emission rates are produced for different vehicle classes and age distributions for specified calendar years. MOBILE5 produces a single set of speed-dependent running emission rates (in grams per mile), whereas MOBILE6 produces dif- ferent speed-dependent emissions for arterials and freeways, along with non–speed-dependent rates for ramps and local roadways. Both models, however, derive their speed correc- tion factors from emission tests on selected driving cycles. As a result, they cannot be reliably used to assess the effects of projects that tend to “smooth” traffic flow, or otherwise alter the speed/acceleration distributions of traffic (effec- tively, engine power demands) from those assumed in the use of the specific driving cycles. Emissions associated with “trip ends” (i.e., excess emissions during starts and evaporative emissions during hot soak, diurnal, and resting loss) can be obtained from these models to assess the effects of changes in the number of trips. 9.3 THE MOVES MODEL Motor Vehicle Emission Simulator (MOVES) is an effort to develop a set of modeling tools for the estimation of emis- sions produced by on-road and nonroad mobile sources. It is intended to include hydrocarbons, CO, NOX, PM, air toxics, and greenhouse gases at various levels of resolution needed for diverse applications of the system. 9.4 THE MEASURE MODEL The MEASURE model is a mobile emission model that estimates the production of carbon monoxide, VOCs, and NOX both spatially and temporally for a region.137 The model is GIS based and employs a vehicle mode of operations emis- sion model. The MEASURE architecture has been laid out so that MEASURE can grow in sophistication as better ana- lytical techniques become available. 9.4.1 Engine Start Emissions Engine start emissions are modeled as “puffs” that occur in the starting zone of the vehicle trip. The grams-per-start emission rates were developed from a re-analysis of the Fed- eral Test Procedure (FTP) database. Later stages of model development will incorporate improvements from studies by the California Air Resources Board (e.g., rates that are a function of soak time and modal activity and allocation of some of the start emissions to the network) and eventually a probabilistic approach with start emission rates modeled as a function of vehicle characteristics, environmental parame- ters, vehicle activity prior to soak, soak time, driver behav- ior, and modal activity during the start period. 9.4.2 On-Network Running Exhaust Emissions The MEASURE model is designed to work with three dif- ferent emission models: MOBILE5A, an aggregate modal emission model, and a load-based modal emission model still under development. This section describes the aggregate modal emission model that is currently operational within MEASURE. A hierarchical tree-based regression analysis was per- formed on the EPA (and other) vehicle emissions database to extract factors that best explained the variations in emissions between drive cycles and vehicles. A total of 700 vehicles and 4,000 vehicle-drive cycle tests were included in the database. The regression equations for predicting emissions include the following variables: acceleration rate, deceleration rate, iner- tial power surrogate, drag power surrogate, cruise speed, and percent time idling. The percent of variation explained by these equations is currently on the order of 17 percent. There is a great deal of “same vehicle” variability that cannot be explained by the model. While ingenious in conceptual approach, the use of regres- sion to develop modal emission equations from experiments that lack information on emissions by modal activity unfor- tunately results in equations of low explanatory power. The advantage of using these equations, though, is that they can be implemented in an aggregate modeling framework. One does not need to resort to traffic microsimulation models to generate the aggregate inputs required by the modal emission equations. 9.4.3 Off-Network Emissions Estimates The MEASURE architecture is set up for three optional modeling approaches similar to the “On-Network Running

Exhaust” model (MOBILE5A, an aggregate modal emission model, or a load-based model). The Off-Network Module estimates vehicle running emissions that occur off of the typ- ical travel demand model network. These emissions mostly occur on minor collector and local roads not included in travel demand models. 9.4.4 Impact of MEASURE on Emission Estimates A recent paper by Hallmark et al.138 illustrates the poten- tial impact of incorporating vehicle modal activity into the emission estimates for traffic-flow improvements. The table from the paper shows that MEASURE predicts about four times the emission reductions for signal coordination (for an individual traffic signal coordinated with adjacent upstream signals) as a traditional analysis using MOBILE5 factors. 9.5 THE NCHRP 25-11 MODAL EMISSION MODEL At the time of the NCHRP 25-21 research project, NCHRP 25-11 was in the final stages of developing a vehicle emission model that is sensitive to mode of operation.139 The model is called CMEM.140, 141 It is designed for integration with micro- simulation models (it reads CORSIM output files) but will not replace MOBILE at the regional scale. The NCHRP project sampled only Tier 0 and Tier 1 light- duty vehicles (cars and small trucks). The CMEM model that was developed as part of that project is based upon 357 sam- ple vehicles (238 cars and 119 trucks; 82 percent of the vehi- 93 cles were registered in California). Tier 2 vehicles and low- emission vehicles (LEVs) are not included in the sample or in the model. The NCHRP project was completed in Decem- ber 1999; however, the University of California, Riverside, has other sponsors to extend CMEM to vehicle types not sampled as part to the NCHRP project. CMEM uses 23 vehicle technology categories (see Table 31). CMEM is a physical, power demand model. It is com- posed of six computation modules. The Engine Power Demand Module converts second-by-second data on desired vehicle speed and acceleration into an estimate of engine power demand. The engine power demand is converted to engine speed (revolutions per minute) and air/fuel ratio by the Engine Speed and Air/Fuel Ratio Modules. The engine power demand, engine speed, and air/fuel ratio are then used to compute the rate of fuel consumption in the Fuel Rate Module. The fuel use rate and the air/fuel ratio are used to compute emissions by the Engine-out Emissions Module. The engine emissions and the air/fuel ratio are then used to compute the tailpipe emissions by the Catalyst Pass Fraction Module (see Figure 14). CMEM is available in three model forms: tables, a batch model, and a graphical user interface (GUI) model. The tables are designed to convert CORSIM microsimulation model output into second-by-second emission estimates. The tables convert vehicle-seconds by speed and acceleration category into estimates of CO, HC, NOx, and fuel consumption. These tables are considered to be less accurate than the batch model or the GUI model since they do not take into account vehicle activity prior to the second under consideration (activity dur- ing the prior second has a significant impact on emissions during the current second). FIN A L R EPO RT Type of Vehicle Technology Category No Catalyst 2-way Catalyst 3-way Catalyst, Carbureted 3-way Catalyst, Fuel Injection, > 50,000 miles, low power/weight ratio 3-way Catalyst, Fuel Injection, > 50,000 miles, high power/weight ratio 3-way Catalyst, Fuel Injection, < 50,000 miles, low power/weight ratio 3-way Catalyst, Fuel Injection, < 50,000 miles, high power/weight ratio Tier 1, > 50,000 miles, low power/weight ratio Tier 1, > 50,000 miles, high power/weight ratio Tier 1, > 50,000 miles, low power/weight ratio Normal-Emitting Cars Tier 1, > 50,000 miles, high power/weight ratio Pre-1979 (≤ 8500 lbs Gross Vehicle Weight) 1979 – 1983 (≤ 8500 lbs Gross Vehicle Weight) 1984 – 1987 (≤ 8500 lbs Gross Vehicle Weight) 1988 – 1993, ≤ 3750 lbs Loaded Vehicle Weight 1988 – 1993, > 3750 lbs Loaded Vehicle Weight Tier 1, Light-Duty Truck 2 or 3 (3751–5750 Loaded Vehicle Weight) Normal-Emitting, Light-Duty Trucks Tier 1, Light-Duty Truck 4 (6001–8500 Loaded Vehicle Weight) Runs Lean (High NOX Emitter) Runs Rich (High HC Emitter) Misfire Bad Catalyst High-Emitting Vehicles Runs Very Rich (Super High HC Emitter) TABLE 31 CMEM vehicle technology categories

94 FI N A L R EP O RT The batch model takes second-by-second vehicle trajectory data (speed and acceleration) and grade data and computes total emissions for the duration of the trajectory. The GUI model is similar to the batch model, only it is implemented in Microsoft ACCESS. CMEM is sensitive to power demand, including the increased likelihood of vehicles going into power enrichment with mild acceleration under high-speed, low-congestion con- ditions. This result parallels the NCHRP 25-6 finding that power enrichment does not appear to be a significant factor in emissions at congested intersections because of the limi- tations on acceleration by queued vehicles. 9.5.1 Vehicle Operating Data Vehicle operating data consist of second-by-second data on vehicle speed, vehicle acceleration, and grade. Also included is the second-by-second power drawn by any accessories, such as air conditioning. 9.5.2 Vehicle Characteristics Vehicle characteristics include vehicle mass (lb), engine displacement (liters), number of cylinders, coastdown power (horsepower), ratio of engine speed to vehicle speed (rpm/mph), maximum torque (foot-pounds), number of gears, gear ratios, maximum power (horsepower), air/tire drag coefficients, and drive train efficiency as a function of speed. Note that ignition timing is not a factor in the analysis. It is indirectly accounted for in the characterization of high ver- sus normal emitter vehicles (e.g., misfire) and the other vehi- cle characteristics related to engine power. Fuel type is also not directly accounted for in the analysis. It is indirectly accounted for by the engine power and the air/fuel ratio. 9.5.3 Engine Power Demand Module The Engine Power Demand Module converts vehicle speed and acceleration at time t into estimates of engine power demand, P(t), and engine torque demand, Q(t), at time t. The specific formulas employed in this module are not described in the January 2000 user’s guide. 9.5.4 Engine Speed Module The Engine Speed Module converts engine power demand, P(t), and engine torque demand, Q(t), into an estimate, N(t), of the engine revolutions per minute at time t. It is a function of the engine parameters, which give power and torque by rpm (P[rpm], Q[rpm]); the gear ratios; and the gear shift schedule. The specific formulas employed in this module are not described in the January 2000 user’s guide. 9.5.4 Air/Fuel Ratio Module The Air/Fuel Ratio Module determines the air/fuel ratio at time t. The air/fuel ratio is assumed to be stoichiometric if the power demand is less than the power threshold above which the engine goes into enrichment. The air/fuel ratio is a linear function of the power demand if the power demand goes into enrichment or is negative. The specific formulas employed in this module are not described in the January 2000 user’s guide: A/F(t) = stoichiometric ratio (if 0 ≤ P(t) ≤ Pr) Equation 47 A/F(t) = linear function of P(t) for P(t) < 0 or P(t) > Pr Equation 48 Where: A/F(t) = air/fuel ratio at time t, P(t) = engine power demand at time t, and Pr = power threshold above which engine goes into enrichment. 9.5.6 Fuel Rate Module The Fuel Rate Module computes the rate of fuel con- sumption at time t based upon the air/fuel ratio, friction loss, engine speed, displacement, and power demand: Equation 49 Equation 50 Where: FR(t) = fuel use rate at time t, A/F(t) = air/fuel ratio, k t k N t( ) ( )= ∗ + −( ) ∗ −[ ]0 1 33 2 10 4 FR( ) ( ) ( ) ( ) ( ) t A F t k t N t D P t m = ∗ ∗ ∗ ∗ +[ ] 1 44 Vehicle Operating Data Vehicle Characteristics Engine Power Demand Engine Speed Air Fuel Ratio Fuel Rate Engine Out Emissions Catalyst Pass Fraction Tailpipe Emissions Figure 14. Comprehensive modal emissions model.

95 FIN A L R EPO RT k(t) = friction loss in kilojoules per revolution of engine and per liter of displacement, k0 = friction factor, N(t) = engine speed in rpm, D = engine displacement in liters, P(t) = engine power demand at time t, and m = engine efficiency. 9.5.7 Engine-Out Emissions Module The fuel rate and the inverse of the air/fuel ratio are used to compute hot-stabilized engine-out emissions: ECO = [C0 ∗ (1 − F/A) + aCO] ∗ FR(t) Equation 51 EHC = aHC ∗ FR(t) + rHC Equation 52 Equation 53 Equation 54 Where: ECO = engine-out carbon monoxide emissions per second, EHC = engine-out hydrocarbon emissions per second, ENOX = engine-out nitrous oxide emissions per second, C0 = calibrated parameter, F/A = inverse of air/fuel ratio at time t, aCO = calibrated parameter, FR(t) = fuel use rate at time t, aHC = calibrated parameter, rHC = calibrated parameter, a2NOX = calibrated parameter, a1NOX = calibrated parameter, and FRNOXr = fuel use rate at the NOX threshold. 9.5.8 Catalytic Pass Fraction The catalytic pass fraction is determined based upon a set of 10 parameters that determine the relationships between catalyst efficiencies and engine-out emissions and fuel/air ratios under hot-stabilized conditions. The specific formulas employed in this module are not described in the January 2000 user’s guide. 9.5.9 Tailpipe Emissions The tailpipe emissions are determined from the engine-out emissions multiplied by the catalytic pass fraction: ENOX FR FR if NOXr= ∗ −( ) ≥ a t A F t 2 1 05 NOX ( ) ( ) . ENOX FR FR if NOXr= ∗ −( ) < a t A F t 1NOX ( ) ( ) .1 05 TCO(t) = ECO(t) ∗ CPFCO(t) Equation 55 THC(t) = EHC(t) ∗ CPFHC(t) Equation 56 TNOX(t) = ENOX(t) ∗ CPFNOX(t) Equation 57 Where: TCO(t) = tailpipe emissions of CO at time t, THC(t) = tailpipe emissions of HC, TNOX(t) = tailpipe emissions of NOX, ECO(t) = engine-out emission rate for CO at time t, CPFCO(t) = catalyst pass fraction for CO at time t, EHC(t) = engine-out emission rate for HC at time t, CPFHC(t) = catalyst pass fraction for HC at time t, ENOX(t) = engine-out emission rate for NOX at time t, and CPFNOX(t) = catalyst pass fraction for NOX at time t. 9.5.10 Cold-Start Emissions Cold-start emissions are estimated by applying a subset of seven model input parameters describing both cold-start catalyst performance and engine-out emissions to the above formulas. 9.6 THE NCHRP 25-6 INTERSECTION CO EMISSION MODEL At the time of the NCHRP 25-21 research, HYROAD, the Hybrid Roadway Intersection Model, was undergoing final revisions under NCHRP Project 25-6. HYROAD is a disag- gregate emission model that models the geographic disper- sion of CO emissions in the vicinity of an intersection. The vehicle demands are given to the model, which then disag- gregates the activity data by vehicle type, modal activity, and distance from the intersection. The model includes three mod- ules: a Traffic Module consisting of a modified version of NETSIM, an Emissions Module that uses regression-derived weights for MOBILE5 driving cycles to generate a speed dis- tribution that most closely matches the speed distribution for the Traffic Module, and a Dispersion Module that incorpo- rates vehicle-induced roadway turbulence on air flow and near-roadway dispersion. The Traffic and Emission Modules can be readily adapted to simulate the emission effects of changes in congestion, subject to the limitations in the under- lying emission factor model (currently MOBILE5). How- ever, the Traffic Module produces detailed speed and accel- eration distribution information, which can be directly used for emission calculations based a “modal” emission model, such as that being developed under NCHRP Project 25-11. These speed and acceleration distributions are currently dis- aggregated by location and signal phase, and further stratifi- cation by vehicle class can be readily accomplished.

HYROAD was still undergoing refinement and testing, but could provide useful results for assessing second-by-second emissions using simulated or measured vehicle speed and power profiles. In addition, statistical analyses of the vehicle testing database from this project may be used to rank the importance of changes in vehicle operating conditions aris- ing from transportation projects. 9.7 THE NCHRP 25-14 HEAVY-DUTY VEHICLE EMISSION MODEL At the time of the NCHRP 25-21 research, NCHRP Proj- ect 25-14 was producing analytical tools for predicting the effects of various transportation planning policies on heavy- duty vehicle activities and the associated emissions. The first 96 phase of this research involved inventorying heavy-duty vehicle usage patterns. 9.8 ASSESSMENT Currently, no single model addresses the range of specific emission processes in sufficient detail to capture all of the effects of traffic-flow improvement projects. At the present time, CMEM (from NCHRP Project 25-11) provides the most detailed and best tested estimates of hot-stabilized vehi- cle exhaust emissions at different speeds and accelerations. Similarly, EMFAC2000 (Version 2.02) provides the most detailed estimates of process-specific evaporative emissions and excess start emissions. FI N A L R EP O RT

FIN A L R EPO RT 97 CHAPTER 10 STRATEGIC APPROACH TO METHODOLOGY This chapter establishes the criteria to be used in develop- ment of the NCHRP 25-21 methodology, evaluates current practice against these criteria, and then evaluates various strategic approaches that might be taken by the proposed methodology to accomplish the project objectives. 10.1 PROJECT OBJECTIVE AND REQUIREMENTS FOR METHODOLOGY The research problem statement for this project gives the following objectives for this research: The objective of this research is to develop and demonstrate, in case study applications, a methodology to predict the short-term and long-term effects of corridor-level, traffic- flow improvement projects on carbon monoxide (CO), volatile organic compounds (VOCs), oxides of nitrogen (NOX), and particulate emissions (PM). The methodology should evalu- ate the magnitude, scale (such as regionwide, corridor, or local), and duration of the effects for a variety of representa- tive urbanized areas. The research problem statement goes on to identify the fol- lowing specific requirements for the methodology to be devel- oped by this project: • The methodology should be able to predict short-term (less than 5 years) and long-term (more than 10 years) air quality effects of completed traffic-flow improve- ment projects. • The methodology should evaluate those effects at the local, corridor, and regional scales. • Examples of traffic-flow improvement projects that the methodology should address are added freeway lanes, arterial widenings, intersection channelization, access management, HOV lanes, signal coordination, transit improvements, ramp metering, and park-and-ride lots. • The methodology should include consideration of the secondary effects of traffic-flow improvements, includ- ing possible changes in emissions resulting from project impacts on land use and on safety and accessibility for pedestrians, bicyclists, and transit users. • To the extent possible, the methodology should be designed to use data sources commonly available to the transportation planning process. Discussions between the NCHRP 25-21 panel and the research team during the development of the augmented work plan yielded the following additional guidance on the research objectives: • While this project will result in analytical methods for assessing short- and long-term air quality and other effects; it is desired that the research team develop a visionary approach that can be applied to the broadest range of issues and options. • The research should focus on analytical methods that can be implemented in a broad range of existing soft- ware used for travel demand modeling. The research report should describe the methods in sufficient detail for analysts to write the necessary job control state- ments, macros, or software to implement the methods. • The potential audience for this research will be broad, including both technical and nontechnical interests. The final product should become a tool for effective decision making in investing transportation resources and should provide both qualitative policy direction and a “state of the practice” methodology for analyzing emission impacts. • There is no expectation for the research team to predict pollutant concentrations or ozone formation resulting from traffic-flow improvement projects. Rather, the team is expected to use the best available emission factors and vehicle operations and activity data to estimate net changes in emissions of ozone precursors, particulates, and CO. • The use of a modal emission model is a critical element of this project. In addition, this project should include the assessment of heavy-duty vehicle emissions. • The land-use submodel need not be as geographically comprehensive or as detailed as HLFM/QRS, but it should be superior to STEP in its treatment of household and employment relocation issues. 10.2 METHODOLOGY EVALUATION CRITERIA Rephrasing the research objectives and breaking them down into questions that can be mostly answered “Yes” and “No” yields the following criteria for evaluating the ability

of existing and new methodologies to accomplish the project objectives: 1. Is the methodology suitable for predicting the long- term (10+ years), the short-term (under 5 years), or both effects of corridor-level traffic-flow improve- ment projects? 2. Is the methodology capable of accurately predicting the magnitude of impacts? 3. Is the methodology capable of predicting the geo- graphic scale of impacts (i.e., regionwide, corridor, or local)? 4. Is the methodology capable of predicting the duration of impacts? 5. Is the methodology suitable for a wide variety of urban- ized areas (small, medium, or large; technologically unsophisticated or sophisticated; data rich or poor)? 6. For which of the following types of traffic projects is the methodology suitable: major capacity increases, operational/access management improvements, or alter- native mode improvements? 7. Does the methodology take into account the secondary land-use impacts of transportation projects to the extent that the projects affect long-term motor vehicle traffic demand? 8. Does the methodology take into account the secondary safety and accessibility impacts of transportation proj- ects on pedestrians, bicyclists, and transit users to the extent that the projects affect mode share and motor vehicle use? 9. Does the methodology require data that are not com- monly available to transportation planners? 10. Can the methodology be implemented in a broad range of travel demand–modeling software? 11. Is the methodology compatible with a motor vehicle modal emission model? 12. Can the methodology assess the impacts of traffic- flow improvement projects on heavy-duty vehicle emissions? 10.3 EVALUATION OF CURRENT PRACTICE AGAINST NCHRP 25-21 OBJECTIVES Current MPO analytical tools for evaluating the air qual- ity impacts of traffic-flow improvements meet relatively few of the NCHRP 25-21 criteria, even when considering the most sophisticated MPOs (see Table 32). No MPO is set up to analyze vehicle mode of operation emissions. Very few MPOs have models to forecast the impacts of traffic-flow improvements on truck activity. None of them take into account the impacts of traffic-flow improvement projects on pedestrian/bicycle/transit accessibility and safety and their effect on vehicle activity. None of them predict short-term impacts (under 5 years), and all are oriented toward predict- ing equilibrium effects, not the duration of impacts. 98 The major shortfalls of current methodologies when com- pared with the NCHRP 25-21 criteria are as follows: • Inability to predict short-term impacts under 5 years. For such short time periods in the future, it is best to predict changes from current conditions rather than relying on a model to predict both existing and future conditions. • Inability to predict the temporal duration of impacts. Models seek equilibrium and do not consider dynamic effects that may accelerate, delay, or prevent equilibrium. • Difficulty of including land-use effects. Land-use mod- els are available, but are crude and difficult to apply, which discourages their use except for a few major tests. • Inability to predict the effects of traffic-flow improve- ments on pedestrian/bicycle/transit access and safety and their consequential impact on vehicle demand patterns. • Inability to estimate vehicle mode of operation. • Paucity of models for predicting heavy-vehicle activity effects of traffic-flow improvements. 10.4 EVALUATION OF STRATEGIC APPROACHES There are three basic strategic approaches to developing a methodology to meet the objectives of NCHRP 25-21. They correspond to three levels of detail with which to attack the research objective (see Figure 15). The first is a macroscopic, areawide sketch-planning approach. The second is a meso- scopic approach equivalent to zones and links level of aggre- gation used by current MPO modeling technology. The final approach is a microscopic approach that evaluates changes in trip making and emissions at the individual household or per- son level. These three strategic approaches are evaluated against the NCHRP 25-21 research objectives in the follow- ing sections. Table 33 summarizes the discussion. 10.5 MACROSCOPIC SKETCH-PLANNING APPROACH The macroscopic sketch-planning approach involves the development of a simple set of procedures, like SPASM, SMITE, or HERS, to predict the regional effects of traffic- flow improvements. Sketch-planning methodologies require a very simple set of input data and employ a limited set of variables (which limits the range of policy questions that can be addressed and limits the ability to take into account local variations) to arrive at estimates of regional average results. This approach has the advantage of simplicity, which makes it a tool more likely to be used by decision makers. The sketch-planning approach, however, fails to meet many of the other NCHRP 25-21 objectives that require a greater level of detail than can be provided by a macroscopic approach. Localized impacts will be difficult to predict reliably with a sketch-planning method that is not sensitive to local condi- tions. A single elasticity based on an analysis of national data FI N A L R EP O RT

is unlikely to be robust in the face of “non-national average” conditions in the project area. For example, the HERS model elasticities make assumptions of route shifting between the nonsampled segments in the system to and from the highway performance monitoring system (HPMS) sampled segments. This assumption, which is incorporated into the long- and short-term elasticities, would be difficult to modify for project- specific conditions where route shifting is expected to be greater or less than the regional average conditions across the nation. 99 Sketch-planning approaches have also been traditionally “equilibrium” oriented, rather than dynamic. They are not reli- able for predicting short-term, non-equilibrium conditions or the duration of non-equilibrium conditions. HERS has both long-term and short-term elasticities, but in both cases the traffic demand is assumed to reach an equilibrium. Duration of the effects is not predicted. The strength of sketch-planning methods is their simple data requirements. The downside to its simplicity is that it is difficult to meaningfully test microscopic effects such as FIN A L R EPO RT Criteria Small MPOs (e.g., COFG, MRCOG, and COMPASS) Medium to Large MPOs (e.g., DVRPC, CATS, and Metro Washington) Advanced-Practice MPOs (e.g., Portland, PSRC, and MTC) 1. Predicts short- and long-term effects MPOs and State DOTs and the analytical procedures they use are generally focused on long-term (10+ years) analyses. Very simple growth factor methods are used for short-term (< 5 years) analyses. 2. Magnitude of impacts Analysis procedures leave out many second-order effects. Procedures generally incorporate most second- order effects. Most sophisticated. Procedures occasionally include third-order effects. 3. Geographic scale of impacts All: local, corridor, regional. All: local, corridor, regional. All: local, corridor, regional. 4. Predicts duration of impacts Analyses have not typically been concerned with dynamics. They generally predict equilibrium impacts. 5. Suitable for small and large MPOs Procedures are suitable, by definition. Procedures are suitable, by definition. Procedures are suitable, by definition. 6. Range of projects covered Few nonhighway projects considered. All projects. All projects. Some more sophisticated pricing and land- use measures considered. 7. Includes land-use effects Not often, if at all. Manually estimated, some model use. Generally DRAM/EMPAL type of models used. Not frequently done due to computational resources required. 8. Includes ped/bike/ transit access/ safety effects No. Generally no. Most take into account transit auto and walk access, but both auto access and nonmotorized access are only crudely modeled. No safety effects. No. One MPO does include mode split effects of nonmotorized accessibility, but does not have procedure to predict how traffic-flow improvements would change accessibility. 9. Uses commonly available data Yes. Generally yes, but a few use more sophisticated data. Most use sophisticated household survey and land- use/accessibility data. 10. Implementable in commonly used software Yes. Yes. Generally yes, but some custom software are used for more sophisticated computations. 11. Compatible with modal emission model No. Models produce crude mean speed by road segment. No. Models produce mean speed by road segment. No. Models produce mean speed by road segment. 12. Heavy-duty vehicle emissions No. No. Some MPOs have truck activity models. Very rare for truck model to be sensitive to traffic-flow improvements. The numbers in the first column correspond to the numbered items in Section 10.2. Second-order effects are trip generation, distribution, mode choice, route choice, and time-of-day effects. Third-order effects are land-use effects. CATS = Chicago Area Transportation Study, COFG = Council of Fresno Governments (in Fresno, California), COMPASS = Community Planning Association of Southwest Idaho, DVRPC = Delaware Valley Regional Planning Commission, MPO = metropolitan planning organization, MRCOG = Midregion Council of Governments (in Albuquerque, New Mexico), MTC = San Francisco Metropolitan Transportation Commission. PSRC = Puget Sound Regional Commission. TABLE 32 Evaluation of current MPO modeling approaches against NCHRP 25-21 criteria

pedestrian and bicycle access effects of traffic-flow improve- ment projects. It is also impossible to meaningfully predict vehicle modal activity. Heavy-duty vehicle activity predic- tions would also require more data than typically employed in sketch-planning methods. Traffic-flow improvement projects tend to be (from the point of view of the entire air basin) a microscopic change to the transportation system. Even 20-year long-range trans- portation plan (LRTP) improvements tend to affect only a small portion of the transportation system. Minor changes to the regional transportation system are difficult to reliably detect within the expected accuracy of a sketch-planning model. The signal to noise ratio is usually too low. Sketch-planning approaches are easy to apply and easy for decision makers to use. They meet some of the project crite- ria for NCHRP 25-21 but are incompatible with modal emis- sion and heavy-duty vehicle emission models. They do not have very many policy-sensitive variables, which limits the range of policies that can be tested. Sketch-planning models have relatively little explanatory power, since the elasticities used in these models combine within them many separate effects. They cannot be calibrated to local or project-specific conditions and, as such, are not sensitive to the impacts of local contexts. 10.6 MESOSCOPIC CONVENTIONAL MODEL APPROACH A mesoscopic approach to achieving the NCHRP 25-21 research objectives would involve various improvements to 100 current conventional MPO models. The improvements can be incorporated into the models or used as a preprocessor or postprocessor to the conventional model. A postprocessor, in the style of IDAS or STEAM, is one way to enhance the analytical reliability and capabilities of conventional models to meet NCHRP 25-21 objectives. This approach requires that a regional model produce one or more trip tables and one or more loaded highway networks. The postprocessor then develops heavy-duty vehicle and vehicle modal activity data for each highway link to meet the NCHRP 25-21 emission model objectives. A preprocessor, such as the DRAM/EMPAL land-use models, might be rec- ommended to enhance the long-range forecasting of conven- tional models. MPOs may find the preprocessors or postprocessors so valuable that the MPOs will incorporate them directly into the model streams. Nevertheless, this approach has not often been the case in the past. HLFM/QRS is the only example of an inte- grated land-use and travel forecasting software. Most agencies run their land-use forecasting models separately from their travel demand models and do so only sparingly because of issues with staff training and resource requirements. The one major disadvantage of pre- or postprocessors, as far as meeting NCHRP 25-21 objectives, is that they cannot conveniently feed back their results to the land-use and household characteristics stage of the travel demand fore- casting process. The postprocessor in particular does not have access to the behavior and characteristics of individual trav- elers and households. The postprocessor works with an aggre- gated base trip table, which the conventional travel demand model produced from the baseline land-use and household FI N A L R EP O RT Le ve l o f A gg re ga t Analysis Stage io n Land Use Travel Demand System Operations Pollutant Emissions In di vi du al STEP CORSIM NCHRP 25-11 NCHRP 25-14 A re a- W id e Zo ne s UrbanSim SMITE/HERS/TCM Tools DRAM/ EMPAL Traditional 4-Step Model Mobile5/6 Macroscopic Mesoscopic Microscopic Figure 15. Strategic approaches.

FIN A L R EPO RT TABLE 33 Evaluation of general analytical approaches Criteria Macroscopic Approach (e.g., HERS, SPASM, and SMITE) Mesoscopic Approach (e.g., STEAM and IDAS) Microscopic Approach (e.g., UrbanSim, TRANSIMS, and CMEM) 1. Predicts short- and long-term effects HERS has separate long- and short-term elasticities. Others have single term. Yes. Generally better at long-term effects. Yes. Especially good at short- term effects. 2. Magnitude of Impacts See Section 2.3 regarding difficulties of developing and applying elasticities. Wide range of elasticities in literature. Regional estimates of demand changes are essentially unvalidatable with field data. Since postprocessor interfaces with regional model, it has many of the detail and accuracy advantages/disadvantages of conventional models. Since interface is “one-way” (from model to postprocessor), postprocessor does not equilibrate as well as if it were part of an improved model. More detail implies (but does not ensure) better accuracy. Sensitive to more specific factors than sketch-planning models. Portions of models are validatable against household survey and count data. 3. Geographic scale of impacts Generally limited to prediction of regional impacts of facility- specific improvements. All levels: local, corridor, regional. Ideal for local, good for corridor, very data intensive when applied at regional level. 4. Predicts duration of impacts No. Assumes demand has reached equilibrium. Not a dynamic model. No. All conventional demand models are equilibrium models. UrbanSim can predict duration, being a dynamic model. Others cannot. 5. Suitable for small and large MPOs Most suitable for small MPOs with limited resources. Also of use to supplement more formal model runs at large MPOs. Simple way to improve model accuracy (for selected model outputs) for small MPOs. Less useful for large MPOs where some postprocessor functions may already be in MPO model. Most likely to be feasible for only very large MPOs. 6. Range of projects covered All projects types, but in less detail than for other approaches. Deviation of local conditions from national average can be problem. Most all project types. Conventional models will be less adept at forecasting impacts of microscopic traffic-flow improvements. All project types. 7. Includes land- use effects Presumably included in long-term elasticity, but no identifiable step for land use alone. Can be included if the user reruns the regional model that produced the land-use forecasts and takes care of manual feedback of results. UrbanSim, yes. 8. Includes pedestrian/bike/ transit access/ safety effects No. Does not forecast impacts of traffic-flow improvement projects on access/safety. No. Some indirect effects might be included in the mode split model. Yes possible, but not currently in TRANSIMS. 9. Uses commonly available data Yes. Generally yes. Might require some specialized data to evaluate certain projects. Requires additional specialized data. 10. Implementable in commonly used software Usually implemented in a spreadsheet. Usually implemented in a custom program. Not generally implementable within standard demand model software. No. Requires specialized software. 11. Compatible with modal emission model No. Completely incompatible. Can be indirectly linked to modal emission models. Can be directly linked to modal emission models. 12. Heavy-duty vehicle emissions No. Generally incompatible. Truck activity not forecasted separately. Yes, but limited. Truck models are comparatively rare. Truck speeds not isolated from cars. Yes, but truck modal activity results have not been validated. The numbers in the first column correspond to the numbered items in Section 10.2. MPO = metropolitan planning organization. PUMS = Census public use microdata samples.

characteristics data. Except for this one limitation, a post- processor meets most of the NCHRP 25-21 objectives. One key issue with the design of a postprocessor is to iden- tify which impacts of traffic-flow improvements are to be modeled inside the conventional travel demand model and which impacts are to be modeled by the postprocessor. For example, STEAM assumes that the conventional model pro- duces demand forecasts that exclude some share of demand inducement. STEAM also assumes that the speed estimates produced by conventional models are sufficiently accurate for estimating base demand, but are not sufficiently accurate for estimating induced demand and user benefits. By way of contrast, IDAS assumes that the conventional model base demand and speed forecasts are accurate for the base case. All demand changes computed by IDAS are assumed to be the result only of changes in the supply conditions caused by ITS projects. There is no correction made for “errors” in the conventional model. The mesoscopic conventional model approach has the advantage of meeting most of the NCHRP 25-21 research objectives, although feedback and equilibration would be difficult. However, this advantage results in a tool that can- not be directly used by most decision makers, but that must be applied instead by modeling experts. The modeling experts then must transmit the results to the decision makers. This limitation may limit the use of the NCHRP 25-21 methodol- ogy by small MPOs. 10.7 MICROSCOPIC APPROACH A microscopic approach involves developing and extend- ing current analytical tools for analyzing the microscopic effects of traffic-flow improvements on individual and house- hold travel behavior and emissions. Examples of microscopic methodologies are UrbanSim, TRANSIMS, STEP, CORSIM, and CMEM. A microscopic modeling approach is the one most consis- tent with the NCHRP 25-21 requirement that the methodol- ogy employ modal activity emission rates such as those pro- duced by CMEM. Only vehicle microsimulation models produce directly the modal activity data required by CMEM. Other approximate methods are available to generate the nec- 102 essary modal activity data. The microscopic approach also follows the recommendations identified in NCHRP 8-33. An improved TRANSIMS model could potentially meet most of the NCHRP 25-21 objectives. The simple household model in TRANSIMS could be replaced with the Portland Tour-Based Model. A heavy-duty vehicle model could be added. TRANSIMS could be linked to UrbanSim to model land-use effects. It could then feed modal activity data to CMEM for modal emissions. Nevertheless, such a massively linked model would be difficult to set up and operate. This massive model would not meet the NCHRP 25-21 objectives of using commonly available data and would not be feasible for medium to small MPOs. A 100-percent microsimulation approach for NCHRP 25-21 is simply not feasible at this time. It would require a massive amount of data and computer resources. Only the very largest MPOs with sufficiently pressing air quality prob- lems and the staff resources to address them would be even remotely likely to use such a tool. 10.8 A BLENDED MICROSCOPIC/ MESOSCOPIC APPROACH The mesoscopic approach that works with conventional travel forecasting models and improves them has many advan- tages. Nevertheless, microscopic analysis has the ability to provide greater sensitivity and accuracy at critical points in the analysis. Thus, it appears that there might be some real advan- tages to blending the mesoscopic and microscopic approaches into a single methodology. Microscopic analysis would be used when sufficient data are available and greater sensitivity is needed. Mesoscopic analysis would be used when one needs to save on analytical and data resources. The recommended NCHRP 25-21 methodology employs this blended approach. It uses mesoscopic analysis to predict the impacts of traffic-flow improvements on travel time and to predict the changes in modal activity. It uses microscopic analysis to analyze household travel behavior responses and modal activity emission rates. Using this hybrid approach saves on the heavy investment that would be required to microsimulate vehicle activity at the regional level and yet pre- serves the strength of a household-level travel behavior model. FI N A L R EP O RT

103 FIN A L R EPO RT CHAPTER 11 RECOMMENDED METHODOLOGY The recommended methodology is designed to answer one fundamental question: “Will a specified traffic-flow improve- ment contribute to improved or worsened air quality locally and at the regional level, in the short term and in the long term?” Repeated exercise of the methodology on various case studies will answer the question, “Under what condi- tions will a specified traffic-flow improvement contribute to improved or worsened air quality?” The methodology is applicable to a broader range of trans- portation improvement projects besides traffic-flow improve- ments. The methodology can be applied to transit improve- ments and projects to reduce traffic capacity. 11.1 RESEARCH OBJECTIVES FOR METHODOLOGY The research statement for NCHRP 25-21 methodology lays out an aggressive set of objectives for the methodology: • The methodology must apply to a wide variety of traffic- flow improvement projects at the local, corridor, and regional scales. • The methodology must predict short-term and long-term impacts. • The methodology must consider not only primary effects but also secondary effects resulting from accessibility and safety impacts on pedestrians, bicyclists, transit users, and land use. • The methodology must have exceptional depth of detail, using a modal emission model, and include an assess- ment of heavy-duty vehicle emissions. • The methodology should be easy to use and designed to use commonly available data sources (to the extent pos- sible). It should be implementable in a broad range of existing travel demand software. The objectives go on to state that the final product of this research should become a tool for effective decision making in investing transportation resources and should provide both qualitative policy direction and a state-of-the-practice method- ology for analyzing emission impacts. The potential audience for this research will be broad, including both technical and nontechnical interests. 11.2 THEORETICAL FOUNDATION The recommended methodology proceeds from the funda- mental theoretical foundation that “nobody travels for the fun of it.” Travel is a derived demand. People travel in order to obtain the ability to participate in activities or to obtain goods that are superior to what they could have done or obtained at their original location. Even sightseers are using the trans- portation to experience a vista they could not see at home. They may say they enjoyed the drive, but what they really enjoyed was what they could see out of the window. This blanket statement excludes individuals who test their vehi- cles or are hired to drive a vehicle. Travel demand is, therefore, not VMT. Travel demand is the schedule of activities by location that travelers would like to pursue that day. In modeling parlance, it is the OD table of person trips for that day by time of day. VMT is merely the most cost-effective method (from the traveler’s point of view) for satisfying that demand. Thus, traffic-flow improvements by reducing average travel times can affect both the total demand for travel and the trav- eler’s choice of the most cost-effective means for satisfying that demand. In addition, some traffic-flow improvements do not change the average travel time but reduce the variance in travel speeds by smoothing out the traffic flow. Thus, it is possible for a traffic-flow improvement to have no effect on demand or on how that demand is satisfied on the street system and yet still have an effect on air quality by smoothing out the “stop-and-go” nature of the trip itself. Finally, a series of traffic-flow improvements can make one portion of the metropolitan area more attractive to growth and new development than older, more established parts of the region. The shifting of growth from centrally located devel- oped areas to undeveloped fringe areas can affect both demand and how it is satisfied on the transportation system. This is a very long-term impact. (Extensive transportation capacity investments in one metropolitan area can also increase the net in-migration to the region, but this effect will not be consid- ered in this research.)

104 FI N A L R EP O RT Thus, this methodology addresses four basic mechanisms by which traffic-flow improvements can influence mobile source emissions: • Operational improvements that smooth out traffic flow and thus reduce acceleration/deceleration events, • Travel time savings and losses on particular routes and modes of travel that influence the traveler’s choice of the most cost-effective means for satisfying their demand to travel, • Travel time savings that increase the total demand for travel, and • Travel time savings that increase the relative attractive- ness and therefore the growth rate of subareas in the region. The first mechanism, operational improvements, will be called the “operations” effect. Traffic-flow improvements may increase the average speed on the facility, and/or they may increase the capacity of the facility prior to affecting travel behavior. Operational improvements will also affect vehicle mode of operation activity by reducing acceleration and deceleration events. The operations effect occurs on the first day that an improvement is opened for traffic. Travelers have not yet had an opportunity to change their demand schedule in response to the travel time savings provided by the improvement. The second and third effects of traffic-flow improvements will be combined into a single “traveler behavior” effect. This effect comes in the months following opening day. As travelers become aware of the improvements, they change route, mode of travel, and departure time to take advantage of them. After the improvement has been in place for suffi- cient time for travelers to change their demand schedule (e.g., the OD table), they will take advantage of the reduced travel costs brought about by the improvement. Traveler behavior effects include changes in destination choice and trip generation (extra trips or stops along the way of a pre- existing trip). The result of the behavior effects will be to par- tially counteract the opening-day travel time improvements. It is assumed that the traveler behavior effects cannot com- pletely eliminate the opening-day travel time improvements, or there is no longer a stimulus to cause the traveler behav- ior effects. The fourth effect of traffic-flow improvements is a redis- tribution of growth (new homes and jobs) to areas within the region that are benefited by the travel time savings attribut- able to a traffic-flow improvement. This will be called the “growth redistribution” effect. It is possible that the traffic- flow improvement might also enhance the relative competi- tiveness of the entire metropolitan region for new jobs and new homes, thus influencing overall growth of the region. However, this global effect is beyond the scope of this research project and methodology. It would require a full- blown socioeconomic forecasting model plus some kind of assumption regarding the pace of traffic-flow improvements in other competing metropolitan areas of the United States, Mexico, and Canada. The research team assumed for the sake of this research project that all competing metropolitan areas have similar policies for implementing traffic-flow improvements and that, thus, a specific set of traffic-flow improvements would likely also be implemented in all metro regions. Thus, the relative competitiveness of the metro regions would be unaffected. The methodology will focus on redistribution impacts within a region, not total growth impacts for the entire region. The foundation of this methodology is that traveler behav- ior response and growth redistribution occur only if the traffic- flow improvement results in a net change in trip travel time. Thus, travel behavior responses and growth redistribution can never completely eliminate travel time savings caused by a traffic-flow improvement. Other possible effects of traffic-flow improvement not related to travel time such as operating cost improvements will be neglected. Vehicle operating cost (which can also affect travel demand) is correlated to travel time and will not be treated separately here, since the research team is not considering toll changes. The marginal effects of reduced acceleration/deceleration events on vehicle wear and tear (and thus vehicle operating costs) also will be neglected. 11.3 OUTLINE OF METHODOLOGY The recommended methodology is a blended macroscopic- microscopic approach composed of five modules: • The “HCM Assignment Module” predicts the highway travel times based upon traffic operations analysis speed- flow curves contained in the 2000 HCM. • The “Traveler Behavior Response Module” uses elas- ticities derived from the Portland Tour-Based Model to predict the impact of travel time changes on trip making by peak period and by mode of travel. • The “Growth Redistribution Module” predicts the impacts of traffic-flow improvements on growth patterns within the region. Subareas within the region that have better than average accessibility improvements will have greater than average growth rates in the region. • The “Vehicle Modal Activity Module” translates the mean speeds and volumes predicted by the previous modules into a distribution of VHT by speed category and acceleration/deceleration rate category. • The “Vehicle Emissions Module” translates the modal activity data into estimates of vehicle emissions. The methodology employs macroscopic approximations of microscopic behavior throughout each of the modules.

105 FIN A L R EPO RT The intent is to obtain a practical methodology that can be employed by a wide range of agencies while retaining as much as possible the behavioral accuracy of a microscopic analytical approach. The proposed methodology predicts the change in demand and vehicle emissions caused by traffic-flow improvements at two points in time: short term (5 to 10 years) and long term (25+ years). Figure 16 provides a flow chart overview of the methodology. The methodology requires as input: • A set of baseline travel demand tables (OD tables) for AM, PM, and off-peak periods; • A set of baseline highway and transit networks for the AM, PM, and off-peak periods; and • The proposed traffic-flow improvement characterized in terms of its impact on mean free-flow speeds and capac- ities in the baseline networks. The first round of analysis (base) assigns the baseline OD tables (by mode of travel and time period) to the baseline (no improvement) transportation networks for each time period and mode. The analysis then computes the mean speed and flow for each highway link. This link information is fed to the Vehicle Modal Activity Module, which outputs tables of vehi- cle activity (VMT) by speed and acceleration/deceleration cat- egory. The modal activity information is fed to the Vehicle Emissions Module (VOC, CO, NOX, and PM), which com- putes the vehicular emissions. The second round of analysis (short term) adds the traffic- flow improvement to the baseline network and computes new vehicle trip travel times for the improved network. The new travel times are compared to the baseline travel times to deter- mine the changes in travel times. The changed travel times are entered into the Traveler Behavior Response Module, which modifies the baseline OD tables to produce revised OD tables. The revised OD tables are assigned to the highway network to Growth Redistribution Module Growth Redistribution Module Traveler Behavior Response Module HCM Assignment Module Traffic-Flow Improvement Baseline OD Tables Baseline Network Assign Traffic Using HCM Curves HCM Assignment Module Vehicle Modal Activity Module Vehicle Modal Activity Module Vehicle Emissions Module Vehicle Emissions Module VMT VHT Assign Traffic Using HCM Curves HCM Assignment Module Results Base Short Term Long Term Assign Revised Traffic Using HCM Curves Revised HCM Assignment Module Assign Revised Traffic Using HCM Curves Revised HCM Assignment Module VMT VHT VMT VHT Traveler Behavior Response Module Traveler Behavior Response Module Assign Revised Traffic Using HCM Curves Revised HCM Assignment Module Figure 16. NCHRP 25-21 methodology.

106 FI N A L R EP O RT produce mean speed and flow for each highway link. The infor- mation is then fed to the Vehicle Modal Activity and Vehicle Emissions Modules to obtain emissions for the short term. The third round of analysis (long term) feeds the short- term results into the Growth Redistribution Module, which computes the impacts of the traffic-flow improvements on the relative growth rates of zones within the region. The revised growth rates are used to redistribute the origins and destinations of the trips in the short-term OD tables. The revised OD tables are then fed back through the Traveler Behavior Response Module one more time to obtain the mean speed and flow for each highway link. The information is then fed to the Vehicle Modal Activity and Vehicle Emissions Modules to obtain emissions for the long term. The methodology generally follows the recommendations of NCHRP 8-33. The methodology is designed to predict the changes due to the traffic-flow improvement projects. It does not predict baseline conditions. Baseline conditions (the base- line OD tables) must be input to the methodology. The methodology does not separately model the demand response of heavy-duty vehicles to traffic-flow improvements. Modeling truck demand response would require a completely separate methodology with separate data requirements. Trucks are presumed to be a fixed percentage of current and future traffic demands in this methodology. The methodology does not forecast socioeconomic changes, traffic condition changes, or emission changes that are due to factors other than traffic-flow improvements. The proposed methodology therefore must be used in conjunction with some other model for predicting future baseline conditions, usually a conventional travel demand model. 11.4 HCM ASSIGNMENT MODULE On the opening day, drivers will experience the maximum travel time savings provided by an improvement project, before it is diminished by changes in vehicle demand. The improved road section will have higher operating speeds and fewer and milder acceleration/deceleration events. If the improvement also increases peak capacity, then more vehicles will be able to pass through the improved segment during the peak hour and potentially impact downstream capacity bottlenecks. The HCM Assignment Module predicts the highway vehi- cle travel time effects of the traffic-flow improvement for a fixed level of demand. Inputting the base demand to the mod- ule is equivalent to predicting travel times for the day that a traffic-flow improvement is first opened to traffic. Travelers have not had time to adjust to the travel time savings, so, at this stage in time, there is no demand response. If future demands are input to the module, then the module will pre- dict future travel times and delays for that level of demand. This module has multiple uses in the methodology, being applied to the base case, short-term, and long-term analyses. The required inputs for the HCM Assignment Module are vehicle OD tables (by mode and time period), the base- line geometric characteristics of the regional highway net- work (facility type, free-flow speeds, capacity characteris- tics, and segment lengths), and similar geometric information for the traffic-flow improvement. The module computes the highway link operating characteristics: volume/capacity and mean speed. The module uses the 2000 HCM Chapter 30 speed-flow curves (these curves are sometimes called the Akcelik curves in the literature) and capacities to estimate the mean speed of traffic on each link of the highway network. A standard static users equilibrium (SUE) assignment of the OD table is per- formed in this module using the HCM curves for each period of the day (typically AM, PM, and off-peak). It should be noted that the travel time savings on the improved segment may be partially compensated by increased delays at downstream bottlenecks. This “downstream” effect of traffic-flow improvements are neglected by this module. (Tests showed that this downstream effect was not significant for the conditions of the PSRC, so this effect was ignored in the methodology. See the later chapter on the derivation of the HCM Assignment Module for more details.) The module computes only highway travel times for mixed- flow and HOV lanes. Transit, bicycle, and pedestrian travel times (if needed) must be computed using some standard travel demand–modeling procedure consistent with the pro- cedure used to estimate the baseline OD tables by time period for each of these non-auto modes of travel. 11.5 TRAVELER BEHAVIOR RESPONSE MODULE Travelers will adjust their demand schedule for travel in response to changes in the travel time required to reach their daily activity locations. Demand responses may include changes in trip lengths (trip distribution), number of trips (trip generation), time of day (peaking), and mode of travel (mode choice). The Traveler Behavior Response Module predicts how travel demand will react to the travel time sav- ings created by traffic-flow improvements. The module computes estimated changes in demand for each entry in the OD table for each mode of travel and each period of the day based on the estimated changes in travel times by mode and by time period. The module employs direct elasticities and cross-elasticities derived from the Port- land Tour-Based Model. (An example of a direct elasticity is the percentage change in HOV demand for each percentage change in HOV travel time. An example of a cross-elasticity is the percentage change in HOV demand for each percent- age change in SOV travel time. Cross-elasticities are also used to account for shifting of travel between peak and off- peak periods for each mode of travel.) Heavy-duty vehicles (e.g., trucks) are presumed to respond in the same manner as light-duty SOVs respond to travel changes in this module. Thus, heavy-duty vehicles are not modeled separately.

107 FIN A L R EPO RT If an MPO already has a tour-based model in place that can predict the impacts of travel time and cost changes on out-of- home trip making, time of day, and mode choice, then that model can be used in place of the simpler Traveler Behavior Response Module described here. The methodology applies to all trips made in the region. As long as through trips are included in the regional OD tables, then the methodology will adjust in response to travel time savings generated by the traffic-flow improvement proj- ects. If through trips are not included in the base regional OD table, then the methodology will be unable to adjust. 11.6 GROWTH REDISTRIBUTION MODULE Significant improvements in transportation infrastructure in one part of the urban region will impact the geographic dis- tribution of housing and job growth in the region over the very long term (25+ years). The region’s ability to retain current residents and reduce emigration may be affected. The total growth rate for the region may also be affected by changing the attractiveness of the region to migrants from other regions. This last effect, however, requires a model at the national level to properly account for migration between regions. Therefore, the overall effect on total regional growth will be excluded from the methodology. The Growth Redistribution Module will predict the very long-term impacts of localized travel time changes (caused by traffic-flow improvements) on the geographic distribution of growth in a metropolitan area. There are already several sophisticated land-use models available (such as UrbanSim) that could be used for the purpose of this module. However, these models require a great deal of specialized economic data and effort to set up for a region (which may be beyond the resources of many MPOs). Where such a model exists in a region, it can be used to predict the long-term effects. Where such a model is not available, the simpler Growth Redistribution Module is proposed for use to approximate the long-range land-use effects of traffic-flow improvements. The Growth Redistribution Module requires that a base- line 20- to 25-year forecast of land-use growth (households and employment changes) be available for the metropolitan area. This baseline forecast should have been prepared either manually or with a model taking into account accessibility changes as well as all of the other factors that commonly affect the distribution of growth within a region. A simple linear regression model is fitted to the baseline land-use fore- cast. The regression module predicts the change in the growth rate in households and employment in each zone of the region as a function of the relative change of accessibility for each zone. Although not sophisticated enough to predict actual growth, the model should be sufficient to predict how small changes in travel time accessibility can affect the pre- dicted baseline growth rate in specific zones of the region. The module presumes that total regional growth will be unaffected by traffic-flow improvements (in other words, the model will not be sensitive to the potential effects of differ- ing levels of regional traffic-flow improvements on the com- petitiveness of regions for attracting new households or jobs). The module predicts only how regional growth might be real- located from marginally less accessible zones to more acces- sible zones within the region. 11.7 VEHICLE MODAL ACTIVITY MODULE The Vehicle Modal Activity Module converts the macro- scopic vehicle activity data produced by the previous mod- ules (VHT and VMT by link, mode, and time period) to micro- scopic modal activity data (VHT by speed and acceleration category). Four tables (Uncongested Freeway, Congested Freeway, Uncongested Arterial, and Congested Arterial) con- taining percentages are used to determine the proportion of total vehicle-hours on each street and freeway segment that are spent in each speed/acceleration category. These tables were derived from microsimulation of vehicle activity on example real-world sections of freeways and arterial streets using the CORSIM model. Additional tables for other facility types and varying ITS and traffic management options can be created using COR- SIM. The creation of such tables was beyond the resources of this research project and was consequently deferred to future research. 11.8 VEHICLE EMISSION MODULE The Vehicle Emission Module converts the passenger car modal activity data into estimates of vehicular emissions. The potential impacts of traffic-flow improvements on heavy- duty vehicle and transit vehicle emissions are neglected. (The necessary information on heavy-duty vehicle emission rates by mode of operation was not available at the time of this research.) Modal emission factors from CMEM and EMFAC2000 are used to produce the emissions estimates. The primary effects of traffic-flow improvement projects are related to speeds and delay along specific corridors. The direct emissions effects include the following: • Running exhaust emissions (due to changes in vehicle speed and acceleration profiles, as well as changes in VMT due to route choice), • Running evaporative emissions (due to changes in total travel time), and • Refueling and CO2 emissions (due to changes in fuel efficiency). CMEM was used to produce running exhaust emission rates for the specified speed and acceleration frequency dis- tributions contained in the modal activity data. There are two secondary effects of traffic-flow improve- ment projects that influence emissions. First, to the extent

108 FI N A L R EP O RT that traffic-flow improvement projects reduce total travel time, there may be some increase in the number of trips made, resulting in additional start emissions. Second, both reduced travel time and increased numbers of trips alter the number and timing of hot soak, diurnal, and resting loss periods for the vehicle. Neither of these effects are included in the cur- rent version of the NCHRP 25-21 methodology. Virtually all emission rates are dependent on ambient tem- perature. This methodology uses an average summer day tem- perature profile for VOC and NOX and an average winter day for CO analyses. 11.8.1 Heavy-Duty Vehicle Emissions Changes in heavy-duty vehicular emissions due to traffic- flow improvements are not explicitly included in the method- ology for two reasons: (1) modal emission rate data were not available for heavy-duty vehicles at the time of the NCHRP 25-21 project and (2) the majority of urban area travel demand models do not explicitly model heavy-duty vehicle activity separately from light-duty vehicles. The proposed methodol- ogy consequently focuses on modeling light-duty vehicle emission changes. 11.8.2 Changes in Emission Control Technologies The emission rates used in the NCHRP 25-21 methodol- ogy can be replaced with new modal emission rates when they become available. The analyst simply substitutes new rate tables categorized by mean speed and acceleration rate for the original CMEM rates.

109 FIN A L R EPO RT CHAPTER 12 DERIVATION OF HCM ASSIGNMENT MODULE The purpose of the HCM Assignment Module was to improve current methods for estimating the travel delay effects of traffic congestion. The approach taken was to replace the conventional Bureau of Public Roads (BPR) equation method still used in many travel demand models with more up-to- date traffic operations research results contained in the 2000 HCM. The module substitutes the following HCM-based information into the SUE traffic assignment step of the travel demand model process: • Free-flow speeds by facility type and area type; • Link capacities by facility type, area type, and other char- acteristics of facility; and • HCM-based Akcelik set of speed-flow curves. A process was also developed for constraining the demands downstream of a bottleneck to the maximum flow rate of the bottleneck. However, testing showed that the improved accu- racy in the estimated delays was not worth the cost of addi- tional computer run times. More than 75 percent of the observed improvement in accuracy could be obtained simply by incorporating improved free-flow speeds, capacities, and speed-flow curves without having to incur the cost of a tripling of computer run times required to complete a peak-period assignment. This conclusion is explained in more detail later in this chapter. 12.1 HCM/AKCELIK SPEED-FLOW EQUATION The mean speed for each segment during the peak period is estimated using the following equations taken from the 2000 HCM. The mean vehicle speed for the link is computed by dividing the link length by the link traversal time. The link traversal time (R) is computed according to the following modified Akcelik equation from the HCM: Where: R = the segment traversal time (hours), R0 = the segment traversal time at free-flow speed (hours), R R D D N T x x J L x N T L= + + + ∗ − + −( ) + ∗ ∗   0 0 2 2 2 20 25 1 1 16 Equation 58 . ( ) D0 = the zero-flow control delay at signals (equals zero if no signals) (hours), DL = the segment delay between signals (equals zero if no signals) (hours), N = the number of signals on the segment (equals one if no signals), T = the expected duration of the demand (length of analy- sis period) (hours), x = the segment demand/capacity ratio, L = the segment length (miles), and J = the calibration parameter. The computation of the free-flow travel time (R0) and sig- nal delay terms (D0, DL) is explained in the following sections. The number of signals (N) on the facility segment excludes the signal at the start of the street segment (if present), because this signal should already have been counted in the upstream segment. Streets are often split into segments (links) starting and ending at signalized intersections. The counting con- vention suggested here avoids double counting of the signals located at the start and end points of each segment. When there are no signals on the facility, N is still set equal to one. This is because N is really the number of delay- causing elements on the facility. Each delay-causing element on the facility adds to the overall segment delay when demand starts to approach and/or exceed capacity at that element or point. Since demand in excess of capacity must wait its turn to enter the facility segment, there is always at least one delay-causing element (i.e., the segment itself) on a facility, even when there are no signals. The more signals there are on a facility, the more points there are where traffic is delayed along the way. This means that a bottleneck section of the facility should be coded as a single link and not arbitrarily split into sub- links. The HCM/Akcelik equation (and the standard BPR equation as well) treats each link as a potential delay-causing bottleneck on the network. Splitting one real-world bottle- neck into three hypothetical links each with the same demand would triple the estimated delay at the bottleneck. The expected duration of demand is set equal to the length of the analysis period. The segment demand/capacity ratio (x) is the ratio of the total demand for the analysis period divided by the total capacity for the period.

110 FI N A L R EP O RT The calibration parameter J is selected so that the traver- sal time equation will predict the mean speed of traffic (aver- aged over the length L of the link) when demand is equal to capacity. It is computed according to the following equation: Equation 59 Where: J = the calibration parameter, Rc = the link traversal time when demand equals capacity (hours), R0 = the free-flow speed traversal time (hours), D0 = the zero-flow control delay (hours), DL = the segment delay (hours), and L = the length of the link (miles). J R R D D L c L = − − −( )0 0 2 2 The values for J, shown in Table 34 and Table 35, repro- duce the mean segment speeds at capacity predicted by the analysis procedures contained in the 2000 HCM. Tables 34 and 35 use the following definitions of facility types from Chapter 5 of the 2000 HCM: • Freeway: A multilane, divided highway with a minimum of two lanes for the exclusive use of traffic in each direc- tion and full control of access without traffic interruption. • Multilane highway: A highway that has at least two lanes in each direction for the exclusive use of traffic, that has no control or partial control of access, and that may have periodic interruptions to flow at signalized intersections no closer than 2 miles apart. • Two-lane highway: A highway that has only one lane in each direction (with or without occasional passing lanes) for the exclusive use of traffic, that has no control SI Units Facility Type Signals Per Km Free-Flow Speed (km/h) Speed at Capacity (km/h) J Freeway n/a 120.0 85.7 1.11E-05 Freeway n/a 110.0 83.9 8.00E-06 Freeway n/a 100.0 82.1 4.75E-06 Freeway n/a 90.0 80.4 1.76E-06 Multi-Lane Hwy n/a 100.0 88.0 1.86E-06 Multi-Lane Hwy n/a 90.0 80.8 1.60E-06 Multi-Lane Hwy n/a 80.0 74.1 9.91E-07 Multi-Lane Hwy n/a 70.0 67.9 1.95E-07 Two-Lane Hwy n/a 110.0 70.0 2.70E-05 Two-Lane Hwy n/a 100.0 60.0 4.44E-05 Two-Lane Hwy n/a 90.0 50.0 7.90E-05 Two-Lane Hwy n/a 80.0 40.0 1.56E-04 Two-Lane Hwy n/a 70.0 30.0 3.63E-04 U.S. Customary Units Facility Type Signals Per mile Free-Flow Speed (mph) Speed at Capacity (mph) J Freeway n/a 75.0 53.3 2.947E-05 Freeway n/a 70.0 53.3 2.003E-05 Freeway n/a 65.0 52.2 1.423E-05 Freeway n/a 60.0 51.1 8.426E-06 Freeway n/a 55.0 50.0 3.306E-06 Multi-Lane Hwy n/a 60.0 55.0 2.296E-06 Multi-Lane Hwy n/a 55.0 51.2 1.821E-06 Multi-Lane Hwy n/a 50.0 47.5 1.108E-06 Multi-Lane Hwy n/a 45.0 42.2 2.174E-06 Two-Lane Hwy n/a 65.0 40.2 9.043E-05 Two-Lane Hwy n/a 60.0 35.2 0.0001385 Two-Lane Hwy n/a 55.0 30.2 0.0002239 Two-Lane Hwy n/a 50.0 25.2 0.0003893 Two-Lane Hwy n/a 45.0 20.2 0.0007484 TABLE 34 Recommended calibration parameter J for freeways and highways

111 FIN A L R EPO RT or partial control of access, and that may have periodic interruptions to flow at signalized intersections no closer than 2 miles apart. • Arterial: A signalized street that primarily serves through traffic and that secondarily provides access to abutting properties, with signals spaced 2 miles or less apart. Arte- rials are divided into classes according to the posted speed limit and signal density criteria shown in Table 36. 12.2 FREE-FLOW SPEEDS The segment traversal time for free-flow conditions (Ro) is computed from the free-flow speed: Equation 60 Where: R L S0 0= SI Units Facility Type Signals Per Km Free-Flow Speed (km/h) Speed at Capacity (km/h) J Arterial Class I 0.333 80 53 2.21E-05 Arterial Class I 1.000 80 31 2.04E-04 Arterial Class I 2.500 80 15 1.25E-03 Arterial Class II 0.500 64 40 4.99E-05 Arterial Class II 1.000 64 28 2.00E-04 Arterial Class II 2.000 64 18 7.91E-04 Arterial Class III 2.000 56 17 8.02E-04 Arterial Class III 3.000 56 13 1.78E-03 Arterial Class III 4.000 56 10 3.18E-03 Arterial Class IV 4.000 48 10 3.17E-03 Arterial Class IV 5.000 48 8 4.99E-03 Arterial Class IV 6.000 48 7 7.11E-03 U.S. Customary Units Facility Type Signals Per Mile Free-Flow Speed (mph) Speed at Capacity (mph) J Arterial Class I 1.000 50 33.1 2.21E-05 Arterial Class I 2.000 50 19.3 2.04E-04 Arterial Class I 4.000 50 9.6 1.25E-03 Arterial Class II 1.000 40 24.8 4.99E-05 Arterial Class II 2.000 40 17.8 2.00E-04 Arterial Class II 3.000 40 11.2 7.91E-04 Arterial Class III 3.000 35 10.9 8.02E-04 Arterial Class III 5.000 35 7.9 1.78E-03 Arterial Class III 6.000 35 6.3 3.18E-03 Arterial Class IV 6.000 30 6.1 3.17E-03 Arterial Class IV 8.000 30 5.0 4.99E-03 Arterial Class IV 10.000 30 4.3 7.11E-03 TABLE 35 Recommended calibration parameter J for signalized streets Arterial Class Posted Speed Limit (SI Units) Signal Density (SI Units) Posted Speed Limit (U.S. Customary Units) Signal Density (U.S. Customary Units) Class I 70-90 km/h 0.3-2.5 signals/km 45-55 mph 0.5-4 signals/mi. Class II 55-70 0.3-3.1 35-45 0.5-5 Class III 50-55 2.5-6.3 30-35 4-10 Class IV 40-50 2.5-12.5 25-35 4-20 Source: Chapter 15, Urban Streets, HCM. Note that there may be instances of overlaps in arterial class definitions. The analyst should consult Chapter 15 of the HCM for additional information on the identification of a specific arterial class. TABLE 36 HCM arterial class criteria

112 FI N A L R EP O RT R0 = free-flow traversal time (hours), L = length (miles), and S0 = the segment free-flow speed (mph). The free-flow speed is the mean speed of traffic when demand is so low that changes in demand do not affect the mean speed of traffic on the segment. For freeways and multi- lane highways, free flow is the mean speed observed when volumes are less than 1,300 vehicles per hour per lane. For signalized streets, the free-flow speed is the maximum mean speed of traffic obtained at any point between signalized intersections for low-volume conditions. The mean speed is computed as the sum of the travel times to traverse the length of the segment, divided into the length of the segment times the number of vehicles in the sample. The following linear equations can be used to estimate free- flow speed based on the posted speed limit for arterials, free- ways, and highways (source: NCHRP Report 387: Planning Techniques to Estimate Speeds and Service Volumes for Plan- ning Applications, Transportation Research Board, Washing- ton, D.C., 1997): For posted speed limits of 50 mph or greater: FFS = 0.88 ∗ PSL + 14 Equation 61 For posted speed limits of less than 50 mph: FFS = 0.79 ∗ PSL + 12 Equation 62 Where: FFS = free-flow speed (mph) and PSL = posted speed limit (mph). 12.3 CAPACITIES Highway link capacities are estimated using the proce- dures contained in the 2000 HCM. The following subsec- tions summarize the information contained in Chapter 30 of the HCM. 12.3.1 Freeways, Multilane Highways, and Two-Lane Highways The following equation is used to compute the capacity of a freeway or highway link at its critical point. The critical point is the point on the link with the lowest throughput capacity. c = Q ∗ N ∗ Fhv ∗ Fp ∗ Fg ∗ PHF Equation 63 Where: c = capacity (vph), Q = the passenger car equivalent (p.c.e.) capacity per hour per lane, N = number of through lanes (ignore auxiliary and “exit only” lanes), Fhv = heavy-vehicle adjustment factor, Fp = driver population adjustment factor, Fg = grade adjustment factor, and PHF = peak-hour factor. Table 37 provides the HCM-recommended passenger car equivalent capacities per lane (Q). See the HCM for appro- priate values for the adjustment factors. 12.3.2 Arterials The capacity of an arterial is determined by examining the through-movement capacity at each signal-controlled inter- section on the arterial link. The intersection with the lowest through capacity determines the overall capacity of the arte- rial link. The following equation is used to compute the one- direction through capacity at each signal: c = S0 ∗ N ∗ fw ∗ fhv ∗ Fg ∗ fp ∗ fbb ∗ fa ∗ fLU ∗ fLT ∗ fRT ∗ fLpb ∗ fRpb ∗ PHF ∗ g/C Equation 64 Where: c = capacity (vph), S0 = ideal saturation flow rate = 1,900 vehicles per hour of green per lane, N = number of lanes, fw = lane-width adjustment factor, PCE Capacity (passenger cars per hour per lane) Free-Flow Speed Freeways Multilane Hwys Two-Lane Hwys 75 mph (112 km/h) 2400 70 mph (104 km/h) 2350 65 mph (96 km/h) 2300 2200 1700 60 mph (88 km/h) 2250 2100 1700 55 mph (80 km/h) 2000 1700 50 mph (70 km/h) 1900 1700 TABLE 37 Passenger car equivalent (PCE) capacities for freeways and highways

fhv = heavy-vehicle adjustment factor, Fg = grade adjustment factor, fp = on-street parking crossing adjustment factor, fbb = local bus adjustment factor, fa = central business district adjustment factor, fLU = lane-use adjustment factor, fLT = left-turn adjustment factor, fRT = right-turn adjustment factor, fLpb = pedestrian/bicycle blockage of left-turn factor, fRpb = pedestrian/bicycle blockage of right-turn factor, PHF = peak-hour factor, and g/C = ratio of effective green time per cycle. See the HCM for appropriate values for the adjustment factors. 12.4 SIGNAL DATA REQUIRED BY HCM/AKCELIK The zero-flow control delay and the between-signal delay are required to estimate speeds for signalized arterial streets. The zero-flow control delay (D0) is computed as follows: Equation 65 Where: D0 = the zero-flow control delay at the signal (hours); N = maximum of one, or the number of signals on the segment; D N C gC0 2 3 600 2 1= ∗ ∗ −( ), DF 113 3,600 = conversion from seconds to hours; g/C = average effective green time per cycle for signals on segment; C = average cycle length for all signals on the segment (seconds); and DF = delay factor, = 0.9 for uncoordinated traffic-actuated signals, = 1.0 for uncoordinated fixed-time signals, = 1.2 for coordinated signals with unfavorable progression, = 0.9 for coordinated signals with favorable progres- sion, and = 0.6 for coordinated signals with highly favorable progression. If the ratio of green time per cycle for the arterial through movement is not known, a default value of 0.44 can be used. Similarly, if the signal cycle length is not known, then a default value of 120 seconds can be used. A survey of local average signal cycle lengths by area type (e.g., downtown, suburban, and rural) may be desirable to establish appropri- ate local default values. The segment delay between signals (DL) is estimated as follows: Equation 66 Where: L = The length of the segment and dL = The delay per mile, given in Table 38. D L dL L= ∗ 60 FIN A L R EPO RT Source: 2000 HCM, Exhibit 15-3, Segment Running Time Per Mile. Table computed by subtracting running time if traveling at free-flow speed from running time shown in exhibit. Segment Delay (secs/mile) Arterial Class: I I I II II II III III IV IV Free-Flow Speed (mph) 55 50 45 45 40 35 35 30 35 30 signal spacing (miles) 0.05 107 0.10 42 35 62 60 0.15 32 21 37 30 0.20 29 25 22 25 14 27 20 0.25 32 28 24 24 20 16 17 7 19 12 0.30 27 23 19 19 12 7 0.40 17 14 14 14 6 2 0.50 8 6 8 8 3 0 0 0 0 0 0 01.00 Segment Delay (secs/km) Arterial Class: I I I II II II III III IV IV Free-Flow Speed (km/h) 88 80 72 72 64 56 56 48 56 48 signal spacing (km) 0.08 n/a n/a n/a n/a n/a n/a n/a n/a n/a 66.9 0.16 n/a n/a n/a n/a n/a n/a 26.3 21.9 38.8 37.5 0.24 n/a n/a n/a n/a n/a n/a 20.1 13.1 23.2 18.8 0.32 n/a n/a n/a 18.1 15.6 13.8 15.7 8.8 17.0 12.5 0.40 19.7 17.5 15.0 15.0 12.5 10.1 10.7 4.4 12.0 7.5 0.48 16.6 14.4 11.9 11.9 7.5 4.5 n/a n/a n/a n/a 0.64 10.3 8.8 8.8 8.8 3.8 1.3 n/a n/a n/a n/a 0.80 4.7 3.8 5.0 5.0 1.9 0.0 n/a n/a n/a n/a 1.60 0.0 0.0 0.0 0.0 0.0 0.0 n/a n/a n/a n/a TABLE 38 Segment delay between signals

12.5 CONSTRAINING DEMAND DOWNSTREAM OF BOTTLENECKS One criticism of conventional travel demand model prac- tice has been that all demand is loaded on the highway net- work, even if capacity bottlenecks might prevent the demand from getting through the network before the end of the analy- sis period. The demand for links downstream of a bottleneck is not reduced to account for demand stored at the bottleneck itself. If demand is greater than capacity at a bottleneck, the driv- ers must wait in line at the bottleneck until it is their turn to go through bottleneck. Figure 17 illustrates this process. A total of 5,000 vehicles per hour wish to go from point “A” to point “B.” They proceed from point “A” to “B” until they hit Bottleneck 1. This bottleneck can carry only 4,000 vehicles per hour, so 1,000 must wait until the next hour before they can proceed. Meanwhile, the 4,000 that can pass through Bottleneck 1 proceed until they hit Bottleneck 2. Bottleneck 2 can carry only 2,500, so 1,500 must wait until the next hour before they can proceed. The remaining 2,500 vehicles are actually able to get from point “A” to point “B” in the first hour. Excess demand is in essence stored within the network. It was originally proposed that the HCM Assignment Module identify the magnitude of this excess demand, “store it,” and carry it over to the next hour. Each peak period is divided into 1-hour time slices. Only the demand that could physically get through all the network bottlenecks within 1 hour would be assigned to the highway network during each 1-hour time slice. The excess demand that could not be served in the first hour is carried over to the next hour. 114 Results for all the 1-hour time slices are then summed to obtain the overall peak-period results. This process is illus- trated in Figure 18. Following this approach, the module would then be able to identify how capacity increases at one bottleneck affect total origin-to-destination travel times for each hour within the peak period. The bottleneck constraint option was tested on the Seattle (PSRC) data set. The bottleneck constraint option required 11 hours of computer time to complete two SUE assignments, one for the 3-hour AM peak-period and one for the 3-hour PM peak-period. In contrast, applying the HCM Assignment FI N A L R EP O RT A B 1 5000 -1000 4000 -1500 2500 2 Cap = 2500 Cap = 4000 Figure 17. Impact of bottlenecks on hourly OD flows. V = demand. c = capacity. First-Hour OD Third-Hour OD Plus Unserved OD (2) Second-Hour OD Plus Unserved OD (1) Equilibrium Assignment Identify Segments where V > c Find Constrained OD so that all V ≤ c Shift Unserved OD to Next Hour Compute Travel Time Using Constrained OD. Add Delay for Unserved OD Equilibrium Assignment Identify Segments where V > c Find Constrained OD so that all V ≤ c Equilibrium Assignment Report Three 1-hour Travel Time Skim Tables for Peak Period Compute Travel Time Shift Unserved OD to Next Hour Compute Travel Time Using Constrained OD. Add Delay for Unserved OD Report Segment V/c Ratios and Speeds for Each Hour of Peak Period Figure 18. Constraining demand for bottleneck effects.

Module without the bottleneck constraint option required only 1 hour of computer time. Table 39 compares the predicted VMT and VHT impacts of a hypothetical 20-year RTP when estimated using differ- ent traffic assignment approaches: • SUE using standard BPR speed-flow equations, • SUE using the HCM/Akcelik speed-flow equations, and • SUE using the HCM/Akcelik speed-flow equations plus downstream bottleneck constraints. In each case, the same future-year OD trip tables were loaded on the networks. Even so, there are significant differ- ences in the predicted impacts of the RTP improvements between SUE assignments using the standard BPR equations and the HCM/Akcelik equations. The standard BPR equa- tions predict a 5-percent increase in VMT and a 9-percent 115 reduction in VHT for the RTP improvements. The HCM/ Akcelik equations, however, predict a 0.5-percent decrease in VMT and a 48-percent decrease in VHT. Adding the bottle- neck constraint option to the assignment process results in only modest changes to the HCM/Akcelik results (a 0.3- percent increase in VMT and a 52-percent decrease in VHT). Although the same demand tables were assigned to the same highway networks each time, the differences in the predicted VMT are due to the differences in the congestion delays pre- dicted by the BPR and HCM equations. The higher congestion delays predicted by the HCM equations cause more traffic to take roundabout routes than predicted by the BPR equations. The result is that the BPR equations predict less VMT than the HCM equations predict for congested conditions. These results suggest that most of the advantages of the HCM/Akcelik equations can be obtained without having to incorporate the bottleneck constraints. FIN A L R EPO RT Assignment Technique Scenario Peak Period Vehicle-Miles Traveled Vehicle-Hours Traveled Speed (mph) RTP Improvements AM 19,468,400 643,040 30.3 PM 27,670,300 977,495 28.3 Total 47,138,700 1,620,535 29.1 No Improvements AM 18,459,100 696,708 26.5 PM 26,472,000 1,088,500 24.3 Total 44,931,100 1,785,208 25.2 Difference 2,207,600 -164,673 3.9 Standard BPR % Difference +4.9% -9.2% RTP Improvements AM 19,618,792 777,708 25.2 PM 28,400,816 1,625,331 17.5 Total 48,019,608 2,403,039 20.0 No Improvements AM 19,669,446 1,297,723 15.2 PM 28,597,942 3,344,435 8.6 Total 48,267,388 4,642,158 10.4 Difference -247,780 -2,239,119 9.6 HCM/Akcelik % Difference -0.5% -48.2% RTP Improvements AM 19,666,651 994,131 19.8 PM 28,326,698 2,019,930 14.0 Total 47,993,349 3,014,061 15.9 No Improvements AM 19,571,285 1,840,165 10.6 PM 28,264,870 4,419,772 6.4 Total 47,836,155 6,259,937 7.6 Difference 157,194 -3,245,876 8.3 HCM/Akcelik With Bottleneck Constraints % Difference +0.3% -51.9% TABLE 39 Impacts of 20-year RTP as estimated using different assignment techniques

116 FI N A L R EP O RT CHAPTER 13 DERIVATION OF TRAVEL BEHAVIOR RESPONSE MODULE This chapter discusses the derivation of the Travel Behav- ior Response Module for this study. The Portland Tour-Based Model was selected as the basis for the travel behavior response model because of its ability to predict both modal and temporal shifts in travel behavior as well as predict the impact on overall out-of-the-home trip making. 13.1 OVERVIEW OF PORTLAND TOUR-BASED MODEL The Portland Tour-Based Model was originally developed as part of a project to analyze road pricing policy alternatives in Portland. (A full description of the Portland tour-based model is given by Mark Bradley Research and Consulting, A System of Activity-Based Models for Portland, Oregon, Washington, D.C.: Travel Model Improvement Program, U.S. Dept. of Transportation, Report No. DOT-T-99-02, U.S. Environmental Protection Agency, 1998. Consult this refer- ence for details on model structure and coefficients.) An overview of the Portland model within a broader context is shown in Figure 19; the Portland Tour-Based Model proper consists of the blocks within the large rectangle. A more detailed look at the Portland model is given in Figure 20, which shows information flows between the dif- ferent submodels. The model system is designed to predict the following: • A full-day activity pattern (primary activity and, for tour activities, subtour pattern), • Times of day (outbound, inbound) for home-based tours, • Primary mode and destination, • Work-based subtours, and • Location of intermediate stops. The Portland model is a conceptual descendant of Greig Harvey’s STEP model, with considerable additional detail. A description of the STEP model and the theory behind the model is presented by Elizabeth Deakin and Greig Harvey in Transportation Pricing Strategies for California: An Assess- ment of Congestion, Emissions, Energy and Equity Impacts: Final Report, prepared for the California Air Resources Board, 1996. The Portland model has several features that distin- guish it from traditional four-step travel models: • Simultaneous modeling of trip generation, time of day, mode choice, and destination choice. Utilities of lower- level choices (e.g., mode and destination choice) are incorporated into the utilities of higher-level choices (e.g., time of day and primary activity pattern). • Application of the model to individual travelers. This approach, known as sample enumeration when applied to travel survey data and more generally as microsimu- lation, is considered to be at the forefront of the cur- rent state of the art in travel modeling. Microsimulation allows the incorporation of detailed household and per- son characteristics that can significantly affect travel behavior, such as presence of children in the household and competition for available cars in the household for different trip purposes. • Use of a synthetic sample to develop the base population to which the model is applied. This approach provides the model with a sufficiently large population so that com- plete trip tables can be produced. Sample enumeration approaches based only on travel surveys generally pro- duce results at a much larger scale, such as superdistrict- to-superdistrict trip movements. The synthetic sam- pling approach has been used for over 25 years. One early application was the development of a database for research on discrete-choice models. See Gerald Duguay, Woo Jung, and Daniel McFadden, “SYNSAM: A Meth- odology for Synthesizing Household Transportation Sur- vey Data,” Berkeley: Urban Travel Demand Forecasting Project, Working paper no. 7618, September 1976. Syn- thetic sampling is currently used in the TRANSIMS model and in the current version of the STEP model. An additional advantage of the synthetic sampling approach is that it enables disaggregation of benefit and cost esti- mates by socioeconomic category, which is often a sig- nificant issue in transportation policy analysis. 13.2 MICROSIMULATION MODEL IMPLEMENTATION 13.2.1 Overall Design The Portland model required nearly 2 years for estimation and implementation. For this study, time constraints limited

117 FIN A L R EPO RT what could be done for model estimation and implementa- tion. Hence, the following guided this study’s implementa- tion of the model: • Model estimation. The Portland model was applied “as- is” to Seattle without re-estimation of the model steps. Where necessary, choice-specific constants would be adjusted so that aggregate model outputs would be suf- ficiently close to observed values. • Model components. The model for this study was only used to estimate primary destination, primary mode, and primary activity time of day. The intermediate stop mod- els were not implemented. (In the Portland model, the intermediate stop models were applied as aggregate adjustments to trip tables developed from the primary tour choice models.) • Model variables. A number of variables required by the Portland model were not readily available from current data maintained in Seattle. Because these variables did not change from alternative to alternative, they were assigned default values for all model runs. • Expansion of synthetic population sample. The syn- thetic sample for the Portland model was developed from forecasts of numbers of households by household size, income, and age of head of household. For appli- cation to Seattle, the available socioeconomic forecasts by household type included percentage multifamily households and households by income quartile. Hence, these forecasts were used to develop the synthetic sam- ple for Seattle. • Modes. Two modes in the Portland model—light rail with walk access and light rail with auto access—were not available in Seattle. They were dropped from the mode/destination choice model. • Time of day. The Portland model provides forecasts for five times of day: early, AM peak, midday, PM peak, and late. The PSRC model network is based on two time peri- ods: AM peak and daily. Some network data on auto travel are available for PM peak. The research team there- fore made the following assumptions: (1) model results would be reported for three time periods—AM peak, PM peak, and off peak; (2) network levels of service for early, midday, and late time periods would be approximated by daily level-of-service values; and (3) PM peak transit level-of-service matrices were approximated by taking transposes of the corresponding AM peak matrices. • Application. As was done in Portland, the original intent was to apply the model by pivoting the model results around a base trip table to develop forecasts for the transportation alternatives. Although the scope of the study required the above sim- plifications, the research team also saw the opportunity to base the model implementation on modern software engi- neering methods. • The model was implemented using object-oriented soft- ware engineering analysis and design methods. This greatly increased the verifiability, maintainability, and extensibility of the model. • A high-precision, random number generator was used, as discussed below. This was done in order to minimize the potential for serial correlations between sets of ran- dom numbers to confound the model outputs. • Simulation in the original Portland model is carried out as a “one-shot” process with a fixed synthetic sample. That approach works only if one can assume that a cross- sectional synthetic population sample will produce the same results as those from repeated population sampling, which is a very strong (and probably unwarranted) assumption. (This is roughly analogous to assuming that a stochastic process is ergodic, i.e., the cross-sectional ensemble average is equal to the average over time. For many physical processes, ergodicity is a strong, but rea- sonable assumption. For socioeconomic processes, it is less clear that ergodicity holds. See Julius S. Bendat and Allan G. Piersol, Random Data: Analysis And Measure- Input • Employment by sector by TAZ • Sample of households • Modal LOS measures Household-based tour model • Primary activity • Secondary tour choice • Time-of-day choice • Mode/destination choice Work-based subtour model Intermediate stop location model (car driver tours only) Decompose tours to trips Output — OD trip matrices by: • Mode • Time of day • Income group Network model (trip assignment by mode and time period) LOS = level of service. TAZ = traffic analysis zone. Figure 19. Portland Tour-Based Model system flow chart.

118 FI N A L R EP O RT ment Procedures, 2nd ed., rev. and expanded, New York: Wiley, 1986.) The model implementation for this study carries out repeated sampling of the population dynami- cally as the program is running. • The Portland model kept track of all combinations of choice probabilities for a given individual in the syn- thetic population and aggregated the results by all of these combinations. The model implementation for this study uses the choice probabilities from the different nests in the model to provide a discrete estimate of the combination of primary activity, times of day, mode, and primary destination for each individual in the syn- thetic sample. • A number of refinements were made to increase execu- tion efficiency and reduce the running time of the model software. 13.2.2 Model Components 13.2.2.1 Population Synthesis The population synthesis component of the model is intended to generate a sample population that, on average, replicates the regional population for the forecast year. The population synthesis procedure is illustrated in Figure 21. Basically, the procedure consists of drawing repeated sam- ples of households from the 1990 U.S. Census Public Use Microsample (PUMS) data for the Seattle region. The population synthesis procedure follows the follow- ing steps: 1. For the given zone, determine the PUMS area (PUMA) that contains the zone. 2. For PUMS households in the PUMA, generate a weighted cross-classification table of households by dwelling unit type (single-/multifamily) and income quartile. 3. Using the Puget Sound Regional Council (PSRC) fore- cast marginal totals of households by single-/multifamily and income quartile, adjust the table using iterative pro- portional fit (IPF). (IPF is a maximum-likelihood pro- cedure used to fit multidimensional tables to fixed mar- ginal totals. In the transportation literature, it is often referred to as a Furness or Fratar procedure. Some sta- tisticians also refer to this as Johnson’s method. See Yvonne M. M. Bishop, Stephen E. Fienberg, and Paul W. Holland, Discrete Multivariate Analysis: Theory and Practice, Cambridge, Mass., MIT Press, 1975.) Network supply data by time of day Synthetic population Zonal population and land-use data Full-Day Activity Pattern Home-Based Tour Times of Day Home-Based Tour Mode and Destination Location of Intermediate Stops (car driver tours only) Output: OD trip matrices by mode, purpose, time of day, and income class Work-Based Subtour Models Predicted tours by purpose and chain type Predicted tours by purpose, chain type, and time of day Predicted tours by purpose, chain type, time of day, and mode Accessibility logsum values by tour purpose and tour type Accessibility logsum values by tour purpose, tour type, times of day, mode, and destination (not used in current version of model) Accessibility logsum values by tour purpose, tour type, and times Figure 20. Information flows in Portland Tour-Based Model.

119 FIN A L R EPO RT 4. For each dwelling unit type/income quartile cell, sam- ple with replacement a fixed number of households from PUMS. 5. Compute a weighted cross-classification table by dwell- ing unit type/income quartile for PUMS households in the sample. 6. Using the adjusted table developed in Step 3, readjust individual household weights so that the weighted total number of households in each cell of the sample equals the forecast number of households in each cell for the zone. 7. Provide the sample households to the travel modeling procedure. 13.2.2.2 Primary Activity Model The primary activity model, illustrated in Figure 22, esti- mates the main activity of the day for the following primary activities: • Work (includes school) on tour, • Maintenance (e.g., shopping and personal business) on tour, • Discretionary (e.g., social and recreational) on tour, • Work at home, • Maintenance at home, and • Discretionary at home. The first three primary activities involve travel; the remain- ing three are for a person who does not make a trip that day. Inputs to the primary activity model include characteristics of the user and logsums from the Mode/Destination and Time-of-Day Modules. 13.2.2.3 Time-of-Day Module The Time-of-Day Module predicts time of day for leaving home and for the return trip. Five times of day are defined; the model ignores trips that extend overnight. The resulting possible combinations of times for leaving and returning home are summarized in Figure 23. The Time-of-Day Module is illustrated in Figure 24. There are separate Time-of-Day Modules for each primary tour type. Inputs to the Time-of-Day Modules include tour type, socioeconomic characteristics of the traveler, and logsums from the mode/destination choice models. 13.2.2.4 Mode/Destination Choice The mode/destination choice models predict simultane- ous choice of mode and destination. A separate model was IPF Recompute HH weights PUMS data for PUMA containing the TAZ TAZ marginal totals: • SF/MF DU • income quartile Weighted HH sample Base tables of HH by DU/ income HH by DU/income for zone Sample households by DU/income category HH by DU/income for sample DU = dwelling unit. HH = household. SF = single family. MF = multifamily. IPF = iterative proportional fit. PUMS = Public Use Microsample. PUMA = Public Use Microsample Area. TAZ = traffic analysis zone. Figure 21. Population synthesis procedure.

developed for each primary tour type. Each model is a multi- nomial logit model with 147 choices, representing a combi- nation of 21 zones by 7 travel modes. The model forms are illustrated in Figure 25. Inputs to the mode destination models include tour type, times of day, mode characteristics (mainly time and cost), and socioeconomic characteristics of the traveler. One significant difference between this implementation and the Portland model is that the model for this study includes only seven modes: • Drive alone, • Drive with passenger, • Car passenger, • Transit with walk access, • Transit with auto access, • Bicycle, and • Walk only. 120 The Portland model includes two additional modes: light rail with walk access and light rail with drive access. The mode/destination choice model is also based on a sample of destination zones. In the application for this proj- ect, the following destination zones were sampled for each application: • The origin zone (zone of residence), • Four zones from the 20th-percentile distance, • Four zones between the 20th- and 40th-percentile distances, • Four zones between the 40th- and 60th-percentile dis- tance, and • Eight zones greater than the 60th-percentile distance. The Portland model uses a “generalized time” for motor- ized (i.e., car and transit) modes, equal to the total time and cost utility divided by the car drive-alone time coefficient. FI N A L R EP O RT PRIMARY ACTIVITY Work on tour Maintenance on tour Discretionary on tour Work at home Maintenance at home Discretionary at home Accessibility logsums from mode/ destination and time of day Source: Mark Bradley, A System of Activity-Based Models for Portland, Oregon, Federal Highway Administration, U.S. Department of Transportation, FHWA-PD-99-003, 1998, p. 13. Return home Leave home Early 0300 – 0659 AM Peak 0700 – 0929 Midday 0930 – 1559 PM Peak 1600 – 1859 Late 1900 – 0259 Early      AM Peak      Midday      PM Peak      Late       = Valid combination.  = Invalid combination. Figure 22. Primary activity model. Figure 23. Time period combinations in Portland Tour-Based Model.

For each motorized mode, the generalized time utility func- tion is a cubic function that decreases sharply at larger val- ues of time, as shown in Figure 26. 13.2.3 Applying the Model In the research team’s original design, the model was to be applied much as it was in Portland: 1. For each zone, select a sample of households. 2. For each household, determine the appropriate weight- ing factor based on household type and zonal data on households by type. 3. For each person over age 16 in each sample household, compute the model choice probabilities. 4. After the choice probabilities are computed, determine the choices for the person based on a random pass through the models. This determination produces a set 121 of choices: primary activity, subtours, time of day (out- bound from home and inbound to home), primary mode/ destination, and secondary destinations. 5. Decompose the tour into trips. 6. Add the trips—weighted by the household weight determined in Step 2—to the appropriate trip table(s). 13.3 DERIVATION OF ELASTICITIES The Portland model has several drawbacks in application, chief of which is the length of time required to operate it on even a high-speed computer. Consequently, the Portland model has been used to develop a set of elasticities for pre- dicting small changes in traveler behavior in response to indi- vidual traffic-flow improvement projects. The model was executed several times on a range of travel time saving alter- natives, and the results were used to fit a set of demand/time elasticities. These elasticities were then incorporated into the NCHRP 25-21 methodology. 13.3.1 Definition of Elasticity 13.3.1.1 Economics Definitions The elasticity of demand for travel Q with respect to its cost c is a dimensionless quantity defined as the proportion- ate change in demand divided by the proportionate change in the price, all other things being equal. Mathematically, this quantity can be expressed as follows: Equation 67 ε ∂ ∂ ∂ ∂ P c Q Q c c Q c c Q Q c = = = → lim log log ∆ ∆ ∆0 FIN A L R EPO RT Tour time of day (15 combinations of 5 time periods) Tour type Accessibility logsums from mode/destination and time of day Source: Mark Bradley, A System of Activity-Based Models for Portland, Oregon, Federal Highway Administration, U.S. Department of Transportation, FHWA- PD-99-003, 1998, p. 16. Figure 24. Time-of-Day Module. Primary destination location and mode Dest. 1 Dest. 2 Dest. N Mode 1 Mode 2 Mode M 7 modes sample 21 destinations Tour type and time of day for both legs (outbound and inbound) Source: Mark Bradley, A System of Activity-Based Models for Portland, Oregon, Federal Highway Administration, U.S. Department of Transportation, FHWA-PD-99-003, 1998, p. 18. Figure 25. Mode/destination choice models.

In the above equation, εP is called the point elasticity because it is evaluated at a single point on the demand curve. Other definitions of elasticity are in common use. When there are data for two points on the demand curve (indexed by 0 and 1), one can define the arc elasticity as follows: Equation 68 Transit operators commonly approximate fare elasticities using the following formula: Equation 69 Where εS is properly called a shrinkage ratio; in this formula, the costs refer to the transit fares. 13.3.1.2 Modal Time Elasticities The concept of elasticity can be extended to other factors that affect demand, such as travel time (which is really another form of user cost). For example, for the demand for a particular mode at a particular time (say, auto drive alone during the AM peak period), one can derive elasticities with respect to the AM peak drive-alone travel time, the AM peak travel times for competing modes, and the drive-alone travel time at other times of day. The economic term for the first of these is called an own elasticity: i.e., it is the elasticity of demand with respect to its own characteristics. The remain- ing elasticities are called cross-elasticities because they depend on the characteristics of other choices. εS Q Q Q c c c = − − 1 0 0 1 0 0 ε A Q Q c c = ( ) ( ) log log 1 0 1 0 122 For this study, the research team adopted the following notation for deriving and applying elasticities: Equation 70 Where: εmpm′p′ = elasticity of demand for travel from origin i to des- tination j by mode m in time period p (denoted by Tmpij ) with respect to travel time origin i to destina- tion j by mode m′ in time period p′ (denoted by tm′p′ij ). For m′ = m and p′ = p, there is an own elasticity; otherwise, the quantity is a (mode or time or mode/time) cross-elasticity. 13.3.2 Deriving Elasticities from the Microsimulation Model 13.3.2.1 Method Using the notation defined in Equation 42, one can write a constant elasticity demand model in the following form: Equation 71 Where the quantities with tildes represent trips and travel times after some change and the other quantities represent base case trips and travel times. This can be converted to a log-log linear model: Equation 72Ln T T t t ij mp ij mp m p mp m p ij m p ij m p ˜ ln ˜      =    ′ ′′ ′ ′ ′ ′ ′∑ ε ˜ ˜T T t t ij mp ij mp ij m p ij m p m p m p mp =     ′ ′ ′ ′ ′ ′ ∏ ′ ′ ε ε ∂ ∂′ ′ ′ ′=m p mp ij mp ij m p T t log log FI N A L R EP O RT -40 -35 -30 -25 -20 -15 -10 -5 0 0 50 100 150 200 250 300 Generalized Time (minutes Ut ili ty ) Maintenance Work Discretionary Figure 26. Generalized utilities as a function of generalized time.

One can therefore estimate the elasticities by observing the quantities and running a set of regressions. For applying the Traveler Behavior Module, the research team defined the following modes and time periods: • Three modes: auto drive alone, auto shared ride, and transit. • Three time periods: AM peak, PM peak, and off peak. Hence, 81 elasticities (own elasticities and cross-elasticities) could theoretically be considered. Practically, it does not appear desirable or feasible to estimate all possible elasticities. The research team therefore made the following assumptions: • Only peak travel times would change for the case stud- ies under consideration. Hence, elasticities with respect to off-peak travel times would not be estimated. • Where both the mode and time period were different from the mode and time period under consideration, the cross-elasticities would not be estimated. In other words, the research team assumed that εmpm′p′ = 0 for both m′ ≠ m and p′ ≠ p. The result of these assumptions is illustrated diagrammat- ically in Table 40. The assumptions result in 30 elasticities (6 own elasticities plus 24 mode or time cross-elasticities) that need to be estimated. 13.3.2.2 Method Given the constant elasticity model discussed above, the approach to generating the necessary data points was straightforward: T T t tijmp ijmp ijm p ijm p, ˜ , , ˜ and ′ ′ ′ ′ 123 1. Define a set of i, j zone pairs to be sampled. These zone pairs were sampled to focus on the areas of interest. For example, given the case study area, the research team focused on movements from within King County to Seattle, from Pierce County to Seattle, and from Snohomish County to Seattle. Movements to and from Kitsap County were ignored because the research team believed that the ferry network may not be adequately represented to treat it alongside bus transit as a tran- sit mode. 2. Pick a particular zone pair with “home” zone i and des- tination zone j. Randomly generate a travel time change in the AM peak period for the auto mode and run the model only for the population within zone i. Store the relative travel time change and the relevant changes in travel by mode and time period as a data point. 3. Repeat Step 2 for different values of change to the travel time. 4. Repeat Steps 2 and 3 for different time periods. 5. Repeat Steps 2–4 for different modes. 6. Repeat Steps 2–5 for different i, j zone pairs. 7. Collect the data points and run regressions on the appropriate variables. 13.3.2.3 Modifications to Original Model Implementation The original implementation of the Portland model was designed to produce trip tables. For the elasticity analysis, the research team wanted to use the model to predict changes in trips by a given mode in a given time period for a specific OD zone pair. The research team also wanted the model to report travel times and trips by mode and time of day in a format that could be easily imported by a statistical analysis package. FIN A L R EPO RT Travel time AM peak PM peak Demand DA SR TR DA SR TR AM peak DA       SR       TR       PM peak DA       SR       TR       Off peak DA       SR       TR        = elasticity to be estimated.  = elasticity assumed to be zero. DA = drive alone. SR = shared ride. TR = transit. TABLE 40 Elasticities to be estimated

The research team found it necessary to make the following modifications to the original Portland model implementation: • Change the reporting module. The original reporting module produced trip tables. The modified design sub- stituted a reporting module that collected travel time and trip statistics for a prespecified origin/destination zone pair. • Change the destination sampling method. The Port- land model uses a discrete-choice model for mode and destination choice and generates a subset of destination zones “on the fly” when the model is applied. Hence, there is no guarantee that the zone pairs of interest will be represented adequately unless the model is run for a pro- hibitively long time. The destination zone sampling rou- tine was therefore changed so that the user could specify that a particular zone be made part of each destination zone sample. This change did not affect the statistical validity of the results because generated tours that did not involve the specified destination zone would be ignored. • Change the tour generation method. Originally, the choice probabilities were computed once for each per- son, and a single tour was generated for that person. For zone pairs and modes with low selection probabilities, the model would have to run for a prohibitively long time in order to get a sufficient number of observations on that zone pair/mode/time period combination. The model was therefore modified to allow the user to spec- ify that more than one sample tour be generated for each sample person once the choice probabilities were com- puted. The sample weight for each person was factored down accordingly. 13.3.3 Application of Elasticity Model The elasticity model would be applied as follows, given a change to a single link or facility: 1. Run a SELECT LINK analysis to determine the OD pairs affected. 2. Use elasticities in combination with time changes to estimate changes in demand by OD pair, mode, and time of day. 3. Reassign the change OD demand estimates to the network. 13.3.4 Issues with the Use of an Elasticity Model A number of issues pose potential problems with a con- stant elasticity model, including the following: 124 • There may be a large number of cross-elasticities when all modes and time periods are considered. Simplifying the problem as discussed above could ignore some impor- tant cross-elasticities. But the research team’s results, dis- cussed below, led the research team to believe that it would not have been possible to develop statistically sig- nificant estimates for cross-elasticities that were ignored. • Using very few elasticities, the research team is trying to capture trip generation (i.e., total number of trips pro- duced), time-of-day shifts, and mode shifts. • Travel time changes are likely highly correlated between different motorized modes and between time periods (at least between the two peak time periods). • It is unlikely that elasticities are constant. Not only are elasticities likely to vary by amount of time change, but they are also likely to vary with percentage time change. Longer-distance trips are likely to experience proportion- ally smaller changes in time than shorter-distance trips. • The Portland model contains variables that might change over time, such as household structure, number of work- ers, auto ownership, and income. Despite the above potential problems with a constant elas- ticity model, the research team believes that the following simplifications are reasonable: • For small travel time changes, the constant elasticity approximation is probably good enough. It can be regarded as a first-order approximation to the demand function. • Capacity improvements are likely to affect the peak periods only. Hence, the main mode shifts are likely to occur during the peak periods, and the research team reasonably ignored off-peak mode shifts. 13.3.5 Proposed Elasticity Model Form The procedure employed to generate the elasticities was to randomly generate a set of travel time changes, calculate the change in demand, then estimate the elasticities using the fol- lowing regression model: Equation 73 where the starred quantities indicate the changed values and the unstarred quantities indicate the base values. To judge how well the constant elasticity assumption works, the following criteria were examined: • Significance of the regression, • t-statistics of the elasticity parameter estimates, log logT T t tij mp ijmp mpm p m p ij mp ij mp∗ ′ ′ ′ ′ ∗( ) = ( )∑ γ FI N A L R EP O RT

• Residual plots (to test for heteroskedasticity, i.e., whether the elasticities might vary with the amount of time dif- ference), and • Overall goodness of fit. 13.4 FINAL ELASTICITIES The final set of elasticities fitted to the Portland Tour- Based Model is shown in Table 41. As shown in the table, a 10-percent decrease in AM peak- period travel time for drive alone would result in the follow- ing predicted demand effects: • A 2.25-percent increase in drive alone during the AM peak, 125 • A 0.37-percent decrease in shared ride during the AM peak, • A 0.36-percent decrease in transit during the AM peak, • A 1.24-percent increase in drive alone during the PM peak, and • A 1.70-percent increase in drive alone during the off-peak. The table shows that travel time savings for drive alone trips in the AM peak would result in an increase in drive alone demand and a decrease in shared ride and transit dur- ing the AM peak. This change illustrates a mode shift effect within the same time period. The table also shows that drive-alone travel time savings in the AM peak will spur increases in drive-alone demand for the other periods of the day. The table shows that changes in PM peak travel times have generally half the effect as changes in AM peak travel times. FIN A L R EPO RT Travel Time AM peak PM peak Demand DA SR TR DA SR TR AM peak DA -0.225 0.030 0.010 -0.024 0.000 0.000 SR 0.037 -0.303 0.032 0.000 -0.028 0.000 TR 0.036 0.030 -0.129 0.000 0.000 -0.007 PM peak DA -0.124 0.000 0.000 -0.151 0.015 0.005 SR 0.000 -0.109 0.000 0.019 -0.166 0.016 TR 0.000 0.000 -0.051 0.018 0.015 -0.040 Off peak DA -0.170 0.000 0.000 -0.069 0.000 0.000 SR 0.000 -0.189 0.000 0.000 -0.082 0.000 TR 0.000 0.000 -0.074 0.000 0.000 -0.014 DA = drive alone. SR = shared ride. TR = transit. Source: Portland Tour-Based Model Applied to PSRC data set. Estimates (shown in italics) appear in the table when statistically significant results could not be generated from the data set. Zero values are shown for cross-elasticities that were deemed (a priori) to be insignificant. TABLE 41 Travel time elasticities

126 FI N A L R EP O RT CHAPTER 14 DERIVATION OF GROWTH REDISTRIBUTION MODULE The Growth Redistribution Module predicts the very long- term impacts of localized travel time changes (caused by traffic-flow improvements) on the geographic distribution of growth in a metropolitan area. There are already several sophisticated land-use models available (such as UrbanSim) that could be used for the purpose of this module. However, these models require a great deal of specialized economic data and effort to be set up for a region, and such data and effort are beyond the resources of many MPOs. Where a sophisticated land-use model exists in a region, it can be used to predict the long-term growth effects. Where a sophisti- cated land-use model is not available, the simplified module described here is proposed for use to approximate the long- term land-use effects of traffic-flow improvements. 14.1 MODULE DESCRIPTION The Growth Redistribution Module requires that a base- line 20- to 25-year forecast of land-use growth (households and employment changes) be available for the metropolitan area. This baseline forecast should have been prepared either manually or with a model, taking into account accessibility changes as well as all of the other factors that commonly affect the distribution of growth within a region. The Growth Redistribution Module consists of a simple linear regression model that is fitted to the baseline forecast. The module predicts the change in the growth rate in house- holds and employment in each zone of the region as a func- tion of the relative change of accessibility for each zone. Although not sophisticated enough to predict actual growth, the module should be sufficient to predict how small changes in travel time accessibility can affect the predicted baseline growth rate in specific zones of the region. The module equa- tion is as follows: Equation 74 Where: LUinew = predicted sum of the number of households and jobs in zone i after the traffic-flow improvement, LUiold = sum of households and jobs in zone i before the traffic-flow improvement, LU LU G AA Ri new i old i new i old= ∗ + ∗ −    CP Ainew = predicted AM peak home-based work accessibil- ity of zone i after the traffic-flow improvement, Aiold = AM peak home-based work accessibility of zone i before the traffic-flow improvement, CP = calibration parameter for the model determined from the linear regression (CP is the slope of the least-squared error line constrained to go through 0), G = ratio of the total predicted number of households in the region after the traffic-flow improvement divided by the number of households in the region before the improvement, and R = ratio of the total predicted accessibility for the region after the traffic-flow improvement divided by the total accessibility for the region before the improvement. The module presumes that total regional growth will be unaffected by traffic-flow improvements (in other words, the module will not be sensitive to the potential effects of differ- ing levels of regional traffic-flow improvements on the com- petitiveness of regions for attracting new households or jobs). The module predicts only how regional growth might be re- allocated from marginally less accessible zones to more acces- sible zones within the region. The marginal change in zonal accessibility is obtained by subtracting the average change in regional accessibility from the zone-specific change in acces- sibility (this is accomplished in Equation 12 by subtracting the ratio R from the ratio of new to old accessibility for each zone i). For similar reasons, the amount of household growth that would have normally occurred in a zone (if the zone had grown at the regional average growth rate) is added to the module-predicted growth rate that is due exclusively to mar- ginal changes in the zonal accessibility (this is accomplished in Equation 12 by adding the ratio G). The effect of the above normalization is that if the ratio of the new accessibility to the old accessibility for a zone is less than the average ratio for the entire region, then the zone’s growth will be less than the regional average. If the zonal accessibility ratio is greater than the average regional acces- sibility ratio, then the zone’s growth will be greater than the regional average. The value of G will normally be 1.00 unless there is a sig- nificant period of time between the “before” and “after”

127 FIN A L R EPO RT traffic-flow improvement dates. The ratio G allows the ana- lyst to account for any baseline growth in the region that might have occurred between the “before” condition and the “after” condition that would have occurred with or without the traffic-flow improvement. CP is the calibration parameter that converts a percentage change in zonal accessibility into a percentage change in zonal growth. It is the slope of the regression line fitted to local data on the correlation between the marginal change in zonal accessibility and the marginal change in zonal growth expressed as the sum of households and jobs. The measure of zonal accessibility (Ai) is the denominator of the trip distribution gravity model for home-based work trips. The denominator is the sum of the weighted travel time impedances to each destination zone in the region. The AM peak-period accessibility for home-base work trips is used as a proxy for total daily accessibility for all trips, based on the presumption that commute accessibility has the greatest effect on housing and job location decisions. Equation 75 Where: Ai = accessibility of zone i, Tj = total trips generated by zone j, and Fij = AM peak travel time impedance for home-based work travel between zone i and zone j. A T Fi j ij j = ∗∑ The impedance is a decreasing function of travel time between zones and takes whatever form was used to calibrate the regional travel demand model. 14.2 MODULE APPLICATION The Growth Redistribution Module is calibrated for each region in which it is applied. Base and future employment and household forecasts are assembled for the region. A lin- ear regression model of the form shown in Equation 74 is fit- ted to the data to obtain the value of CP. The fitted equation is then used to predict how individual zones will deviate from the regional average growth rate based upon changes in zonal accessibility from the base condition. The following paragraphs illustrate such an application of the module to the Seattle metropolitan area. The Puget Sound Regional Council (PSRC) provided household and employ- ment forecasts for the years 1990 and 2020. These forecasts had been produced through a combination of inventory (for 1990) and land-use modeling (using DRAM/EMPAL) with modifications made in response to local agency input. Accessibility generally improved between the 1990 and 2020 PSRC forecasts; however, some zones experienced sig- nificant changes in accessibility between 1990 and 2020 that varied a great deal from the average (see Figure 27, which plots the percentage change in accessibility for approximately the first 790 of the PSRC zones). -100.0% 0.0% 100.0% 200.0% 300.0% 400.0% 500.0% 1 25 49 73 97 12 1 14 5 16 9 19 3 21 7 24 1 26 5 28 9 31 3 33 7 36 1 38 5 40 9 43 3 45 7 48 1 50 5 52 9 55 3 57 7 60 1 62 5 64 9 67 3 69 7 72 1 74 5 76 9 79 3 Zone % C ha ng e in A cc es si bi lit y Figure 27. PSRC zonal accessibility changes between 1990 and 2020.

128 FI N A L R EP O RT The zonal accessibilities for each mode of travel were reported out from the EMME2 in which the PSRC model is implemented. The reports were then imported into a spread- sheet, which was used to compute the differences between 1990 and 2020 and to fit a regression line to the data. A least- squared error regression line was fitted to the 832 zonal data points (see Figure 28). The line was forced through zero. The slope was 0.72, and the resulting correlation coefficient was 67.99 percent. 14.3 EQUILIBRATION The initial design for the Growth Redistribution Module included an equilibration step because changes in long-term growth patterns were expected to influence travel behavior, which in turn would affect travel times, which in turn would affect long-term growth patterns. The equilibration step was implemented using the method of successive averages (MSA). Equation 76 Where: n = the current iteration number Tnij = number of trips from i to j computed from the cur- rent iteration (n), T n T n n Tijn ijn ijn+ −= + + + 1 11 1 1( ) ( ) Tijn + 1 = number of trips from i to j to be used in the next iteration (n + 1), and Tijn − 1 = number of trips from i to j computed from the pre- vious iteration (n − 1). However, tests with a null case (where no traffic-flow improvements were coded in the test network) demonstrated the tendency of the software implementation of the methodol- ogy to diverge from the theoretically correct result (no change in the network should cause no change in the demand). The software implementation of the model initially reproduced the theoretical “No change” result for the null case for the first few iterations of the MSA, but then diverged in suc- ceeding iterations. Efforts to track down the cause of this software implemen- tation deviation from the theoretically correct “no-change” result identified two possible causes: • In the Travel Behavior Response Module, the ratio of new travel time to old travel time is raised to a power specified by the appropriate Portland elasticity. It is pos- sible that the mathematical algorithm used in the soft- ware to compute the value of a ratio raised to a power may have some rounding problems that cause the soft- ware to output a value slightly different than 1.00 for one raised to a power, thus resulting in nonunitary growth factors for the OD table. y = 0.72x R2 = 0.6799 -200% 0% 200% 400% 600% 800% 1000% -100% 100% 300% 500% 700% 900% Percent Change Accessibility Beyond Average Change Pe rc en t C ha ng e Jo bs +D w el lin gs B ey on d Av er ag e Ch an ge Figure 28. Calibration of long-term module to PSRC data.

• At one point in the growth redistribution algorithm, it is necessary to reallocate the trips in the baseline OD table to the zones predicted to have higher-than-average growth rates (and take away these trips from the slower growth zones). This reallocation is accomplished through a matrix-balancing routine that applies factors to each of the rows and columns of the base OD table. These factors are applied iteratively until a convergence criterion is reached. It is possible that this balancing routine left some very small changes in the table even when factors of 1.00 were applied to all of the row and column totals. None of these possible causes could be actually observed within the precision of the results reported by the software in printed outputs. Factors of 1.00 raised to a power were reported as 1.00 within the number of significant digits pro- vided in the output. Similarly, the reported number of trips for each cell of the OD table was found to be identical before 129 and after the travel behavior response for each cell of the OD table. The total reported trips in the table also remained unchanged (to the nearest hundredth of a trip reported in the output). However, computing the squared error between the “before” and “after” trip tables found a squared error of one- thousandth of one trip between the two supposedly identical trip tables after two iterations. This small error was magni- fied in succeeding iterations until after six iterations it reached several hundred trips. Based on the above results, it was determined that the number of iterations of the Travel Behavior Module and Growth Redistribution Module should be as limited as pos- sible to avoid software-rounding problems. The final method- ology consists of one iteration of the Travel Behavior Mod- ule to obtain medium-term results, plus one iteration of the Growth Redistribution Module and another iteration of the Travel Behavior Module to obtain the long-term results. Equi- libration of iterations was dropped from the methodology. FIN A L R EPO RT

130 FI N A L R EP O RT CHAPTER 15 DERIVATION OF MODAL ACTIVITY MODULE The purpose of the Modal Activity Module is to calculate the VHT by mode of operation (i.e., cruise, idle, and accelera- tion/deceleration), which is defined by speed and acceleration category. The estimates of vehicle activity are then used with modal emission factors (e.g., the University of California, Riverside/NCHRP 25-1) to produce the emission estimates. 15.1 METHODOLOGY DEVELOPMENT The methodology for estimating modal activity is largely based on previous research conducted by the investigator under the sponsorship of CARB. (See Skabardonis. A., “A Modeling Framework for Estimating Emissions in Large Urban Areas,” Transportation Research Record 1587, 1997; and Skabardonis A., “Feasibility and Demonstration of Net- work Simulation Techniques for Estimation of Emissions in a Large Urban Area,” Final Report, prepared for the Califor- nia Air Resources Board, DHS Inc., 1994.) This research pro- duced a set of relationships through microscopic simulation that determine the proportion of the time spent Tij on a net- work link i in driving mode j as a function of the link’s type: Tij = F(link type, v/c) Equation 77 Where: link type = the link classification based on the design, traf- fic, and control characteristics and v/c = the volume-to-capacity ratio. The link classification (i.e., type) was based on typical link classifications employed in planning and operational studies (e.g., facility types) and on key design/operational character- istics (e.g., number of lanes, free-flow speed, and signal spac- ing). Thirty-three link types were selected. The relationships were developed through processing of simulated vehicle tra- jectories using the INTRAS (predecessor of FRESIM) and TRAF-NETSIM microscopic simulation models. 15.1.1 Freeways Figure 29 shows the speed distributions for freeways derived in the CARB study based on INTRAS simulations supple- mented by field data from the Interstate 880 freeway floating car runs. Under undersaturated traffic conditions, most of the time was spent traveling at the free-flow speeds. Freeway con- nectors and weaving areas had lower speeds than basic free- way sections had. Under oversaturated traffic conditions, about 45 percent of the time was spent traveling at speeds less than 40 mph. Recent floating car data from the Interstate 680 freeway were obtained and analyzed. The data were collected for 3 days in the AM peak period along the southbound direction of Interstate 680 freeway. The 20-mile freeway section is congested for most of the AM peak period. Figure 30 shows a typical speed contour plot from the field data. Figure 31 shows a speed-distance profile from a floating car run. The data were analyzed to determine vehicle activity for uncongested conditions (v/c < 1), bottleneck locations (v/c = 1), and congested conditions (v/c > 1). Figure 32 shows three-dimensional plots of the percent time spent, speed, and acceleration for each traffic regime. Figure 33 shows the average speed and acceleration distributions for uncongested conditions. Figure 34 shows a comparison of simulated and measured speed distributions for uncongested conditions. The simu- lated results agree with the field data, taking into considera- tion that the simulated values are based on the trajectories of all vehicles in the traffic stream and field data are from test cars traveling at the lane next to the median. This figure also shows that one can use a single distribution of time spent versus speed using the ratio of speed divided by free-flow speed. Therefore, one may account for differences in design characteristics of freeway facility types using different free- flow speed and the normalized time-spent speed/acceleration relationships. Figure 35 shows the speed distributions for bottleneck locations and congested conditions. The data show a clearer picture of the effect of traffic conditions on the vehicle activ- ity than the simulated values shown in Figure 29 because the simulated trajectories include data from both bottleneck loca- tions and congested sections. Further analysis indicated no significant differences in vehicle activity in congested sec- tions with different average travel times. Based on the above analysis, the CARB relationships for freeway facilities were updated and replaced as follows:

FIN A L R EPO RT 0 5 10 15 20 25 30 35 0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 SPEED (MPH) % T IM E SP EN T ID=1 FWY-TO-FWY ID=2, FWY ID=11, V/C>1 Figure 29. Speed distribution freeways—CARB simulation. I-680 SB TACH RUNS SPEED CONTOUR MAP October 2 (Tuesday) Section number Row Summary Min Avg Max 5:00 49 61 65 65 63 68 60 66 64 66 60 59 61 58 46 47 28 48 56 59 61 59 62 61 62 63 64 64 62 64 65 64 64 62 60 62 28 60 68 5:15 42 65 65 63 61 66 61 63 55 59 62 56 38 39 18 36 39 44 45 56 59 59 62 61 62 63 64 64 62 64 65 64 64 62 60 62 18 56 66 5:30 50 65 66 66 66 66 66 62 43 12 11 26 51 51 49 43 48 43 53 53 56 58 58 57 60 63 65 65 63 61 62 61 69 72 63 41 11 55 72 5:45 51 61 61 59 60 61 55 24 27 22 34 34 41 44 33 27 33 51 50 53 53 58 58 61 65 65 65 65 65 65 65 65 32 58 55 49 22 51 65 6:00 50 64 64 65 64 65 44 36 23 22 28 30 46 41 31 24 28 45 46 52 56 59 56 53 59 61 61 61 59 58 61 57 58 57 54 56 22 50 65 6:15 63 64 64 64 64 29 34 20 19 21 22 26 51 37 29 22 22 39 52 53 50 57 53 54 57 60 60 59 59 59 62 59 60 60 59 54 19 48 64 6:30 51 65 65 66 65 48 29 21 21 19 35 28 44 23 32 23 31 39 35 46 48 51 48 57 60 61 58 56 55 60 64 60 62 63 64 53 19 47 66 6:45 46 62 62 63 61 42 18 33 14 17 28 35 31 36 19 19 25 38 47 40 41 48 35 45 44 40 44 45 42 50 59 64 64 64 63 53 14 43 64 7:00 51 65 66 66 65 47 17 35 17 25 34 31 36 27 29 26 30 47 17 25 31 37 27 39 27 19 29 34 28 40 62 57 58 55 48 48 17 39 66 7:15 51 63 63 53 31 49 22 32 15 17 24 34 46 39 11 25 27 24 39 10 36 48 24 34 24 26 22 25 31 46 62 63 61 65 63 48 10 38 65 7:30 53 65 64 60 47 53 26 36 16 10 28 37 48 13 13 14 12 31 14 11 26 30 23 30 15 21 21 23 33 33 62 65 58 55 55 53 10 35 65 7:45 55 66 66 66 64 57 30 39 17 13 23 26 33 11 11 13 15 21 17 11 16 12 23 24 18 25 24 24 25 40 62 63 60 56 53 41 11 34 66 8:00 48 63 63 65 64 64 59 37 10 16 17 14 19 9 8 13 15 21 17 13 18 11 22 21 19 21 22 21 26 34 58 62 60 57 55 48 8 33 65 8:15 46 63 65 63 63 63 58 21 14 10 13 20 25 11 7 12 18 11 21 14 20 11 21 21 19 21 22 21 26 34 58 62 60 57 55 48 7 33 65 8:30 47 65 65 65 65 66 49 28 17 7 22 22 11 13 8 15 22 34 17 14 12 25 16 18 20 17 19 18 26 27 54 60 60 59 57 55 7 33 66 8:45 48 63 64 64 62 63 62 25 18 11 5 24 20 9 10 8 18 39 19 11 15 12 23 17 17 21 28 26 24 38 59 57 60 62 60 53 5 34 64 9:00 47 63 65 65 65 65 65 65 19 12 10 17 16 7 9 13 26 8 21 15 17 18 16 25 20 18 23 22 36 45 61 66 57 60 54 40 7 35 66 9:15 58 65 66 67 66 67 66 61 16 13 16 15 10 11 13 10 21 20 21 10 17 29 16 20 19 22 17 26 21 30 60 64 62 67 67 66 10 36 67 9:30 50 65 65 66 65 64 65 60 35 9 21 28 31 8 11 7 17 8 24 16 15 27 18 25 19 30 35 23 30 42 61 62 63 63 57 49 7 37 66 9:45 44 64 67 58 63 63 63 65 32 17 31 17 28 10 18 8 8 26 38 26 4 31 21 30 16 21 16 20 30 36 59 59 60 64 65 65 4 37 67 10:00 49 65 66 66 65 66 66 66 65 45 28 23 25 18 14 22 14 18 21 23 20 15 26 18 20 22 26 16 28 28 51 62 55 64 64 60 14 39 66 10:15 43 61 62 62 61 62 62 62 64 64 63 37 25 13 14 20 28 35 26 30 35 34 31 30 27 25 14 21 28 44 68 60 61 60 55 48 13 43 68 10:30 50 61 65 65 64 65 65 65 64 51 46 44 45 40 36 42 40 42 33 45 38 26 23 48 31 31 29 25 37 45 59 60 62 61 60 57 23 48 65 10:45 48 63 64 63 63 63 63 64 64 56 64 65 64 62 58 63 65 67 44 21 37 56 47 49 51 28 42 39 28 42 61 59 64 63 58 53 21 54 67 11:00 52 66 65 65 65 65 65 65 65 66 70 66 66 63 60 64 66 68 67 67 67 67 63 64 64 62 61 62 63 62 66 65 66 67 60 58 52 64 70 Column Summary Overall Row Summary Min 42 61 61 53 31 29 17 20 10 7 5 14 10 7 7 7 8 8 14 10 4 11 16 17 15 17 14 16 21 27 51 57 32 55 48 40 Min 4 24 61 Avg 50 64 65 64 62 59 51 46 33 27 32 33 37 28 23 25 28 35 34 31 34 37 35 39 36 36 37 37 39 46 61 62 60 61 58 53 Avg 23 43 65 Max 63 66 67 67 66 68 66 66 65 66 70 66 66 63 60 64 66 68 67 67 67 67 63 64 65 65 65 65 65 65 68 66 69 72 67 66 Max 60 66 72 13 14 379 10 11 12 15 16 175 6 7 8Interval Start 2 3 40 1 18 19 20 32 3321 22 23 29 34 35 3624 25 26 27 28 30 31 Figure 30. Interstate 680 speed contour map.

• v/c < 1: time spent/speed/acceleration based on simula- tion normalized by speed divided by the free-flow speed. • v/c = 1: time spent/speed/acceleration based on the field data. • v/c > 1: time spent/speed/acceleration based on the field data. 15.1.2 Freeway Ramps The same process was used to develop relationships as in basic freeway sections. Field data were analyzed and com- pared with the CARB simulated values. For volumes less than capacity, the simulated values agree closely with field data when one normalizes using the free-flow speed. There were no field data available for oversaturated traffic conditions and for metered ramps. The existing CARB rela- tionships may be used, but they need to be verified as appro- priate through field data and additional simulations. 15.1.3 Arterials The CARB relationships for uncongested conditions were evaluated and updated using the same approach as in freeway 132 facilities, but normalized for free-flow speed. No changes were made for oversaturated traffic conditions. These rela- tionships are based on g/C ratios typical for arterial streets. The relationships were developed based on simulated opti- mal timing plans (favorable signal progression), but they explicitly account for the quality of progression. 15.2 MODAL OPERATIONS TABLES Tables of proportion of VHT spent by operating mode have been developed for the Modal Operations Module (see Tables 42 through 45). The tables are applied as follows: 1. Identify facility type and whether the volume-to-capacity ratio exceeds 1.00. 2. Select the appropriate table. 3. Multiply row-heading (leftmost column) percentages of free-flow speed by facility free-flow speed to obtain speeds that will be predicted by the table. 4. Multiply VHT by proportions in the table to obtain the number of vehicle-hours spent in each mode of opera- tion for the facility. FI N A L R EP O RT 0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 100 1 2 3 4 5 6 7 8 9 11 12 13 14 15 16 17 18 19 20 21 DISTANCE (miles) SP EE D (m ph ) Figure 31. Speed-distance profile—floating car run Interstate 680.

133 FIN A L R EPO RT 0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 -10 -4 2 8 0 5 10 15 20 % TIME SPENT SPEED (mph) ACCELERATION (mph/sec) v/c < 1.00 0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 -10 -4 2 8 0 2 4 6 8 10 % TIME SPENT SPEED (mph) ACCELERATION (mph/sec) v/c = 1.00 0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 -10 -4 2 8 0 2 4 6 8 10 % TIME SPENT SPEED (mph) ACCELERATION (mph/sec) v/c > 1.00 Figure 32. Measured (Interstate 680) percent time spent speed acceleration.

FI N A L R EP O RT 0 5 10 15 20 25 0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 SPEED (MPH) % T IM E SP EN T 0 10 20 30 40 50 60 -10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 ACCELERATION (MPH/SEC) % T IM E SP EN T Figure 33. Field-measured speed and acceleration distributions—uncongested freeways.

135 FIN A L R EPO RT 0 5 10 15 20 25 30 35 40 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 SPEED/FFS % T IM E SP EN T Field 1 mph Simulation ID=1 Simulation ID=2 Field 5 mph Figure 34. Comparison of measured and simulated speed distributions—uncongested freeways.

FI N A L R EP O RT 0 2 4 6 8 10 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 52 54 56 58 60 62 64 66 68 70 72 74 76 78 80 SPEED (mph) % T IM E SP EN T 0 2 4 6 8 10 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 52 54 56 58 60 62 64 66 68 70 72 74 76 78 80 SPEED (mph) % T IM E SP EN T Figure 35. Speed distributions—bottleneck locations and congested freeways.

137 FIN A L R EPO RTACCELERATION (mph/sec) Spd/FreSpd -10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 - - - - - - - - - - - 0.0065 - - - - - - - - - - 0.0167 - - - - - - - - 0.0003 0.0007 0.0008 0.0007 0.0002 - - - - - - - - 0.0333 - - - - - - - - 0.0003 0.0007 0.0008 0.0007 0.0002 - - - - - - - - 0.0500 - - - - - - - - 0.0003 0.0007 0.0008 0.0007 0.0002 - - - - - - - - 0.0667 - - - - - - - - 0.0003 0.0007 0.0008 0.0007 0.0002 - - - - - - - - 0.0833 - - - - - - - - 0.0003 0.0007 0.0008 0.0007 0.0002 - - - - - - - - 0.1000 - - - - - - - 0.0002 0.0004 0.0006 0.0008 0.0008 0.0002 - - - - - - - - 0.1167 - - - - - - - 0.0002 0.0004 0.0006 0.0008 0.0008 0.0002 - - - - - - - - 0.1333 - - - - - - - 0.0002 0.0004 0.0006 0.0008 0.0008 0.0002 - - - - - - - - 0.1500 - - - - - - - 0.0002 0.0004 0.0006 0.0008 0.0008 0.0002 - - - - - - - - 0.1667 - - - - - - - 0.0002 0.0004 0.0006 0.0008 0.0008 0.0002 - - - - - - - - 0.1833 - - - - - - 0.0002 0.0002 0.0004 0.0008 0.0012 0.0010 0.0004 0.0002 - - - - - - - 0.2000 - - - - - - 0.0002 0.0002 0.0004 0.0008 0.0012 0.0010 0.0004 0.0002 - - - - - - - 0.2167 - - - - - - 0.0002 0.0002 0.0004 0.0008 0.0012 0.0010 0.0004 0.0002 - - - - - - - 0.2333 - - - - - - 0.0002 0.0002 0.0004 0.0008 0.0012 0.0010 0.0004 0.0002 - - - - - - - 0.2500 - - - - - - 0.0002 0.0002 0.0004 0.0008 0.0012 0.0010 0.0004 0.0002 - - - - - - - 0.2667 - - - - - - 0.0002 0.0002 0.0004 0.0010 0.0014 0.0010 0.0004 0.0002 - - - - - - - 0.2833 - - - - - - 0.0002 0.0002 0.0004 0.0010 0.0014 0.0010 0.0004 0.0002 - - - - - - - 0.3000 - - - - - - 0.0002 0.0002 0.0004 0.0010 0.0014 0.0010 0.0004 0.0002 - - - - - - - 0.3167 - - - - - - 0.0002 0.0002 0.0004 0.0010 0.0014 0.0010 0.0004 0.0002 - - - - - - - 0.3333 - - - - - - 0.0002 0.0002 0.0004 0.0010 0.0014 0.0010 0.0004 0.0002 - - - - - - - 0.3500 - - - - - - 0.0002 0.0002 0.0004 0.0010 0.0016 0.0012 0.0004 - - - - - - - - 0.3667 - - - - - - 0.0002 0.0002 0.0004 0.0010 0.0016 0.0012 0.0004 - - - - - - - - 0.3833 - - - - - - 0.0002 0.0002 0.0004 0.0010 0.0016 0.0012 0.0004 - - - - - - - - 0.4000 - - - - - - 0.0002 0.0002 0.0004 0.0010 0.0016 0.0012 0.0004 - - - - - - - - 0.4167 - - - - - - 0.0002 0.0002 0.0004 0.0010 0.0016 0.0012 0.0004 - - - - - - - - 0.4333 - - - - - - 0.0002 0.0002 0.0004 0.0012 0.0018 0.0014 0.0004 - - - - - - - - 0.4500 - - - - - - 0.0002 0.0002 0.0004 0.0012 0.0018 0.0014 0.0004 - - - - - - - - 0.4667 - - - - - - 0.0002 0.0002 0.0004 0.0012 0.0018 0.0014 0.0004 - - - - - - - - 0.4833 - - - - - - 0.0002 0.0002 0.0004 0.0012 0.0018 0.0014 0.0004 - - - - - - - - 0.5000 - - - - - - 0.0002 0.0002 0.0004 0.0012 0.0018 0.0014 0.0004 - - - - - - - - 0.5167 - - - - - - 0.0002 0.0002 0.0004 0.0014 0.0022 0.0016 0.0004 - - - - - - - - 0.5333 - - - - - - 0.0002 0.0002 0.0004 0.0014 0.0022 0.0016 0.0004 - - - - - - - - 0.5500 - - - - - - 0.0002 0.0002 0.0004 0.0014 0.0022 0.0016 0.0004 - - - - - - - - 0.5667 - - - - - - 0.0002 0.0002 0.0004 0.0014 0.0022 0.0016 0.0004 - - - - - - - - 0.5833 - - - - - - 0.0002 0.0002 0.0004 0.0014 0.0022 0.0016 0.0004 - - - - - - - - 0.6000 - - - - - - - 0.0002 0.0004 0.0020 0.0022 0.0020 0.0004 - - - - - - - - 0.6167 - - - - - - - 0.0002 0.0004 0.0020 0.0022 0.0020 0.0004 - - - - - - - - 0.6333 - - - - - - - 0.0002 0.0004 0.0020 0.0022 0.0020 0.0004 - - - - - - - - 0.6500 - - - - - - - 0.0002 0.0004 0.0020 0.0022 0.0020 0.0004 - - - - - - - - 0.6667 - - - - - - - 0.0002 0.0004 0.0020 0.0022 0.0020 0.0004 - - - - - - - - 0.6833 - - - - - - 0.0002 0.0002 0.0004 0.0020 0.0024 0.0022 0.0002 - - - - - - - - 0.7000 - - - - - - 0.0002 0.0002 0.0004 0.0020 0.0024 0.0022 0.0002 - - - - - - - - 0.7167 - - - - - - 0.0002 0.0002 0.0004 0.0020 0.0024 0.0022 0.0002 - - - - - - - - 0.7333 - - - - - - 0.0002 0.0002 0.0004 0.0020 0.0024 0.0022 0.0002 - - - - - - - - 0.7500 - - - - - - 0.0003 0.0003 0.0006 0.0030 0.0035 0.0033 0.0003 - - - - - - - - 0.7667 - - - - - - 0.0001 0.0001 0.0002 0.0021 0.0038 0.0021 0.0002 - - - - - - - - 0.7833 - - - - - - 0.0001 0.0001 0.0003 0.0033 0.0059 0.0033 0.0003 - - - - - - - - 0.8000 - - - - - - 0.0001 0.0001 0.0003 0.0033 0.0059 0.0033 0.0003 - - - - - - - - 0.8167 - - - - - - 0.0001 0.0001 0.0002 0.0026 0.0046 0.0026 0.0002 - - - - - - - - 0.8333 - - - - - - 0.0001 0.0001 0.0002 0.0027 0.0048 0.0027 0.0002 - - - - - - - - 0.8500 - - - - - 0.0001 0.0001 0.0001 0.0003 0.0055 0.0156 0.0055 0.0003 - - - - - - - - 0.8667 - - - - - 0.0001 0.0001 0.0001 0.0002 0.0040 0.0113 0.0040 0.0002 - - - - - - - - 0.8833 - - - - - 0.0001 0.0001 0.0001 0.0003 0.0052 0.0146 0.0052 0.0003 - - - - - - - - 0.9000 - - - - - 0.0001 0.0001 0.0001 0.0003 0.0065 0.0184 0.0065 0.0003 - - - - - - - - 0.9167 - - - - - 0.0001 0.0001 0.0001 0.0004 0.0073 0.0207 0.0073 0.0004 - - - - - - - - 0.9333 - - - - - - - - 0.0003 0.0070 0.0225 0.0080 0.0004 0.0001 - - - - - - - 0.9500 - - - - - - - - 0.0004 0.0092 0.0294 0.0105 0.0006 0.0002 - - - - - - - 0.9667 - - - - - - - - 0.0004 0.0104 0.0333 0.0119 0.0006 0.0002 - - - - - - - 0.9833 - - - - - - - - 0.0005 0.0127 0.0406 0.0145 0.0008 0.0003 - - - - - - - 1.0000 - - - - - - - - 0.0017 0.0417 0.1339 0.0476 0.0025 0.0008 - - - - - - - 1.0167 - - - - - - - - 0.0006 0.0116 0.0385 0.0153 0.0006 0.0006 - - - - - - - 1.0333 - - - - - - - - 0.0003 0.0055 0.0184 0.0073 0.0003 0.0003 - - - - - - - 1.0500 - - - - - - - - 0.0001 0.0019 0.0064 0.0026 0.0001 0.0001 - - - - - - - 1.0667 - - - - - - - - - 0.0005 0.0018 0.0007 - - - - - - - - - 1.0833 - - - - - - - - - 0.0002 0.0008 0.0003 - - - - - - - - - 1.1000 - - - - - - - - - - - - - - - - - - - - - Note: entries are proportion of total vehicle-hours on link that fall in each speed/acceleration category. Columns are acceleration rate category in units of miles per hour per second. Rows are speed category expressed as a ratio of the link free-flow speed. Spd/FreSpd = ratio of speed over free-flow speed. TABLE 42 Vehicle modal activity table for uncongested freeways

138 FI N A L R EP O RT ACCELERATION (mph/sec) Spd/FreSpd -10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 0.0000 - - - - - 0.0002 0.0004 0.0008 0.0017 0.0025 0.0443 - - - - - - - - - 0.0167 - - - - - 0.0001 0.0001 0.0005 0.0010 0.0015 0.0013 0.0019 - - - - - - - - 0.0333 - - - - - - 0.0002 0.0006 0.0008 0.0023 0.0025 0.0010 0.0017 - - - - - - - 0.0500 - - - - - 0.0001 0.0002 0.0003 0.0011 0.0030 0.0049 0.0016 0.0010 0.0010 - - - - - - 0.0667 - - - - - 0.0001 0.0006 0.0006 0.0023 0.0054 0.0122 0.0033 0.0013 0.0005 0.0007 - - - - - 0.0833 - - - - - 0.0002 0.0004 0.0005 0.0022 0.0057 0.0163 0.0056 0.0020 0.0006 0.0003 0.0002 - - - - 0.1000 - - - - - 0.0001 0.0002 0.0010 0.0018 0.0044 0.0126 0.0053 0.0013 0.0006 0.0003 0.0001 - - - - 0.1167 - - - - - - 0.0003 0.0007 0.0015 0.0039 0.0095 0.0052 0.0017 0.0005 0.0002 0.0001 - - - - 0.1333 - - - - - 0.0001 0.0003 0.0006 0.0017 0.0038 0.0091 0.0049 0.0015 0.0006 0.0002 - - - - - 0.1500 - - - - - 0.0001 0.0004 0.0007 0.0018 0.0039 0.0080 0.0042 0.0018 0.0006 0.0003 - - - - - 0.1667 - - - - - 0.0001 0.0003 0.0009 0.0022 0.0051 0.0100 0.0061 0.0023 0.0007 0.0002 0.0001 - - - - 0.1833 - - - - - - 0.0003 0.0007 0.0014 0.0044 0.0090 0.0046 0.0017 0.0006 0.0002 - - - - - 0.2000 - - - - - 0.0001 0.0003 0.0007 0.0016 0.0040 0.0094 0.0047 0.0021 0.0006 0.0001 - - - - - 0.2167 - - - - - 0.0002 0.0006 0.0009 0.0026 0.0071 0.0126 0.0068 0.0031 0.0008 0.0002 - - - - - 0.2333 - - - - - - 0.0004 0.0005 0.0016 0.0047 0.0085 0.0056 0.0023 0.0006 0.0001 - - - - - 0.2500 - - - - - 0.0001 0.0003 0.0009 0.0024 0.0054 0.0122 0.0054 0.0027 0.0006 0.0002 - - - - - 0.2667 - - - - - 0.0001 0.0002 0.0005 0.0016 0.0043 0.0089 0.0053 0.0022 0.0007 0.0001 0.0001 - - - - 0.2833 - - - - - 0.0001 0.0003 0.0008 0.0018 0.0050 0.0113 0.0052 0.0022 0.0008 0.0001 - - - - - 0.3000 - - - - 0.0001 - 0.0002 0.0007 0.0015 0.0048 0.0106 0.0057 0.0021 0.0004 0.0001 - - - - - 0.3167 - - - - 0.0001 - 0.0003 0.0008 0.0017 0.0044 0.0107 0.0056 0.0020 0.0006 0.0001 - - - - - 0.3333 - - - - - - 0.0003 0.0007 0.0021 0.0053 0.0125 0.0066 0.0026 0.0008 0.0001 - - - - - 0.3500 - - - - - 0.0001 0.0004 0.0005 0.0013 0.0045 0.0125 0.0059 0.0024 0.0004 - - - - - - 0.3667 - - - - 0.0001 0.0001 0.0002 0.0004 0.0025 0.0069 0.0144 0.0075 0.0025 0.0008 - - - - - - 0.3833 - - - - - 0.0001 0.0002 0.0004 0.0012 0.0038 0.0107 0.0056 0.0020 0.0004 0.0001 - - - - - 0.4000 - - - - - - 0.0003 0.0004 0.0015 0.0047 0.0093 0.0050 0.0020 0.0003 0.0001 - - - - - 0.4167 - - - - - - 0.0001 0.0002 0.0012 0.0043 0.0115 0.0059 0.0018 0.0003 0.0001 - - - - - 0.4333 - - - - - 0.0002 0.0002 0.0004 0.0014 0.0042 0.0130 0.0066 0.0014 0.0003 0.0001 - - - - - 0.4500 - - - - - - 0.0002 0.0003 0.0012 0.0034 0.0092 0.0045 0.0017 0.0003 - - - - - - 0.4667 - - - - - - 0.0002 0.0004 0.0009 0.0033 0.0091 0.0046 0.0011 0.0002 - - - - - - 0.4833 - - - - - 0.0001 0.0002 0.0004 0.0010 0.0033 0.0076 0.0045 0.0014 0.0003 - - - - - - 0.5000 - - - - 0.0001 0.0001 0.0002 0.0003 0.0012 0.0045 0.0105 0.0055 0.0014 0.0003 0.0001 - - - - - 0.5167 - - - - - 0.0001 0.0002 0.0004 0.0013 0.0034 0.0109 0.0051 0.0013 0.0003 - - - - - - 0.5333 - - - - - - 0.0001 0.0002 0.0005 0.0025 0.0077 0.0038 0.0008 0.0002 0.0001 - - - - - 0.5500 - - - - - - 0.0002 0.0003 0.0007 0.0028 0.0067 0.0039 0.0012 0.0002 - - - - - - 0.5667 - - - - - 0.0001 0.0001 0.0004 0.0006 0.0019 0.0063 0.0032 0.0007 0.0001 - - - - - - 0.5833 - - - - - 0.0001 - 0.0003 0.0006 0.0015 0.0047 0.0029 0.0009 0.0001 - - - - - - 0.6000 - - - - - 0.0001 0.0002 0.0002 0.0006 0.0023 0.0080 0.0037 0.0009 0.0001 - - - - - - 0.6167 - - - - - - 0.0002 0.0003 0.0006 0.0020 0.0064 0.0033 0.0005 0.0001 - - - - - - 0.6333 - - - - - - 0.0001 0.0001 0.0006 0.0015 0.0057 0.0026 0.0008 0.0001 - - - - - - 0.6500 - - - - - - - 0.0002 0.0009 0.0022 0.0054 0.0036 0.0011 0.0002 - - - - - - 0.6667 - - - - - - - 0.0002 0.0001 0.0007 0.0022 0.0012 0.0003 0.0001 - - - - - - 0.6833 - - - - - - 0.0001 0.0001 0.0002 0.0009 0.0035 0.0014 0.0003 - - - - - - - 0.7000 - - - - - - - 0.0001 0.0006 0.0011 0.0033 0.0017 0.0006 0.0001 - - - - - - 0.7167 - - - - - - 0.0001 0.0001 0.0004 0.0013 0.0027 0.0018 0.0004 0.0001 - - - - - - 0.7333 - - - - - - - 0.0001 0.0005 0.0015 0.0035 0.0019 0.0006 - - - - - - - 0.7500 - - - - - - 0.0001 0.0002 0.0003 0.0013 0.0044 0.0023 0.0004 - - - - - - - 0.7667 - - - - - 0.0001 0.0001 0.0002 0.0002 0.0020 0.0050 0.0021 0.0003 0.0001 - - - - - - 0.7833 - - - - - - 0.0001 0.0002 0.0006 0.0016 0.0046 0.0023 0.0007 0.0001 - - - - - - 0.8000 - - - - - - - 0.0002 0.0002 0.0014 0.0048 0.0017 0.0006 - - - - - - - 0.8167 - - - - - - 0.0001 0.0001 0.0002 0.0013 0.0043 0.0017 0.0003 0.0001 - - - - - - 0.8333 - - - - - - - 0.0001 - 0.0004 0.0019 0.0006 0.0001 - - - - - - - 0.8500 - - - - - - - - - - - - - - - - - - - - 0.8667 - - - - - - - - - - - - - - - - - - - - 0.8833 - - - - - - - - - - - - - - - - - - - - 0.9000 - - - - - - - - - - - - - - - - - - - - 0.9167 - - - - - - - - - - - - - - - - - - - - 0.9333 - - - - - - - - - - - - - - - - - - - - 0.9500 - - - - - - - - - - - - - - - - - - - - 0.9667 - - - - - - - - - - - - - - - - - - - - 0.9833 - - - - - - - - - - - - - - - - - - - - 1.0000 - - - - - - - - - - - - - - - - - - - - 1.0167 - - - - - - - - - - - - - - - - - - - - 1.0333 - - - - - - - - - - - - - - - - - - - - 1.0500 - - - - - - - - - - - - - - - - - - - - 1.0667 - - - - - - - - - - - - - - - - - - - - 1.0833 - - - - - - - - - - - - - - - - - - - - 1.1000 - - - - - - - - - - - - - - - - - - - - Note: entries are proportion of total vehicle-hours on link that fall in each speed/acceleration category. Columns are acceleration rate category in units of miles per hour per second. Rows are speed category expressed as a ratio of the link free-flow speed. Spd/FreSpd = ratio of speed over free-flow speed. TABLE 43 Vehicle modal activity table for congested freeway sections

139 FIN A L R EPO RTACCELERATION (mph/sec) Spd/FreSpd -10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 0.000 - - - - - - - - - - 0.2006 - - - - - - - - - 0.0286 - - - - - 0.0001 - - 0.0008 0.0005 0.0009 0.0005 0.0001 0.0002 0.0001 0.0005 - - - - 0.0571 - - - - - 0.0002 - - 0.0015 0.0010 0.0018 0.0009 0.0002 0.0004 0.0001 0.0011 - - - - 0.0857 - - - - - 0.0002 - - 0.0015 0.0010 0.0018 0.0009 0.0002 0.0004 0.0001 0.0011 - - - - 0.1143 - - - - - 0.0002 - - 0.0015 0.0010 0.0018 0.0009 0.0002 0.0004 0.0001 0.0011 - - - - 0.1429 - - - - - 0.0002 - - 0.0015 0.0010 0.0018 0.0009 0.0002 0.0004 0.0001 0.0011 - - - - 0.1714 - - - - - 0.0002 0.0001 0.0003 0.0012 0.0009 0.0020 0.0012 0.0005 0.0006 0.0006 0.0005 0.0002 - - - 0.2000 - - - - - 0.0002 0.0002 0.0003 0.0014 0.0011 0.0023 0.0014 0.0005 0.0007 0.0007 0.0006 0.0002 - - - 0.2286 - - - - - 0.0002 0.0002 0.0003 0.0014 0.0011 0.0023 0.0014 0.0005 0.0007 0.0007 0.0006 0.0002 - - - 0.2571 - - - - - 0.0002 0.0002 0.0003 0.0014 0.0011 0.0023 0.0014 0.0005 0.0007 0.0007 0.0006 0.0002 - - - 0.2857 - - - - - 0.0002 0.0002 0.0003 0.0014 0.0011 0.0023 0.0014 0.0005 0.0007 0.0007 0.0006 0.0002 - - - 0.3143 - - - - - 0.0002 0.0004 0.0004 0.0013 0.0010 0.0016 0.0010 0.0004 0.0011 0.0011 0.0004 0.0003 - - - 0.3429 - - - - - 0.0002 0.0004 0.0004 0.0012 0.0009 0.0015 0.0010 0.0004 0.0010 0.0011 0.0004 0.0003 - - - 0.3714 - - - - - 0.0002 0.0004 0.0004 0.0012 0.0009 0.0015 0.0010 0.0004 0.0010 0.0011 0.0004 0.0003 - - - 0.4000 - - - - - 0.0002 0.0004 0.0004 0.0012 0.0009 0.0015 0.0010 0.0004 0.0010 0.0011 0.0004 0.0003 - - - 0.4286 - - - - - 0.0002 0.0004 0.0004 0.0012 0.0009 0.0015 0.0010 0.0004 0.0010 0.0011 0.0004 0.0003 - - - 0.4571 - - - - - 0.0002 0.0005 0.0003 0.0012 0.0009 0.0018 0.0011 0.0005 0.0028 0.0002 - 0.0003 - - - 0.4857 - - - - - 0.0002 0.0006 0.0003 0.0013 0.0010 0.0020 0.0012 0.0006 0.0031 0.0002 - 0.0003 - - - 0.5143 - - - - - 0.0002 0.0006 0.0003 0.0013 0.0010 0.0020 0.0012 0.0006 0.0031 0.0002 - 0.0003 - - - 0.5429 - - - - - 0.0002 0.0006 0.0003 0.0013 0.0010 0.0020 0.0012 0.0006 0.0031 0.0002 - 0.0003 - - - 0.5714 - - - - - 0.0002 0.0006 0.0003 0.0013 0.0010 0.0020 0.0012 0.0006 0.0031 0.0002 - 0.0003 - - - 0.6000 - - - - - 0.0001 0.0004 0.0004 0.0011 0.0009 0.0020 0.0022 0.0019 0.0025 - - - - - - 0.6286 - - - - - 0.0002 0.0005 0.0004 0.0012 0.0009 0.0022 0.0024 0.0020 0.0027 - - - - - - 0.6571 - - - - - 0.0002 0.0005 0.0004 0.0012 0.0009 0.0022 0.0024 0.0020 0.0027 - - - - - - 0.6857 - - - - - 0.0002 0.0005 0.0004 0.0012 0.0009 0.0022 0.0024 0.0020 0.0027 - - - - - - 0.7143 - - - - - 0.0002 0.0005 0.0004 0.0012 0.0009 0.0022 0.0024 0.0020 0.0027 - - - - - - 0.7429 - - - - - 0.0001 0.0005 0.0003 0.0010 0.0009 0.0043 0.0065 0.0025 0.0006 - - - - - - 0.7714 - - - - - 0.0002 0.0006 0.0003 0.0013 0.0012 0.0054 0.0081 0.0031 0.0008 - - - - - - 0.8000 - - - - - 0.0002 0.0006 0.0003 0.0013 0.0012 0.0054 0.0081 0.0031 0.0008 - - - - - - 0.8286 - - - - - 0.0002 0.0006 0.0003 0.0013 0.0012 0.0054 0.0081 0.0031 0.0008 - - - - - - 0.8571 - - - - - 0.0002 0.0006 0.0003 0.0013 0.0012 0.0054 0.0081 0.0031 0.0008 - - - - - - 0.8857 - - - - - 0.0001 0.0003 0.0002 0.0010 0.0020 0.0209 0.0175 0.0010 - - - - - - - 0.9143 - - - - - 0.0001 0.0003 0.0002 0.0012 0.0023 0.0253 0.0198 0.0011 - - - - - - - 0.9429 - - - - - 0.0002 0.0005 0.0004 0.0019 0.0039 0.0395 0.0330 0.0018 0.0001 - - - - - - 0.9714 - - - - - 0.0001 0.0004 0.0003 0.0015 0.0031 0.0316 0.0264 0.0015 0.0001 - - - - - - 1.0000 - - - - - 0.0001 0.0004 0.0003 0.0015 0.0031 0.0316 0.0264 0.0015 0.0001 - - - - - - 1.0286 - - - - - 0.0001 0.0003 0.0003 0.0011 0.0016 0.0214 0.0172 0.0002 - - - - - - - 1.0571 - - - - - - 0.0001 0.0001 0.0005 0.0007 0.0098 0.0079 0.0001 - - - - - - - 1.0857 - - - - - - 0.0001 0.0001 0.0005 0.0007 0.0098 0.0079 0.0001 - - - - - - - 1.1143 - - - - - - 0.0001 0.0001 0.0005 0.0007 0.0098 0.0079 0.0001 - - - - - - - 1.1429 - - - - - - 0.0001 0.0001 0.0005 0.0007 0.0098 0.0079 0.0001 - - - - - - - 1.1714 - - - - - 0.0001 0.0001 0.0001 0.0003 0.0002 0.0073 0.0038 - - - - - - - - 1.2000 - - - - - - - - 0.0001 0.0001 0.0027 0.0014 - - - - - - - - 1.2286 - - - - - - - - 0.0001 0.0001 0.0027 0.0014 - - - - - - - - 1.2571 - - - - - - - - 0.0001 0.0001 0.0027 0.0014 - - - - - - - - 1.2857 - - - - - - - - 0.0001 0.0001 0.0027 0.0014 - - - - - - - - 1.3143 - - - - - - - - - 0.0001 0.0015 0.0011 - - - - - - - - 1.3429 - - - - - - - - - 0.0001 0.0008 0.0006 - - - - - - - - 1.3714 - - - - - - - - - 0.0001 0.0008 0.0006 - - - - - - - - 1.4000 - - - - - - - - - - 0.0006 0.0005 - - - - - - - - 1.4286 - - - - - - - - - - 0.0006 0.0005 - - - - - - - - 1.4571 - - - - - - - - - - - - - - - - - - - - 1.4857 - - - - - - - - - - - - - - - - - - - - 1.5143 - - - - - - - - - - - - - - - - - - - - 1.5429 - - - - - - - - - - - - - - - - - - - - 1.5714 - - - - - - - - - - - - - - - - - - - - 1.6000 - - - - - - - - - - - - - - - - - - - - 1.6286 - - - - - - - - - - - - - - - - - - - - 1.6571 - - - - - - - - - - - - - - - - - - - - 1.6857 - - - - - - - - - - - - - - - - - - - - 1.7143 - - - - - - - - - - - - - - - - - - - - 1.7429 - - - - - - - - - - - - - - - - - - - - 1.7714 - - - - - - - - - - - - - - - - - - - - 1.8000 - - - - - - - - - - - - - - - - - - - - 1.8286 - - - - - - - - - - - - - - - - - - - - 1.8571 - - - - - - - - - - - - - - - - - - - - 1.8857 - - - - - - - - - - - - - - - - - - - - Note: entries are proportion of total vehicle-hours on link that fall in each speed/acceleration category. Columns are acceleration rate category in units of miles per hour per second. Rows are speed category expressed as a ratio of the link free-flow speed. Spd/FreSpd = ratio of speed over free-flow speed. TABLE 44 Vehicle modal activity table for uncongested arterials

140 FI N A L R EP O RT Spd/FreSpd -10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 0.0000 - - - - - - - - - - 0.5317 - - - - - - - 0.0286 - - - - - 0.0001 - - 0.0013 0.0006 0.0007 0.0002 0.0001 0.0003 0.0001 0.0009 - - 0.0571 - - - - - 0.0003 - - 0.0025 0.0012 0.0013 0.0003 0.0003 0.0006 0.0002 0.0017 - - 0.0857 - - - - - 0.0003 - - 0.0025 0.0012 0.0013 0.0003 0.0003 0.0006 0.0002 0.0017 - - 0.1143 - - - - - 0.0003 - - 0.0025 0.0012 0.0013 0.0003 0.0003 0.0006 0.0002 0.0017 - - 0.1429 - - - - - 0.0003 - - 0.0025 0.0012 0.0013 0.0003 0.0003 0.0006 0.0002 0.0017 - - 0.1714 - - - - - 0.0001 0.0001 0.0007 0.0017 0.0008 0.0018 0.0012 0.0005 0.0009 0.0010 0.0004 0.0002 - 0.2000 - - - - - 0.0001 0.0001 0.0008 0.0019 0.0009 0.0019 0.0013 0.0006 0.0010 0.0011 0.0005 0.0003 - 0.2286 - - - - - 0.0001 0.0001 0.0008 0.0019 0.0009 0.0019 0.0013 0.0006 0.0010 0.0011 0.0005 0.0003 - 0.2571 - - - - - 0.0001 0.0001 0.0008 0.0019 0.0009 0.0019 0.0013 0.0006 0.0010 0.0011 0.0005 0.0003 - 0.2857 - - - - - 0.0001 0.0001 0.0008 0.0019 0.0009 0.0019 0.0013 0.0006 0.0010 0.0011 0.0005 0.0003 - 0.3143 - - - - - 0.0002 0.0009 0.0007 0.0013 0.0007 0.0012 0.0008 0.0002 0.0017 0.0014 0.0003 0.0004 - 0.3429 - - - - - 0.0002 0.0009 0.0007 0.0012 0.0006 0.0012 0.0007 0.0002 0.0016 0.0013 0.0003 0.0004 - 0.3714 - - - - - 0.0002 0.0009 0.0007 0.0012 0.0006 0.0012 0.0007 0.0002 0.0016 0.0013 0.0003 0.0004 - 0.4000 - - - - - 0.0002 0.0009 0.0007 0.0012 0.0006 0.0012 0.0007 0.0002 0.0016 0.0013 0.0003 0.0004 - 0.4286 - - - - - 0.0002 0.0009 0.0007 0.0012 0.0006 0.0012 0.0007 0.0002 0.0016 0.0013 0.0003 0.0004 - 0.4571 - - - - - 0.0002 0.0010 0.0005 0.0012 0.0006 0.0007 0.0008 0.0008 0.0040 0.0002 0.0001 0.0003 - 0.4857 - - - - - 0.0002 0.0011 0.0006 0.0014 0.0007 0.0008 0.0008 0.0009 0.0045 0.0002 0.0001 0.0003 - 0.5143 - - - - - 0.0002 0.0011 0.0006 0.0014 0.0007 0.0008 0.0008 0.0009 0.0045 0.0002 0.0001 0.0003 - 0.5429 - - - - - 0.0002 0.0011 0.0006 0.0014 0.0007 0.0008 0.0008 0.0009 0.0045 0.0002 0.0001 0.0003 - 0.5714 - - - - - 0.0002 0.0011 0.0006 0.0014 0.0007 0.0008 0.0008 0.0009 0.0045 0.0002 0.0001 0.0003 - 0.6000 - - - - - 0.0002 0.0009 0.0003 0.0012 0.0005 0.0010 0.0035 0.0027 0.0019 - - 0.0001 - 0.6286 - - - - - 0.0002 0.0010 0.0003 0.0012 0.0006 0.0011 0.0037 0.0029 0.0020 - - 0.0001 - 0.6571 - - - - - 0.0002 0.0010 0.0003 0.0012 0.0006 0.0011 0.0037 0.0029 0.0020 - - 0.0001 - 0.6857 - - - - - 0.0002 0.0010 0.0003 0.0012 0.0006 0.0011 0.0037 0.0029 0.0020 - - 0.0001 - 0.7143 - - - - - 0.0002 0.0010 0.0003 0.0012 0.0006 0.0011 0.0037 0.0029 0.0020 - - 0.0001 - 0.7429 - - - - - 0.0001 0.0006 0.0003 0.0005 0.0007 0.0082 0.0075 0.0019 0.0001 - - - - 0.7714 - - - - - 0.0001 0.0008 0.0004 0.0007 0.0009 0.0110 0.0101 0.0025 0.0002 - - - - 0.8000 - - - - - 0.0001 0.0008 0.0004 0.0007 0.0009 0.0110 0.0101 0.0025 0.0002 - - - - 0.8286 - - - - - 0.0001 0.0008 0.0004 0.0007 0.0009 0.0110 0.0101 0.0025 0.0002 - - - - 0.8571 - - - - - 0.0001 0.0008 0.0004 0.0007 0.0009 0.0110 0.0101 0.0025 0.0002 - - - - 0.8857 - - - - - 0.0001 0.0004 0.0002 0.0009 0.0005 0.0069 0.0094 0.0005 - - - - - 0.9143 - - - - - - 0.0002 0.0001 0.0004 0.0002 0.0031 0.0042 0.0002 - - - - - 0.9429 - - - - - 0.0001 0.0003 0.0002 0.0007 0.0004 0.0051 0.0070 0.0004 - - - - - 0.9714 - - - - - 0.0001 0.0002 0.0001 0.0006 0.0003 0.0041 0.0056 0.0003 - - - - - 1.0000 - - - - - 0.0001 0.0002 0.0001 0.0006 0.0003 0.0041 0.0056 0.0003 - - - - - 1.0286 - - - - - 0.0001 0.0001 0.0001 0.0003 0.0003 0.0029 0.0033 - - - - - - 1.0571 - - - - - - - - 0.0001 0.0001 0.0011 0.0012 - - - - - - 1.0857 - - - - - - - - 0.0001 0.0001 0.0011 0.0012 - - - - - - 1.1143 - - - - - - - - 0.0001 0.0001 0.0011 0.0012 - - - - - - 1.1429 - - - - - - - - 0.0001 0.0001 0.0011 0.0012 - - - - - - 1.1714 - - - - - - - - - 0.0001 0.0004 0.0009 - - - - - - 1.2000 - - - - - - - - - - 0.0001 0.0002 - - - - - - 1.2286 - - - - - - - - - - 0.0001 0.0002 - - - - - - 1.2571 - - - - - - - - - - 0.0001 0.0002 - - - - - - 1.2857 - - - - - - - - - - 0.0001 0.0002 - - - - - - 1.3143 - - - - - - - - - - - - - - - - - - 1.3429 - - - - - - - - - - - - - - - - - - 1.3714 - - - - - - - - - - - - - - - - - - 1.4000 - - - - - - - - - - - - - - - - - - 1.4286 - - - - - - - - - - - - - - - - - - 1.4571 - - - - - - - - - - - - - - - - - - 1.4857 - - - - - - - - - - - - - - - - - - 1.5143 - - - - - - - - - - - - - - - - - - 1.5429 - - - - - - - - - - - - - - - - - - 1.5714 - - - - - - - - - - - - - - - - - - 1.6000 - - - - - - - - - - - - - - - - - - 1.6286 - - - - - - - - - - - - - - - - - - 1.6571 - - - - - - - - - - - - - - - - - - 1.6857 - - - - - - - - - - - - - - - - - - 1.7143 - - - - - - - - - - - - - - - - - - 1.7429 - - - - - - - - - - - - - - - - - - 1.7714 - - - - - - - - - - - - - - - - - - 1.8000 - - - - - - - - - - - - - - - - - - 1.8286 - - - - - - - - - - - - - - - - - - 1.8571 - - - - - - - - - - - - - - - - - - 1.8857 - - - - - - - - - - - - - - - - - - 1.9143 - - - - - - - - - - - - - - - - - - 1.9429 - - - - - - - - - - - - - - - - - - Note: entries are proportion of total vehicle-hours on link that fall in each speed/acceleration category. Columns are acceleration rate category in units of miles per hour per second. Rows are speed category expressed as a ratio of the link free-flow speed. Spd/FreSpd = ratio of speed over free-flow speed. TABLE 45 Vehicle modal activity table for congested arterials

FIN A L R EPO RT 141 CHAPTER 16 DERIVATION OF VEHICLE EMISSION MODULE This chapter describes the recommended method by which the emission effects of traffic-flow improvements will be esti- mated. There are several central concepts that guide the selec- tion of the specific methods described. These concepts relate to both (1) the effect of traffic-flow improvement projects on vehicle activity and (2) the state of knowledge and avail- able modeling tools for emission estimation. An outline of the emission analysis methodology is then presented, followed by specific descriptions of the specific models and data require- ments for the various emission processes. 16.1 OVERVIEW OF EMISSION ESTIMATION METHODOLOGY The underlying concept for traditional on-road emission inventory development using composite emission factors expressed in grams per mile can be thought of as “traffic on roads.” That is, the fundamental processes affecting emis- sions can be decomposed to roadway segments and charac- terized by the nature of traffic occurring on them. The emis- sion effects of traffic-flow improvement projects arise from a variety of factors that go beyond segment-based analysis. Although second-by-second vehicle operations are important, route choice and trip-making behavior also influence total emissions. For this reason, the research team recommends that the emission estimation methodology be based on a “vehicles making trips” concept. Under this approach, traffic- flow improvement projects will be evaluated by identifying the number of vehicles whose activity is influenced by the project and by characterizing the effect of the project on the vehicle’s trip characteristics. Currently, no single model addresses the range of spe- cific emission processes in sufficient detail to capture the effects of traffic-flow improvement projects. At the present time, the CMEM (NCHRP 25-11) model provides the most detailed and best tested estimates of hot-stabilized vehicle exhaust emissions at different speeds and accelerations. Simi- larly, EMFAC2000 (Version 2.02) provides the most detailed estimates of process-specific evaporative emissions and excess start emissions. The methodology proposed here relies on emission rate estimates from these two models. As described previously, no currently available models address either heavy- duty vehicle emissions or PM emissions at the same level of detail as CMEM. MOBILE6 includes comparable detail to that of EMFAC- 2000 (Version 2.02) and may be used in its place. MOBILE6 closely follows EMFAC2000 in its treatment of start and evaporative emissions, relying in some cases on the same databases and statistical models. The primary effects of traffic-flow improvement projects relate to speeds and delay along specific corridors. The direct emissions effects include • Running exhaust emissions (due to changes in vehicle speed and acceleration profiles, as well as changes in VMT due to route choice), • Running evaporative emissions (due to changes in total travel time), and • Refueling and CO2 emissions (due to changes in fuel efficiency). The CMEM (NCHRP 25-11) model can produce running exhaust emission and fuel consumption rates for user-specified speed and acceleration frequency distributions (SAFDs). These SAFDs, expressed in the number of vehicle-seconds of operation falling within specified ranges of speeds and accelerations, can be generated in a number of ways. For urban areas, a set of driving cycles were developed for MOBILE6 that were intended to characterize facility-class–specific driv- ing patterns on local roads, on-ramps, arterials, and freeways. Different cycles were developed for different levels of service (LOSs) on arterials and freeways. These cycles can be used to provide a nominal estimate of urban SAFDs from available VMT and LOS data. To evaluate traffic-flow improvement project effects, a two-step process is needed. First, the base case vehicle activ- ity affected by the project (expressed as the vehicle-seconds in each category of the SAFD) is identified and removed from the regional SAFD. Second, a new SAFD for the affected traf- fic is developed. Running exhaust emissions and fuel con- sumption estimates are calculated directly using CMEM. Run- ning evaporative emission rates are produced not by CMEM, but by EMFAC2000 (Version 2.02) on a gram/hour basis. These rates can be directly applied to the total vehicle-seconds of operation for base case and traffic-flow improvement proj- ect case SAFDs to evaluate running-loss VOC emission effects.

Traffic-flow improvement projects (particularly ramp meter- ing) can potentially cause enrichment events. CMEM was designed to directly address such events, but requires more detailed inputs for emission calculation. If the nature of traffic- flow improvement effects are expected to include increases in sustained accelerations, second-by-second vehicle trajecto- ries (i.e., distance traveled at each second while within the cor- ridor) are needed for emission analysis. Microscale simulation can provide vehicle trajectories at this level of detail, but ques- tions exist regarding the representativeness of second-by- second acceleration results. For specific projects, empirical vehicle trajectory inputs may provide more accurate emis- sion estimates. In either case, emissions for each vehicle (or for each of several “representative” vehicles) moving along a corridor can be modeled with CMEM and aggregated to obtain total project effects. If sustained accelerations do not occur, the emissions calculated using detailed time-series inputs are effectively identical to those calculated using com- bined SAFDs as model inputs. There are two secondary effects of traffic-flow improve- ment projects that influence emissions. First, to the extent that traffic-flow improvement projects reduce total travel time, there may be some increase in the number of trips made, resulting in additional start emissions. Second, both reduced travel time and increased numbers of trips alter the number and timing of hot soak, diurnal, and resting loss peri- ods for the vehicle. Given that the number of vehicles within a region is not affected by traffic-flow improvement projects, and that the average time each vehicle spends parked each day remains relatively constant, it is the potential increase in start and hot soak emissions that will be most important. EMFAC2000 (Version 2.02) produces start emissions on a grams-per-start basis for soak times of 10 through 1,440 min- utes. It also produces hot soak emissions on a grams-per-1- hour-soak basis. Inputs required to estimate base case emis- sions are the number of starts and the distribution of soak times through the day for the region of interest. Traffic-flow improvement project effects require the explicit identifica- tion of new trips and of additional intermediate destinations (i.e., trip chaining). Duration of soak times will be particu- larly important for the destinations of new trips or interme- diate destinations. The research team anticipates that these soak times may be shorter than average, as the new destina- tions are likely to be brief errands, rather than major new activities. Assumptions may be needed regarding the timing of new trips as they affect soak time distributions for diurnal and resting evaporative emissions. The following notation is used to describe the specific cal- culation approach for estimating total emissions and emission changes resulting from traffic-flow improvement projects: QX = total emissions for process X in grams; qX = emission rate (typically g/s) for process X (argu- ments, such as qX(m), indicate dependence on a fac- 142 tor such as speed or soak time, where m is the index value for that factor); v(ij) = vehicle speed and acceleration frequency distribu- tion (cumulative vehicle-seconds) for the ith speed category and j th acceleration category; and s(k) = number of vehicle starts following a park of dura- tion within the kth soak time range. Process indicators are as follows: S = starts, R = running exhaust, E = running losses (evaporative), K = hot soak, L = diurnal, G = resting loss, and F = fuel-related (refueling and CO2 exhaust). Virtually all rates depend on ambient temperature, and this dependence is not shown. In practice, an average summer day temperature profile will be used (or average winter day for CO analyses), resulting in an implicit dependence of each equation on time of day (from the fact that rates for hour h will be calculated based on the assumed ambient temperature for hour h). Start and running exhaust emissions have sepa- rate rates for VOC, NOX, and CO. Fuel-related emissions have separate rates for VOC and CO2. All other rates are VOC only. Start and running exhaust PM and running road dust and tire and brake wear PM emissions also have separate rates. 16.2 ESTIMATION OF START EXHAUST EMISSIONS Start emission rates qS(k) vary with soak time. Thus, Equation 78 Where: ES = the start exhaust emissions and qS = the emission rate in grams per start for starts follow- ing a soak time of the kth duration, as provided by EMFAC2000 (Version 2.02). A nominal soak time distribution can be assumed for regional travel totals, but the traffic-flow improvement proj- ect effects on new starts and the soak time distributions for new starts must be explicitly estimated. For example, new starts arising from an additional shopping destination on a return HBW trip will likely have a distribution of soak times in the 10-, 20-, or 30-minute range, rather than the regional average, which is likely to be longer than 4 hours. The start exhaust emission rates provided by the EMFAC- 2000 and MOBILE6 models presume a regional average dis- tribution of soak times. This macroscopic assumption is not E q k s kS S k = ∗∑ ( ) ( ) FI N A L R EP O RT

compatible with the microscopic emissions analysis proposed for this project using CMEM. Additional research would be required to develop a model of the microscopic vehicle soak time impacts of traffic-flow improvements so that CMEM rates for cold starts could be applied as adjustments to the CMEM running exhaust emis- sions. Consequently, this methodology currently neglects the differences in the cold-start emissions between build and no- build cases for a traffic-flow improvement project. 16.3 ESTIMATION OF RUNNING EXHAUST EMISSIONS Vehicle speed- and acceleration-indexed running exhaust emission rates qR(i,j) can be produced by the CMEM model in units of grams per second. Thus, Equation 79 Where ER equals running exhaust emissions and the identifi- cation of the joint speed-acceleration frequency distribution v can represent regional totals or the specific vehicle activity affected by the traffic-flow improvement projects. CMEM calculates emission rates for feasible values of vehicle speeds and accelerations based on vehicle weight and engine power output. The development of SAFDs in the Traffic Module must be constrained to these feasible values. Otherwise, emis- sions will be underestimated, as vehicles will be assumed to travel at higher-than-achievable speeds (and for shorter time periods) than would actually be the case. Heavy-duty vehicle running exhaust emissions will not be treated in this methodology for several reasons. Only limited preliminary data are available from NCHRP Project 25-14 on speed and acceleration effects on heavy-duty NOX emissions, and even fewer data are available for other pollutants. Also, the Traffic Module will not be able to produce reliable estimates of the changes in SAFDs for heavy-duty vehicle activity, and such changes may well be negligible for many traffic-flow improve- ment projects. Consequently, the methodology will neglect the differences in the heavy-duty vehicle activity between build and no-build cases for a traffic-flow improvement project. Similarly, no modeling tools are currently available that accurately characterize acceleration effects on PM emissions (except as embodied in driving cycle–based rate measure- ments for different average speeds). Consequently, the meth- odology will neglect the differences in the heavy-duty vehicle activity between build and no-build cases for a traffic-flow improvement project. 16.4 ESTIMATION OF OFF-CYCLE EMISSIONS If traffic-flow improvement projects influence the fre- quency of sustained hard accelerations, such as metering traf- E q i j v i jR R ij = ⋅∑ ( , ) ( , ) 143 fic on freeway on-ramps, second-by-second vehicle trajec- tory data are needed as inputs to CMEM for running exhaust emissions. These are distance vector inputs of form d = (d(1), d(2), . . . , d(n)) for an n-second trajectory of length d(n). CMEM directly produces total trajectory emissions: ER = QR(d) Equation 80 Where: Qr = the function (as implemented in CMEM) that calcu- lates Er based on the distance vector and the vehicle characteristics (weight, engine displacement, etc.). Emissions over multiple trajectories and vehicle types must be summed for the subset of vehicle activity of interest. Rep- resentative trajectories can be used, but individual vehicle emissions must be calculated and aggregated to obtain fleet- average effects. Initial CMEM runs were conducted using both time-series and speed- and acceleration-indexed emis- sion calculations to determine if enrichment effects are sig- nificant. It was found that while there were significant effects, speed- and acceleration-indexed emission rates would satis- factorily capture most of the impacts of traffic-flow improve- ment projects. 16.5 ESTIMATION OF RUNNING EVAPORATIVE EMISSIONS Running evaporative emission rates (qE) are calculated internally in EMFAC2000 (Version 2.02) for different trip durations, but are currently only output on a gram-per-vehicle- hour basis. Traffic-flow improvement projects are unlikely to significantly alter the average duration of trips, and the func- tional form of trip-duration dependence of running evapora- tive emissions is not highly sensitive to trip duration. There- fore, running evaporative emissions (EE) can be calculated as Equation 81 However, because CMEM does not include running evap- orative emissions, the impact of traffic-flow improvements on running evaporative emissions is currently not included in the NCHRP Project 25-21 methodology. 16.6 ESTIMATION OF HOT SOAK, DIURNAL, AND RESTING EVAPORATIVE EMISSIONS The so-called “trip-end” evaporative emissions depend primarily on ambient temperatures. Each trip end generates a hot soak, and park times longer than 1 hour produce diur- nal or resting evaporative emissions depending on whether ambient temperatures are rising, constant, or falling. The change in total vehicle operating time betweenv i j i j ( , ) , ∑ E q v ijE E ij = ∑* ( ) FIN A L R EPO RT

scenarios results directly in an increase or decrease in resting loss or diurnal emission times, since the total number of vehi- cles within a region is assumed to be constant. For purposes of evaluating traffic-flow improvement project effects, one can focus on the change in vehicle operating time, multiply- ing this difference by the appropriate diurnal or resting loss rate, qL or qG. Hot soak emissions require additional information because they are specifically associated with new trips or trip chain- ing. Base case hot soak emissions are calculated based on an assumed distribution of soak times by hour of day: Equation 82 Where: EK = hot soak, diurnal, and resting evaporative emissions, s(k) = the number of trips ending that will have a soak time of duration index k, and qK(k) = the fraction of a 1-hour hot soak emission that is associated with a soak time of duration index k (constant for soak times longer than 1 hour). If a significant fraction of soak times for added trips are less than 1 hour, then the nonlinear nature of qK (k) for short soak times should be explicitly treated. The available data do not provide information on how the number of starts will be impacted by traffic-flow improve- ment projects. Especially difficult would be finding data on how traffic-flow improvement projects influence soak times. Consequently this methodology will neglect the differences in the number of starts and soak times between build and no- build cases for a traffic-flow improvement project. 16.7 ESTIMATION OF FUEL-DEPENDENT EMISSIONS CMEM directly calculates fuel consumption from engine load for (1) both speed- and acceleration-indexed vehicle activity inputs or (2) time-series vehicle activity inputs. Thus, the evaporative emissions associated with fuel use can be directly calculated from the grams-per-second fuel consump- tion rates as Equation 83 Where: EF = fuel-dependent emissions, qF (i,j) = rate of fuel-dependent emissions, and C = a factor derived from EPA AP-42 (or other refuel- ing emission factors) and any unit conversion fac- tors needed for fuel density and vapor pressure. E C q i j v i jF F ij = ∑* ( , ) * ( , ) E q k s kK K k = ∑ ( ) ( ) 144 The same equation can be used for CO2 emissions if C is derived from the carbon mass fraction of fuel and conversion factors for CO2 molecular weight. Because the necessary activity data are lacking, evapora- tive emissions will be neglected in this methodology. 16.8 ESTIMATION OF PM10 EMISSIONS PM10 emissions could be estimated using EMFAC2000 rates per VMT for each speed category by functional road class. However, because CMEM, upon which the NCHRP 25-21 methodology is based, does not include PM10 emis- sions, the impact of traffic-flow improvements on PM10 emissions is currently not included in the NCHRP 25-21 methodology. 16.9 ESTIMATION OF HEAVY-DUTY VEHICLE EMISSIONS Changes in heavy-duty vehicle emissions due to traffic- flow improvements are not explicitly included in the method- ology for two reasons. One, modal emission rate data were not available for heavy-duty vehicles at the time of the NCHRP 25-21 project. Two, the majority of urban area travel demand models do not explicitly model heavy-duty vehicle activity separately from light-duty vehicles. The proposed methodology consequently focuses on modeling light-duty vehicle emission changes. The primary heavy-duty vehicle emissions of interest are NOX. These are currently estimated using rates keyed to the mean speed and vehicle type. As such, a weighted-average emission rate can be used for light- and heavy-duty vehicles combined and applied to total VMT based on an estimated percentage of heavy-duty vehicles in the vehicle fleet. How- ever, results from NCHRP Project 25-14, “Heavy-Duty Emis- sion Factors,” were not ready in time for this research. 16.10 FINAL VHT-BASED EMISSION RATES Two sets of emission rates tables are included in the meth- odology. The first table, taken from EMFAC2000, is based on average trip speeds. The second set of rates is based on time spent by light-duty vehicles at specific acceleration/ deceleration rates and speeds during a trip. The tables con- vert VHT into appropriate emissions in grams. 16.11 TREATMENT OF EMISSION RATE UPDATES Revised CMEM modal emission rates reflecting new emis- sion control technology and new fuel standards can be sub- stituted into the methodology if and when they become avail- able. The analyst simply substitutes the new rates by modal activity category for Tables 46 through 49. FI N A L R EP O RT

145 FIN A L R EPO RT Speed (mph) THC CO NOx Speed (mph) THC CO NOx 1 14.67607 159.5444 11.61292 41 24.80151 528.7634 58.42341 2 14.67607 159.5444 11.61292 42 25.15197 539.0394 60.04349 3 14.67607 159.5444 11.61292 43 25.50243 549.3153 61.66357 4 14.67607 159.5444 11.61292 44 25.85289 559.5912 63.28365 5 14.67607 159.5444 11.61292 45 26.20335 569.8672 64.90373 6 15.84475 179.0649 13.29768 46 26.75994 583.4598 66.83860 7 17.01342 198.5855 14.98243 47 27.31652 597.0524 68.77347 8 18.18209 218.1060 16.66719 48 27.87311 610.6450 70.70834 9 19.35076 237.6266 18.35195 49 28.42970 624.2376 72.64321 10 20.51943 257.1471 20.03670 50 28.98628 637.8302 74.57808 11 20.91987 270.0474 21.37240 51 29.82726 656.7163 76.93856 12 21.32031 282.9476 22.70810 52 30.66824 675.6025 79.29904 13 21.72074 295.8479 24.04379 53 31.50923 694.4886 81.65952 14 22.12118 308.7481 25.37949 54 32.35021 713.3747 84.02000 15 22.52162 321.6484 26.71519 55 33.19119 732.2609 86.38048 16 22.61334 331.0057 27.87704 56 34.43988 759.6495 89.32167 17 22.70507 340.3631 29.03890 57 35.68856 787.0381 92.26286 18 22.79679 349.7205 30.20075 58 36.93725 814.4268 95.20406 19 22.88851 359.0779 31.36260 59 38.18594 841.8154 98.14525 20 22.98024 368.4353 32.52446 60 39.43462 869.2041 101.08640 21 22.97946 376.0226 33.62934 61 41.28965 910.5624 104.83420 22 22.97869 383.6099 34.73422 62 43.14467 951.9207 108.58200 23 22.97791 391.1972 35.83910 63 44.99969 993.2791 112.32980 24 22.97714 398.7845 36.94398 64 46.85472 1034.6370 116.07750 25 22.97636 406.3718 38.04887 65 48.70974 1075.9960 119.82530 26 22.98913 413.3314 39.18121 66 48.70974 1075.9960 119.82530 27 23.00189 420.2911 40.31356 67 48.70974 1075.9960 119.82530 28 23.01466 427.2508 41.44591 68 48.70974 1075.9960 119.82530 29 23.02742 434.2105 42.57826 69 48.70974 1075.9960 119.82530 30 23.04018 441.1702 43.71061 70 48.70974 1075.9960 119.82530 31 23.12495 448.3715 44.93947 71 48.70974 1075.9960 119.82530 32 23.20971 455.5729 46.16833 72 48.70974 1075.9960 119.82530 33 23.29448 462.7743 47.39719 73 48.70974 1075.9960 119.82530 34 23.37924 469.9757 48.62605 74 48.70974 1075.9960 119.82530 35 23.46400 477.1770 49.85491 75 48.70974 1075.9960 119.82530 36 23.66141 485.4391 51.24460 76 48.70974 1075.9960 119.82530 37 23.85882 493.7012 52.63428 77 48.70974 1075.9960 119.82530 38 24.05623 501.9633 54.02396 78 48.70974 1075.9960 119.82530 39 24.25364 510.2254 55.41365 79 48.70974 1075.9960 119.82530 40 24.45105 518.4875 56.80333 80 48.70974 1075.9960 119.82530 THC = total hydrocarbons. CO = carbon monoxide. NOX = oxides of nitrogen. TABLE 46 EMFAC2000 light-duty vehicle emission rates (grams/hour)

146 FI N A L R EP O RT Speed Acceleration (mph/sec) (mph) -10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 0 0.002658 0.002696 0.002734 0.00277 0.002815 0.00287 0.002936 0.002973 0.003021 0.003079 0.003148 0.003227 0.00319 0.003152 0.003114 0.003076 1 0.002677 0.002718 0.002758 0.002797 0.002846 0.002905 0.002976 0.003017 0.00307 0.002965 0.003611 0.004525 0.005438 0.006488 0.007662 0.008949 2 0.002696 0.002739 0.002782 0.002824 0.002876 0.00294 0.003016 0.003061 0.003118 0.002851 0.004073 0.005822 0.007686 0.009823 0.01221 0.014822 3 0.002715 0.002761 0.002807 0.002851 0.002907 0.002975 0.003057 0.003106 0.003168 0.002912 0.004172 0.005968 0.007879 0.010071 0.01252 0.015199 4 0.002734 0.002782 0.002831 0.002877 0.002937 0.00301 0.003097 0.00315 0.003217 0.002974 0.00427 0.006113 0.008071 0.010319 0.012829 0.015577 5 0.002753 0.002804 0.002855 0.002904 0.002968 0.003046 0.003138 0.003195 0.003267 0.003036 0.004372 0.006267 0.008271 0.010584 0.013162 0.015984 6 0.002772 0.002826 0.00288 0.002931 0.002998 0.003081 0.003179 0.00324 0.003318 0.003099 0.004473 0.006421 0.00847 0.010849 0.013495 0.01639 7 0.002791 0.002847 0.002904 0.002958 0.003029 0.003116 0.003221 0.003286 0.003369 0.003199 0.004639 0.006647 0.008797 0.011255 0.013995 0.016985 8 0.002809 0.002869 0.002928 0.002985 0.00306 0.003152 0.003262 0.003332 0.003421 0.003299 0.004805 0.006874 0.009123 0.011661 0.014495 0.017579 9 0.002828 0.00289 0.002952 0.003012 0.003091 0.003188 0.003304 0.003379 0.003473 0.003454 0.005077 0.007247 0.009612 0.012262 0.016072 0.019497 10 0.002847 0.002912 0.002977 0.003039 0.003121 0.003224 0.003346 0.003425 0.003526 0.003609 0.005349 0.00762 0.0101 0.012864 0.017649 0.021414 11 0.002866 0.002933 0.003001 0.003066 0.003152 0.00326 0.003389 0.003473 0.00358 0.003831 0.005681 0.008066 0.010651 0.013562 0.018357 0.022567 12 0.002885 0.002955 0.003025 0.003093 0.003183 0.003296 0.003432 0.003521 0.003634 0.004053 0.006013 0.008512 0.011202 0.01426 0.019065 0.02372 13 0.002885 0.002955 0.003025 0.003094 0.003185 0.0033 0.00344 0.003534 0.003654 0.004294 0.006375 0.009001 0.011812 0.015005 0.020031 0.026457 14 0.002885 0.002955 0.003025 0.003094 0.003186 0.003304 0.003448 0.003548 0.003675 0.004536 0.006738 0.009489 0.012423 0.01575 0.020998 0.029195 15 0.002885 0.002955 0.003025 0.003094 0.003188 0.003307 0.003454 0.003557 0.003687 0.004175 0.006286 0.009011 0.012397 0.016558 0.021999 0.032044 16 0.002885 0.002955 0.003025 0.003094 0.003189 0.003311 0.003459 0.003565 0.003699 0.003815 0.005835 0.008534 0.012372 0.017365 0.023001 0.034894 17 0.002885 0.002955 0.003025 0.003095 0.00319 0.003314 0.003465 0.003575 0.003714 0.003918 0.006039 0.008855 0.012567 0.018326 0.024525 0.037048 18 0.002885 0.002955 0.003025 0.003095 0.003192 0.003318 0.003472 0.003585 0.003728 0.00402 0.006243 0.009176 0.012761 0.019287 0.026049 0.039203 19 0.002885 0.002955 0.003025 0.003096 0.003194 0.003322 0.00348 0.003599 0.003747 0.00415 0.006484 0.009548 0.013303 0.02048 0.028191 0.041716 20 0.002885 0.002955 0.003025 0.003096 0.003195 0.003326 0.003488 0.003613 0.003766 0.00428 0.006725 0.009921 0.013844 0.021672 0.030333 0.04423 21 0.002885 0.002955 0.003025 0.003097 0.003198 0.003331 0.003498 0.003629 0.003787 0.00444 0.00701 0.010353 0.014466 0.023047 0.032765 0.047201 22 0.002885 0.002955 0.003025 0.003097 0.0032 0.003336 0.003508 0.003645 0.003808 0.0046 0.007294 0.010786 0.015089 0.024421 0.035196 0.050173 23 0.002885 0.002955 0.003025 0.003098 0.003202 0.003341 0.00352 0.003663 0.003831 0.004781 0.007615 0.011274 0.015786 0.02489 0.038191 0.053409 24 0.002885 0.002955 0.003025 0.003098 0.003205 0.003347 0.003532 0.00368 0.003853 0.004962 0.007936 0.011762 0.016483 0.025359 0.041187 0.056644 25 0.002885 0.002955 0.003025 0.003099 0.003208 0.003353 0.003545 0.003699 0.003877 0.004673 0.007614 0.011577 0.01727 0.027457 0.044686 0.060157 26 0.002885 0.002955 0.003025 0.0031 0.003211 0.003359 0.003558 0.003718 0.0039 0.004384 0.007292 0.011392 0.018056 0.029554 0.048185 0.063669 27 0.002885 0.002955 0.003025 0.003101 0.003214 0.003366 0.003571 0.003737 0.003924 0.004506 0.007568 0.011896 0.018792 0.031874 0.052215 0.067293 28 0.002885 0.002955 0.003025 0.003101 0.003217 0.003373 0.003585 0.003756 0.003948 0.004628 0.007845 0.012399 0.019527 0.034194 0.056244 0.070917 29 0.002885 0.002955 0.003025 0.003102 0.00322 0.003381 0.003599 0.003776 0.003972 0.004768 0.008165 0.012974 0.020779 0.037172 0.059341 0.073925 30 0.002885 0.002955 0.003025 0.003103 0.003224 0.003389 0.003613 0.003796 0.003996 0.004908 0.008485 0.013549 0.022031 0.040151 0.062437 0.076934 31 0.002885 0.002955 0.003025 0.003104 0.003228 0.003397 0.003628 0.003816 0.004021 0.005069 0.008846 0.014246 0.026044 0.055502 0.083352 0.09644 32 0.002885 0.002955 0.003025 0.003105 0.003231 0.003406 0.003643 0.003836 0.004045 0.005231 0.009208 0.014942 0.030058 0.070853 0.104266 0.115947 33 0.002885 0.002955 0.003025 0.003106 0.003235 0.003415 0.003659 0.003857 0.00407 0.005415 0.0096 0.015646 0.034014 0.075459 0.10678 0.116582 34 0.002885 0.002955 0.003025 0.003107 0.00324 0.003425 0.003675 0.003878 0.004095 0.005598 0.009993 0.016351 0.03797 0.080066 0.109294 0.117217 35 0.002885 0.002955 0.003025 0.003108 0.003244 0.003435 0.003691 0.0039 0.00412 0.0058 0.010414 0.017107 0.042012 0.085436 0.111487 0.117473 36 0.002885 0.002955 0.003025 0.003109 0.003248 0.003446 0.003708 0.003921 0.004146 0.006003 0.010836 0.017863 0.046054 0.090806 0.113679 0.117729 37 0.002885 0.002955 0.003025 0.003111 0.003252 0.003456 0.003724 0.003942 0.00417 0.006221 0.011284 0.019046 0.050309 0.094876 0.115585 0.117879 38 0.002885 0.002955 0.003025 0.003112 0.003256 0.003467 0.003741 0.003963 0.004195 0.00644 0.011732 0.020228 0.054563 0.098945 0.117491 0.11803 39 0.002885 0.002955 0.003025 0.003113 0.003262 0.003479 0.003759 0.003986 0.00422 0.006678 0.012211 0.021496 0.05862 0.102344 0.117605 0.118131 40 0.002885 0.002955 0.003025 0.003115 0.003267 0.003491 0.003777 0.004008 0.004245 0.006916 0.01269 0.022763 0.062677 0.105743 0.117719 0.118233 41 0.002885 0.002955 0.003025 0.003116 0.003272 0.003503 0.003794 0.00403 0.00427 0.007158 0.013193 0.024449 0.066554 0.107877 0.117829 0.118409 42 0.002885 0.002955 0.003025 0.003118 0.003278 0.003516 0.003812 0.004051 0.004294 0.0074 0.013696 0.026134 0.07043 0.11001 0.117938 0.118585 43 0.002885 0.002955 0.003025 0.003119 0.003283 0.003528 0.00383 0.004072 0.004317 0.007675 0.014233 0.028355 0.074045 0.111712 0.118044 0.118629 44 0.002885 0.002955 0.003025 0.003121 0.003289 0.00354 0.003847 0.004094 0.00434 0.007951 0.01477 0.030575 0.07766 0.113413 0.11815 0.118674 45 0.002885 0.002955 0.003025 0.003123 0.003296 0.003553 0.003864 0.004111 0.004358 0.007571 0.014581 0.033722 0.08189 0.115578 0.118223 0.118712 46 0.002885 0.002955 0.003025 0.003125 0.003302 0.003566 0.00388 0.004128 0.004377 0.00719 0.014392 0.036869 0.086119 0.117744 0.118296 0.118751 47 0.002885 0.002955 0.003025 0.003127 0.003309 0.003582 0.003902 0.004156 0.004381 0.007407 0.014936 0.040112 0.091233 0.117894 0.118592 0.118837 48 0.002885 0.002955 0.003025 0.003129 0.003315 0.003598 0.003925 0.004183 0.004386 0.007624 0.01548 0.043356 0.096347 0.118044 0.118887 0.118924 49 0.002885 0.002955 0.003025 0.003131 0.003322 0.003614 0.003946 0.004209 0.00436 0.007874 0.016025 0.046748 0.098776 0.118163 0.118963 0.118996 50 0.002885 0.002955 0.003025 0.003134 0.003329 0.00363 0.003967 0.004235 0.004335 0.008124 0.016569 0.05014 0.101206 0.118282 0.119039 0.119068 51 0.002885 0.002955 0.003025 0.003136 0.003335 0.003646 0.003988 0.004225 0.004315 0.008402 0.017218 0.053512 0.104209 0.118369 0.119097 0.119122 52 0.002885 0.002955 0.003024 0.003138 0.003342 0.003661 0.004008 0.004216 0.004294 0.00868 0.017866 0.056883 0.107212 0.118456 0.119154 0.119177 53 0.002885 0.002955 0.003024 0.003141 0.003349 0.003676 0.004027 0.0042 0.004278 0.008991 0.018885 0.061015 0.109287 0.118535 0.119205 0.119225 54 0.002885 0.002955 0.003024 0.003143 0.003356 0.003692 0.004047 0.004185 0.004262 0.009303 0.019905 0.065148 0.111362 0.118613 0.119256 0.119274 55 0.002885 0.002955 0.003024 0.003146 0.003364 0.003707 0.004067 0.004173 0.00425 0.009647 0.02097 0.068089 0.112713 0.118971 0.119301 0.119318 56 0.002885 0.002955 0.003024 0.003149 0.003372 0.003722 0.004087 0.00416 0.004237 0.009991 0.022035 0.07103 0.114065 0.119329 0.119347 0.119361 57 0.002885 0.002955 0.003024 0.003152 0.00338 0.003738 0.004091 0.00415 0.004227 0.010388 0.023019 0.073712 0.116439 0.119372 0.119388 0.119401 58 0.002885 0.002955 0.003024 0.003155 0.003388 0.003753 0.004096 0.00414 0.004216 0.010785 0.024003 0.076395 0.118812 0.119415 0.11943 0.119441 59 0.002885 0.002955 0.003025 0.003159 0.003397 0.003768 0.004087 0.004132 0.004208 0.011203 0.025289 0.080088 0.119213 0.119789 0.119825 0.119854 60 0.002885 0.002955 0.003025 0.003163 0.003405 0.003783 0.004079 0.004124 0.0042 0.011621 0.026575 0.083781 0.119614 0.120164 0.12022 0.120267 61 0.002885 0.002955 0.003026 0.003167 0.003415 0.003798 0.004072 0.004117 0.004193 0.012076 0.028303 0.08672 0.11971 0.120203 0.120256 0.120301 62 0.002885 0.002955 0.003026 0.003171 0.003424 0.003813 0.004065 0.00411 0.004186 0.012531 0.03003 0.089659 0.119807 0.120242 0.120293 0.120335 63 0.002885 0.002955 0.003027 0.003175 0.003435 0.003828 0.004059 0.004105 0.004393 0.013026 0.032123 0.092225 0.119899 0.120276 0.120325 0.120365 64 0.002885 0.002955 0.003028 0.00318 0.003445 0.003844 0.004054 0.0041 0.0046 0.013521 0.034216 0.094791 0.119991 0.12031 0.120357 0.120396 65 0.002885 0.002955 0.003029 0.003185 0.003459 0.003862 0.004047 0.004094 0.004742 0.014057 0.037802 0.09815 0.120064 0.120344 0.120388 0.120425 TABLE 47 CMEM light-duty vehicle hyrocarbon emission rates (grams/hour)

147 FIN A L R EPO RT Speed Acceleration (mph/sec) (mph) -10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 0 0.028047 0.028482 0.028916 0.029297 0.029973 0.030957 0.032259 0.033447 0.034957 0.03679 0.038944 0.041419 0.040984 0.04055 0.040116 0.039681 1 0.028264 0.02873 0.029195 0.029604 0.030337 0.031408 0.032827 0.034132 0.03579 0.034819 0.046809 0.069906 0.08806 0.106167 0.125502 0.145875 2 0.028482 0.028978 0.029474 0.029912 0.030701 0.031859 0.033395 0.034817 0.036623 0.032847 0.054673 0.098394 0.135136 0.171784 0.210888 0.252068 3 0.028699 0.029226 0.029753 0.03022 0.031068 0.032316 0.033975 0.03552 0.037481 0.03399 0.056689 0.101211 0.138787 0.176257 0.216237 0.258334 4 0.028916 0.029474 0.030033 0.030529 0.031436 0.032774 0.034554 0.036223 0.038339 0.035134 0.058704 0.104028 0.142439 0.18073 0.221585 0.264599 5 0.029133 0.029722 0.030312 0.030838 0.031806 0.033238 0.035146 0.036945 0.039226 0.036318 0.060684 0.10697 0.143501 0.185463 0.227265 0.271271 6 0.02935 0.029971 0.030591 0.031147 0.032176 0.033702 0.035738 0.037667 0.040113 0.037501 0.062664 0.109912 0.144563 0.190196 0.232945 0.277942 7 0.029567 0.030219 0.03087 0.031457 0.03255 0.034175 0.036344 0.038412 0.041031 0.039142 0.064794 0.110489 0.15009 0.19462 0.24136 0.287548 8 0.029784 0.030467 0.031149 0.031767 0.032924 0.034648 0.036951 0.039157 0.041949 0.040783 0.066923 0.111066 0.155617 0.199045 0.249774 0.297153 9 0.030002 0.030715 0.031428 0.032078 0.033303 0.035129 0.037573 0.039926 0.042903 0.043185 0.070557 0.11468 0.162242 0.206266 0.348428 0.41193 10 0.030219 0.030963 0.031707 0.03239 0.033681 0.035611 0.038196 0.040696 0.043857 0.045588 0.07419 0.118295 0.168867 0.213487 0.447081 0.526707 11 0.030436 0.031211 0.031987 0.032702 0.034064 0.036104 0.038838 0.041495 0.044853 0.048782 0.078812 0.123638 0.173964 0.224166 0.440598 0.564039 12 0.030653 0.031459 0.032266 0.033015 0.034446 0.036597 0.03948 0.042295 0.045849 0.051976 0.083433 0.128981 0.179062 0.234845 0.434115 0.601371 13 0.030653 0.031459 0.032266 0.033019 0.034493 0.03673 0.039742 0.042727 0.046493 0.055636 0.088675 0.135375 0.185885 0.243104 0.449627 0.730662 14 0.030653 0.031459 0.032266 0.033023 0.03454 0.036864 0.040005 0.04316 0.047138 0.059297 0.093917 0.141769 0.192708 0.251363 0.46514 0.859953 15 0.030653 0.031459 0.032266 0.033034 0.034582 0.036962 0.040179 0.043431 0.047524 0.055207 0.090169 0.142488 0.201027 0.25945 0.476636 1.057539 16 0.030653 0.031459 0.032266 0.033045 0.034624 0.03706 0.040353 0.043701 0.047911 0.051117 0.086422 0.143206 0.209345 0.267537 0.488132 1.255126 17 0.030653 0.031459 0.032266 0.033057 0.034672 0.037172 0.040551 0.044008 0.048348 0.052658 0.08933 0.147124 0.213806 0.281838 0.521418 1.396491 18 0.030653 0.031459 0.032266 0.03307 0.034719 0.037283 0.04075 0.044315 0.048785 0.054199 0.092238 0.151042 0.218266 0.296139 0.554704 1.537856 19 0.030653 0.031459 0.032266 0.033085 0.034773 0.037412 0.041009 0.044756 0.049381 0.056202 0.095788 0.155977 0.226882 0.323353 0.617387 1.704427 20 0.030653 0.031459 0.032266 0.0331 0.034827 0.037541 0.041269 0.045197 0.049977 0.058206 0.099339 0.160912 0.235498 0.350567 0.680069 1.870999 21 0.030653 0.031459 0.032266 0.033117 0.034898 0.037698 0.041582 0.045689 0.050621 0.060715 0.103593 0.166882 0.244733 0.381279 0.763732 2.111163 22 0.030653 0.031459 0.032266 0.033134 0.03497 0.037854 0.041895 0.046182 0.051265 0.063224 0.107846 0.172851 0.253969 0.411991 0.847395 2.351327 23 0.030653 0.031459 0.032266 0.033153 0.03505 0.038031 0.042259 0.046722 0.051953 0.066021 0.112615 0.17967 0.26321 0.47856 1.006917 2.601943 24 0.030653 0.031459 0.032266 0.033172 0.035131 0.038208 0.042623 0.047261 0.052641 0.068819 0.117384 0.18649 0.272451 0.545129 1.166439 2.852558 25 0.030653 0.031459 0.032266 0.033194 0.03522 0.038405 0.043023 0.047835 0.053358 0.065894 0.117043 0.193401 0.283011 0.617672 1.388613 3.105985 26 0.030653 0.031459 0.032266 0.033215 0.03531 0.038603 0.043423 0.048408 0.054075 0.062969 0.116703 0.200311 0.29357 0.690215 1.610787 3.359413 27 0.030653 0.031459 0.032266 0.03324 0.03541 0.038822 0.043843 0.048997 0.054803 0.064945 0.121222 0.20854 0.316367 0.772146 1.927233 3.558146 28 0.030653 0.031459 0.032266 0.033264 0.035509 0.039042 0.044263 0.049587 0.055531 0.066921 0.125741 0.216768 0.339165 0.854078 2.243678 3.756879 29 0.030653 0.031459 0.032266 0.033291 0.035618 0.039285 0.044703 0.050192 0.05627 0.069154 0.130954 0.225943 0.377428 1.04598 2.4849 3.824236 30 0.030653 0.031459 0.032266 0.033318 0.035728 0.039528 0.045143 0.050798 0.057008 0.071388 0.136167 0.235117 0.415691 1.237882 2.726122 3.891593 31 0.030653 0.031459 0.032266 0.033348 0.035847 0.039795 0.045605 0.05142 0.057758 0.073924 0.14199 0.245964 0.55502 2.441528 4.409657 5.124772 32 0.030653 0.031459 0.032266 0.033378 0.035966 0.040063 0.046067 0.052043 0.058507 0.076459 0.147813 0.25681 0.69435 3.645173 6.093192 6.357952 33 0.030653 0.031459 0.032266 0.033411 0.036096 0.040356 0.046551 0.052681 0.059265 0.079425 0.154055 0.267552 0.917787 3.920339 6.170853 6.394732 34 0.030653 0.031459 0.032266 0.033444 0.036226 0.040649 0.047034 0.05332 0.060023 0.08239 0.160298 0.278295 1.141225 4.195504 6.248513 6.431511 35 0.030653 0.031459 0.032266 0.033481 0.036366 0.04097 0.047541 0.053976 0.06079 0.085646 0.16688 0.289615 1.491987 4.572735 6.285364 6.445984 36 0.030653 0.031459 0.032266 0.033517 0.036506 0.041291 0.048048 0.054631 0.061557 0.088901 0.173462 0.300935 1.842749 4.949966 6.322216 6.460458 37 0.030653 0.031459 0.032266 0.033557 0.036631 0.041615 0.048552 0.055276 0.062304 0.092351 0.180323 0.350712 2.329718 5.301288 6.382241 6.468102 38 0.030653 0.031459 0.032266 0.033597 0.036756 0.041939 0.049057 0.055921 0.06305 0.0958 0.187183 0.400489 2.816687 5.652609 6.442266 6.475747 39 0.030653 0.031459 0.032266 0.03364 0.036917 0.042314 0.049604 0.056598 0.063816 0.099501 0.194386 0.448877 3.001184 5.854157 6.446879 6.479942 40 0.030653 0.031459 0.032266 0.033684 0.037078 0.042689 0.050151 0.057276 0.064582 0.103202 0.201589 0.497264 3.185683 6.055706 6.451492 6.484138 41 0.030653 0.031459 0.032266 0.033731 0.03725 0.043068 0.050692 0.057936 0.065317 0.106758 0.209033 0.531855 3.277467 6.168096 6.456062 6.5064 42 0.030653 0.031459 0.032266 0.033778 0.037422 0.043447 0.051234 0.058596 0.066052 0.110314 0.216476 0.566446 3.369252 6.280485 6.460631 6.528662 43 0.030653 0.031459 0.032266 0.033829 0.037604 0.043829 0.051767 0.059235 0.066748 0.114437 0.224304 0.639643 3.484976 6.303052 6.465085 6.530388 44 0.030653 0.031459 0.032266 0.033879 0.037787 0.04421 0.052301 0.059873 0.067445 0.118559 0.232131 0.71284 3.6007 6.32562 6.469538 6.532115 45 0.030653 0.031459 0.032266 0.033943 0.037988 0.044601 0.0528 0.060402 0.068005 0.115489 0.240296 0.876628 3.844832 6.384479 6.470284 6.533744 46 0.030653 0.031459 0.032266 0.034006 0.03819 0.044992 0.053298 0.060931 0.068565 0.112419 0.24846 1.040417 4.088964 6.443338 6.47103 6.535373 47 0.030653 0.031459 0.032266 0.034075 0.038392 0.046587 0.056188 0.065067 0.073945 0.122485 0.263922 1.289181 4.703476 6.455601 6.513739 6.545506 48 0.030653 0.031459 0.032266 0.034144 0.038594 0.048182 0.059078 0.069202 0.079326 0.132551 0.279383 1.537945 5.317988 6.467863 6.556447 6.55564 49 0.030653 0.031459 0.032266 0.034218 0.038804 0.049771 0.061928 0.073295 0.084662 0.141913 0.292048 1.799182 5.468068 6.479035 6.567725 6.567054 50 0.030653 0.031459 0.032266 0.034292 0.039013 0.051361 0.064778 0.077388 0.089998 0.151274 0.304713 2.060419 5.618147 6.490208 6.579003 6.578468 51 0.030653 0.031459 0.032256 0.034361 0.039222 0.052934 0.067592 0.081443 0.089294 0.152942 0.315395 2.416013 5.763675 6.494328 6.585236 6.584744 52 0.030653 0.031459 0.032246 0.034431 0.03943 0.054507 0.070406 0.085498 0.088589 0.15461 0.326077 2.771608 5.909202 6.498448 6.591469 6.59102 53 0.030653 0.031459 0.03224 0.034511 0.039651 0.056077 0.073215 0.083953 0.086315 0.158401 0.371649 2.840733 6.106389 6.499936 6.595277 6.594852 54 0.030653 0.031459 0.032234 0.03459 0.039872 0.057648 0.076024 0.082408 0.084041 0.162192 0.417221 2.909858 6.303576 6.501425 6.599084 6.598685 55 0.030653 0.031459 0.032234 0.034681 0.040108 0.059216 0.077493 0.081054 0.082756 0.167163 0.461675 3.125623 6.320415 6.553251 6.601884 6.601498 56 0.030653 0.031459 0.032233 0.034771 0.040343 0.060784 0.078962 0.079699 0.08147 0.172135 0.506128 3.341387 6.337254 6.605077 6.604683 6.604311 57 0.030653 0.031459 0.032237 0.034873 0.040594 0.062349 0.078017 0.078853 0.080713 0.178873 0.536202 3.48152 6.447301 6.607379 6.606992 6.606624 58 0.030653 0.031459 0.032242 0.034975 0.040845 0.063914 0.077072 0.078006 0.079955 0.185611 0.566276 3.621652 6.557348 6.609681 6.609301 6.608937 59 0.030653 0.031459 0.032253 0.03509 0.041113 0.065475 0.076429 0.077463 0.079516 0.192368 0.5896 4.071856 6.706609 6.756441 6.756442 6.756446 60 0.030653 0.031459 0.032264 0.035204 0.04138 0.067037 0.075786 0.076921 0.079077 0.199126 0.612923 4.522059 6.85587 6.903201 6.903583 6.903955 61 0.030653 0.031459 0.032281 0.035331 0.041665 0.068594 0.075337 0.076575 0.078848 0.206551 0.67175 4.718489 6.865068 6.909377 6.909732 6.910086 62 0.030653 0.031459 0.032298 0.035459 0.041951 0.070151 0.074888 0.07623 0.078618 0.213977 0.730576 4.914918 6.874267 6.915552 6.915882 6.916216 63 0.030653 0.031459 0.032323 0.0356 0.042277 0.071724 0.074587 0.076034 0.079409 0.222057 0.810598 4.965501 6.882252 6.920411 6.920721 6.921045 64 0.030653 0.031459 0.032347 0.035741 0.042604 0.073298 0.074285 0.075838 0.0802 0.230138 0.890619 5.016084 6.890237 6.925269 6.925561 6.925874 65 0.030653 0.031459 0.03238 0.035896 0.044123 0.073073 0.073884 0.07565 0.08228 0.238931 1.066209 5.396034 6.896957 6.929721 6.929997 6.930302 TABLE 48 CMEM light-duty vehicle CO emission rates (grams/hour)

148 FI N A L R EP O RT Speed Acceleration (mph/sec)(mph) -10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 0 0.00071 0.000712 0.000715 0.000716 0.000734 0.000782 0.000868 0.000982 0.001126 0.001299 0.001501 0.001732 0.001729 0.001726 0.001724 0.001721 1 0.000711 0.000714 0.000717 0.000718 0.000738 0.000791 0.000886 0.001011 0.001168 0.001117 0.002465 0.004046 0.005485 0.007039 0.008712 0.010503 2 0.000712 0.000716 0.000719 0.000719 0.000742 0.0008 0.000903 0.00104 0.001211 0.000935 0.00343 0.00636 0.00924 0.012351 0.015699 0.019285 3 0.000714 0.000717 0.000721 0.000721 0.000746 0.000809 0.000922 0.00107 0.001256 0.001004 0.003575 0.006584 0.009528 0.012711 0.016141 0.019816 4 0.000715 0.000719 0.000722 0.000723 0.00075 0.000819 0.00094 0.0011 0.001301 0.001073 0.00372 0.006808 0.009816 0.013072 0.016582 0.020346 5 0.000717 0.00072 0.000724 0.000725 0.000754 0.000829 0.00096 0.001132 0.001348 0.001147 0.003875 0.007046 0.010133 0.013458 0.017057 0.020919 6 0.000718 0.000722 0.000726 0.000727 0.000759 0.00084 0.00098 0.001164 0.001396 0.001222 0.004029 0.007285 0.01045 0.013845 0.017532 0.021493 7 0.000719 0.000724 0.000728 0.000729 0.000763 0.000851 0.001001 0.001199 0.001446 0.001345 0.004261 0.007614 0.010857 0.014367 0.018152 0.022229 8 0.000721 0.000725 0.000729 0.000731 0.000768 0.000862 0.001022 0.001233 0.001497 0.001468 0.004493 0.007942 0.011264 0.014889 0.018771 0.022966 9 0.000722 0.000727 0.000731 0.000733 0.000773 0.000874 0.001045 0.00127 0.001551 0.001753 0.004925 0.0085 0.011939 0.015685 0.020202 0.024612 10 0.000724 0.000728 0.000733 0.000735 0.000778 0.000886 0.001068 0.001306 0.001605 0.002038 0.005357 0.009058 0.012613 0.016481 0.021632 0.026258 11 0.000725 0.00073 0.000735 0.000737 0.000783 0.0009 0.001093 0.001346 0.001663 0.002408 0.005864 0.009707 0.01339 0.017407 0.02251 0.027521 12 0.000726 0.000731 0.000737 0.000739 0.000788 0.000913 0.001117 0.001385 0.001721 0.002778 0.006371 0.010355 0.014167 0.018334 0.023388 0.028784 13 0.000726 0.000731 0.000737 0.000739 0.000792 0.000925 0.001142 0.001426 0.001781 0.003201 0.006941 0.01108 0.015037 0.019366 0.024615 0.03061 14 0.000726 0.000731 0.000737 0.00074 0.000796 0.000937 0.001166 0.001466 0.001841 0.003623 0.007511 0.011805 0.015907 0.020397 0.025841 0.032436 15 0.000726 0.000731 0.000737 0.00074 0.000799 0.000946 0.001183 0.001492 0.001877 0.003016 0.00689 0.0112 0.01647 0.021469 0.027123 0.036146 16 0.000726 0.000731 0.000737 0.000741 0.000803 0.000956 0.001199 0.001517 0.001913 0.002408 0.006269 0.010596 0.017033 0.022542 0.028404 0.039856 17 0.000726 0.000731 0.000737 0.000741 0.000807 0.000967 0.001219 0.001546 0.001954 0.002631 0.006621 0.0111 0.017074 0.023872 0.029932 0.043863 18 0.000726 0.000731 0.000737 0.000742 0.000811 0.000978 0.001238 0.001575 0.001995 0.002853 0.006973 0.011604 0.017114 0.025202 0.03146 0.04787 19 0.000726 0.000731 0.000737 0.000743 0.000815 0.00099 0.001262 0.001616 0.00205 0.003115 0.007374 0.012176 0.017578 0.02657 0.033189 0.050153 20 0.000726 0.000731 0.000737 0.000744 0.00082 0.001003 0.001287 0.001657 0.002106 0.003377 0.007776 0.012748 0.018042 0.027939 0.034917 0.052437 21 0.000726 0.000731 0.000737 0.000745 0.000827 0.001018 0.001317 0.001704 0.002166 0.003674 0.008225 0.013388 0.018837 0.029514 0.037043 0.0548 22 0.000726 0.000731 0.000737 0.000745 0.000834 0.001034 0.001346 0.00175 0.002226 0.00397 0.008675 0.014027 0.019633 0.031089 0.039168 0.057164 23 0.000726 0.000731 0.000737 0.000747 0.000842 0.001051 0.001381 0.0018 0.00229 0.004291 0.009167 0.014732 0.020586 0.032628 0.046087 0.059711 24 0.000726 0.000731 0.000737 0.000748 0.000849 0.001069 0.001415 0.001851 0.002354 0.004612 0.009658 0.015437 0.02154 0.034167 0.053007 0.062259 25 0.000726 0.000731 0.000737 0.000749 0.000858 0.001088 0.001453 0.001905 0.002421 0.00417 0.009259 0.015249 0.022603 0.034959 0.054569 0.065133 26 0.000726 0.000731 0.000737 0.000751 0.000867 0.001107 0.001491 0.001958 0.002488 0.003729 0.00886 0.01506 0.023666 0.03575 0.056131 0.068007 27 0.000726 0.000731 0.000737 0.000752 0.000876 0.001128 0.00153 0.002013 0.002555 0.003986 0.009312 0.015788 0.024719 0.037164 0.060508 0.071253 28 0.000726 0.000731 0.000737 0.000754 0.000886 0.001149 0.00157 0.002068 0.002623 0.004244 0.009763 0.016515 0.025771 0.038578 0.064885 0.0745 29 0.000726 0.000731 0.000737 0.000756 0.000896 0.001172 0.001611 0.002124 0.002691 0.004528 0.010271 0.017333 0.027154 0.04665 0.068097 0.079059 30 0.000726 0.000731 0.000737 0.000757 0.000906 0.001195 0.001652 0.002181 0.00276 0.004812 0.010779 0.018151 0.028537 0.054721 0.07131 0.083618 31 0.000726 0.000731 0.000737 0.000759 0.000918 0.001221 0.001695 0.002238 0.002829 0.005121 0.011334 0.019389 0.032247 0.069823 0.090544 0.09832 32 0.000726 0.000731 0.000737 0.000762 0.000929 0.001246 0.001738 0.002296 0.002899 0.005431 0.011888 0.020628 0.035958 0.084926 0.109778 0.113022 33 0.000726 0.000731 0.000737 0.000764 0.000941 0.001273 0.001783 0.002355 0.002969 0.005795 0.012474 0.021645 0.043996 0.089076 0.111044 0.113725 34 0.000726 0.000731 0.000737 0.000766 0.000954 0.001301 0.001828 0.002414 0.003039 0.006159 0.01306 0.022661 0.052035 0.093227 0.112309 0.114429 35 0.000726 0.000731 0.000737 0.000769 0.000967 0.001331 0.001875 0.002475 0.00311 0.006513 0.013675 0.02375 0.053835 0.099019 0.113086 0.114694 36 0.000726 0.000731 0.000737 0.000772 0.00098 0.001361 0.001922 0.002535 0.00318 0.006867 0.01429 0.024839 0.055635 0.104812 0.113862 0.114959 37 0.000726 0.000731 0.000737 0.000775 0.000992 0.001391 0.001968 0.002595 0.003249 0.007237 0.014933 0.0262 0.059043 0.106868 0.114481 0.115094 38 0.000726 0.000731 0.000737 0.000779 0.001004 0.001421 0.002015 0.002654 0.003318 0.007608 0.015575 0.02756 0.062452 0.108925 0.1151 0.115228 39 0.000726 0.000731 0.000737 0.000782 0.00102 0.001456 0.002066 0.002717 0.003389 0.008 0.016252 0.028914 0.068435 0.110739 0.115208 0.115334 40 0.000726 0.000731 0.000737 0.000786 0.001035 0.001491 0.002116 0.00278 0.00346 0.008393 0.016928 0.030267 0.074417 0.112553 0.115316 0.11544 41 0.000726 0.000731 0.000737 0.00079 0.001051 0.001526 0.002167 0.002841 0.003528 0.00874 0.017624 0.032984 0.079901 0.112976 0.11542 0.115564 42 0.000726 0.000731 0.000737 0.000794 0.001068 0.001562 0.002217 0.002902 0.003596 0.009087 0.01832 0.035701 0.085384 0.113399 0.115524 0.115688 43 0.000726 0.000731 0.000737 0.000799 0.001085 0.001597 0.002267 0.002962 0.003661 0.009512 0.019061 0.0371 0.090015 0.113918 0.115623 0.115771 44 0.000726 0.000731 0.000737 0.000804 0.001103 0.001633 0.002316 0.003021 0.003726 0.009937 0.019802 0.038499 0.094646 0.114437 0.115723 0.115853 45 0.000726 0.000731 0.000737 0.00081 0.001123 0.00167 0.002363 0.003071 0.003778 0.009502 0.019581 0.045323 0.099275 0.115095 0.115804 0.115925 46 0.000726 0.000731 0.000737 0.000817 0.001143 0.001708 0.002411 0.003121 0.003831 0.009067 0.01936 0.052146 0.103905 0.115754 0.115885 0.115996 47 0.000726 0.000731 0.000737 0.000824 0.001162 0.001783 0.002533 0.003285 0.004038 0.009658 0.020316 0.055965 0.106478 0.116131 0.11631 0.116366 48 0.000726 0.000731 0.000737 0.000831 0.001181 0.001858 0.002655 0.00345 0.004245 0.010249 0.021272 0.059784 0.109052 0.116508 0.116734 0.116736 49 0.000726 0.000731 0.000737 0.000839 0.001201 0.001933 0.002773 0.003611 0.004261 0.010497 0.022051 0.062572 0.11115 0.117239 0.117453 0.117454 50 0.000726 0.000731 0.000737 0.000847 0.001221 0.002007 0.002892 0.003772 0.004276 0.010744 0.022829 0.065361 0.113248 0.11797 0.118173 0.118173 51 0.000726 0.000731 0.000736 0.000855 0.001242 0.002081 0.003008 0.003733 0.004076 0.011035 0.023839 0.068341 0.115 0.118388 0.118583 0.118582 52 0.000726 0.000731 0.000736 0.000863 0.001262 0.002156 0.003124 0.003695 0.003876 0.011326 0.024849 0.071321 0.116751 0.118806 0.118993 0.118992 53 0.000726 0.000731 0.000736 0.000871 0.001284 0.002229 0.003239 0.003576 0.00375 0.011728 0.026264 0.07117 0.117085 0.119093 0.119273 0.119272 54 0.000726 0.000731 0.000736 0.00088 0.001305 0.002302 0.003355 0.003457 0.003624 0.01213 0.02768 0.071018 0.11742 0.11938 0.119553 0.119551 55 0.000726 0.000731 0.000736 0.000889 0.001328 0.002375 0.003286 0.003375 0.003539 0.012611 0.028856 0.074127 0.117975 0.119679 0.119764 0.119763 56 0.000726 0.000731 0.000736 0.000899 0.001351 0.002448 0.003217 0.003294 0.003455 0.013093 0.030033 0.077235 0.11853 0.119978 0.119976 0.119975 57 0.000726 0.000731 0.000736 0.000909 0.001374 0.002521 0.003158 0.003235 0.003395 0.01376 0.031211 0.08091 0.119344 0.12015 0.120149 0.120147 58 0.000726 0.000731 0.000736 0.000919 0.001398 0.002594 0.003099 0.003177 0.003336 0.014427 0.032389 0.084586 0.120157 0.120323 0.120321 0.12032 59 0.000726 0.000731 0.000737 0.00093 0.001423 0.002666 0.003055 0.003134 0.003294 0.015032 0.034573 0.096071 0.12467 0.124824 0.124825 0.124826 60 0.000726 0.000731 0.000738 0.000941 0.001449 0.002738 0.003011 0.003092 0.003252 0.015637 0.036756 0.107557 0.129183 0.129325 0.129328 0.129333 61 0.000726 0.000731 0.000739 0.000953 0.001475 0.002809 0.002977 0.00306 0.003222 0.016295 0.037851 0.110544 0.129464 0.12959 0.129594 0.129598 62 0.000726 0.000731 0.00074 0.000966 0.001502 0.002881 0.002944 0.003028 0.003192 0.016953 0.038946 0.11353 0.129745 0.129856 0.129859 0.129864 63 0.000726 0.000731 0.000741 0.000979 0.001532 0.002931 0.002919 0.003005 0.003491 0.017665 0.042293 0.115898 0.129976 0.130071 0.130074 0.130079 64 0.000726 0.000731 0.000743 0.000992 0.001562 0.002981 0.002894 0.002982 0.00379 0.018377 0.045641 0.118266 0.130208 0.130286 0.130289 0.130295 65 0.000726 0.000731 0.000745 0.001007 0.00163 0.002923 0.002858 0.002955 0.003955 0.019146 0.050374 0.121041 0.130421 0.130488 0.130491 0.130497 TABLE 49 CMEM light-duty vehicle NOX emission rates (grams/hour)

FIN A L R EPO RT 149 CHAPTER 17 VALIDATION OF METHODOLOGY This chapter presents the results of various examinations of the validity of the recommended methodology. 17.1 VALIDATION OBJECTIVES The objective of validation is to verify the results predicted by the methodology against field measurements of the impacts of traffic-flow improvements on air quality. This objective is very difficult to achieve because of the subtlety and perva- siveness of the impacts of traffic-flow improvements. As described previously, a traffic-flow improvement impacts more than just the traffic on the facility itself. It impacts traf- fic flows and speeds on numerous nearby facilities and impacts trip making and mode choice. It is therefore very difficult to identify a bounded domain for conducting “before and after” measurements of the traffic impacts of the traffic-flow improvements. In addition, the desire to capture the full long- term effects of a traffic-flow improvement requires a very long time frame (30 years or so) for measuring the “after” effects. A confounding problem for measuring the validity of the methodology is the research team’s inability to “control” for the impacts of factors extraneous to the methodology that the research team is trying to validate. Other unrecorded traffic- flow improvements in the area, demographic changes, and economic changes are all factors that can impact the mea- sured results in the field, cannot be controlled by the investi- gator, and are external to the methodology. Finally, since this methodology is new, previous efforts to measure the impacts of traffic-flow improvements on air quality did not gather the necessary data for this new method- ology and did not control for the factors external to this methodology. Thus, hard numeric validation of this new methodology against numeric results in the field is not feasible without launching a new data collection effort specifically tailored to the features of this new methodology. A new data collection effort will also require extensive resources since the method- ology is designed to track the “regional” air quality impacts of a specific traffic-flow improvement. What can be done is to review the performance of the methodology on a series of case studies and compare the results to expectations about the direction and magnitude of the impacts based upon theory. Thus, if the methodology pre- dicts a net travel time reduction for the facility and a conse- quent increase in usage of the facility, the methodology is following the research team’s expectations that reductions in travel costs (in this case time) result in increases in demand for the facility. While theory quite clearly guides the research team on the likely impacts of traffic-flow improvements on facility usage, theory is less clear about systemwide impacts (since there are many counteracting behavioral mechanisms). Increases in usage of a single facility may be partially or completely coun- teracted by decreases in parallel facilities. The predictions of the methodology for the systemic impacts of traffic-flow improvements can be compared with the results of previous studies of the systemwide impacts of traffic-flow improvements. The recommended methodology can be exer- cised on a set of case studies and the resulting predicting of travel behavior converted into equivalent elasticities. The elas- ticities output by the methodology can then be compared with the ranges of elasticities produced by various previous studies of the impacts of capacity improvements on VMT. However, the previous studies (since they used simpler methodologies) did not collect and document the “before and after” values of the many factors required by the proposed methodology. Consequently, the comparison of elasticities can be only on an “order of magnitude” basis, and even when there are large differences, there may be legitimate undocu- mented reasons for the differences. 17.2 EVALUATION AGAINST EXPECTATIONS FOR FACILITY-SPECIFIC IMPACTS The recommended methodology was exercised on 12 case studies (see Table 50). These case studies were selected spe- cifically to illustrate the application of the recommended methodology to a wide range of traffic-flow improvement projects (many of which have not previously been consid- ered) and to show how the impacts of an improvement proj- ect can vary widely depending on the base conditions prior to the improvement. The PSRC 2020 travel model database for the Seattle- Tacoma metropolitan area was selected as the test-bench for the case studies. The methodology was applied to each case study. More information about the specifics of the application

of the methodology to each case study can be found in the user’s guide, which is published as the second part of this report. The AM peak-period, peak-direction travel times and volumes predicted by the methodology were compared with the base case (i.e., no change) travel times and volumes and divided by the base times and volumes to obtain the percent- age change for each case. The percentage volume change was divided by the percentage time change to obtain the equiva- lent demand/travel time elasticity predicted by the methodol- ogy for each case. The results are shown in Table 51. As can be seen in Table 51, five case studies resulted in pre- dicted travel time savings on the facility, as expected. Three of the case studies had no predicted travel time savings, also as expected. The null case had no change in the network, so 150 there should be no time savings. Case 1, the rural freeway lane addition case, had no congestion prior to the lane addition, so the addition of a lane would be expected to have no travel time impacts. Case 3a, the addition of an HOV lane to an uncon- gested freeway, also resulted in no predicted travel time sav- ings. Again, this result was expected, because the original freeway had plenty of excess capacity. The addition of more excess capacity to a preexisting excess capacity situation results in no improvement in freeway operating speeds. For the five case studies where facility-specific travel time savings were reported, various increases in peak-direction, peak-period traffic volumes were also predicted by the meth- odology. Again, the results were as expected. Travel time sav- ings in all cases resulted in predicted increases in facility FI N A L R EP O RT Case # Case Study Title Description 0 Null (No Change) This is the “base” network for all of the other case studies. The “Null Case” was created from a duplicate of the base network for the purpose of verifying the accuracy of the software implementation of the methodology. No change in network should result in no change in behavior. 1 Add Fwy Lane Rural (uncongested) This case involves the addition of one continuous through lane in each direction to 7.6 miles of freeway located in a rural mountainous area. The freeway was uncongested in the base case; thus, this capacity increase had no impact on freeway operating speeds. It was included to demonstrate the impacts of capacity improvements that result in no performance improvements for the facility. 2 Add Fwy Lane Urban (modest congestion) This case is the addition of one continuous through lane in each direction to 6.6 miles of two-lane (in each direction) freeway in an urban location. The improvement reduced the peak-period, peak-direction volume/capacity ratio from 94% to 78%, thus improving travel speeds. 3a Add Fwy HOV Lane (uncongested) This case added a pair of median HOV lanes (one lane in each direction) to 2.05 miles of urban freeway. The freeway was generally uncongested (volume/capacity < 89%) before the addition of the HOV lanes. The HOV lanes in this case consequently had no effect on freeway operating speeds. 3b Add Fwy HOV Lane (congested) This case added a pair of median HOV lanes (one lane in each direction) to 2.00 miles of congested urban freeway. Peak-period, peak- direction volume/capacity ratios ranged between 94% and 96% before the addition of the HOV lanes. The HOV lanes in this case increased freeway operating speeds. 4 Add Arterial Lane This case added one through lane in each direction to 10.1 miles of a suburban/rural highway. 5 Access Management This case involved conversion of a 10.1-mile stretch of rural highway with uncontrolled driveway access into an urban expressway with access consolidated at signalized intersections. 6 Intersection Channelization Improvement This case added left-turn and right-turn lanes to each of four approaches of an uncongested urban signalized intersection. 7 Signal Coordination This case coordinated six traffic signals on a 0.54-mile stretch of urban arterial. Free-flow speeds were improved 10%. 8 Transit Improvement This case doubled the frequency of service for a 23.55-mile radial bus route serving a central business district. Headways were improved from 30 minutes to 15 minutes. 9 Park-and-Ride Lot This case added a park-and-ride lot 5 miles outside of the central business district on a major freeway serving downtown with a dedicated HOV lane with existing express bus service. 10 30 Years of Improvements This case added 1,683 lane-miles of highway capacity improvements, plus many bus and ferry transit improvements contained in a typical 20- year regional transportation plan, plus those improvements completed 10 years prior to adoption of the regional transportation plan. The intent of this case is to show the impacts of 30 years’ worth of traffic-flow and transit improvements. TABLE 50 Description of case studies

usage. In several cases, the volume/travel time elasticity exceeded 100 percent. The elasticities of demand with respect to travel time were all negative and quite large, as expected. In one case, Case 3a, no change in facility travel time still resulted in a modest 3-percent increase in facility usage. The mere provision of an additional facility in the freeway for HOVs resulted i