National Academies Press: OpenBook

Representing Freight in Air Quality and Greenhouse Gas Models (2010)

Chapter: Chapter 3 - Evaluation of Current Methods

« Previous: Chapter 2 - Application of Freight Emissions
Page 30
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 30
Page 31
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 31
Page 32
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 32
Page 33
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 33
Page 34
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 34
Page 35
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 35
Page 36
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 36
Page 37
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 37
Page 38
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 38
Page 39
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 39
Page 40
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 40
Page 41
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 41
Page 42
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 42
Page 43
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 43
Page 44
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 44
Page 45
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 45
Page 46
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 46
Page 47
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 47
Page 48
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 48
Page 49
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 49
Page 50
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 50
Page 51
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 51
Page 52
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 52
Page 53
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 53
Page 54
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 54
Page 55
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 55
Page 56
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 56
Page 57
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 57
Page 58
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 58
Page 59
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 59
Page 60
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 60
Page 61
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 61
Page 62
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 62
Page 63
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 63
Page 64
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 64
Page 65
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 65
Page 66
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 66
Page 67
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 67
Page 68
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 68
Page 69
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 69
Page 70
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 70
Page 71
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 71
Page 72
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 72
Page 73
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 73
Page 74
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 74
Page 75
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 75
Page 76
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 76
Page 77
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 77
Page 78
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 78
Page 79
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 79
Page 80
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 80
Page 81
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 81
Page 82
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 82
Page 83
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 83
Page 84
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 84
Page 85
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 85
Page 86
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 86
Page 87
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 87
Page 88
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 88
Page 89
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 89
Page 90
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 90
Page 91
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 91
Page 92
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 92
Page 93
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 93
Page 94
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 94
Page 95
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 95
Page 96
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 96
Page 97
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 97
Page 98
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 98
Page 99
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 99
Page 100
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 100
Page 101
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 101
Page 102
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 102
Page 103
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 103
Page 104
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 104
Page 105
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 105
Page 106
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 106
Page 107
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 107
Page 108
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 108
Page 109
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 109
Page 110
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 110
Page 111
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 111
Page 112
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 112
Page 113
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 113
Page 114
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 114
Page 115
Suggested Citation:"Chapter 3 - Evaluation of Current Methods." National Academies of Sciences, Engineering, and Medicine. 2010. Representing Freight in Air Quality and Greenhouse Gas Models. Washington, DC: The National Academies Press. doi: 10.17226/14407.
×
Page 115

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.

30 The objective of this chapter is to evaluate the current methods used to generate air emissions information from freight transportation activities, including the following: • Brief description of the state of the practice for the calcula- tion of emissions related to freight transportation, • Analysis of the strengths and weaknesses of the main methods and models, • Assessment of process uncertainty related to the main methods and models, and • Assessment of parameter uncertainty related to the inputs required by the main methods and models. This chapter is organized by transportation mode, includ- ing heavy-duty trucks, rail, ocean-going vessels, harbor craft, cargo handling equipment, and aircraft. The following three additional subsections, which are not dependent on mode, are also included: (1) a discussion of general methods for emis- sion calculations, (2) an evaluation of methods and models that estimate freight emissions at the national scale, and (3) an evaluation of how emissions estimates are used in air quality dispersion models, health risk assessments, and other appli- cations. To the extent possible, each mode-specific subsection is divided by geographic scale. This chapter is a combination of the results from Tasks 2 and 4. In Task 2, the project team examined the current state of practice for estimating freight transportation emissions of criteria pollutants, air toxics, and GHGs. In Task 4, the proj- ect team evaluated the current practices for estimating freight emissions, including both freight activity estimates and freight emission factors. 3.1 General Methods Although some methods and models are mode-specific, there are standard methods that can be applied to all modes. As illustrated in Equation 1, freight emissions are generally the product of freight activity (e.g., fuel consumed, energy generated, or vehicle miles traveled) and emission factors (in grams of pollutant per measure of freight activity). Depending on data availability and the complexity of ana- lytical methods, emissions might be calculated separately by vehicle type or other factors that affect emission factors (e.g., average speed, road level of service), and added up to a total by pollutant. With the exception of GHGs, which are summed by multiplying their respective emissions by their global warm- ing potential, the emissions of other pollutants are always reported separately. Some emissions models incorporate both measures of freight activity and emission factors and output total emissions, while other emissions models are used to extract emission factors only. 3.1.1 Greenhouse Gases Transportation sources emit different gases that contribute to global warming, including CO2, CH4, N2O, and hydrofluo- rocarbons (HFCs). Carbon dioxide is by far the most preva- lent GHG emitted by transportation sources. According to the EPA GHG Inventory, nationally, more than 95% of trans- portation GHG emissions were in the form of CO2 in 2004, when measured in terms of global warming potential (i.e., CO2 equivalent emissions). (1) The remainder of transportation GHG emissions took the form of N2O, 2.2%; CH4, 0.1%; and HFCs, 2.3%. Note that GHG emissions typically are reported in terms of CO2 equivalent to provide a common unit of mea- sure. Other GHGs are converted into CO2 equivalent on the basis of their global warming potential, which is defined as the cumulative radiative forcing effects of a gas over a specified time horizon in comparison to CO2. Radiative forcing is the change in balance between radiation entering the Earth’s at- mosphere and radiation being emitted back into space. Emissions Freight Activity Emission Factor E= × ( quation 1) C H A P T E R 3 Evaluation of Current Methods

Given the importance of CO2, it is usually appropriate and acceptable for transportation GHG analyses to focus solely on this gas, particularly if resources are limited and if the analysis is designed to provide a general indication of GHG impacts. A summary of the fuel types commonly used by various modes is provided in Exhibit 3-1. Carbon Dioxide The calculation procedures for estimating CO2 from on- road and nonroad sources are conceptually the same, since CO2 is released in direct proportion to fuel consumption, with differences in the amount of emissions by fuel type. The carbon content of a specific fuel (e.g., diesel) is the same re- gardless of what mode consumes it (e.g., trucks, locomotives, ships). However, the tools available to analyze emissions from nonroad sources differ from those that can be used for exclu- sively assessing on-road emissions. Moreover, state and local transportation agencies often have limited data on fuel con- sumption by nonroad modes. The amount of CO2 produced is a product of the amount of fuel combusted, the carbon content of the fuel, and the fraction of carbon that is oxidized when the fuel is com- busted. A simple formula for the calculation of CO2 for each fuel is as shown in Equation 2. Fuel combustion (in gallons for liquid fuels or cubic feet for natural gas) is converted into units of energy (Btus). The carbon content of fuel varies by type of fuel and is usually ex- pressed in terms of units of carbon per Btu. The fraction of the carbon oxidized is a lesser consideration since it has tra- ditionally been assumed to be 99% for all fossil fuel combus- tion. (Recent analyses conducted for EPA suggest that the ox- idation fraction for light-duty gasoline vehicles is virtually CO emitted Fuel Combusted Carbon Content C2 = × oefficient Fraction Oxidized 44 Equati× × ( )12 ( on 2) 100%; EPA recently recommended use of the 100% fraction for transportation for its international reporting.) The factor 44/12 is the weight of CO2 in relation to the amount of car- bon in the fuel, assuming all carbon burned eventually oxi- dizes to form CO2. Some carbon in fossil fuels is emitted in the form of carbon monoxide, which swiftly decays into CO2, and volatile organic compounds, which also decay into CO2. Consequently, the key analysis that needs to be conducted to estimate CO2 is to determine the amount of fuel consumed by fuel type (e.g., motor gasoline, diesel, jet fuel, compressed natural gas). Although conceptually simple, this calculation in practice is quite complex since transportation agencies do not typi- cally collect data to track vehicle fuel consumption by fuel type. In a limited number of cases, fuel data are available and can be used directly in calculating CO2. For instance, for GHG inventory development, state fuel tax records are often used to estimate motor vehicle fuel consumption and CO2. The availability of direct measures of fuel consumption, how- ever, is generally limited for transportation agencies, and fuel consumption estimates may not be available at all for project- level, corridor, or regional analysis. Transportation modeling generally focuses on estimat- ing vehicle-miles traveled (VMT) for motor vehicles, or passenger-miles traveled (PMT) for transit and nonroad modes. Given the primary use of VMT as a metric for trans- portation activity, the other key factor necessary to estimate vehicle fuel consumption is vehicle fuel economy (miles per gallon). Many factors influence vehicle fuel economy, includ- ing the following: • The mix of travel by vehicle type and model year; • Vehicle operating characteristics, such as speeds and accel- erations, and amount of idling; and • Other factors, like vehicle maintenance, tire pressure, and air conditioner use. 31 Exhibit 3-1. Fuel types commonly used by different transportation modes. Fuel Heavy- Duty Trucks Rail WaterborneVessels Cargo Handling Equipment Aircraft Motor Gasoline Diesel (Distillate) Jet Fuel Aviation Gasoline Residual Fuel Electricity Other Fuels* *Other fuels include compressed natural gas (CNG), liquefied petroleum gasoline (LPG), and other alternative fuels.

The relationships between these factors and fuel economy are not simple. For instance, the implications of vehicle oper- ating speeds on fuel consumption are not linear and depend on vehicle type and size. Consequently, an approach that as- sumes an average fuel economy by vehicle category will not accurately account for the effects of transportation projects that address vehicle speeds and traffic flow. The effects of ve- hicle operating speeds on fuel economy also vary based on the model year and age of the vehicle; for instance, studies of vehicle fuel economy taken during the 1990s show less of a drop-off in vehicle fuel economy above 55 miles per hour than in similar studies of vehicles during the 1970s and 1980s, due to vehicle design changes affecting aerodynamics and engine operating efficiency, among other factors. (38) As a result, an approach that assumes a standard formula for the level of fuel consumed per mile at a certain vehicle speed will not accurately account for the effects of changes in vehicle designs over time. Nitrous Oxide and Methane Like CO2, N2O and CH4 are released during fossil fuel con- sumption, but in much smaller quantities. CH4 emissions are greater from alternative-fuel vehicles such as LNG trucks that store natural gas as a cryogenic liquid. To prevent build-up of pressure, gases are vented from the cryogenic tank, leading to fugitive emissions of CH4. However, since the market share of LNG vehicles is very small, these fugitive emissions do not impact the overall transportation GHG inventory. The emis- sions rates of N2O and CH4 are not directly proportional to fuel consumption. N2O and CH4 emissions rates per mile are affected by vehicle emissions control technologies. The newest motor vehicle emission control technologies produce significantly less N2O and CH4 than do early emission control technologies—for instance, for a gasoline powered automo- bile, a vehicle with LEV technology produces only about one-third the N2O emissions of a vehicle with Tier I emission controls. According to EPA, (1) N2O and CH4 from on-road sources declined by over 20% between 2000 and 2003 while VMT rose. As a result, emission factors for on-road vehicles are usually presented in per mile units, and analyses of these pollutants require information on VMT and the distribu- tion of miles by vehicle type (e.g., automobile, light-duty truck, heavy-duty truck), fuel type (e.g., gasoline, diesel), and technology type (e.g., Tier 0, Tier I, LEV). Knowing the emissions control technology used by vehicles is very im- portant for generating accurate results. A simple formula for the calculation of N2O or CH4 emissions for each individual vehicle/fuel/technology type is as shown in Equation 3. Emissions VMT Vehicle, Fuel, Technology Type= ( ) × (Equation 3) Emission Factor Vehicle, Fuel, Technology Type( ) For nonroad modes, N2O and CH4 are generally assumed to be proportional to fuel consumption, making the calcula- tion relatively simple. However, with the introduction of emission control technologies to nonroad sources, such as retrofits of diesel transportation construction equipment, more detailed analysis by control technology type may be needed to accurately address the impacts of these technologies on N2O and CH4. HFCs and Other Gases HFCs are synthetic chemicals that are used in vehicle air con- ditioning and refrigeration systems as alternatives to ozone- depleting substances being phased out under the Montreal Protocol. Leakage of HFCs during equipment operation, ser- vicing, and disposal also contributes to GHGs, so the level of HFCs released depends on factors such as air conditioning use and amount of refrigerated transport. Finally, the transportation sector also contributes to emis- sions of several other compounds that are believed to have an indirect effect on global warming. These include ozone, car- bon monoxide, and aerosols. Ozone traps heat in the atmos- phere and prevents a breakdown of CH4, but its lifetime in the atmosphere varies from weeks to months, making it difficult to estimate net radiative forcing effects. CO indirectly affects global warming by reacting with atmospheric constituents that would otherwise destroy CH4 and ozone. Aerosols are small airborne particles or liquid droplets that have both di- rect and indirect effects on global warming. The most promi- nent aerosols are sulfates and black carbon, or soot. Sulfate aerosols also have some cooling effect by reflecting light back into space. Scientists have not yet been able to quantify the impact of ozone, carbon monoxide, or aerosols with reason- able certainty; thus, these compounds are not included in re- porting GHG emissions. 3.1.2 Criteria Air Pollutants and Air Toxics Emissions of criteria air pollutants and air toxics are not di- rectly proportional to fuel consumption, with emissions rates per mile being affected by vehicle emissions control technolo- gies. Therefore, emission factors for on-road vehicles are usu- ally presented in per mile units, and analyses of these pollu- tants require information on VMT and the distribution of miles by vehicle type (e.g., automobile, light-duty truck, heavy- duty truck), fuel type (e.g., gasoline, diesel), and technology type (e.g., Tier 0, Tier I, LEV). Knowing the emissions con- trol technology used by vehicles is very important for gener- ating accurate results. Equation 3 shows a simple formula for the calculation of criteria air pollutants and air toxics for each individual vehicle/fuel/technology type. 32

For nonroad modes, the calculation of emissions of criteria air pollutants is similar but the measure of freight activity might be different (e.g., ton-miles in the case of line-haul rail). The main difference between criteria air pollutants and air toxics is data availability. Although most models include emission factors for all criteria air pollutants, the same is not true for air toxics due to a lack of data. Instead, many models estimate emissions from air toxics based on comparative ratios from criteria air pollutants. 3.2 National Methods At the national level, several inventories measure emissions associated with the transportation sector (see Exhibit 3-2). Each methodology discussed here is specific to classes of pol- lutants: greenhouse gas emissions are quantified in the EPA GHG Inventory, (1) and criteria air pollutants and air toxics are quantified in the NEI. (2) Both the EPA GHG Inventory and the NEI capture nationwide emissions across economic sectors; in addition to transportation, these inventories in- clude industrial, commercial, and residential emission sources. Because the methodologies of these inventories are consider- ably broader than the mode-specific methodologies con- tained in Sections 3.3 through 3.8 of this report, they are detailed independently of the modal analyses. This section discusses the strengths, weaknesses, inputs, and results of the EPA GHG Inventory and the NEI. 3.2.1 Summary of Methods and Models The purpose of EPA’s national inventories is to capture na- tional emissions across all sources and to allocate emissions to each sector. Although both the GHG Inventory and NEI report detailed emissions estimates, they differ in the com- plexity of analytical methods. The GHG Inventory uses a con- sistent methodology to calculate emissions for each category, but the NEI relies on unique methodologies across modes. The GHG Inventory primarily relies on fuel consumption data to calculate emissions. The inventory allocates emissions to each transportation mode, and to subcategories within each mode, according to fuel consumption and fuel type. Total GHG emissions are calculated as a function of the car- bon content of each fuel. Although the GHG Inventory does not disaggregate freight and non-freight emissions, it lists modal categories in sufficient detail to make such disaggrega- tion possible, albeit while introducing uncertainties into the calculations. Although the GHG Inventory uses a straightforward ap- proach to calculating emissions, the NEI methodology is com- paratively more complex. First, the NEI analyzes a greater number of pollutants than the GHG Inventory: 6 criteria pol- lutants and up to 188 air toxics. In addition, because the emis- sions of these pollutants depend on vehicle type, age, and activity, the NEI relies on separate methodologies for each transportation mode. Finally, the NEI has much more geo- graphic detail than the GHG Inventory. Although the latter only presents emissions at the national level, the former allo- cates emissions to the state and county level. For these rea- sons, the NEI methodology is presented here in much greater detail than the GHG Inventory. 3.2.2 EPA GHG Inventory Methodology In accordance with the 1992 United Nations Framework Convention on Climate Change, EPA produces an annual as- sessment of national greenhouse gas emissions, which spans several industries and economic sectors including transporta- tion. The analysis is based on methodologies, guidelines, and best practices established by the Intergovernmental Panel on Climate Change (IPCC), most recently updated in 2006. (39) Regarding transportation, the EPA GHG analysis calculates GHG emissions by measuring fossil fuel consumption in each transportation mode. Because emissions are broken down by mode rather than activity, the EPA inventory does not di- rectly quantify emissions associated with freight movement. The GHG Inventory accounts for emissions of three green- house gases: CO2, CH4, and N2O. The GHG Inventory does not measure HFC emissions. Although both CH4 and N2O have a greater global warming potential than CO2 (the global warming effect of CH4 is 21 times greater than that of CO2, and the effect of N2O is 310 times greater) their level of emis- sions is so small that their overall effect is negligible in this analysis. In the transportation sector, CO2 accounts for 98.4% of all greenhouse gases. (40) Since transportation emissions of CO2 are caused by the combustion of fossil fuels, such as gasoline, diesel, aviation fuel, and marine bunker oil, the CO2 emissions inventory is calculated by measuring fuel con- sumption from each mode. The GHG Inventory measures and reports greenhouse gas emissions on an annual basis at the national scale. Since 33 Exhibit 3-2. List of national methods. Method/Model Type Geographic Scale Pollutants Freight/Passenger EPA GHG Inventory Method National GHG Both NEI Method National, State, County CAP, HAP Both

emissions are not allocated to the state and county level, the inventory is less data intensive, and only requires aggregated national fuel consumption data. This makes the methodology less complex and reduces uncertainties from collecting and aggregating local data, but introduces additional uncertain- ties in allocating national data to the transportation sector, and to each mode individually. Since GHG emissions are re- ported for the analysis year, the GHG Inventory does not fur- ther break down the result by season or by month. Although there is no analysis of future years, the GHG Inventory in- cludes a comparison of current year emissions to past year emissions, back to 1990. Inventory Structure The EPA GHG Inventory is structured according to emis- sions category, including energy production, industrial processes, agriculture, and land-use change. Energy produc- tion accounts for the majority of emissions. In 2007, 80% of nationwide GHG emissions were due to fossil fuel combus- tion, and 26% of nationwide emissions were due to fossil fuels used in transportation. Within the transportation sector, emissions are divided by fuel type and subdivided by mode and vehicle type. Gasoline, consumed mainly by passenger cars and light-duty trucks, is the largest contributor to trans- portation emissions, followed by diesel fuel, consumed by heavy-duty trucks and rail, and jet fuel. Although the inven- tory does not separate freight and non-freight emissions, the specificity of the vehicle subcategories allows for freight emis- sions to be summed together across modes. Allocations are then checked against “bottom-up” fuel use data when avail- able, such as railroad fuel consumption data from the Surface Transportation Board. (41) GHG emissions are calculated using fuel-based (rather than activity-based) emission factors derived from the carbon intensity of each fuel. To perform this calculation, EPA col- lects data on total fuel sales and allocates fuel to each subcat- egory. The GHG methodology includes the steps shown in Exhibit 3-3. Although the process of calculating total fossil fuel emis- sions is straightforward, the allocation steps introduce uncer- tainties into the methodology. Each allocation, first to the transportation sector, then to each mode and vehicle type, re- quires additional assumptions and estimates. Although the GHG Inventory methodology does not further allocate emis- sions to freight and non-freight sources, this allocation can be made using additional assumptions about the freight mix of heavy-duty trucks, rail, and commercial aircraft. Summary of Strengths and Weaknesses. A summary of GHG Inventory methodology strengths and weaknesses is provided in Exhibit 3-4. Analysis of Process Uncertainty. In the EPA GHG Inven- tory methodology, the greatest elements of uncertainty are present in the allocation of GHG emissions to the transporta- tion sector and subsequently to individual modes. The sector- level allocation is achieved with a top-down approach that measures activity across all economic sectors. In comparison, the allocation across modes is achieved with a bottom-up ap- proach, which applies and compares activity levels for each mode. The uncertainty resulting from each allocation is dis- cussed here. A more thorough analysis of the parameter uncer- tainty surrounding each data set can be found in Section 3.2.4. Although data on total fuel use are considered accurate, the allocation of fuel consumption data to end-use sector relies on a variety of economic and activity measurements, which may reduce the accuracy of the allocation. Since each metric has its own sources of error, the allocation of fuel using sev- eral metrics creates further uncertainty. 34 Exhibit 3-3. GHG Inventory methodology. Determine total consumption by fuel type and sector Total fuel sales, available from the Energy Information Administration (EIA) are allocated by economic sector (e.g., industrial, commercial, transportation). Data for the overall allocation are supplemented by industry surveys and other end-use consumption metrics. Adjust transportation consumption based on activity measures EPA reconciles the transportation fuel allocation with VMT and other activity data from FHWA, AAR, and other sources. This “bottom-up” analysis serves as a check for fuel consumption and is the basis for the allocation among transportation modes. Allocate GHG emissions to transportation sector CO2 emissions are calculated based on the carbon content of each fuel; CH4 and N2O emissions are calculated based on an activity-based emissions factor. Allocate transportation emissions to each mode and vehicle subcategory For on-road vehicles, emissions are assigned based on VMT data from FHWA, specific to vehicle type. Nonroad data are assigned based on data from AAR, FAA, EIA, and other sources.

Within the transportation sector, fuel use is allocated to each mode through a comparison of modal activity factors. However, since each activity data set (i.e., VMT for on-road, ton-miles for rail) uses separate sources and methodologies, the margin of error in each data source is difficult to compare; it is not clear how the uncertainty in one would compare to the other. Although these uncertainties do not affect the quantification of emissions from the transportation sector, they do affect both the modal breakdown and the estimate of freight versus non-freight emissions. Uncertainties in allocation are partially addressed by com- paring the “top-down” allocation to “bottom-up” fuel con- sumption data. For example, railroads report fuel consump- tion data to the Surface Transportation Board. This value is compared against the determined allocation to identify the magnitude of discrepancy. This approach acts as a partial check to mitigate uncertainties in allocating fuel consumption. Although the GHG Inventory does not separate freight- related emissions from the total transportation inventory, it does present vehicle-specific emissions in sufficient detail to allow an estimation of freight emissions. However, this esti- mate requires a different approach for each mode, and relies on external assumptions about the proportion of freight versus passenger travel. For example, on-road categories include both gasoline and diesel medium- and heavy-duty trucks, and the aircraft category specifies emissions from commercial aircraft. 3.2.3 EPA National Emissions Inventory The NEI documents total emissions of criteria pollutants and air toxics nationwide. This database catalogs emissions from point, non-point, and mobile sources, with each trans- portation mode analyzed independently within the mobile analysis. Depending on the mode, emissions are determined using one of several possible methods: by applying computa- tional models, by combining activity data with emission fac- tors, or by scaling prior emission inventories by a growth factor. This section discusses how the NEI calculates modal emissions, and analyzes the strengths and weaknesses of each approach. An evaluation of analytical models (e.g., MOBILE6, NONROAD), as applied to each transportation mode, is discussed in Sections 3.3. to 3.8. Consistent with EPA’s mandates in the Clean Air Act as amended in 1990, the NEI measures nationwide emissions of 6 criteria pollutants and up to 188 air toxics. The measured criteria pollutants include CO, SOX, NOX, and PM. (There are two additional criteria air pollutants: lead and ozone— a secondary pollutant formed by the combination of HC and NOx.) Measured air toxics include 188 defined compounds. (42) However, not all HAPs are estimated by the mobile source methodologies. For example, the National Mobile In- ventory Model (NMIM) only produces inventories of 50 HAPs for on-road and nonroad sources. 35 Exhibit 3-4. Analysis of strengths and weaknesses—EPA GHG Inventory methodology. Criteria Strengths Weaknesses Representation of physical processes Methodology does not rely on models of physical processes for calculation of GHG emissions. EFs for CH4 and N2O are based on vehicle test data; EFs may become inaccurate if vehicle technology, maintenance, operations change. Model sensitivity to input parameters Since methodology is based on fuel use and activity data, it is not affected by changes in vehicle operations, maintenance, or environment. Ability to incorporate effects of emission reduction strategies Methodology does not forecast alternative scenarios to show benefits of emission reduction strategies. Representation of future emissions Methodology does not predict future trends in GHG emissions, but it does report historical emissions beginning in 1990. Consideration of alternative vehicle/fuel technologies Methodology captures GHG benefits of alternative fuels by including unique GHG EFs for each fuel type. Data quality High data quality. Fuel consumption and activity factors are industry standard; EFs for CH4 and N2O directly measured from vehicle tests. Spatial variability Methodology is only applied at the national level. It does not measure emissions at the regional or local level. Temporal variability Methodology is not subject to temporal fluctuations, since it measures emissions at the national scale. Endorsements Methodology is endorsed by EPA, UN IPCC.

The NEI produces inventory data for a wide range of geo- graphic scales, including the national, state, and county levels. This range of data presentation allows the inventory to inform air quality analyses by local, state, and federal government agencies as well as private industry. However, depending on the pollutant source, emissions data may be accurate at one geographic extent but inaccurate in other scopes or regions. For example, when source emissions are calculated in a “top- down” analysis, inventories at the state and regional level are apportioned from the national inventory. This process may introduce errors depending on the apportioning methodol- ogy and available data. Alternatively, inventories collected using a “bottom-up” approach may be accurate in certain regions with thorough data, but inaccurate in regions with little available data. These errors propagate to larger scopes as regional inventories are aggregated to the state and national level. A more thorough discussion of uncertainties in inven- tory apportionment and aggregation is presented later in this chapter. The NEI presents emissions data with a limited temporal scope. The inventory is calculated on an annual basis and is not broken down using a seasonal or monthly timeframe. In addition, the inventory is only published for the current year, and does not forecast emissions for future years. However, EPA publishes historical comparisons of the current-year in- ventory to past NEIs, (43) as well as a long-range analysis of emission trends from the year 1900. (44) EPA publishes the NEI on a three-year cycle; the most re- cent NEI was published in 2005 for the 2002 analysis year. In addition to the summary reports of emissions statistics, the NEI data are also distributed in database form. (45) As of 2009, EPA is finalizing the 2005 NEI, and collecting data for the 2008 analysis year inventory. NEI Structure and Methodologies The NEI is a comprehensive nationwide inventory from all stationary and mobile emission sources (see Exhibit 3-5). The breadth of data collection and modeling require unique methodological approaches for many emission sources, leading to a tiered or bottom-up structure for assembling the inventory. NEI calculations are separated into three components: point sources, non-point sources, and mobile sources. This section focuses on the mobile source compo- nent, because it includes all emissions from freight trans- portation. However, a brief discussion of point and non- point sources is included here in order to illuminate the scope of the NEI. Mobile Source Emissions All transportation-related emissions are captured within the mobile source component of the NEI. This category includes emissions from on-road vehicles, nonroad vehicles (cargo han- dling equipment), locomotives, commercial marine vessels, and aircraft. Each mode has a different approach to measur- ing emissions and apportioning the national inventory to states and counties. Although the inventory for each mode is calcu- lated independently, emissions from on-road and nonroad sources are grouped within the National Mobile Inventory Model (NMIM), a meta-model that collects input data and processes results for the two modes. 36 Exhibit 3-5. Structure of NEI methodology for mobile source emissions. EPA National Emissions Inventory (NEI) Mobile Sources Point Sources (industrial, commercial, etc.) Non-Point Sources (area sources) National Mobile Inventory Model (NMIM) On-Road Model: MOBILE6 Nonroad (CHE) Model: NONROAD Locomotive Model: none Apply EFs to fuel consumption data Commercial Marine (OGV, harbor craft) Model: none Carry forward prior inventory work Aircraft Model: EDMS Data from FAA LTO database

The methodological approach for each mode varies de- pending on the quality of data and tools available. Depend- ing on the mode, emissions are calculated by applying analyt- ical models, combining activity data with emission factors, or applying a growth factor to the results of prior inventories. As a result, the strengths, weaknesses, and uncertainties in the NEI mobile source inventory vary by mode. A summary of the methodologies is as follows: • On-road emissions, which include all heavy-duty trucks, are calculated using EPA’s MOBILE6 model, combined with nationwide data on vehicle activity from FHWA. When states provided alternate activity or other model inputs, the state-level data were used in place of EPA inputs. Emissions are allocated to the county level using NMIM. • The inventory for nonroad emissions, which include cargo handling equipment, is calculated using EPA’s NMIM, which calculates emissions through the NONROAD model. When states provided alternate model inputs, the state- level data were used in place of EPA inputs. Emissions are allocated to the county level using NMIM. • The locomotive emissions inventory is developed by com- bining locomotive fuel-use data from DOE with published criteria pollutant emission factors (EFs). HAP emissions are calculated by applying speciation profiles to VOC and PM estimates. Emissions are allocated to the county level using rail network data from U.S.DOT. • The 2002 NEI inventory for commercial marine vessels (harbor craft, inland vessels, and ocean-going vessels) was based on emissions estimations produced for “marine diesel regulations for 2000.” Port emissions were disaggregated based on cargo volume, and underway emissions were dis- aggregated based on United States Army Corps of Engi- neers (U.S. ACE) waterway data. When state data were available, they were used in place of EPA inputs. HAP emissions were calculated by applying speciation profiles to VOC and PM estimates. • Emissions from commercial aircraft are calculated by ap- plying airport activity data to FAA’s Emissions Disper- sion Modeling System (EDMS) model. HAP emissions are calculated by applying speciation profiles to VOC and PM estimates. The EDMS inventory is measured on the county-level scale; state and national emissions are calcu- lated by aggregating project-level emissions. • The mobile component of the NEI does not include emis- sions from pipelines. NEI On-Road Methodology The NEI mobile source component includes a methodol- ogy for calculating criteria pollutant and HAP emissions asso- ciated with on-road vehicles, including heavy-duty vehicles. The methodology collects county-level vehicle data, calculates emissions using the MOBILE6 model, and allocates the result- ing emissions inventory to the state and county level. The process is achieved using the National Mobile Inven- tory Model (NMIM), which operates above MOBILE6, pre- processing input data and postprocessing emission results. NMIM contains a database of all county-level information re- quired to run the emissions model. The NMIM County Data- base, used for both on-road modeling with MOBILE6 and nonroad modeling with NONROAD, contains detailed infor- mation on vehicle activity, fleet mix, and infrastructure. This information, in addition to county-level meteorological and fuel data, comprises a complete set of data inputs for MOBILE6. In the postprocessing phase, NMIM combines emissions re- sults from MOBILE6 with nonroad emissions, and reallocates the resulting inventory to the state and county level in a form consistent with other components of the NEI. Where states provide alternate inputs into NONROAD, these values are used in place of the default NMIM inputs. Although the general approach used in this methodology has remained consistent since 1990, details of its application continue to evolve. In December 2008, EPA issued updated guidance for the on-road methodology used for the 2005 NEI. The new methodology is more consistent than in prior years, and relies on the MOBILE6 model to compute on-road emis- sions throughout the United States, Puerto Rico, and the Vir- gin Islands. In past years, when state emissions inventories were available, notably in California, Colorado, and Oregon, the state inventories were used in place of EPA emissions cal- culations. However, the 2005 NEI methodology still gives precedence to state-level VMT and activity data when avail- able. More information about the development and valida- tion of the NEI can be found in the 2002 National Emission Inventory (NEI) Preparation Plan—Final. (46) Summary of Strengths and Weaknesses. A summary of NEI on-road methodology strengths and weaknesses is pro- vided in Exhibit 3-6. Sources of Uncertainty. The NEI on-road methodology introduces uncertainty into several aspects of the approach, from data collection to inventory assessment. This section focuses on uncertainties unique to the NEI method, includ- ing uncertainties in allocating emissions across geographic scales and uncertainties in disaggregating emissions into freight and non-freight inventories. Uncertainties associated with MOBILE6 and estimation of truck VMT are discussed in Section 3.3. Uncertainty also exists in the way that NMIM aggregates emissions results from the project-level level. The approach used in NMIM introduces uncertainties about the accuracy of state and national emissions. NMIM uses county-level data 37

sets to calculate local emissions inventories; as such, the ac- curacy of state and national emissions results depends on the accuracy of county data. Although EPA maintains default data sets on each county, local agencies have the opportunity to supplement or replace EPA values with more accurate county-specific data. Since the accuracy of emissions inven- tories varies by county, any county-level errors will propagate upward when local results are aggregated to the state and national level. Using data outputs from NMIM, the NEI methodology al- lows users to disaggregate freight emissions at a high degree of detail. NMIM reports annual emissions by pollutant and by vehicle category. The specified vehicle types are referenced from the MOBILE6 model, and include light-duty vehicles (passenger cars), light-duty trucks, medium-duty trucks, and heavy-duty trucks. Data are further allocated to gasoline and diesel categories. For example, freight emissions can be deter- mined by selecting emissions from certain vehicle classes, such as Class 8B heavy-heavy-duty trucks. Because county- level data are typically not reported with the same level of de- tail, the NEI relies on MOBILE6’s default VMT distribution among truck classes, which is based on national default pa- rameters. National parameters are a poor surrogate for local parameters since the distribution of VMT by truck classes should not be consistent across different counties. As a re- 38 sult, there are uncertainties associated with disaggregating county data to a level that is more detailed than was originally reported. NEI Rail Methodology In the NEI, EPA divides rail transportation into the follow- ing five categories: • Line-haul service (Class I), • Regional and local service (Class II/III), • Railyard, • Passenger, and • Commuter. Freight transportation is represented in the first three cat- egories, with the majority of emissions generated by line-haul transportation. Unlike the methodologies for on-road, nonroad, and air- craft emissions, the NEI rail methodology does not rely on analytical models to calculate an emissions inventory. In- stead, emissions are calculated directly from industry-wide fuel usage data, and combined with fuel-based emissions fac- tors. Data on rail fuel consumption, reported by EIA, are al- located to individual rail categories according to established Exhibit 3-6. Summary of strengths and weaknesses—NEI on-road methodology. Criteria Strengths Weaknesses Representation of physical processes Method represents physical processes through MOBILE6 model. Sensitivity to input parameters Method utilizes detailed facility-level and meteorological data to account for operational fluctuations in emissions. Method has significant data reporting requirements at the county level; relies on state and local agencies for accurate input. Flexibility Method is flexible enough to be applicable to all counties. Data reporting requirements are high, in fixed format. Ability to incorporate effects of emission reduction strategies Method can include county-level emission inspection and maintenance programs. Representation of future emissions Method does not predict future emissions. Consideration of alternative vehicle/fuel technologies Method can incorporate alternative fuels, low emission vehicles. Data quality EPA performs data checks and follows up with states and local agencies regarding discrepancies. Spatial variability Method incorporates variations in altitude and meteorology by county. Temporal variability Method only calculates annual emissions. It does not evaluate emissions or fluctuations on seasonal or monthly scales. Review process Draft NEI made available for public and peer review, comment, revisions. Endorsements EPA.

category ratios developed for the NEI; this fuel allocation is examined more closely in Section 3.4.4. Since California re- quires low-sulfur fuel for in-state locomotives, the emis- sions calculations are performed separately, although the same methodology is employed. Since the emissions inventory is created using national data, it must be distributed to the state and county level using a top- down approach. Emissions are allocated to counties based on county-level rail activity data, which is provided by the Bureau of Transportation Statistics (BTS). A GIS analysis allocates traffic on rail segments to each county. The inventory of railyard emissions is allocated spatially using a separate ap- proach, in which emissions are allocated to urban counties containing Class I railyards. Summary of Strengths and Weaknesses. A summary of NEI rail methodology strengths and weaknesses is provided in Exhibit 3-7. Sources of Uncertainty. The most accurate data for rail emission calculations are where fuel is purchased and added to locomotives, as well as rail activity (in ton-miles) at the state level. The burn ratio in gallons per ton-mile, and the al- location of rail activity to regions are the least known param- eters. The NEI rail methodology introduces two principal sources of uncertainty into emissions calculations. One in- stance, discussed in this section, occurs when the NEI distrib- utes rail consumption data to each rail category (i.e., line- haul, Class II/III). Additional sources of uncertainty, including 39 the challenges of allocating emissions to the county level using BTS activity data, are not unique to the NEI and are examined more fully in Section 3.4. This methodology relies on fuel sales data to estimate rail emissions, which—while simplifying the analysis—adds chal- lenges in data collection. Fuel consumption data are most read- ily available at the national level for the entire rail industry and are reported annually by EIA. However, an accurate represen- tation of rail emissions requires more detailed fuel consump- tion data for each rail company. Although aggregated fuel consumption information is available from the Surface Trans- portation Board, more detailed data are often unavailable, because many companies view fuel consumption as propri- etary information. To distribute fuel consumption among each rail category, EPA devised Source Classification Code (SCC) Ratios, or activity correction factors that express the ratio of fuel usage attributable to each rail class. For example, EPA determined through an analysis outside the NEI that Class I line-haul rail accounts for 85% of rail fuel consumption, and allocates fuel use to Class I according to this ratio. However, EPA’s methodology for developing these SCC Ratios is poorly documented, and it is difficult to evaluate their accuracy. If the values were developed using a limited data set, then they may introduce considerable uncertainty into the analysis. NEI Commercial Marine Vessel Methodology The commercial marine vessel (CMV) methodology ac- counts for emissions from marine transportation. It is broken Exhibit 3-7. Summary of strengths and weaknesses—NEI rail methodology. Criteria Strengths Weaknesses Representation of physical processes Does not address physical processes; applies average EF to fuel consumption. Sensitivity to input parameters Insensitive to all parameters aside from fuel consumption and freight volume. Flexibility Method has little flexibility in data sources and parameters. Ability to incorporate effects of emission reduction strategies None. Representation of future emissions Method does not forecast future emissions. Consideration of alternative vehicle/fuel technologies Uses California-specific data to account for cleaner fuel. Data quality Emission factors based on EPA locomotive standards. (47) Spatial variability Does not account for geographic variations in terrain, speeds. Temporal variability Method only calculates annual emissions. It does not evaluate emissions or fluctuations on seasonal or monthly scales. Review process Draft NEI made available for public and peer review, comment, revisions. (46) Endorsements EPA.

down into different categories based upon engine size as shown in Exhibit 3-8. Category 1 and 2 CMVs include “all boats and ships used either directly or indirectly in the conduct of commerce or military activity.” (48) CMVs can range from 20-ft charter boats to 1,000-ft tankers and military vessels. Although the majority of marine vessels are included in this source cate- gory, recreational marine vessels are classified as nonroad ve- hicles and included in the nonroad category. Category 1 and 2 CMV emissions inventories for years 2005 and 2002 are based on emission estimates EPA performed for the Draft Regulatory Impact Analysis Control of Emissions from Compression-Ignition Marine Engines. (49) This document uses a bottom-up approach to quantify total marine emis- sions. First, an engine inventory is built using data on engine sales and scrappage, to which annual load factors are applied in order to calculate total marine engine activity. Total CMV emissions are calculated by combining activity levels with emission factor standards set in the RIA. This emission inven- tory was carried forward to NEI 2002 and 2005. Before allocating the inventory to the county level, the NEI methodology divides emissions by mode of operation: in/near port and underway operation. The disaggregation follows EPA SIP guidance that 75% of distillate fuel and 25% of resid- ual fuel is consumed while in/near port. (50) This separation into port emissions and underway emissions allows a more precise geographic allocation. The method for allocating emis- sions geographically is more complex than with on-road or nonroad vehicles, since CMV emissions impact only selected counties. The port emissions are allocated among the 150 largest U.S. ports, based on total port traffic. Underway emis- sions are allocated through a GIS-based approach that over- lays shipping lanes and waterways with county borders. Based on this analysis, emissions are allocated to counties based on ton-miles of cargo in adjacent waterways. Port and waterway data were supplied by the Army Corps of Engineers, and GIS data were supplied by BTS. Category 3 NEI inventory includes emissions from both propulsion and auxiliary engines. The inventories include both near-port emissions as well as the inter-port (under- way) emissions from these vessels when operating away from port in U.S. waters. The boundaries for vessels oper- ating in the oceans generally extend from the U.S. coastline to the 200 nautical mile limit of the Exclusive Economic Zone (EEZ). For ships operating in the Great Lakes, the boundary extends out to the international boundary with Canada. Emissions were developed separately for near-port and un- derway emissions. For near-port emissions, inventories for 2002 were developed for 89 deep water and 28 Great Lakes ports in the United States. The Waterway Network Ship Traf- fic, Energy, and Environmental Model (STEEM) provides emissions from ships traveling in shipping lanes between and near individual ports. (51) Near-port inventories were per- formed in a manner similar to the mid-tier methodology dis- cussed in Section 3.5.2. These emissions were married with the STEEM data, and replaced the less accurate near-port estimates in STEEM. Port call data came from the Army Corps of Engineers’ entrances and clearances data set, which is also discussed in Section 3.5.2. Where state agencies had developed a state-wide CMV emissions inventory, these values were given precedence over EPA calculations. As more states perform their own invento- ries, this inclusion leads to a less consistent overall method- ology. In the 2002 NEI, 26 states submitted statewide inven- tories, and the NEI methodology was applied to emissions in the remaining states and territories. Summary of Strengths and Weaknesses. A summary of NEI (46) marine methodology strengths and weaknesses is provided in Exhibit 3-9. Sources of Uncertainty. Category 1 and 2 inventories rely on engine counts determined from Power Systems Research. This database does not determine how many engines are on each vessel or accurately determine usage or load factors as discussed in Section 3.5.3. Category 3 data rely on foreign cargo movements and a somewhat streamlined methodology that uses detailed data from typical ports to estimate emissions at other ports. This is discussed in detail in Section 3.5.2. 40 Exhibit 3-8. EPA marine compression-ignition engine categories. Category Specification Use ApproximatePower Ratings 1 Gross Engine Power ≥ 37 kW*Displacement < 5 liters per cylinder Small harbor craft and recreational propulsion < 1,000 kW 2 Displacement ≥ 5 and < 30 liters per cylinder OGV auxiliary engines, harbor craft, and smaller OGV propulsion 1,000 – 3,000 kW 3 Displacement ≥ 30 liters per cylinder OGV propulsion > 3,000 kW * EPA assumes that all engines with a gross power below 37 kW are used for recreational applications and are treated separately from the commercial marine category.

NEI Nonroad Methodology The NEI nonroad category encompasses a wide array of vehicles—essentially all motorized vehicles and equipment that are not normally operated on public roads. This category also excludes locomotives, commercial marine vessels, and aircraft, which are analyzed separately. The nonroad category extends to a variety of fuel types, including diesel, gasoline, compressed natural gas (CNG), and liquefied petroleum gas (LPG). The fol- lowing types of vehicles are included in the nonroad analysis: • Freight cargo handling equipment (CHE); • Airport ground support equipment (GSE); • Recreational vehicles and equipment (marine and land based); • Farm and construction machinery; and • Industrial, commercial, and lawn and garden equipment. The approach employed in this methodology is similar to the approach in the on-road methodology: activity, engine mix, and fuel data are collected at the county level, emissions are calculated using EPA’s NONROAD model, and the result- ing national inventory is apportioned to the state and county levels. As in the on-road inventory, data collection and dis- aggregation is handled by NMIM, which operates above NONROAD. NMIM formats county data into input files for NONROAD, runs the model, and processes the results to be consistent with other components of the NEI. Where states provide alternate inputs into NONROAD, these values are used in place of the default NMIM inputs. The nonroad methodology has evolved as better tools and data have emerged. The NONROAD model was first applied to this category in 2001 for the 1996 inventory. In past years, state emissions inventories were used in place of EPA cal- culations, notably in California, Pennsylvania, and Texas. Inventories for years prior to 1996 were developed retro- actively using NONROAD. Summary of Strengths and Weaknesses. A summary of NEI (46) nonroad methodology strengths and weaknesses is provided in Exhibit 3-10. Sources of Uncertainty. There are sources of uncertainty in all nonroad methodologies in terms of data collection, equipment emission factors, and other factors. These topics are discussed in more detail in Section 3.6. NEI Aircraft Methodology The NEI aircraft methodology captures emissions from all domestic and international aircraft operating within the United States. Aircraft are classified by EPA into four cate- gories: commercial, air taxi, general aviation, and military. This analysis focuses on emissions from commercial aircraft used to carry freight, passengers, or both. Commercial aircraft tend to be large, powered by jet engines, and operate at large 41 Exhibit 3-9. Summary of strengths and weaknesses—NEI marine methodology. Criteria Strengths Weaknesses Representation of physical processes Relies on port call activity for Category 3. None. Methodology does not include physical processes in inventory for Category 1 and 2. Sensitivity to input parameters None. Methodology based on estimate of engine inventory for Category 1 and 2. Relies on port call data from U.S. ACE and STEEM for Category 3. Flexibility None. Methodology relies on inventory constructed for year 2000 for Category 1 and 2. Relies on top 117 ports for Category 3. Ability to incorporate effects of emission reduction strategies None. Representation of future emissions None. Consideration of alternative vehicle/fuel technologies None. Does not consider benefits from new fuels. Data quality Emissions calculations rely on assumptions related to equipment inventory. Spatial variability Allocates emissions locally according to county- level marine activity. Temporal variability None. Review process Draft NEI made available for public and peer review, comment, revisions. Endorsements EPA.

Airport statistics are input into the EDMS model, which combines LTO data with emissions factors that are specific to each type of aircraft and each phase of the LTO cycle. EDMS uses default time-in-mode (TIM) values to determine total time spent by aircraft in each LTO mode, and calculates emis- sions using measured emissions factors. See Section 3.7 for more information on EDMS. Summary of Strengths and Weaknesses. A summary of NEI (46) aircraft methodology strengths and weaknesses is provided in Exhibit 3-11. Sources of Uncertainty. Sources of uncertainty are dis- cussed in Section 3.8. 3.2.4 Evaluation of Parameters The GHG Inventory and NEI methodologies include an analysis of all transportation modes; as such, many data inputs required for these analyses are the same as inputs re- quired for each modal methodology, as discussed in sub- sequent chapters. This section focuses on parameters that are unique to the EPA national methodologies. These parameters are all unique to the process of measuring and allocating fuel consumption at the national level. airports. Emissions from these aircraft are calculated by com- bining airport activity data with FAA’s EDMS emissions model. Since the inventory is estimated independently for each airport, emissions are allocated to the county level by default. State and national emissions are calculated by aggre- gating county-level emissions. Data on airport activity is measured in terms of the Landing and Takeoff (LTO) cycle, a five-mode approach consisting of the following: • Approach: period beginning when aircraft enters the pol- lutant “mixing zone,” typically at an altitude of 3,000 ft, until landing; • Taxi/idle-in: time spent after landing until aircraft is parked at the gate and engines turned off; • Taxi/idle-out: period from engine startup to takeoff; • Takeoff: time spent after takeoff that lasts until the aircraft reaches 500 to 1,000 ft; • Climbout: period following takeoff that concludes when aircraft passes out of mixing zone. LTO data is collected in Airport Activity Statistics of Cer- tificated Air Carriers, (52) which captures statistics for all domestic carriers. Each LTO cycle is correlated with an air- port location, carrier, and aircraft type. 42 Exhibit 3-10. Summary of strengths and weaknesses—NEI nonroad methodology. Criteria Strengths Weaknesses Representation of physical processes Does not address physical processes; applies average EF to equipment inventory. Sensitivity to input parameters Flexibility Although the NONROAD model has default equipment distributions, local agencies can submit county-specific data. Ability to incorporate effects of emission reduction strategies Does not incorporate inspection/maintenance profiles or other strategies. Representation of future emissions Does not forecast future emissions. Consideration of alternative vehicle/fuel technologies Can incorporate alternative fuels such as LPG and CNG/LNG. Data quality Can incorporate county-level data submitted by local agencies. Quality of data will vary depending on locality. Default parameters may not capture spatial variations. Spatial variability Does not account for the effect of geography on emissions estimates. Temporal variability Method only calculates annual emissions. It does not evaluate emissions or fluctuations on seasonal or monthly scales. Review process Draft NEI made available for public and peer review, comment, revisions. Endorsements EPA.

The parameters discussed in this section are used in allo- cating fuel consumption to the transportation sector and to individual modes, and are used in one or both of the EPA na- tional methods. A summary of parameters is presented in Exhibit 3-12, and more detailed information is provided in the pedigree matrix (Exhibit 3-13) and the subsequent qual- itative discussion. Pedigree Matrix A pedigree matrix, provided in Exhibit 3-13, for data qual- ity assessment assigns quantitative scores to all parameters in- cluded in Exhibit 3-12. The criteria to assign scores in the pedigree matrix are included in Appendix A. Parameters Used in Fuel Consumption Calculations Since transportation emissions in the EPA GHG Inventory are due to the combustion of fossil fuels, the primary input into the inventory is data on fuel consumption within each mode. Although some data sources capture fuel use in indi- vidual modes (e.g., rail), EPA chooses a methodology that measures nationwide fuel consumption and allocates fuel use to economic sectors such as industrial, residential, and trans- portation. This approach has several benefits: it relies on comprehensive fuel data available from EIA, and it accurately measures GHG emissions due to fuel use for the nation as a whole. However, the process introduces uncertainties when fuel use and GHG emissions are assigned to the transportation 43 Exhibit 3-11. Summary of strengths and weaknesses—NEI aircraft methodology. Criteria Strengths Weaknesses Representation of physical processes Accounts for variations in emissions between aircraft engines, between LTO modes. Sensitivity to input parameters Sensitive to activity by type of aircraft Flexibility Can include changes in activity at any airport Ability to incorporate effects of emission reduction strategies None. Representation of future emissions None. Consideration of alternative vehicle/fuel technologies None. Data quality FAA maintains detailed activity (LTO) records. Spatial variability Accounts for activity and fleet mix at each airport. Does not incorporate local meteorology. Temporal variability None. Emissions reported annually. Review process Draft NEI made available for public and peer review, comment, revisions. Endorsements EPA. Exhibit 3-12. Parameters for the national inventories. Parameter Methods/Models GeographicScale Pedigree Matrix Qualitative Assessment Quantitative Assessment Fuel Supply Data GHG Inventory National Economic Sector Activity Data GHG Inventory National Modal Activity Data GHG Inventory, NEI National Fuel Carbon Content GHG Inventory National Modal Emissions Factors NEI National Marine Equipment Inventory NEI National Rail GIS Data NEI National Local Nonroad Equipment Inventory NEI National On-Road Fleet Mix NEI National Key: indicates that a parameter is analyzed in the way denoted by the column: indicates that the parameter is not discussed in the way denoted by the column.

sector and to individual modes. This section qualitatively analyzes the assumptions made when estimating modal fuel consumption. Measuring Nationwide Fuel Use—Fuel Supply Data. The combustion of fossil fuels accounted for 83% of nation- wide GHG emissions in 2007. (53) Since emissions from fuel combustion constitute the vast majority of the inventory, the need to accurately measure nationwide fuel use is paramount. EPA measures total fuel consumption in the United States using data from EIA, primarily the agency’s Monthly Energy Review, and additional petroleum product detail. The Monthly Energy Review reports data on both fuel production (petro- leum imports, domestic production, and refining) as well as consumption (by fuel and end-use sector). The fuel produc- tion data conforms to a reporting convention promulgated by IPCC and the International Energy Agency (IEA), in which data are presented in a top-down format. This structure ag- gregates data on fuel production and distribution to assess fuel use, referred to as “apparent consumption.” These data are used by the GHG Inventory as the first step in allocating fuel consumption. This step in the process contains few uncertainties com- pared to subsequent steps. The collection of national fuel data contains few assumptions, since EIA has comprehensive ac- cess to primary sources of information. Larger uncertainties occur in the following steps in which national data are allo- cated to the transportation sector and individual modes. Assigning Fuel Use to the Transportation Sector— Economic Sector Activity Data. After collecting national fuel consumption data, the EPA methodology distributes fuel use among economic sectors, to determine the GHG emis- sions attributable to each sector. As part of this step, EPA rec- onciles the results of a top-down approach, based on EIA data, with the results of a bottom-up approach, based on in- dustry activity measurements. Consistent with IPCC guide- lines, the bottom-up (or sectorial approach) relies on several data points, including consumption data by EIA and end-use energy consumption surveys such as the Manufacturing En- ergy Consumption Survey, which is conducted every four years. Additional information is used to adjust fuel consump- tion for the transportation sector: EPA builds an activity- based estimate of fuel consumption from modal data, includ- ing FHWA statistics for on-road activity and AAR statistics for rail activity. Several potential sources for error exist in applying this method to allocate fuel consumption to the transportation sector. These include • Consumption data, often collected in the form of fuel ex- penditures, may distort true fuel usage. For example, collec- tion methods may focus on large, more efficient consumers, and bypass smaller entities that may use comparatively more fuel. In addition, data based on fuel prices may be biased as larger consumers can often leverage lower prices due to high purchasing volume. • Transportation activity data, collected for each mode inde- pendently by separate agencies, may contain different biases and errors due to differing methodologies. Further, activity sets may be incomplete for modes with limited information such as commercial vessels and nonroad equipment. 44 Exhibit 3-13. Pedigree matrix—national parameters. Parameter Im pa ct on R es ult Ac qu isi tio n M eth od In de pe nd en ce Re pr es en tat ive ne ss Te m po ra l C or re lat ion Ge og ra ph ic Co rre lat ion Te ch no lo gic al Co rre lat ion Ra ng e o f V ar iat ion Fuel Supply Data 5 1 1 1 1 1 N/A 1 Economic Sector Activity Data 4 3 Varies 2 Varies 1 N/A 5 Modal Activity Data 4 3 Varies 2 1 1 N/A 5 Fuel Carbon Content 5 1 1 1 1 1 1 1 Modal Emissions Factors 4 2 3 2 3 2 2 Varies Marine Equipment Inventory 2 3 3 4 3 2 N/A 5 Rail GIS Data 2 2 1 1 2 1 1 2 Local Nonroad Equipment Inventory 2 3 3 2 3 5 N/A Varies On-Road Fleet Mix 2 2 2 2 1 2 N/A 2

Assigning Fuel Use to Vehicle Types—Modal Activity Data. This stage of distribution allocates total transporta- tion fuel consumption to individual modes and sub-allocates to vehicle types. The modal distribution is completed using a combination of data from the activity analysis conducted in the prior step and EIA data on individual fuel types. These two data sources serve to confirm or reconcile differences in reporting. For example, rail fuel statistics are reported by AAR based on company surveys, while the same data are reported by EIA based on responses from fuel distributors. Similar comparisons can be conducted for aviation fuel and marine bunker fuel. The distribution of gasoline and diesel fuel is more complex, as the fuels are used in several modes, but is conducted using activity data. However, the distribution of fuel within vehicle types requires additional data and as- sumptions. For on-road vehicles, the distribution among pas- senger cars, light-duty trucks, and medium- and heavy-duty trucks is completed using detailed VMT data from FHWA. Similarly, aircraft consumption can be separated into com- mercial and other sources using FAA flight records. However, there is no comparable detailed source of information for dis- tributing fuel use among categories of marine vessels. To the extent that GHG inventories by vehicle subcategory can be used to inform an analysis of an individual vehicle type, this step in the methodology can introduce additional uncertainties into future analyses. Sources of error include • For on-road vehicles, this step requires data on vehicle fuel types (gasoline versus diesel) as well as activity by vehicle type. Since these two data sources are maintained by sepa- rate agencies, EIA and FHWA, respectively, their category definitions and relationships may not align. This issue is magnified when considering alternative fuels with a small vehicle share, such as CNG and LPG. The GHG Inventory does not disaggregate alternative fuel usage to vehicle categories. • Uncertainty exists in activity data in the nonroad category, including the comparative activity of mobile nonroad ve- hicles versus stationary nonroad equipment. This creates added challenges in correctly allocating fuel use and emis- sions to the nonroad mobile category. The GHG Inventory does not distinguish between emissions from construction equipment and agricultural machinery, and emissions from nonroad trucks. 3.3 Heavy-Duty Trucks This section includes (1) a brief documentation of the cur- rent practice and methodologies for calculating emissions from heavy-duty trucks, (2) a summary of the strengths and weaknesses of such methods, and (3) an analysis of uncer- tainty associated with these methods, as well as with the pa- rameters used in the emission calculations. Although the estimation of truck emissions is conceptually simple (i.e., emissions are the product of freight activity and emission factors), the analytical procedures for emission estimation can be quite complex depending on the goals of the analysis and the level of data and resource availability. Exhibit 3-14 summarizes the main methods and models to estimate truck emissions. 3.3.1 Evaluation of Emission Models Despite the high number of existing emission models, this section focuses on the four most widely used models. EPA’s MOBILE6 and CARB’s EMFAC2007 are the two approved models for State Implementation Plan (SIPs), conformity analyses, and project-level analyses under NEPA and CEQA, respectively. EPA’s MOVES2009, which brings many methodological improvements over MOBILE6, is cur- rently in draft form, but will eventually replace MOBILE6. The Comprehensive Modal Emissions Model (CMEM), de- veloped by UC Riverside, is the most established micro-scale emissions model. MOBILE6 MOBILE6 is an emission factor model designed by EPA to produce motor vehicle emission factors for use in trans- portation analyses, including SIP development, transportation 45 Exhibit 3-14. List of truck methods and models. Method/Model Type GeographicScale Pollutants Freight/Passenger MOBILE6 Model All All Both MOVES2009 Model All All Both EMFAC2007 Model All All Both CMEM Model Local All Both Regional Method Method Regional All Both Local Method Method Local All Both

conformity, and project-level analysis required under NEPA. It can be used at any geographic level within the United States. With the release of MOBILE6 in 2001 came several im- provements regarding heavy-duty vehicles (HDVs) over its previous version, MOBILE5: (1) increase in the number of HDV categories, (2) addition of off-cycle NOx impacts as a re- sult of control strategies that optimize fuel economy over emissions (i.e., defeat device issue), and (3) incorporation of 2004 and 2007 HDV emission standards, including the use of low-sulfur fuel starting in 2006. (54) Summary of Strengths and Weaknesses. A summary of MOBILE6 strengths and weaknesses is provided in Exhibit 3-15. Analysis of Process Uncertainty. The emissions rates generated by MOBILE6 require a multitude of input assump- tions that can either be MOBILE6’s national defaults or user- specified parameters. MOBILE6 is particularly sensitive to assumptions regarding vehicle age, VMT by vehicle class, aver- age speeds, and temperature. (55) This discussion focuses on the following key issues: (1) emission factors, (2) truck age distri- 46 Exhibit 3-15. Strengths and weaknesses—MOBILE6. Criteria Strengths Weaknesses Representation of physical processes EFs incorporate effects of vehicle average speeds. EFs are based on engine testing (rather than chassis dynamometer testing). EFs are based on a single driving cycle. Model assumes that brake and tire EFs, which are based on passenger cars, are the same for HDVs. PM EFs are based solely on heavier truck classes; therefore PM emissions from the lighter classes of HDVs might be overestimated. There are concerns about how speed correction factors capture speed/congestion effects on emissions. Data on age distribution, mileage accumulation rates, and fuel ratios are not available for all truck categories. Model does not consider high emitters or mal maintenance and tampering for HDVs. Model does not consider start emissions for diesel vehicles. Model sensitivity to input parameters Number of engine starts and soak time cannot be modified by user. Other than HC, CO, and NOx, emissions of other pollutants are not sensitive to vehicle average speed or facility type. EFs do not take air conditioning effects into account. Model does not consider effects of road grade or pavement quality. Model flexibility Few inputs are required. National default parameters (VMT mix by vehicle class, vehicle age distribution) can be overridden by local estimates. Ability to incorporate effects of emission reduction strategies Model is able to capture the effects of strategies that change truck VMT, vehicle average speed (for HC, CO, NOx), fleet average age. Model is not able to capture the effects of strategies that affect pavement quality or congestion level. Representation of future emissions EFs can be estimated up to 2050. There are concerns as to whether the assumptions used to estimate future EFs are still in line with latest vehicle technology trends. Data quality Data based upon engine testing and conversion factors are applied to calculate grams per mile. These conversion factors are fixed by weight class and may not be representative of heavy-duty freight trucks. Spatial variability There are concerns as to whether national defaults are representative of regional and local parameters. Temporal variability Review process There have been many independent analyses and reviews of MOBILE6. Endorsements MOBILE6 is the required model for SIPs and conformity analyses.

bution and mileage accumulation, (3) how truck speeds affect truck emissions, (4) high emitters, (5) diesel fraction, (6) start emissions and soak time, and (7) classification of trucks. Emission Factors—General. The emission factors in MOBILE6 were based on engine test data submitted by man- ufacturers as part of the certification process (in g/bhp-hr). (54) As a result, emission factors needed to be converted (to grams/mile) based on fuel density, brake-specific fuel con- sumption (BSFC), and fuel economy. Because of the wide variation in gross vehicle weight, fuel economy, horsepower ratings, and transmission types, the gram/mile emission fac- tors derived from engine test data had much higher uncertain- ties than those calculated by vehicle dynamometer testing. This report (54) also compared MOBILE6 emission factors for HDVs with chassis dynamometer data. Results indicated that HC and CO emissions in MOBILE6 matched well with available test data, while NOx emissions seem to be overesti- mated for older models (before 1979) and underestimated for newer models (1994 and later). MOBILE6 is not designed to measure second-by-second emission rates, but it relies on specific driving cycles to gen- erate emission rates. For light-duty vehicles, there are differ- ent driving cycles assumed for each of the four facility types. For heavy-duty trucks however, all vehicle categories are based on the same driving cycle—the Federal Test Procedure (FTP) transient cycle—independently of the facility type. Ad- ditionally, MOBILE6 does not incorporate the effects of road grade, actual vehicle weight, or vehicle aerodynamics, all of which have a strong effect on emission factors. Emission Factors—PM. Previous research indicated sev- eral deficiencies in the estimation of PM emission factors in MOBILE6, mainly from carrying over the algorithms from PART5, the previous model for PM emission factors. (56) PART5 is believed to underestimate emissions from real ve- hicles, primarily because it is based on low-mileage, proper functioning vehicles, and does not consider high emitters to the same degree. (57) MOBILE6 accounts for the implementation of the 2007 PM emission standards for HDVs, which require the imple- mentation of low-sulfur diesel fuel (15 ppm limit) and a 90% reduction in exhaust PM emission standards for HDVs. (58) The assumptions associated with brake and tire PM emissions were not affected. A significant shortcoming of MOBILE6 is that it assumes the same brake and tire PM emission factors (in grams/mile) for all vehicle classes. Because these factors were developed from passenger car testing, brake and tire PM emissions from HDVs are likely underestimated. This is the case because brake and tire wear should be proportional to the energy required to stop a vehicle, which, in turn, is a func- tion of vehicle weight and speed. PM idling emission rates in MOBILE6 (reported in grams/ hour) are based on the heavier classes of HDVs, so MOBILE6 likely overestimates idle PM rates from lighter classes of HDVs. Additionally, these rates are not corrected for diesel sulfur content, nor do they account for more stringent PM standards in the 2007 rule. Emission Factors—Air Toxics. Data on air toxics emis- sions from HDVs are very sparse, and emission factors used in MOBILE6 are based on very few data points. (58) The imple- mentation of the 2007 standards, which are likely to require particulate filters, will certainly increase the margin of error of current air toxic emission factors for HDVs in MOBILE6. Truck Age Distribution and Mileage Accumulation. The default truck age distribution and mileage accumulation in MOBILE6 were developed based on a report that estimated truck age distribution in 1996 from vehicle registration data. (59) Mileage accumulation was estimated from the 1992 TIUS. EPA developed exponential fit curves to convert the 1996 truck age distribution to other years, but the mileage accumulation rates in 1996 were used for the remaining years. This adds a degree of uncertainty in the analysis of emissions, because mileage accumulation rates are likely to evolve over time. The default parameters for truck age distribution and mileage accumulation are important to the extent that emission factors vary by truck model year. Exhibit 3-16 illustrates the relative difference in emission factors of CO2, NOx, CO, HC, and PM10 relative to a 1981 HDV8b truck. In comparison to other pollutants, CO2 emission factors are not very sensitive to truck model year and remain constant after 1996. Other pollutants’ emission factors are very sensitive to truck model year. As a result, assumptions regarding truck age distribu- tion and mileage age distribution have a large impact on fleet- average emission factors and on total emissions. Average Speed. Although MOBILE6 does not enable user- customized driving cycles, speed correction factors are used to differentiate emissions of HC, CO, and NOx by vehicle av- erage speed. For heavy-duty trucks, MOBILE6 inherited the same speed correction factors from MOBILE5, as opposed to light-duty vehicles, for which adjusted speed correction fac- tors were developed. The uncertainties associated with the use of speed correction factors to adjust emission factors by vehicle average speed are discussed in Section 3.3.4. High Emitters. Having been identified as one of the main issues in MOBILE6, correctly representing the share of high emitters is challenging for many reasons, including (1) the number of high emitters is relatively small, (2) the range in emissions is quite large, and (3) owners of high emitters are typically reluctant to submit their vehicles to testing. (55) 47

MOBILE6 incorporates correction factors to account for high emitters, and despite many criticisms about the underlying methodology, it is a step in the right direction. However, such correction factors are applied to light-duty vehicles only, so high emitter heavy-duty trucks are not considered. The effects of tampering and mal maintenance in heavy-duty vehicles also are disregarded. Diesel Fraction. Diesel fraction, defined as the share of diesel vehicles in a particular vehicle category, is important since emission rates are different for diesel and gasoline- powered vehicles. Although users can input specific diesel fractions for each model year within each vehicle category, this is rarely done due to a lack of project-specific informa- tion. As a result, most analyses rely on default values provided in MOBILE6. The main source of uncertainty relates to the fact that MOBILE6 assumes that diesel fractions for vehicles of model years later than 1996 have the same diesel fraction as a 1996 model year. (60) Although this is not an issue for Class 8 trucks, which are virtually all diesel powered, diesel fraction for Classes 2b through 7 have varied quite substan- tially from 1972 to 1996. Start Emissions and Soak Time. Start emissions are those that occur immediately after a cold engine start; soak time rep- resents the time between when the engine is turned off and the next time it is restarted. Emission rates for heavy-duty gaso- line vehicles include engine starts, and the number of engine starts and the soak time distribution cannot be adjusted by the user (such adjustment is possible for light-duty vehicles). (60) Because the trip length can be quite different for different truck trips, the inability to customize start emissions can add uncertainty to emission rates. MOBILE6 does not consider start emissions for any diesel vehicles, thus adding another set of uncertainties to the emission calculations. Classification of Trucks. Although MOBILE6 includes 16 categories of heavy-duty trucks, data on age distribution, mileage accumulation rates, and fuel ratios are not available for all truck categories. (59) There are only 7 categories for registration distributions by age and only 18 categories for av- erage annual mileage accumulation rates by age. As a result, some weight classes were combined (Classes 4 and 5 as well as Classes 6 and 7), and it was assumed that such classes had the same age distribution. EMFAC2007 Developed by the California Air Resources Board (ARB), EMFAC is the approved emissions model in California, and it is used for SIP development, conformity analysis, and other analyses that are typically conducted using MOBILE6 in other states. The model produces emission rates and inventories for criteria air pollutants, CO2, and CH4. Air toxics can be speci- ated using CARB factors. EMFAC2007 produces emission calculations at the county, regional, and state levels, for past, current, and future years. The overall approach for emission calculations in EMFAC and MOBILE6 is very similar, where regression analyses of primary data sets are used to calculate emission rates and cor- rection factors. Because the approaches are very similar, the main potential limitations in accuracy are the same as those described in the discussion of MOBILE6. One main differ- ence is the way that EMFAC2007 handles off-cycle emissions due to the defeat device. MOBILE6 allows the user to specify how fast the device will be removed, while EMFAC2007 makes 48 Exhibit 3-16. MOBILE6’s sensitivity to truck model year (HDV8b). - 0.2 0.4 0.6 0.8 1.0 1891 3891 5891 7891 9891 1991 3991 59 91 7991 9991 1002 3002 Em is si on F ac to rs R el at iv e to 1 98 1 Truck Model Year CO2 NOx CO HC PM

assumptions that the device will not be removed in 1994 to 1998 trucks. At its core, the development of EMFAC2007 was based on the inclusion of area-specific activity data for various regions within California, including vehicle registration, mileage ac- cumulation, vehicle age distributions, and VMT, as well as temperature and humidity profiles. Summary of Strengths and Weaknesses. A summary of EMFAC2007 strengths and weaknesses is provided in Exhibit 3-17. Analysis of Process Uncertainty. Development of Heavy-Duty Truck Emission Factors. EMFAC2007 updated heavy-duty truck emission factors and speed correction factors based on new data obtained through the CRC E55/E59 Project, whose objective was to reduce the uncertainty of heavy-duty truck emission factors by quanti- fying PM emissions in the South Coast Air Basin to support emission inventory development and to quantify the influ- ence of tampering and mal maintenance (T&M) on heavy- duty emissions. (61) The update of heavy-duty emission factors was a signifi- cant improvement. In EMFAC’s previous version, heavy-duty truck emission factors were developed from testing of various engines on an engine dynamometer rather than of the entire vehicle on a chassis dynamometer. (55) As a result, emission factors needed to be converted (to grams/mile) based on fuel density, BSFC, and fuel economy. Because of the wide varia- tion in gross vehicle weight, fuel economy, horsepower rat- ings, and transmission types, the gram/mile emission factors derived from engine test data had much higher uncertainties than those calculated by vehicle dynamometer testing. Although the resulting database is the largest available, the fleet is still too small to accurately characterize changes be- tween model years. Another source of uncertainty in this method is that all emissions were measured on a very limited number of driving cycles. 49 Exhibit 3-17. Analysis of strengths and weaknesses—EMFAC2007. Criteria Strengths Weaknesses Representation of physical processes Updated EFs based on chassis dynamometer testing from CRC E55/E59 project. EFs incorporate effects of average trip speeds. Emission factors are based on a very limited number of driving cycles. Not enough data on EF to accurately differentiate among different truck model years. There are concerns about how speed correction factors capture speed/congestion effects on emissions. Relies on “average trip” drive cycle, rather than facility-specific information. Model sensitivity to input parameters Model does not consider effects of road grade or pavement quality. Model flexibility Few inputs are required. County-based default parameters (VMT mix by vehicle class, vehicle age distribution) are included in the model, but can be overridden by local estimates. Ability to incorporate effects of emission reduction strategies Model is able to capture the effects of strategies that change truck VMT, vehicle average speed (for HC, CO, NOx), fleet average age. Model is not able to capture the effects of strategies that affect pavement quality. Representation of future emissions EFs can be estimated up to 2050. There are concerns as to whether the assumptions used to estimate future EFs are still in line with latest vehicle technology trends. Consideration of alternative vehicle/fuel technologies Only indirectly through input of different EFs. Data quality California-specific default data. Other regions can tailor EMFAC using other values. Spatial variability County-based input parameters differentiate results. Cannot be applied to facility level. Valid only at county and state level. Temporal variability Truck VMT distribution based on outdated data. Review process There have not been many independent analyses and reviews of EMFAC. Endorsements EMFAC is the required model for SIPs and conformity analyses within California.

Characterization of Congestion and Modal Emissions. EMFAC2007 uses trip-based speed correction factors (rather than facility-based correction factors in MOBILE6). Trip- based speed correction factors can be appropriate for the development of regional emission inventories, but they fall short when the objective is to estimate local or project-level emissions. This is the case since the outputs from travel de- mand models include speed at the link level and not at the trip level. Therefore, adjusting emissions at the link level with speed correction factors at the trip level is not consistent, and is an important source of uncertainty. Additionally, EMFAC2007 does not incorporate the effects of road grade, actual equip- ment weight, or equipment aerodynamics, all of which have a strong effect on emission factors. In order to assess the degree of uncertainty associated with the use of speed correction factors in the development of emis- sion factors in EMFAC2007, a comparison was done with emission factors generated by a modal approach (i.e., second- by-second approach) where the driving cycles developed by Sierra Research were adapted for heavy-duty trucks. (62) In Exhibit 3-18, the black line represents EMFAC2007 emission factors, while the other lines represent modal emission factors on freeways and arterials, respectively. Both approaches pro- vide comparable results for uncongested freeways at high speeds, but very different results for congested freeways and arterials. Unlike MOVES, EMFAC2007 differentiates GHG emissions based on average trip speed but does not consider congestion explicitly. Additionally, EMFAC2007 does not differentiate among different roadway types. For instance, a vehicle with an average speed of 30 mph could be traveling along an uncongested arterial, or along a congested freeway. Although the emission factors under these two scenarios are very different, EMFAC2007 cannot differentiate between them. The modal approach however, has the ability to dif- ferentiate these two scenarios, thus creating the differences between the two models. Truck VMT Distribution. Because EMFAC2007 is also intended to estimate emissions inventories at the county level, it relies on a methodology to allocate VMT information (re- ported by Council of Governments [COGs] and MPOs) to specific vehicle categories. VMT estimates are provided by travel demand models and validated by traffic count data. VMT estimates are generated at different levels of resolution, and heavy-duty truck VMT is typically not provided separately. In contrast to its previous version, which allocated VMT to each vehicle category based on registration data, EMFAC2007 allocates VMT based on (estimated) travel data. (63) The pri- mary data source was a 1999 Caltrans heavy-duty truck sur- vey, which was used to estimate the fraction of heavy-duty truck VMT traveled in each county in California, as well as mileage accumulation rates and truck age distribution. The survey included origin and destination information but not route, so the latter had to be estimated based on shortest-path algorithms. To reduce the uncertainty in this method, these routes were validated with actual truck routes collected by GPS data. A second source of validation included other annual publications by Caltrans. A statistical comparison of these two data sources indi- cated that results were consistent for one year. In order to es- 50 Exhibit 3-18. Emission factor comparison for heavy-duty trucks: EMFAC2007 vs. modal approach. - 1,000 2,000 3,000 4,000 5,000 6,000 0 10 20 30 40 50 60 70 80 CO 2 EF (g ra ms /m ile ) Speed (mph) Moves Moves Moves Art C- Moves Art Moves Moves Moves Fwy Moves EMFAC

timate truck VMT distribution for earlier and later years, only one of the data sources was used, which accounted for differ- ent growth rates in different counties. MVSTAFF, which is maintained by Caltrans, predicts statewide VMT based on a variety of model inputs including socioeconomic parameters (e.g., population, income, economic growth rates), as well as the past 25 years of vehicle registration information. Natu- rally, there are uncertainties associated with such estimates, given that vehicle registration and socioeconomic patterns might not be the best indicators of heavy-duty truck patterns in the state. The truck VMT in each area was calculated by taking the product of the registered truck population (by model year), out of state fraction, and accumulation mileage rates. Truck VMT was then redistributed to specific counties based on the methodology previously described. There are a few sources of uncertainty associated with the process of estimating truck VMT, as follow: • Based on a 1998 study (64), it is assumed that 25% of trucks are out-of-state trucks, and a factor of 1.33 is applied to total state VMT. There are two main issues with this method: – It assumes that accumulation mileage rates for in-state trucks are the same for out-of-state trucks. This issue might be resolved with the Interstate Registration Pro- gram, which will reevaluate accumulation mileage rates for in-state and out-of-state trucks. – It assumes a percentage of out-of-state trucks based on a single study at a given point in time. • Accrual rates by truck age are based on the 1992 Truck In- ventory and Use Survey, again another snapshot in time. Other sources of uncertainty relate to the way VMT infor- mation is provided by COGs and MPOs. Although some juris- dictions provide explicit VMT estimates for heavy-duty trucks (e.g., SCAG), the majority of organizations provide total VMT unclassified by vehicle type. So the allocation of VMT to the specific categories of heavy-duty trucks is uncertain. Truck Age Distribution. EMFAC2007 relies on a statewide model year distribution for heavy-duty trucks, which is gen- erated based on registration data that CARB receives annu- ally from the DMV. Although this method is likely to be rea- sonable for statewide analyses, the model year distribution could diverge from the state average in isolated counties. For example, there is a significant amount of drayage traffic, which is typically moved by an older fleet, in proximity to the Ports of Los Angeles and Long Beach. Therefore, the use of a statewide model year distribution would not be representa- tive of the actual fleet. Another source of uncertainty is that the gross vehicle weight (GVW) assignment to each truck is not based on DMV registration information (because such information is often not available), but through cross-checking vehicle identifica- tion number with and vehicle reference books, which only in- dicates the manufacturer-specified GVW, and not the actual average GVW. MOVES2009 MOVES is EPA’s most recent emission model, which will eventually replace MOBILE6 and NONROAD when fully im- plemented. Its most current version—MOVES2009—has recently been released. It calculates emissions of GHGs, crite- ria air pollutants, and some air toxics from highway vehicles, and it allows multiple scale analysis—from modal emission analyses to NEI estimation. An uncertainty analysis of MOVES is challenging since it is still in draft version, and it is not yet approved for official use. As a result, there have not been many studies or analyses re- lated to MOVES. Because MOVES is still under development, it is important to define whether one should analyze MOVES in its current form or the version of MOVES when fully im- plemented. Such distinction will be present throughout the analysis. The main improvements MOVES offers in comparison to MOBILE6 can be summarized as follows: • Employs a “modal” emission rate approach “as a prelude to finer-scale modeling”; (65) • Relies primarily on second-by-second data to develop emis- sions rates, which better represents the physical processes from heavy-duty vehicles, including the ability to model cold starts and extended idling; • Is designed to work with transparent databases, which can be modified and updated depending on the user’s needs; • Includes energy consumption, N2O, and CH4 explicitly; • Uses a graphical user interface. Summary of Strengths and Weaknesses. MOBILE6 has been highly scrutinized, and many of its pitfalls are tentatively addressed in the development of MOVES. At the same time, MOVES is still under development, and the lack of available studies prevents a more comprehensive uncertainty analysis. As a result, the analysis of strengths and weaknesses of MOVES will be based on the main differences over MOBILE6 that relate to the representation of heavy-duty truck emissions (see Exhibit 3-19). Analysis of Process Uncertainty. Binning Approach. MOVES uses a binning approach to calculate modal emissions, and unique source bins are 51

52 Exhibit 3-19. Comparison of MOVES and MOBILE6. Criteria MOBILE6 MOVES Comment Micro-scale analysis When fully implemented Geographic Scale Macro-scale analysis Both models enable the estimation of regional and national emission inventories. Criteria air pollutants Both models include all criteria air pollutants. Greenhouse gases Incomplete MOVES adds energy consumption, N2O, and CH4 explicitly. Air toxics When fully implemented. Air Pollutants Life-cycle emissions When fully implemented, MOVES will integrate with GREET to provide well-to-wheels emissions. Ability to consider user- specified driving cycles MOVES employs a “modal” emission rate approach that will allow users to model emissions on a second-by- second basis based on user-specified driving cycles. Emission factors based on actual in-use emissions MOBILE6 uses engine certification data while MOVES uses second-by second vehicle emission rates. Extended idling Cold starts Vehicle Weight Ability to consider different HDT categories Because MOVES classifies heavy-duty vehicles based on how VMT/fuel data are reported, it provides fewer HDT categories than MOBILE does. Representation of physical processes Ability to consider different facility types MOVES expands the number of available facility types. Estimation of Future Calendar Years Representation of future emissions Consideration of alternative vehicle/fuel technologies Incomplete Relationship Database MOVES is designed to work with transparent databases, which can be modified and updated depending on the user’s needs. Ability to incorporate effects of emission reduction strategies Because MOVES is based on a modal approach, it is more capable of capturing the effects of many emission reduction strategies, such as improvements in pavement quality, reduction in congestion, etc. Graphical user interface Model flexibility Uncertainty assessment When fully implemented, MOVES will enable the assessment of uncertainty based on the uncertainty of some inputs. Review process Although MOBILE6 has been highly scrutinized, the final version of MOVES has not been released yet. Key: indicates that a parameter is analyzed in the way denoted by the column: indicates that the parameter is not discussed in the way denoted by the column.

differentiated by characteristics that significantly influence fuel/energy consumption and emissions. (66) At its most dis- aggregated level, emissions can be calculated by • Geography: the entire United States, at the county level; • Facility Types: including off-network roads, rural and urban restricted access roadways (i.e., freeways and inter- states), and rural and urban roads with unrestricted access; • Time Spans: energy/emission output by hour of the day for calendar years 1990 and 1999 through 2050, with options to run at more aggregate month or year levels; • Vehicle Types: all highway vehicle sources, including six heavy-duty truck categories (i.e., refuse, single-unit short- haul, single-unit long-haul, combination short-haul, com- bination long-haul, mobile home). All vehicle types are further subdivided according to fuel type, engine technol- ogy, loaded weight, and engine size. • Energy/Emission Outputs: energy consumption (e.g., total energy, petroleum-based energy, and fossil fuel-based en- ergy), N2O, CH4, atmospheric CO2, CO2 equivalent, total gaseous hydrocarbons, CO, NOx, and PM; • Emissions Processes: running, start, extended idle (e.g., heavy-duty truck “hotelling”), well-to-pump, brake wear, tire wear, evaporative permeation, evaporative fuel vapor venting, and evaporative fuel leaks. Running emissions are further subdivided in vehicle-specific power and instanta- neous speed bins. This method produces 15 bins defined by combinations of speed and vehicle-specific power. Idle and decelerations are considered separately, resulting in 17 total bins. Development of Emission Factors. MOVES provides methodological improvements over MOBILE6 as it relates to the development of emission factors for heavy-duty trucks. The emission factors in MOVES rely upon second-by-second emission data, which allows a much broader range of data to be used in the development of emission rates. Emissions data were compiled from previous EPA test programs and from several external sources, including the Coordinating Research Council (CRC), UC Riverside, Texas Department of Trans- portation, University of Texas, and West Virginia University. EPA contracted with Eastern Research Group (ERG) to assist in the acquisition, quality checks, and compilation of data collected by outside parties. The information included in the second-by-second emis- sion data was used to develop energy rates for each vehicle type. Each data point was allocated to a bin, which was char- acterized by vehicle type, instantaneous speed, and vehicle- specific power. All measurements falling into each bin were then averaged. The end result of this process was a table con- taining energy rates (in kJ per hour) and coefficients of vari- ation by bin. The strengths of this approach included the use on real trucks (as opposed to engine testing), driving cycles based on real-world conditions over a wide range of operating condi- tions, and the inclusion of actual deterioration and mainte- nance. The sample, which was of relatively small size (100 trucks of 30 model years), was biased to older (and potentially dirt- ier) trucks with unknown maintenance history (or degree of tampering). Additionally, driving cycles were not randomly sampled. Because there were some bins without data, supplemental methods were used to “fill the holes.” After an evaluation of different methods for hole filling, two methods were selected: (1) the use of PERE (Physical Emission Rate Estimator), which models fuel consumption on a second-by-second basis ac- cording to a power demand equation, and (2) interpolation of neighboring cells populated with data. Although light-duty vehicles were well covered (i.e., over 90% of bins were filled with primary data), there were rela- tively more holes in bins associated with heavy-duty trucks. Single-unit trucks had 65% of bins filled, and combination trucks had less than 36% of bins filled. In particular, the heavi- est truck classes (over 60,000 lbs) were very poorly repre- sented, so there are concerns related to the validity of such factors. Relational Database. MOVES relies on a relational data- base that contains default information for the entire United States. The data for this database come from many sources in- cluding EPA, Census Bureau vehicle surveys, FHWA travel data, as well as other federal, state, local, industry, and aca- demic sources. The database is transparent, so users can mod- ify the data with updated local inputs, which might be more appropriate for analyses at the project or regional level. CMEM UC Riverside’s Comprehensive Modal Emissions Model (CMEM) estimates vehicle emissions at the micro-scale level. It uses a parameterized physical approach that breaks down the entire combustion process into different components that correspond to physical phenomena associated with vehicle operation. Particular emphasis was taken to model the effects of road grade, variable ignition timing, and truck platoon sce- narios, where aerodynamic effects can provide a significant benefit in terms of fuel savings. The UC Riverside team also simulated instantaneous fuel consumption in a number of actual heavy-duty trucks to calibrate their model. CMEM relies on second-by-second input data including instantaneous speed and road grade, as well as on detailed ve- hicle configuration (e.g., engine power rating, aerodynamic coefficient, rolling resistance coefficient, transmission, weight). As a result, CMEM is generally used for project-level analyses 53

where a high degree of confidence is needed for a particular scenario. Summary of Strengths and Weaknesses. A summary of CMEM strengths and weaknesses is provided in Exhibit 3-20. Analysis of Process Uncertainty. CMEM’s analysis of model uncertainty developed by UC Riverside was divided into the following three areas: • Emissions Measurement Variability: although measure- ment instruments were calibrated prior to each HDDV test, a certain degree of inherent emission measurement vari- ability always exists. The instrument precision varied from less than 0.5% for CO2 to just under 11% for NOx; • Vehicle Operation Variability: it was found that small dif- ferences in driving the specified driving cycles accounted for 5% to 10% variability in emissions. Following specified driving cycles is typically impacted by other vehicles on the road, road grade, wind conditions, and safety concerns; • Vehicle Sampling Variability: although there is little data to estimate vehicle-to-vehicle variability, data from CARB indicate that there is considerable variability in emissions across different model years and equipment manufacturers. 3.3.2 Evaluation of Regional Methods On-road vehicle emission inventories developed by MPOs and state air quality agencies are the most detailed when com- pared to other transportation modes. Trucking emissions are typically calculated as part of the total on-road vehicle emis- sions estimation process. Because on-road vehicles are one of the largest sources of pollutant emissions, and because of Transportation Conformity determination requirements, the process used for estimating on-road vehicle activity and emis- sions is often more complex than for other transportation 54 Exhibit 3-20. Analysis of strengths and weaknesses—CMEM. Criteria Strengths Weaknesses Representation of physical processes CMEM measures fuel and emissions rates on a second-by- second basis according to a set of input parameters that describe the vehicle, driving cycle, and road facility. CMEM’s main advantage over MOVES/PERE is that it considers vehicle operational history effects (i.e., how the last seconds of operations affect fuel consumption/emissions). Model development was not dependent on pre-specified driving cycles. Model sensitivity to input parameters The model outputs are sensitive to all parameters that have a strong effect on fuel consumption and emissions (e.g., vehicle characteristics, fuel characteristics, engine specifications, road grade, second-by-second driving cycle). Model flexibility A driving cycle might not be representative of average traffic mix. If the goal of the analysis is to represent a mix of vehicles and traffic conditions, computation requirements can be heavy. Ability to incorporate effects of emission reduction strategies Model can measure individual project-level impacts, such as changes in congestion levels, use of HOV lanes, incident management programs, traffic signal coordination. Representation of future emissions Modeler needs to know exactly the effects of future scenarios on all input parameters to the model. Consideration of alternative vehicle/fuel technologies Spatial variability Because CMEM is a micro-scale emissions model, it is well set up to capture the variability of emissions based on local conditions (e.g., road grade, pavement quality, ambient temperature). Review process Uncertainty analysis is performed specifically for heavy-duty truck module. Endorsements “Research grade” model—not established for industry use.

sources. All large metropolitan areas develop detailed esti- mates of VMT and on-road emissions by vehicle class and roadway functional class. For emission inventory purposes, some regions rely on the MPO travel demand forecasting model to determine VMT and vehicle speeds, calibrating the model to observed traffic counts. Other regions estimate VMT directly from traffic counts. Emission factors are developed using EPA’s MOBILE6 model or, in California, CARB’s EMFAC model. Development of emission factors requires regionally specific information on inspection and maintenance (I/M) programs, fuel charac- teristics, temperature information, vehicle age distribution, and vehicle mileage accumulation by model year. A previous report has summarized the methods to estimate freight emissions at six metropolitan areas, namely Baltimore, Chicago, Dallas-Fort Worth, Detroit, Houston, and Los Angeles. (67) All six study regions use a similar methodology to estimate on-road vehicle emissions, which can be summa- rized in the following steps: 1. The region’s MPO uses a four-step travel demand model to estimate base year and future year traffic volumes by link. In some cases, the model estimates truck trips inde- pendent of passenger vehicle trips (i.e., independent truck trip generation and trip distribution modules). In other cases, the models estimate only passenger vehicle trips, and truck volumes are calculated as a percentage of pas- senger vehicle volumes. 2. As required by EPA, the MPO adjusts the travel model traffic volumes based on observed traffic counts. In this way, the model is calibrated to reflect base year condi- tions as accurately as possible. 3. The MPO estimates traffic volumes on local roads that are not represented in a travel model. Some MPOs do this estimation themselves (e.g., the Baltimore MPO); others rely on local roadway VMT provided by the state DOT (e.g., the Detroit MPO). 4. Daily traffic volumes by link are disaggregated to hourly volumes, using observed traffic counts. 5. Model traffic volumes at the link level are allocated to major vehicle types, based on traffic count information. 6. VMT is summed by vehicle type and facility type. 7. The MOBILE6 model requires VMT by 16 different ve- hicle types. Most regions do not have VMT or traffic count information at this level of detail, so they rely on the MOBILE6 defaults to apportion VMT into these 16 vehicle types. 8. Hourly speeds are estimated for each link. Because emis- sion factors vary with vehicle speed, the distribution of VMT by speed can have an important effect on emissions. MPOs use equations that compare link-level volume and capacity to estimate speed. 9. MOBILE6 input scripts are developed for information such as fuel Reid vapor pressure (RVP), engine tamper- ing levels, inspection and maintenance programs, and ve- hicle emission standards. If emissions are being calculated for a specific day or month, MOBILE also requires input information for factors such as maximum and minimum temperature and sunrise and sunset times. 10. MOBILE6 produces emission factors and VMT weight- ing factors, typically for each county, urban/rural area, and roadway functional type. VMT is multiplied by the appropriate emission factors to determine emissions. In California, emissions are estimated using the EMFAC model developed by CARB. It is typically assumed that any heavy-duty truck (i.e., any truck over 8,500 lbs GVW) is a “freight truck.” In reality, there are heavy-duty trucks that do not move freight. Some exam- ples of non-freight heavy-duty trucks are utility trucks used for service and repair of utility infrastructure, construction trucks (e.g., winches, concrete mixers and equipment transport vehi- cles), urban garbage haulers, tow trucks, service industry trucks used primarily to transport equipment, and daily rental trucks. Because it is virtually impossible to separate the activity and emissions of non-freight heavy-duty trucks from freight trucks, and because non-freight heavy-duty trucks are relatively in- significant compared to freight trucks, generally no attempt is made to distinguish between the two. Summary of Strengths and Weaknesses. A summary of strengths and weaknesses of regional MPO methods is pro- vided in Exhibit 3-21. Analysis of Process Uncertainty. The analysis of process uncertainty of this regional method is captured within the discussion of parameter uncertainty, including the following: • Estimation of truck VMT by travel demand models; • Use of average speed information; • Use of emission factors; because the emission factors are estimated with either EMFAC (for California) or MOBILE (for the remaining states), the analysis of uncertainty asso- ciated with the estimation of emission factors is included in the discussion of these two models. 3.3.3 Evaluation of Local/Project Methods Typically, the calculation of freight emissions at the local or project level can rely on more accurate estimates of freight activity, which is generally estimated in VMT. The emission factors are generally extracted from the same models used in national or regional approaches, but they are commensurate with the level of detail included in activity data. For example, 55

if activity data includes traffic volumes at different speed bins, then emission factors can be estimated based on these same speed bins. The approach used to calculate freight emissions at the local level can be summarized in five steps: 1. Configuration of vehicle types Because emission models have their own vehicle classifi- cation system, agencies need to understand which specific vehicle types should be considered in an estimation of emissions from heavy-duty trucks. Exhibit 3-22 includes the vehicle types that are considered in the three main emission models. 2. Determination of vehicle activity Truck activity is characterized in terms of VMT and idling hours. For analyses that do not include the effects of speed and congestion on emissions, aggregate measures of VMT by vehicle type in the study area are sufficient for the cal- culation of emissions. If VMT for each vehicle type is not available, the state or county average VMT distribution (i.e., travel fractions) can be used as a surrogate method. More sophisticated analyses include speed and conges- tion effects on emissions. In those cases, VMT by vehicle type and average speed are determined for each roadway link in the study area. Because congestion levels can vary quite rapidly, the definition of time periods is important. To properly evaluate the effects of congestion on GHG emissions, VMT should be determined at different time periods during the day. There is no standard method to determine truck idling hours and, ideally, project-level data are collected. 3. Determination of road level of service and driving cycles For those analyses that include the effects of congestion on emissions, congestion levels are characterized for each roadway segment in all project scenarios. Level of service (LOS) characterizes congestion levels and is the primary 56 Exhibit 3-21. Analysis of strengths and weaknesses—regional MPO method. Criteria Strengths Weaknesses Representation of physical processes Travel demand models are calibrated by current traffic counts. Overall, there are many concerns related to the accuracy of travel demand models in estimating truck VMT. Truck VMT data are not disaggregated into all truck categories in MOBILE6. There are concerns about whether MOBILE6 and EMFAC can accurately capture congestion effects through average speed. Travel demand models do not calculate average speed directly, but rather estimate it through traffic volume and road capacity. Model sensitivity to input parameters The level of detail associated with truck travel activity from travel demand models is not commensurate with the level of detail required by emissions models. Ability to incorporate effects of emission reduction strategies Model is able to capture the effects of strategies that change truck VMT, vehicle average speed (for HC, CO, NOx), fleet average age. Representation of future emissions Future emissions can be represented to the extent that travel demand models can forecast truck VMT. Consideration of alternative vehicle/fuel technologies Typically, there are none. Data quality Depending on the region, truck VMT are estimated as a share of passenger vehicle VMT, otherwise they are estimated through land-use categories as a function of employment. Spatial variability Travel demand models are specific to a given region of interest. Temporal variability Some regions have travel demand models that have the ability to model traffic in different periods throughout the day. Most regions have travel demand models that are based on a 24-h period. This method does not typically capture speed variations within the hour. Endorsements This is the method that most MPOs rely on to calculate regional emissions.

measurement used to determine the operating quality of a roadway segment or intersection. Methods applied to cal- culate LOS are provided in the Highway Capacity Manual (68), which is the industry standard that guides roadway operational analyses. The derivation of emission factors that take road LOS into account depends on the development of customized driving cycles, which consist of a series of data points representing the speed of a vehicle versus time, usually on a second-by- second basis. Because the development of project-specific driving cycles is time and resource intensive, standard driving cycles can be used as a surrogate method. An EPA research project developed a set of driving cycles under a variety of congestion levels for different road types. (62) The representation of congestion and LOS by a driving cycle is often criticized, since traffic patterns and delay can vary substantially within the same LOS. Additional research could indicate alternative methods to consider congestion in the evaluation of emissions from on-road sources. There also has been criticism against the methodology used by Sierra Research in the development of their driving cycles. 4. Calculation of emission factors One of two models is generally used to calculate emission factors, namely EMFAC in California, and MOBILE6 else- where. Depending on the analysis, the emission factors ex- tracted by these models can represent specific truck types, model years, fuel types, and engine technologies. For more sophisticated analyses that include the effects of conges- tion, emission factors also can depend on average speed. In order to consider customized driving cycles in the esti- mation of emission factors, a modal emission model needs to be used. CMEM, which was developed by UC Riverside under an EPA contract, is arguably the most established modal emission model. However, MOVES was designed to enable micro-scale emissions analyses, and also can be used in analyses that consider customized driving cycles. MOBILE6 and EMFAC do not consider different driving cycles explicitly (GHG emissions in MOBILE6 are insen- sitive to speed). Idling emission factors are generally calculated for the lowest possible speed in grams of pollutant per mile, and multiplied by that speed to estimate an emission factor in grams per hour. 5. Calculation of emissions Emissions are calculated by multiplying freight activity by the appropriate emission factors. Summary of Strengths and Weaknesses. The analysis of strengths and weaknesses for local methods will vary signifi- cantly depending on the method utilized for truck activity es- timation. If a travel demand model is used, then the strengths and weaknesses will be similar to those described in the re- gional model. The following section describes the strengths and weaknesses when other methods are applied. Analysis of Process Uncertainty. Estimation of Truck VMT. Some project-level analyses estimate truck VMT based on regional travel demand mod- els, whose uncertainties are discussed in the regional method. However, many local/project level analyses rely on project- specific data, which is more accurate than data estimated by models. Even though there will be variation between esti- mated and actual truck traffic, this is a source of uncertainty inherit to project-level analysis, and it is beyond the scope of this analysis to provide methods to more accurately estimate truck VMT at the project level. However, other sources of uncertainty that could be improved are • Estimation of truck weight: this is key since emissions are highly dependent on truck weight. • Determination of truck specifications: if project-level analy- ses rely on more specific truck configurations, the emission factors need to be consistent with the modeled truck. The most important elements to characterize trucks involve truck class, engine power, gross weight, and fuel type. Truck Age Distribution. Many project-level analyses still rely on the national average vehicle age distribution, which 57 Exhibit 3-22. Heavy-duty truck types. MOBILE6 EMFAC2007 MOVES2009 Class 2b HDV (8,501-10,000 lbs GVWR) Class 3 HDV (10,001-14,000 lbs GVWR) LHDT Light heavy-duty trucks (8,501-14,000 lbs GVWR) Class 4 HDV (14,001-16,000 lbs GVWR) Class 5 HDV (16,001-19,500 lbs GVWR) Class 6 HDV (19,501-26,000 lbs GVWR) Class 7 HDV (26,001-33,000 lbs GVWR) MHDT Medium heavy-duty trucks (14,001-33,000 lbs GVWR) Class 8a HDV (33,001-60,000 lbs GVWR) Class 8b HDV (>60,000 lbs GVWR) HHDT Heavy heavy-duty trucks ( > 33,000 lbs GVWR) Single-Unit Short-Haul Trucks Single-Unit Long-Haul Trucks Combination Short-Haul Trucks Combination Long-Haul Trucks

sometimes is not a good proxy for local vehicle age distribu- tions. For example, if a project is associated with a specific type of traffic, it will more likely focus on a group of carriers that will tend to use a fleet of trucks whose age range is nar- rower. For example, long-distance trucks that transport time- sensitive cargo tend to be newer, while drayage fleets that transport international cargo between port terminals and local facilities tend to be older than the national average fleet. Determination of Driving Patterns. Project-level analy- ses that still rely on average emission factors are implicitly as- suming average driving patterns that might be representative of national patterns, but not necessarily of local driving con- ditions. For example, if project-related traffic occurs solely at night, when traffic flows are generally smooth, or solely dur- ing peak times, when traffic flows are usually interrupted, average traffic patterns will probably not provide a good rep- resentation of actual driving patterns. For those projects that do estimate project-specific driving patterns, the following issues might arise: • There is usually a high degree of variation in traffic pat- terns, so considerable resources need to be spent in order to develop a mix of driving patterns that provide a good representation of actual driving conditions. • Modal emission models rely on entire driving cycles to es- timate emission factors, and it is very rare that project-level analyses have the resources to develop a number of driving cycles that will provide a good representation of project- related driving patterns. More typically, project-level analy- ses rely on traffic volumes and road capacity information to determine average speed and road LOS. • The representation of road LOS and average speeds by spe- cific driving cycles is often criticized because there can be a high degree of variation in driving patterns even within the same LOS, especially for the more congested levels of service; • The emission analyses that rely on the standard definitions of road LOS require the use of driving cycles that represent such levels of service. To date, the driving cycles from Sierra Research (62) are the only ones that were developed with the aim of representing the standard levels of service defined by the Highway Capacity Manual. There are criti- cisms of the validity of the statistical methods used by Sierra Research in the development of those cycles. Further, these cycles were developed for light-duty vehicles, not heavy- duty trucks. To date, there are no driving cycles developed for heavy-duty trucks that aim to characterize different road levels of service. • Another issue is the time resolution of the analysis. For congestion patterns to be properly characterized, time res- olution needs to be evaluated in shorter time periods, usu- ally less than one hour, and ideally less than 30 minutes. Determination of Emission Factors. Emission models such as MOBILE6 and EMFAC are not able to generate emis- sion factors that rely on customized driving cycles. For project- level analyses that characterize congestion by developing a customized driving cycle, a modal emission model is necessary. Two examples are CMEM and MOVES, whose uncertainties are discussed in Section 3.3.2. The development of composite emission factors, which de- pend on the distributions of truck model year, engine tech- nology, fuel type, and vehicle weight to characterize the proj- ect truck fleet average, is generally impacted by the fact that the development of such distributions often relies on very lim- ited information. The use of default distributions included in emissions models brings the issue of whether such distribu- tions are representative of local scenarios. 3.3.4 Evaluation of Parameters Exhibit 3-23 includes a list of parameters used in the meth- ods and models previously described. Pedigree Matrix A pedigree matrix (see Exhibit 3-24) for data quality assess- ment assigns quantitative scores to most of the parameters in- cluded in Exhibit 3-23. The criteria to assign scores in the pedigree matrix are included in Appendix A. Truck VMT In addition to emission factors, a total measure of truck VMT is the parameter with the biggest impact on emissions. It is recognized that current methodologies do not provide estimates of truck VMT with a reasonable degree of accuracy for emission calculation purposes. The main issues relate to (1) how truck movements are represented in travel demand models, (2) how truck trip generation data are developed, and (3) the level of detail included with truck VMT. Estimation of Truck VMT in Travel Demand Models. All large metropolitan areas develop detailed estimates of VMT by vehicle class and roadway functional class. For emission inventory purposes, some regions rely on the MPO travel demand forecasting model to determine VMT and vehicle speeds, calibrating the model to observed traffic counts. Other regions estimate VMT based directly on traffic counts. Many MPOs use a four-step travel demand model to esti- mate base year and future year traffic volumes by link. In some cases, the model estimates truck trips independent of passenger vehicle trips (i.e., independent truck trip genera- tion and trip distribution modules). In other cases, the mod- els estimate only passenger vehicle trips, and truck volumes 58

59 Exhibit 3-23. Parameters. Parameter Methods/Models GeographicScale Pedigree Matrix Qualitative Assessment Quantitative Assessment Truck VMT All All VMT Share by Truck Type All All VMT Share by Time of Day All Regional/Local Truck Age Distribution All All Mileage Accumulation All All Distribution of Emission Control Technology All All Truck Fuel Type Distribution All All Average Speed MOBILE6, EMFAC2007 Regional/Local Driving Cycles CMEM Local Emission Factors All All Classification of Truck Types All All Road Grade CMEM Local Empty Miles All All Key: indicates that a parameter is analyzed in the way denoted by the column: indicates that the parameter is not discussed in the way denoted by the column. Exhibit 3-24. Pedigree matrix—truck parameters. Parameter Im pa ct on R es ult Ac qu isi tio n M eth od In de pe nd en ce Re pr es en tat ive ne ss Te m po ra l C or re lat ion Ge og ra ph ic Co rre lat ion Te ch no lo gic al Co rre lat ion Ra ng e o f V ar iat ion Truck VMT 5 2 1 Varies 1 1 1 4 VMT Share by Truck Type 4 3 3 5 3 2 1 4 VMT Share by Time of Day 2 3 1 5 Varies 2 1 4 Truck Age Distribution 4 2 3 Varies 3 1-2 1 3 Mileage Accumulation 3 2 3 5 3 2 1 3 Truck Fuel Type Distribution 3 3 3 N/A N/A N/A 1 2 Average Speed 3 3 1 1 1 1 1 3 Driving Cycles 3 2 1 4 3 2 1 5 Emission Factors 5 2 1 Varies 3 2 1 5 Classification of Truck Types 3 1 1 N/A N/A N/A 1 3 Empty Miles 3 3 3 5 3 2 1 5

are calculated as a percentage of passenger vehicle volumes. In both cases, many MPOs recognize that the methods to es- timate truck VMT are less sophisticated than those used for passenger VMT. Thus, the uncertainties associated with truck VMT are higher when compared to passenger VMT. Travel demand models use a computerized representation of the regional roadway system that includes all freeways and arterials but typically few or no local streets. This is probably not a big concern for heavy-duty trucks, since just a small share of truck miles are traveled on local roads. Truck Trip Generation Data. Truck trip generation data are used to estimate truck traffic patterns and, consequently, truck VMT. A previous NCHRP report summarized the cur- rent state of practice on the development of truck trip genera- tion data. (69) Conclusions point out that the state of the prac- tice in truck trip generation data are primitive when compared to passenger trip generation data. Therefore, new truck trip data collection methods, capable of better characterizing truck flows at the metropolitan level, need to be developed. (70) Most states and MPOs have not developed truck travel de- mand models, and most often truck traffic is estimated as a fixed percentage of total vehicle flows. There currently are no well-accepted methods of estimating truck trip generation rates, and those models that do utilize some type of method- ology typically estimate truck trip generation rates through land-use categories as a function of employment. Land uses are generally collected by surveys. Sources of errors include the following: • Land-use categories are very broad, and there is a high degree of variability of trip rates within these categories, as well as from region to region; • Land-use categories were originally developed to correlate with the movement of people not freight; • Inaccuracy of self-administered travel diary surveys (re- spondents can be concerned about revealing confidential in- formation and distrusting of government) and small travel survey samples due to low response rates; • Inappropriateness of employment as an explanatory variable—many experts indicate that industrial output is a better indicator of truck trip generation rates than employ- ment, since labor productivity varies widely within indus- try category (within the same land-use category), from firm to firm, and over time; • Lack of a consistent truck classification system—typical approaches include GVW, configuration (e.g., single-unit, combination), and number of axles, but there is no ac- cepted methodology, which makes it difficult to compare trip generation rates; and • A high degree of variability in the underlying economic ac- tivities that generate truck activity, which makes it chal- lenging to apply truck trip generation rates outside of the local area where the data collection took place. When truck trip generation data are obtained from traffic counts, the accuracy of the equipment and the selection of count locations are the most important parameters to deter- mine data uncertainty. Most studies in the literature estimate rates based on small samples (fewer than 10 observations), with high variability from site to site. For projects where truck VMT is estimated from commodity- based models, the number of truck trips is usually calculated by converting total tonnage transported by truck into truck trips by a payload conversion factor. These methods tend to underestimate urban trips, since they do not account for trip chaining nor local and delivery activity. They also exclude construction, service, and utility-related truck trips, which are not captured in commodity flows. Using commodity-based models in regional applications generates challenges because flows are generally allocated to traffic analysis zones (TAZ) using employment shares by industry, and employment data by industry at the TAZ level is difficult to obtain. Level of Activity Detail. The level of detail associated with truck travel activity from current travel demand models is not commensurate with the level of detail required by emis- sions models, which ideally require detailed activity informa- tion disaggregated by truck type, truck weight, model year, fuel type, engine technology, ambient temperature, road type, average speed, and fuel type, among others. If only aggregate estimates of truck VMT are available, average distributions, which might not be representative of regional or local condi- tions, must be used to estimate an average emission factor. Truck VMT data generally used in emission analyses at the national level rely on information from FHWA’s Highway Statistics, which in turn is based on data obtained by the High- way Performance Monitoring System (HPMS). The HPMS provides data that characterize the extent, condition, perfor- mance, use, and operating characteristics of the nation’s high- ways. States are required to report annually to FHWA aggregate estimates of VMT in collector and local roads, which account for over 15% of total highway VMT in the United States. Cur- rent practices used by the states to report local VMT estimates vary significantly and are not typically documented properly. However, because the vast majority of heavy-duty truck traf- fic occurs along arterials and larger facilities, the uncertainty associated with freight VMT should be smaller than for pas- senger VMT. Truck VMT Share by Truck Type Another source of uncertainty is that truck VMT needs to be disaggregated into the different truck categories in emis- 60

sion models. For example, with MOBILE6, truck VMT data need to be disaggregated into eight classes. If only total VMT is estimated, then the data need to be disaggregated into 16 vehicle classes. Because most regions do not have VMT or traffic count information at this level of detail, they rely on the MOBILE6 defaults to apportion VMT into these vehicle classes. Because there is a wide variation in VMT distribution across vehicle categories, the use of national average travel fractions to apportion VMT to specific vehicle categories is certainly a weak method that could add significant uncer- tainty to emissions estimates. Truck VMT Share by Time of Day The estimation of truck VMT by time of day is important for emission analysis because average speed and congestion levels, which can be important inputs, can be very different in peak versus off-peak periods. Additionally, ambient temper- ature is also an important input for some pollutants, espe- cially for NOx, which has a strong impact on ground ozone levels. In most current truck travel demand models, 24-h trip generation rates are disaggregated into time periods based on traffic counts from different time periods. Because of the lim- ited number of traffic counts, there is uncertainty in the algo- rithms used to apply the share determined by each count location to those links where traffic counts are not available. Truck Age Distribution Because emission factors vary by model year, a composite emission factor needs to be developed based on truck age dis- tribution. As previously mentioned, project-level and regional analyses typically rely on national age distributions, which bring uncertainty into emissions analyses since the accuracy of these analyses depends on how well national age distribu- tions reflect local and regional fleets. The variability in truck age distribution nationwide is important to the extent that emission factors vary by truck model year. The discussion of MOBILE6 includes how emission factors vary by model year. Although emissions of CO2 are not very sensitive to truck model year, the emissions of criteria air pollutants are gener- ally very sensitive to truck model year. Distribution of Emission Control Technology Diesel emission control technology is broken down into the following four categories: • Uncontrolled: generally trucks built prior to 1990 would be considered uncontrolled because no federal heavy-duty emission standard existed before 1990. Emission standards started in California for heavy-duty engines in 1987. • Moderate: those trucks with engines meeting standards starting in 1990 and continuing on through 2003. This is because emission control in these engines was mostly due to engine modifications such as better fuel injection, tur- bocharger improvements, combustion cylinder geometry improvements, and use of after coolers. • Advanced: in 2004 most engines required exhaust gas re- circulation to control NOx emissions. • After treatment: for trucks with engines built in 2007 and later, these will require catalyzed diesel particulate filters and other catalytic devices to reduce NOx emissions. The distribution of emission control technology is usually built into the emission factors and takes into account engines in a given model year that meet future or prior emission stan- dards due to averaging, banking, and trading. Areas with sig- nificant amounts of engines that meet future emission stan- dards can provide errors in the emission factors. Average Speed The most common method used to represent congestion or driving patterns in emission models is by assuming an av- erage speed at each roadway link. The implicit assumption is that average speed is a good proxy for congestion. Both MOBILE6 and EMFAC develop base emission rates for vari- ous truck classes using standard driving cycles. These base rates are then adjusted to a particular average speed and road type by using speed correction factors. There are four impor- tant sources of uncertainty with this method: • The use of average speed is not the best method to repre- sent driving patterns. The development of MOVES, which relies on a modal approach, is an indication of the short- comings associated with the characterization of driving patterns by a single estimate of average speed. • Travel demand models do not calculate average speed di- rectly, but use estimates of traffic volume and road capacity to estimate average speed. Speed/volume relationships are not always very accurate, and are sometimes adjusted so that modeled traffic volumes match observed volumes. (55) • Average speeds are estimated at an hourly basis, so this method does not capture speed variations within the hour, which can be quite significant especially during peak times. • In the case of MOBILE6, emission factors only vary by speed for HC, CO and NOx emissions. Other pollutants are insensitive to speed variations and do not represent real world conditions. More accurate methods of characterizing driving patterns are the use of emission factors that are based on specific road levels of service, or on a combination of vehicle-specific power and instantaneous speed. 61

Driving Cycles Emission factors for MOBILE6 and EMFAC2007 are de- veloped based on emissions testing on a standardized driving cycle such as FTP. A great amount of research has been de- voted to the development of driving cycles that reflect actual driving and, as a result, the Heavy-Duty Diesel Test Cycle (HDDTC) was developed by the California Air Resources Board. However, the question still remains as to whether a single driving cycle is able to provide enough information for the development of accurate speed correction factors. Modal emissions models such as MOVES and CMEM also rely on prespecified driving cycles, but the development of emission factors does not depend on speed correction factors, but on a combination of vehicle-specific power and instan- taneous speed. The use of different driving cycles also can reduce the uncertainty associated with the development of emission factors. Emission Factors The analysis of emission factors is discussed under each specific emission model. Classification of Truck Types The classification of trucks is important because (1) truck trip generation rates depend on how trucks are defined and (2) emission rates have a strong dependence on equipment type. Depending on the study, trucks might be classified based on their gross weight, number of axles, or configuration (single-unit, combination). Such variance in classification systems prevents the development of trip generation rate aver- ages across studies. As a result, the number of sample studies for a given classification system is small, which increases the uncertainties associated with trip generation rates. Another issue is that the heavy-duty truck categories in MOBILE6 and EMFAC2007 do not match the categories reported under HPMS. As a result, the process of mapping truck categories between these systems is not always straight- forward. For example, HPMS currently characterizes heavy- duty trucks in two categories, namely single-unit trucks and combination trucks, as opposed to eight categories in MOBILE6, and three categories in EMFAC2007, in both cases according to gross vehicle weight. This issue is being resolved in MOVES since it categorizes heavy-duty trucks according to the same classification system used by HPMS. Road Grade The effects of road grade are not incorporated into MOBILE6 or EMFAC2007. Although this is not a freight-related issue per se, the effects of road grade on emissions are more pro- nounced on heavy-duty trucks than on light-duty vehicles. The importance of road grade on heavy-duty emissions can be evaluated from current modal emissions models. A previ- ous study that evaluated truck movements over 23 different corridors concluded that fuel consumption increased by 10% to 35% as a result of grades. (71) Assuming that fuel con- sumption is a good proxy for emissions, the impacts of road grade are significant on emissions as well. Empty Miles Empty miles refer to the need for empty equipment to be relocated to places where it is required. Because additional fuel is consumed (and emissions generated) in the movement of empty truck equipment, ideally, empty miles should be considered in emission analyses. In national analyses, where truck VMT is estimated based on HPMS data, empty movements are captured as well as loaded movements, since traffic measurements do not differ- entiate trucks based on their cargo. In regional and project- level analyses however, empty movements need to be consid- ered separately. The incorporation of empty miles is very challenging because of a lack of data. Public data sources with aggregate information about empty miles exist. For example, VIUS provides data on empty mileage for different truck types. (Other than a truck’s home-base state and the share of miles driven within and outside of the home-base state, there is no information in VIUS that could indicate empty mileage in specific corridors.) More accurate empty factors for specific lanes and commodities could be obtained directly from truck- ing companies, but that is generally unlikely due to confiden- tiality issues. It is possible that transportation rates could re- flect empty miles, but it is difficult to disaggregate the impacts of empty miles from other factors such as supply and demand, labor markets, and equipment availability. Due to these chal- lenges, many analyses simply disregard empty movements. A recent study from FRA (71) estimated the impacts of empty miles on fuel consumed in different truck movements, with a fuel penalty between 9% and 21% in fuel efficiency. 3.4 Rail This section includes (1) a brief documentation of the cur- rent practice and methodologies for calculating emissions from freight rail, (2) a summary of the strengths and weak- nesses of such methods, and (3) an analysis of uncertainty as- sociated with these methods, as well as with the parameters used in the emission calculations. Topics covered include streamlined and detailed methods of estimating rail activity, emission factors, and total emissions at the national, regional, and project-level geographic scales. Most rail emission method- 62

ologies combine fuel-based emission factors with measured or calculated fuel consumption to determine total emissions. However, as data availability varies over different geographic scales, different methodologies are required. Independently of the geographic scale, rail operations are typically categorized in switch and line-haul due to different activity patterns and equipment configurations. Line-haul operations refer to the movement over long distances, gener- ally with newer and more powerful locomotives than switch operations, and tend to idle less. Switch activities refer to the assembling and disassembling of trains at railyards, sorting of rail cars, and delivery of empty rail cars to terminals. Switch operations involve short-distance movements, significant idling, and older equipment. Most rail methodologies rely on fuel consumption data to determine emissions. Detailed fuel consumption data are typ- ically considered sensitive information by railroads. However, nationwide aggregate fuel consumption data, which are based on 100% reporting for Class I railroads, are available from in- dustry or government agencies (i.e., Association of American Railroads, Energy Information Administration, state agencies, private companies via surveys). When fuel consumption data are not available for the region of interest, it must be estimated either by apportioning fuel consumption from a larger geo- graphic area (top-down) or by aggregating fuel consumption from individual rail movements (bottom-up). Both methods require measurements of rail activity. Because the rail sector has fewer metrics of activity when compared to other modes, methods for calculating emissions tend to be overly simplified or overly complex, with the atten- dant uncertainties and inaccuracy. Streamlined, or top-down, methods determine emissions based on publicly available data on fuel consumption at the state or national level, and appor- tion emissions to the state or county level using an available activity metric, such as traffic density or mileage of active track. Detailed, or bottom-up, methods calculate fuel consumption either by measuring freight movements or surveying individ- ual railroad companies. Both approaches are discussed in this section. Exhibit 3-25 includes the summary of methods to calculate rail emissions. 3.4.1 Evaluation of Emission Models The calculation of rail emissions does not typically rely on a specific emission model. In some isolated cases, train simula- tion software also can be used to estimate fuel consumption on a given rail line. The best well-known train simulation software in the United States is possibly the Train Energy Model (TEM) developed by the Transportation Technology Center for the Association of American Railroads. It is a single train simula- tor for long-haul trains along specific routes, and was designed to calculate journey time and fuel use. Simulation model out- puts are typically compared against real-world scenarios in order to calibrate the model and adjust the coefficients. Like most train simulation models, TEM relies on a set of train re- sistance equations originally developed by W. J. Davis in 1926. (72) These equations quantify train resistance based on train weight, speed, number of axles, train composition, track cur- vature, and grade. Fuel consumption can be derived from train resistance. Since then, the equations have been adapted to more recent standards, accounting for updated rail equipment and operational requirements. The use of train simulation software enables the most accurate results, but requires activity data at a level that is not typically available to most agencies. 3.4.2 Evaluation of Regional Methods Typically, there is little or no published information on railroad activity available for a specific region. Thus, state and regional air quality agencies must obtain railroad activity data directly from the railroad companies. Railroad companies often are reluctant to provide detailed fuel consumption or activity data due to concerns over distributing sensitive infor- mation. Even when these data are provided, they often are not reported with a high level of detail, due in part to the railroad company procedures for maintaining such data. 63 Exhibit 3-25. Rail methods. Method GeographicScale Pollutants EPA GHG Inventory National GHG Locomotive National Emissions Inventory (NEI) National CAP and toxics Line-Haul Emissions by Traffic Density Regional/Local All Line-Haul Emissions by Active Track Regional/Local All Switch Emissions by Number of Switchers or Hours Regional/Local All Line-Haul/Switch Emissions by Employees Regional/Local All Line-Haul/Switch Emissions by Time in Mode Local All Line-Haul Emissions at Marine Terminals Local All

Methods to quantify regional rail emissions can be divided in the following types: (1) line-haul emissions by traffic density, (2) line-haul emissions by active track, (3) switch emissions by number of switchers or hours of operation, and (4) line-haul/ switch emissions by number of employees. Line-Haul Emissions by Traffic Density EPA’s guidance for regional inventory preparation pro- vides an approach that estimates line-haul rail fuel consump- tion by means of traffic density. (73) In the National Emission Inventory (NEI), previously described in Section 3.2.3, EIA’s estimates of national rail fuel consumption are multiplied by EPA’s national locomotive emission factors. (74–75) Na- tional rail emissions can be apportioned to individual coun- ties based on their share of traffic density (gross ton-miles). County traffic density is obtained from the National Trans- portation Atlas Database (NTAD), which includes traffic density data for each track in the United States. (76) To main- tain the confidentiality of railroad data, the NTAD does not contain actual traffic density, but six ranges of traffic density, of which the medians are used for emission calculations. (77) A similar method relies on statewide data, which can be used in place of national data. Each freight railroad that op- erates in a state/region is asked to report gross ton-miles (GTM) by county, as well as total fuel consumption in the state. If a railroad is able to provide this information, the statewide line-haul fuel use is apportioned to counties in di- rect proportion to the GTM. Sometimes the railroads per- form this fuel use allocation using their own estimate of fuel use per GTM. Another variation of the same method relies on more project-level data. According to the formula in Equation 4, fuel consumption is determined by dividing traffic density (in GTV) by the systemwide fuel consumption index, measured in gross ton-miles per gallon. A systemwide fuel consumption index can be determined for each individual railroad by dividing its annual traffic den- sity by its annual fuel consumption, and these two parame- ters can be obtained from published Surface Transportation Board (STB) data. This method also is based on the appor- tionment of fuel use by GTM, but it relies on more specific data, which can be obtained from each of the participating railroads. The fuel use estimates for each railroad are summed, with the result being an estimate of total railroad fuel use by county. Fuel Consumption gallons Rail Traffic Den ( ) = sity gross ton-miles Fuel Consumption Inde ( ) x gross ton-miles per gallon Equation 4) ( ) ( Emission factors (in grams/gallon) are applied to the fuel use figures to estimate annual emissions. Using a constant fuel consumption index, which is equiva- lent to apportioning fuel use by GTM, is an inaccurate method for most regional and project-level emission applications be- cause it ignores key local factors such as grade, equipment type (which influences aerodynamic coefficients, and payload to tare ratios), and possibly congestion. All of these factors can have a substantial effect on fuel consumption per ton-mile, as indicated in a recent study from FRA. (71) Correction fac- tors for grade and commodity group can be used to minimize the uncertainty associated with the use of a single measure of fuel efficiency. There have also been questions about the accu- racy of county-level GTM data reported by railroads. As indicated by a previous study, a good example of the po- tential shortcomings of such an approach is its application in California. (77) The two Class I railroads that operate in Cali- fornia, Union Pacific and Burlington Northern Santa Fe, pri- marily offer intermodal service over relatively hilly terrain in the Sierra Nevada Mountains. Their national operations however, are dominated by coal trains operating at relatively level terrain. Because coal trains are much more fuel efficient than inter- modal trains, system fuel consumption index is a very poor indicator of regional fuel consumption index in California. The FRA study and other analyses have estimated meas- ures of rail fuel efficiency for different types of trains, lanes, and commodities, so it is possible to determine a range of variation in terms of fuel consumption index (Exhibit 3-26). Correction factors to adjust the systemwide fuel consump- tion index in EPA’s guidance were developed by Sierra Re- search. (78) Such correction factors adjust for the steepness of terrain as well as the proportion of bulk rail traffic. Although these factors account for the effects of the most important pa- rameters on rail fuel efficiency, there are concerns about the validity of such factors given that they were estimated based on outdated data from a single study. Additionally, it is un- certain to what extent such correction factors are used in emissions studies. The use of fuel consumption indexes that are specific to a given lane, train type, and commodity, such as those included in the FRA study, provide a more accurate measure of train fuel efficiency. 64 Rail Equipment Min Max Double-Stack 523 849 Mixed 367 691 Auto Rack 542 620 Exhibit 3-26. Range of rail fuel efficiency (gross ton-miles/gallon).

Summary of Strengths and Weaknesses. The analysis of strengths and weaknesses is provided in Exhibit 3-27. Line-Haul Emissions by Active Track For railroads that are not able to report GTM, mileage of active track is used as a proxy. If the railroad is able to report statewide line-haul fuel use, fuel use is apportioned to coun- ties in direct proportion to the railroad’s track mileage by county. If the railroad cannot report statewide fuel use, national-level fuel use (as reported by the Association of American Railroads) is apportioned to state and county based on track mileage. Like the previous method, fuel use estimates for each railroad are summed, resulting in an estimate of total fuel use by county. Emission factors (in grams/gallon) are ap- plied to the fuel use to estimate annual emissions. The main shortfall to this methodology is that active track is almost certainly not an accurate proxy for fuel use. In most regions, some rail lines are used much more heavily than others. Thus, using track length to apportion fuel con- sumption to the county level probably results in significant inaccuracies. Summary of Strengths and Weaknesses. The analysis of strengths and weaknesses is provided in Exhibit 3-27. Switch Emissions by Number of Switchers or Hours EPA and CARB utilize a simplified approach to estimate emissions at individual railyards, whose emissions are added in regional studies. Each freight railroad that operates in a re- gion is asked to report the number of switch yard locomotives they operate, by county or by individual yard. Some railroads also are able to provide hours of switch locomotive use by county or yard. Railroads are asked to report the average annual fuel consumption rate (in gallons per locomotive per year) of their switch yard locomotives. If railroads can- not provide this rate, a rate is assumed based on EPA guid- ance or on information from other railroads. Switch yard locomotive fuel use is then calculated by applying a fuel consumption rate to the number of switch yard locomo- tives, assuming an average locomotive duty cycle. Fuel use estimates are summed, and emission factors (in grams/gallon) are applied to the fuel use to estimate annual emissions from switch locomotives. 65 Criteria Emissions by Traffic Density Emissions by Active Track Emissions by Number of Switch Locomotives or Hours Emissions by Employee Representation of physical processes Weakness: Depending on the quality of input data, this method can provide an inaccurate estimate of regional emissions if it assumes that emissions are proportional to gross ton- miles. This assumption ignores the fact that emissions also depend on type of rail equipment, commodity, terrain level, and logistics requirements. Weakness: This method provides a very inaccurate estimate of regional emissions because it assumes that emissions are proportional to active track. This ignores the dependence of emissions on track utilization, rail equipment, commodity, terrain level, and logistics requirements. Weakness: This method assumes no variation in terms of the number of operating hours per switch locomotive (if that information is not provided by the railroads). It also assumes the same duty cycle across different yards. Weakness: This method is very inaccurate because it does not consider differences in duty cycles, operating hours, commodity carried, equipment types, terrain, and labor productivity. Method sensitivity to input parameters Weakness: These methods are only dependent on one input. Method flexibility Strength: This method can be used with either national or statewide data. Strength: If local data are not available by the participating railroads, surrogate data from average estimates can be used. Representation of future emissions Strength: Because this method relies on data that is published annually and that can be forecasted based on economic projections, emissions can also be forecasted. Data quality Weakness: Because of railroad confidentiality, the NTAD only provides ranges of traffic density. There have also been concerns about the accuracy of county- level GTM data reported by railroads. Strength: If number of switch locomotives is used, the process of data collection should be straightforward and accurate. Weakness: If number of hours is used, data quality can vary widely because there are no standards related to data collection. Spatial variability Weakness: These two methods do not provide a good representation of the differences across geographies because they ignore the impacts of terrain grade on emissions. Weakness: This method does not provide a good representation of the differences across railyards because it assumes the same duty cycle and, sometimes, the same number of hours per switch locomotive. Temporal variability Weakness: Because these methods rely on aggregate data, they do not provide any indication on how emissions are distributed across different months, weeks, or days. The only exception is if the number of hours of operation is collected at different time periods. Review process Weakness: There have not been any studies comparing regional/local emissions from these two methods versus other methods. Weakness: Recent emission inventories completed by railroads show large differences in operating hours and fuel use by switch locomotive. The difference in operating hours is between 0% to110%, and the difference in fuel use per locomotive is between -32 to +41%. Weakness: There have not been any studies comparing regional/local emissions from this method to those of other methods. Endorsements Strength: Based on EPA guidance Strength: Based on EPA and CARB guidance. Exhibit 3-27. Analysis of strengths and weaknesses—comparison of methods.

This method assumes no variation in terms of the number of operating hours per switch locomotive (if that information is not provided by the railroads) or the locomotive duty cycle across different yards. As indicated by a recent study, recent emission inventories completed by railroads to support CARB’s railyard health risk assessment show large differences in operating hours and fuel use by switch locomotive. As for operating hours, the difference between the detailed studies and those utilizing the standard methodology ranged from 0% to almost 110%, while the difference in fuel use per loco- motive ranged from −32% to 41%. (77) Summary of Strengths and Weaknesses. The analysis of strengths and weaknesses is included in Exhibit 3-27. Line-Haul/Switch Emissions by Employees Class II and III railroads (short line and switch railroads) are often unable to provide the information described above (e.g., number of switch locomotives, hours of operation). In some regions (such as Chicago), the number of Class II/III railroads in operation is considered too large to make surveys of individ- ual companies practical. In these cases, fuel consumption can be estimated by obtaining the number of employees of the rail- road by county (using a commercial employment database such as Dun & Bradstreet) and a ratio of fuel consumption per employee. This method does not take into consideration that different railroads carry different commodities on different types of trains over varying terrain—all of which are factors that have a strong effect on fuel efficiency. Additionally, this method also assumes that labor productivity is the same among rail- roads, which is also a questionable assumption. Summary of Strengths and Weaknesses. The analysis of strengths and weaknesses is provided in Exhibit 3-27. 3.4.3 Evaluation of Local/Project-Level Methods The previous section included methods that could estimate emissions at the regional and local level but that generally do not rely on specific project-level data. This section includes those methods that are based on local inputs. Local/project-level analyses that rely on detailed activity data from participating railroads result in more accurate rail emissions than regional analyses do. Line-Haul/Switch Emissions by Time in Mode Rail activity can be measured in number of operating hours in each notch for each type of train traveling on each route or operating at each railyard. Railroads can obtain such infor- mation from locomotive event recorders, which record time spent on each throttle notch, and train performance model- ing software (e.g., Train Energy Model). This is by far the most accurate method, but it relies on detailed information from railroads, which do not always have the resources to collect (or are willing to share) such information. Studies that rely on this type of methodology are gener- ally performed very sporadically due to the intense resource requirements—an example is the Booz-Allen study (79) in California. Updates to such studies, like those done by CARB, typically are based on growth factors that are applied equally to all routes. (80) The use of growth factors is related to sev- eral shortcomings: (1) some growth factors are based on U.S. economic growth and are not specific to California, (2) growth factors will not reflect changes in train and commodity mix, train length, and locomotive power, all of which have a strong effect on locomotive duty cycles and the time spent on each throttle notch. (77) In particular, intermodal traffic has in- creased at an annualized rate of 3.9% from 1990 to 2005, well above the 3.1% annual increase in rail on average. (81) There- fore, it is highly unlikely that the train or commodity mix will remain constant over time. The use of event recorder data to get time-in-notch and fuel consumption can be done to extrapolate data from a few trains to the line average. Summary of Strengths and Weaknesses. An analysis of strengths and weaknesses is provided in Exhibit 3-28. Line-Haul Emissions at Marine Terminals Although EPA guidelines (82) recommend that line-haul locomotive activity be measured in terms of fuel consump- tion, the estimation of rail-related emissions at port emission inventories typically take an alternative approach to better reflect line-haul operations within marine terminals. Since line-haul locomotives move over very short distances within marine terminals, rail activity is measured in hours of opera- tion. Because line-haul emission factors can be expressed in terms of horsepower-hour, rail activity can be calculated in the same unit, as shown in Equation 5. In a detailed inventory, all inputs to this equation are ob- tained from the participating railroads, which otherwise need to be estimated. If local estimates are not available, the num- ber of containerized trains can be calculated based on the number of TEUs, train capacity, an average utilization rate, Line-Haul Rail Activity bhp-hr Number of Trai( ) = ns Locomotives per Train Hours at Port Averag × × × e Load Factor Average Locomotive Horsepower Eq× ( uation 5) 66

plus a ratio of empty miles. EPA’s best practices guidance for port-related emission inventories provides default assump- tions for the other inputs based on previous inventories, but relying on average inputs ignores the operational differences among different ports. As a result, the difference between using default assumptions and using local assumptions could be a factor of up to three times, based on a comparison with emission inventories done by the ports of Los Angeles, (83) Long Beach, (84) and Seattle and Tacoma. (85) A more accurate method to quantify line-haul rail activity at marine terminals is to use event recorders to measure fuel burnt per train mile within a port. Summary of Strengths and Weaknesses. An analysis of strengths and weaknesses is provided in Exhibit 3-29. 3.4.4 Evaluation of Parameters Exhibit 3-30 includes a list of parameters used in the meth- ods and models. These parameters are described throughout this section. Pedigree Matrix. Exhibit 3-31 provides a pedigree matrix for data quality assessment that assigns quantitative scores to all parameters included in Exhibit 3-30. The criteria to assign scores in the pedigree matrix are included in Appendix A. Fuel Consumption Class I railroads are required to report fuel use to the Sur- face Transportation Board (STB) via Schedule 700 of the R1 Annual Report. As a result, the fuel use data published by the Association of American Railroads (AAR) is based on 100% reporting. Even then, there have been questions about the accuracy of fuel consumption data reported by railroads. For example, the fuel use in Texas reported by railroads for 2001 (220 million gallons) is less than half the locomotive fuel sales for the state as reported by DOE (504 million gallons) for that year. Some of this discrepancy can be explained by the fact that railroads often purchase fuel in one state and then consume that fuel in another. Unfortunately, there are no mechanisms to verify the fuel consumption data reported by railroads. Additionally, there is little correlation between fuel purchases and fuel consumption in a state because locomotives can travel long distances between fuel pur- chases. Note that Class II and III railroads are not required to report fuel use. 67 Criteria Strengths Weaknesses Representation of physical processes Provided that data are available to represent the local conditions (grade, equipment type, duty cycles), this is the most accurate method to estimate rail emissions. Method sensitivity to input parameters Method flexibility This method relies on very detailed data requirements. Ability to incorporate effects of emission reduction strategies Because key input parameters are captured in this method, it is generally possible to analyze the effects of emission reduction strategies. Representation of future emissions Consideration of alternative vehicle/fuel technologies This method can capture the effects of the use of hybrid switch locomotives. Data quality Data quality can vary significantly depending on the specific data collection process. Spatial variability If detailed local data are provided, this method gives a good degree of spatial variability. Temporal variability This method does not provide any indication on how emissions are distributed across different months, weeks, or days. Review process Endorsements Exhibit 3-28. Analysis of strengths and weaknesses—emissions by time in mode.

68 Criteria Strengths Weaknesses Representation of physical processes If local estimates of rail activity are not available, the use of default assumptions could result in large uncertainties due to operational differences across different marine terminals. Method sensitivity to input parameters Method flexibility This method can be used with either local estimates or national default assumptions. Ability to incorporate effects of emission reduction strategies Representation of future emissions Because this method relies on cargo data to estimate rail activity, economic indicators can be used to forecast emissions. Consideration of alternative vehicle/fuel technologies Data quality Spatial variability If detailed local data are provided, this method provides a good degree of spatial variability. Temporal variability This method does not provide any indication on how emissions are distributed across different months, weeks, or days. Review process Endorsements Exhibit 3-29. Analysis of strengths and weaknesses—emissions at marine terminals. Parameter Methods/Models GeographicScale Pedigree Matrix Qualitative Assessment Quantitative Assessment Fuel Consumption National National Locomotive Duty Cycles All (explicit in regional/local) Regional/Local Emission Factors All All Locomotive Type All (explicit in local) All Locomotive Tier Distribution All All Empty Miles All Local Traffic Density Emissions by Traffic Density Regional/Local Miles of Active Track Emissions by Active Track Regional/Local Number of Switch Locomotives Emissions by Switchers Regional/Local Hours by Switch Locomotive Emissions by Hours Regional/Local Number of employees Emissions by Employees Regional/Local Key: indicates that a parameter is analyzed in the way denoted by the column: indicates that the parameter is not discussed in the way denoted by the column. Exhibit 3-30. Rail parameters.

Although self-reported fuel consumption estimates are con- sidered the most accurate data source available, this accuracy could be improved by reconciling top-down (i.e., fuel con- sumption through fuel sales data) and bottom-up (i.e., fuel consumption through activity data) approaches. Rail Activity Rail fuel use needs to be estimated based on rail activity if accurate fuel sales data are not available or are not represen- tative of fuel burned in a geographic area. The estimation of rail activity in gross ton-miles or number of hours is exam- ined in the previous discussion of methods. Locomotive Duty Cycles A locomotive duty cycle is a usage pattern expressed as the percentage of time spent in each of the throttle notches. The 1998 rulemaking was based on two duty cycles—one for line- haul and one for switch—which were development based on industry data. (86) Line-haul data were based on 2,475 hours operated by 63 trains from five Class I railroads across many regions in the country. Without more information about the process of sampling and development of an average cycle, it is reasonable to assume that there were enough data points to provide a good representation of an average duty cycle. Switch duty cycle data came from two local railroads with over 300 hours of operation by eight trains. The relatively small number of switch locomotives and railroads brings concerns about the statistical representation of an average switch duty cycle. Additionally, the variation of the percent- age of time in each throttle notch was very high for both line- haul and switch cycles, as illustrated in Exhibit 3-32. Such high variation is the main reason why the use of an average duty-cycle is a poor substitute for regional or local data. These cycles were developed before the widespread use of idle con- trol devices in locomotives, so updated cycles should incor- porate those effects. Emission Factors Generally, locomotive emission factors are based on EPA’s 1992 emission inventory guidance. (87) Documentation since then has provided updated rail emission factors based on more recent emission standards for locomotives, including EPA’s 1998 Regulatory Support Document, and the Sierra Research work published in 2004. (86, 88) The most recent emission factors for locomotives are included in EPA’s 2008 Regula- tory Impact Assessment, (89) which includes new emission standards for Tier III and Tier IV locomotives. The RIA doc- umentation also provides baseline emission rates for NOx, PM, HC, and CO in 2008, which are based on average duty 69 Parameter Im pa ct on R es ult Ac qu isi tio n M eth od In de pe nd en ce Re pr es en tat ive ne ss Te m po ra l C or re lat ion Ge og ra ph ic Co rre lat ion Te ch no lo gic al Co rre lat ion Ra ng e o f V ar iat ion Fuel Consumption 5 1 3 1 1 1 5 Locomotive Duty Cycles 4 Varies 3-4 Varies 4 5 5 Emission Factors 5 2 3 3-4 4 5 5 Locomotive Type 3 Varies 4 3-4 4 5 4 Locomotive Tier Distribution 4 Varies 4 Varies 1 5 3 Equipment Type 4 1 4 Varies 1 Varies 4 Empty Miles 3 4 3 1 1 2 5 Traffic Density 5 1-2 3-4 1 1 1 Varies Miles of Active Track 5 1 3 1 1 1 1 Number of Switch Locomotives 5 1 4 1 1 1 1 Hours by Switch Locomotive 5 2-3 4 1 1 1 3 Number of employees 5 1 3 1 1 1 1 Exhibit 3-31. Pedigree matrix—rail parameters.

cycles for switch and line-haul locomotives. (90) However, the emission rates for Tier II and older locomotives are still based on the previous rulemaking document. Baseline emission rates (NOx, PM10, HC, and CO) by loco- motive type and throttle notch were developed based on data provided by locomotive manufacturers (GM and EMD), and EPA weighted these data by the average duty cycles to esti- mate average baseline emission rates. Exhibit 3-33 summa- rizes the variation in emission rates for NOx and PM. For the line-haul cycle, the highest emission rates were roughly twice the lowest rate, while for the switch cycle the highest rates were about four times higher than the lowest rate. This wide discrepancy is strictly related to the measurement of emission rates and is not influenced by the variation in duty cycles that was previously examined. Therefore, the errors embedded in both parameters will be added and propagated through the calculation of rail emissions at the regional or local level. The emission rates for different locomotive tiers were based on the expected emission reduction compared to the baseline rates. Tier III will need electronic common rail fuel injection systems as well as better oil control. These electronic systems should reduce the amount of uncertainty in emissions factors for these engines. Tier IV will most likely need selective cat- alytic reduction (SCR). Additional complexities exist in tam- pering and mal maintenance as well as whether the urea tanks are filled. Significant swings in emissions can occur if tamper- ing or mal maintenance occurs. In most analyses of rail emissions, emission factors are converted from g/bhp-hr to g/gal by applying a factor of 20.8 bhp-hr/gal for line haul, and 18.5 bhp-hr/gallon for switchers. This assumes a constant brake-specific fuel con- sumption (BSFC) of 0.341 lb/bhp-hr for line-haul and 0.383 lb/ bhp-hr for switchers. These average BSFCs were determined through certification test data, but BSFC tends to vary depend- ing on engine size as well as notch setting. Errors in emission factors can result if the locomotives have different duty cycles than those included in the certification tests. However, signif- icant changes to emission factors typically occur when there are high variations in the share of time spent in notches 5 through 8 versus time in idle. The emission factor for CO2 tends to be the most accurate because CO2 emissions are proportional to fuel consumption. PM2.5 emission factors can be calculated by assuming that they represent a fixed percentage of PM10 emissions. EPA rec- ommends the use of 97% based upon an analysis done for the NONROAD model. This was based upon engines using 500 ppm sulfur diesel fuel and may be different for engines using higher sulfur content. PM10 emission factors reflect the emission rates expected from locomotives operating on fuel with sulfur levels at 3,000 ppm, so it is important that regional and local analyses obtain information about the sulfur con- tent of diesel fuel used in locomotives. EPA estimates that the PM10 emission rate for locomotives operating on nominally 500 and 15 ppm sulfur fuel will be 0.05 and 0.06 g/bhp-hr 70 Line-Haul Duty Cycle Switch Duty Cycle Throttle Notch Average Lowest Highest Average Lowest Highest Idle 38.0 77 1 59.8 82 23 Dynamic Brake 12.5 41 0 N/A N/A N/A 1 6.5 23 0 12.4 18 7 2 6.5 23 0 12.3 18 7 3 5.2 13 2 5.8 20 1 4 4.4 11 1 3.6 17 1 5 3.8 12 0 3.6 15 0 6 3.9 11 0 1.5 10 0 7 3.0 18 0 0.2 1 0 8 16.2 39 0 0.8 4 0 Source: U.S. Environmental Protection Agency (1998): Locomotive Emission Standards, Regulatory Support Document. Exhibit 3-32. Duty cycle variation (% time in throttle notch). Line-Haul Duty Cycle Switch Duty Cycle Pollutant Average Lowest Highest Average Lowest Highest PM 0.32 0.22 0.41 0.44 0.22 0.86 NOx 13.0 10.3 18.2 17.4 9.2 33.1 Source: U.S. Environmental Protection Agency (1998): Locomotive Emission Standards, Regulatory Support Document. Exhibit 3-33. Baseline emission rates (g/bhp-hr).

lower than the PM10 emission rate for locomotives operating on 3,000 ppm sulfur fuel, respectively. (89) Emissions of SO2 are relatively accurate, and can be cal- culated through a mass balance approach, since it can be reasonably assumed that most of the sulfur in the fuel will be converted to SO2 (the rest will be emitted as particulate matter). Locomotive Type Most analyses of rail emissions depend on emission rates developed by EPA as part of the rulemaking. As previously in- dicated, these emission rates were based on measurements from 63 locomotives of three types, and a large variation was observed between the highest and lowest emission measure- ments of the same locomotive type. However, the variation of the minimum measurements across the three locomotive types was not as high, with measurements of NOx for the line- haul cycle ranging from 10.3 to 11.5 g/bhp-hr, and from 0.22 to 0.25 for PM. Similar variations were observed for maxi- mum measurements. The variations across locomotive types for the switch cycle were higher, especially for the maximum measurements, which ranged from 15.8 to 33.1 for NOx, and from 0.39 to 0.86 for PM. These differences are not an issue for analyses where fuel consumption data can be obtained directly. However, for those analyses where fuel use is estimated based on activity data rather than fuel consumption data, variations in locomotive type can increase the difference between actual and modeled emission factors. Locomotive Tier Distribution Locomotive tier distribution is certainly an important fac- tor when deriving a composite emission factor, since emission rates are widely different across locomotive tiers (with the exception of CO), as shown in Exhibit 3-34. Therefore, it is important to obtain the correct locomotive tier distribution from participating railroads when estimating regional and local emissions. Equipment Type Train type has a strong effect on fuel consumption and, consequently, on emissions. Two factors influence this corre- lation, namely the ratio between payload and total car weight (payload plus tare weight), and train aerodynamic resistance. Rail cars with a low ratio between payload and total car weight will have lower fuel efficiency when measured in terms of revenue ton-miles/gallon. A study from FRA evaluated (71) the differences in fuel efficiency among different types of rail cars. For example, auto haulers, with ratios between payload and total car weight ranging between 25% and 30%, have rel- atively poor fuel efficiency in comparison to other equipment types. In contrast, tank cars and covered hoppers have ratios above 75%, which explains higher fuel efficiencies in compar- ison to other equipment types. Empty Miles More sophisticated analyses also can account for fuel consumed in empty movements by applying an empty fac- tor to Equation 5. If local estimates are not available, data from the R1 report can be used to estimate empty ratios by rail car type. (91) Empty miles refer to the miles spent to get empty equip- ment to places where it is needed. Because additional fuel is consumed (and emissions generated) in the movement of empty rail cars, ideally, empty miles should be considered in emission analyses. In national analyses, where fuel use is estimated based on information reported by Class I railroads, empty movements are captured as loaded movements. In regional and local analyses however, empty movements need to be considered separately since fuel use is estimated from rail activity. The incorporation of empty miles is very challenging due to lack of data and the complexity of the logistics of empty move- ments. Public data sources with aggregate information about empty miles exist. For example, data from the R1 report can be used to estimate empty ratios by rail car type. However, due to these challenges, many analyses simply disregard empty movements. A recent study from FRA estimated the 71 Line-Haul Locomotives Switch Locomotives Tier PM10 NOx HC PM10 NOx HC Remanufactured Tier 0 0.20 6.70 0.29 0.23 10.62 0.57 Remanufactured Tier I 0.20 6.70 0.29 0.23 9.90 0.57 Remanufactured Tier II 0.08 4.95 0.13 0.11 7.30 0.26 Tier III 0.08 4.95 0.13 0.08 5.40 0.26 Tier IV 0.015 1.00 0.04 0.015 1.00 0.08 Source: EPA (2008). Exhibit 3-34. Emission rates for line-haul and switch locomotives (g/bhp-hr).

impacts of empty miles on fuel consumed in different rail movements, with a fuel penalty between 4% and 29% in fuel efficiency. (71) 3.5 Waterborne/Ocean-Going Vessels Cargo movements by marine vessels include ocean-going vessels (OGVs) and barge movements pushed by tugs or tows. OGVs are discussed in this section, followed by a dis- cussion of tug/tow movements at ports and inland rivers in Section 3.6. Emissions from OGVs are usually determined at and around ports because these are the entrances and clearances of cargo into the regions of modeling interest. They are estimated using information on number of calls at a particular port, engine power, load factors, emission factors, and time in like modes. The current practice to calculate emissions from OGVs is to use energy-based emission factors together with activity profiles for each vessel. The bulk of the work involves deter- mining representative engine power ratings for each vessel and the development of activity profiles for each ship call. Using this information, emissions per ship call mode can be determined using Equation 6. Where E = Emissions (grams [g]), P = Maximum Continuous Rating Power (kilowatts [kW]), LF = Load Factor (percent of vessel’s total power), A = Activity (hours [h]), and EF = Emission Factor (grams per kilowatt-hour [g/kWh]). 3.5.1 Summary of Methods and Models There are three basic methods for calculating emissions from OGVs at ports, namely (1) detailed methodology where considerable information is gathered regarding ships enter- ing and leaving a given port, (2) a mid-tier method that uses some detailed information and some information from sur- rogate ports, and (3) a more streamlined method in which detailed information from a surrogate port is used to estimate E P LF A EF (Equation 6)= × × × emissions at a “like” port. The detailed methodology requires significant amounts of data and resources and produces the most accurate results. The mid-tier and streamlined methods require less data and resources but produce less accurate re- sults. (9) Exhibit 3-35 lists these three methods. There are no current publicly available models for calcu- lating OGV emissions at ports. Most researchers use one of the three methods described here to estimate emissions at ports. A list of recent mid-tier and detailed inventories is pro- vided in Exhibit 3-36. 3.5.2 Evaluation of Methods and Models Since all of the current methods and models estimate emis- sions at ports, the geographic distinctions (i.e., national, re- gional, and local/project scale analyses) are less meaningful than in other sectors. Generally, to estimate national OGV emissions, all major ports are modeled and emissions added together. For a regional approach, such as that done by CARB for estimating California marine vessel emissions, a similar approach is taken where emissions at the major California ports are estimated and then added together. The difference really relies upon whether a detailed or streamlined method is used for the individual ports and the data that are collected. Detailed Methodology In the detailed methodology, emissions from OGVs are es- timated from detailed information on ship calls at a given port together with detailed ship characteristics, time and speed in each mode, load factors, and emission factors. The more de- tailed the information collected, the more accurate the results. Each parameter, as well as its potential biases and errors, is discussed in the following subsections. Calls. The most accurate information for the number of calls comes from the local port Marine Exchange or Port Authority (MEPA). MEPAs generally record vessel name, IMO number, date and time of arrival, and date and time of depar- ture. Larger MEPAs also record flag of registry; ship type; pier/wharf/dock (PWD) names; dates and times of arrival and departure from various PWDs, anchorages, next ports; cargo type; cargo tonnage; activity description; draft; vessel 72 Method Geographic Scale Pollutants Freight/Passenger Detailed Methodology All All Both Mid-Tier Methodology All All Both Streamlined Methodology All All Both Exhibit 3-35. OGV methods.

dimensions; and other information. Generally MEPAs record every ship that enters or leaves a port but do not record those that stop at private terminals outside the port authority juris- diction. On a national or regional level, not counting these calls can lead to underestimation of emissions related to OGVs for the area. A second source of call data is U.S. ACE entrances and clear- ances data. The Maritime Administration (MARAD) maintains the Foreign Traffic Vessel Entrances and Clearances Database, which contains statistics on U.S. foreign maritime trade. Data are compiled during the regular processing of statistics on for- eign imports and exports. The database contains information on the type of vessel, commodities, weight, customs districts and ports, and origins and destinations of goods. There are several drawbacks to using U.S. ACE entrances and clearances data. First, it does not contain any call TIM information. Average TIM and speeds need to be used with the U.S. ACE data to perform a mid-tier or streamlined analysis. Second, it only represents foreign cargo movements. Thus domestic traffic, defined in the Jones Act (106) as U.S. ships delivering cargo from one U.S. port to another U.S. port, is not accounted for in the database. Ship calls where no cargo is loaded or unloaded are also excluded. However, U.S. flagged ships carrying cargo from a foreign port to a U.S. port or from a U.S. port to a foreign port are accounted for in the U.S. ACE entrances and clearances database since these are considered foreign cargo movements. Although at most ports domestic commerce is carried out by Category 2 ships, there are a few exceptions, particularly on the West Coast. Unfor- tunately, there is little or no readily available information on domestic trips, so determining this without direct port input is difficult. Third, the entrances and clearances data does not always match MEPA data because it does not differentiate between public and private terminals at a port. This is impor- tant because a port authority may not have jurisdiction over private terminals. A recent study found that the U.S. ACE entrances and clearances data accounted for over 90% of the emissions from Category 3 ships calling on U.S. ports. (107) For a national or regional level analysis, not counting U.S. Jones Act ships could result in an large underestimation of emissions if the region is on the West Coast. From a local level, including ship calls that are not part of a port authority 73 Port YearPublished Data Year Method Pollutants Contractor* Selected Alaska Ports (92) 2006 2002 Mid-Tier SO2, NOx, PM10, PM2.5, CO, NH3, VOC Pechan Beaumont/Port Arthur (93) 2004 2000 Detailed NOx, CO, HC, PM10, SO2 Starcrest Charleston (94) 2008 2005 Detailed NOx, TOG, CO, PM10, PM2.5,SO2 Moffatt & Nichol Corpus Christi (95) 2003 1999 Detailed NOx, VOC, CO ACES Houston (96) 2009 2007 Detailed NOx, VOC, CO, PM10, PM2.5,SO2, CO2 Starcrest Great Lakes (Ports of Cleveland, OH and Duluth, MN) (97) 2006 2004 Detailed HC, NOx, CO, PM10, PM2.5, SO2 Lake Carriers Assoc. Lake Michigan Ports (98) 2007 2005 Mid-Tier NOx, PM10, PM2.5, HC, CO, SOx Environ Los Angeles (99) 2008 2007 Detailed NOx, TOG, CO, PM10, PM2.5,SO2, DPM, CO 2, CH4, N2O Starcrest Long Beach (100) 2009 2007 Detailed NOx, TOG, CO, PM10, PM2.5,SO2, DPM, CO2, CH4, N2O Starcrest New York/New Jersey (101) 2008 2006 Detailed NOx, VOC, CO, PM10, PM2.5,SO2, CO2, N2O, CH4 Starcrest Oakland (102) 2008 2005 Detailed NOx, ROG, CO, PM, SOx Environ Portland (103) 2005 2004 Mid-Tier NOx, HC, CO, SOx, PM10,PM2.5, CO2, 9 Air Toxics Bridgewater Consulting Puget Sound** (104) 2007 2005 Detailed NOx, TOG, CO, PM10, PM2.5,SO2, DPM, CO2, CH4, N2O Starcrest San Diego (105) 2008 2006 Detailed NOx, TOG, CO, PM10, PM2.5,SO2, DPM Starcrest Notes: * Starcrest = Starcrest Consulting Group LLC, ACES = Air Consulting and Engineering Solutions; Environ = Environ International Corp. ** Includes the Ports of Anacortes, Everett, Olympia, Port Angeles, Seattle, and Tacoma. Exhibit 3-36. Recent port inventories.

jurisdiction could result in an overestimation of emissions for that port. Power. Determination of ship propulsion power is fairly straightforward using Lloyd’s Ship Register data. Lloyd’s data are produced by Lloyd’s Register-Fairplay Ltd., head- quartered in Surrey, England. (108) Lloyd’s data contains information on ship characteristics that are important for preparing detailed marine vessel inventories. These data include the following: • Name, • Ship Type, • Build Date, • Flag, • Dead weight tonnage (DWT), • Vessel service speed, and • Engine power plant configuration and power. All data are referenced to both ship name and IMO num- ber. Only the IMO number is a unique identifier for each ship because the name of a ship can change. Lloyd’s insures many of the OGVs on an international basis and, for these vessels, the data are quite complete. For other ships using a different insurance certification authority, the data are less robust. Using Lloyd’s data to determine propulsion power should lead to fairly accurate emissions calculations. Auxiliary engine power also can be determined from Lloyd’s data, but many records are missing this information. Best prac- tices dictate using ratios of auxiliary to propulsion power from a CARB survey (109) based upon ship type to determine total ship auxiliary power. Although on a large scale this will lead to fairly accurate emission determinations, on a local level, ship auxiliary power to propulsion power ratios may vary by ship size and thus be less accurate for a smaller port. Load Factor. Load factors are expressed as a percent of the vessel’s total propulsion or auxiliary power. At service or cruise speed, the propulsion load factor is assumed to be 83%. At lower speeds, the Propeller Law should be used to estimate ship propulsion loads, based on the theory that propulsion power varies by the cube of speed as shown in Equation 7. Where LF = Load Factor (percent), AS = Actual Speed (knots), and MS= Maximum Speed (knots). When ships move against significant river currents, the actual speed in Equation 7 should be calculated based upon the following: for vessels traveling with the river current, the LF AS MS (Equation 7)= ( )3 actual speed should be the vessel speed minus the river speed; for vessels traveling against the river current, the actual speed should be the vessel speed plus the river speed. Because of the stall speed of a ship, load factors are assumed not to fall below 2%. There are several assumptions made here. First, the cruise speeds listed in Lloyd’s data are 94% of maximum speed used in Equation 7. Starcrest, in their 2001 Port of Los Angeles inventory (110) determined that service speed varied from 83.3% to 100% of maximum speed for 28 ships surveyed. The average of those surveyed was 94%. Thus, propulsion cruise load factors could vary from 57.8% to 100% resulting in a possible over- or underestimation of emissions. The second assumption is that the Propeller Law holds true for all conditions and propeller designs. The basic Propeller Law assumes a fixed pitch propeller and free sailing in calm waters. Wind and water currents, heavy seas, fouling, and other factors can increase the amount of load necessary, while improved propellers, ship hull design and other factors can reduce the power required to move at a given speed. Thus, propulsion load factors calculated using the Propeller Law can result in potential errors in emission calculations. Since the Propeller Law is used to derive main engine load factors for cruise, RSZ and maneuvering modes, uncertainty in this approach propagates to emissions calculations in those OGV activity modes. Current auxiliary engine load factors came from interviews conducted with ship captains, chief engineers, and pilots dur- ing Starcrest’s vessel boarding programs. (83) Auxiliary load factors are specified by ship type and time in mode. Because ships vary in generating needs, auxiliary load factors can vary from ship to ship. Overstating the auxiliary load factor can re- sult in an overestimation of emissions, while understating the auxiliary load factor can result in an underestimation of emis- sions. In a large inventory (or several inventories to comprise a regional or national analysis) it is likely that these factors balance out. Activity. OGV activity is usually broken into like modes that have similar speed and load characteristics. Vessel move- ments for each call are described by using four distinct TIM calculations. A call combines all four modes, while a shift nor- mally occurs as maneuvering. Each TIM is associated with a speed and, therefore, an engine load that has unique emis- sion characteristics. Although there will be variability in each vessel’s movements within a call, these TIMs allow an average description of vessel movements at each port. TIMs should be calculated for each vessel call occurring in the analysis year over the waterway area covered by the corresponding MEPA. TIMs are described in Exhibit 3-37. Cruise speed (also called service speed) is listed in Lloyd’s data and generally taken as 94% of the maximum service speed. Distances from the maximum port boundary to either the 74

RSZ or the breakwater are used with the cruise speed to de- termine cruise times into and out of the port (however, not all ports have a physical breakwater, and for those without, an imaginary breakwater needs to be defined). Some MEPAs record which route was used to enter and leave the port, and this information can be used to determine the actual dis- tances the ships travel. Determining the actual distance and speed during cruise mode is the most accurate method. Speeds and locations can be determined using the Automatic Identification System (AIS), which at least two services track. (Lloyd’s AISLive and VesselTracker.com). Less accurate meth- ods include assuming the ship travels at service speed during the cruise portion and estimating the distance the vessel travels in that mode. Reduced speed zone TIM also is an estimation based on average ship speed and distance. Starcrest refers to this TIM as “transit” in their inventory documents. Generally, the RSZ starts when a ship enters the U.S. coastline such as a shipping channel, river, or bay where speeds need to be reduced for navigational purposes. The RSZ ends at the port entrance. Pilots can provide average ship speeds for a precautionary or reduced speed zone. Again, such speeds are estimates and more accurate results can be gained from using AIS. Maneuvering time in mode is estimated based on the dis- tance a ship travels from the port entrance to the PWD. Aver- age maneuvering speeds vary from 3 to 8 knots depending on direction and ship type. Outbound speeds are usually greater than inbound speeds because the ship does not need to dock. Ships go from half speed to dead slow to stop during maneu- vering. Time in mode varies depending on the location of, and the approach to, the destination terminal and turning require- ments of the vessel. Best practice is to determine maneuvering times from conversations with pilots. Again, the maneuvering time will vary by ship size, currents, traffic, and other factors. Accuracy in determining maneuvering times can affect calcu- lations of hotelling time as discussed in the next paragraph. Hotelling can be calculated by subtracting time spent maneuvering into and out of a PWD from the departure time minus the arrival time into a port. If possible, anchorage time (time at anchorage within the port but not at a PWD) should be broken out from time at a PWD. Some MEPAs record shifts as well and this will allow for further refinements in maneuvering time. Other methods to determine hotelling include conversations with pilots. During hotelling, the main propulsion engines are off, and only the auxiliary engines are operating, unless the ship is cold ironing. Hotelling times also can be determined from pilot records of vessel arrival and departure times when other data are not available. Actual hotelling times should be calculated for each individual port because hotelling is generally a large portion of the emissions 75 Summary Table Field Description Call A call is one entrance and one clearance from the MEPA area. Shift A shift is a vessel movement within the MEPA area. Shifts are contained in calls. Although many vessels shift at least once, greater than 95% of vessels shift three times or less within most MEPA areas. Not all MEPAs record shifts. Cruise (h/call) Time at service speed (also called sea speed or normal cruising speed) usually is considered to be 94% of maximum speed and 83% of maximum continuous rating (MCR). This is calculated for each MEPA area from the port boundary to the breakwater or reduced speed z one. The breakwater is the geographic marker for the change from open ocean to inland waterway (usually a bay, river, or channel). Reduced Speed Zone* (RSZ) (h/call) Time in the MEPA area at a speed less than cruise and greater than maneuvering. This is the maximum safe speed the vessel uses to traverse distances within a waterway leading to a port. Reduced speeds can be as high as 15 knots in the open water of the Chesapeake Bay, but tend to be about 9 to 12 knots in most other areas. Some ports are instituting RSZs to reduce emissions from OGVs as they enter the port. Maneuver (h/call) Time in the MEPA area between the port entrance and the pier/wharf/doc k (PWD). Maneuvering within a port generally oc curs at 5 to 8 knots on average, with slower speeds maintained as the ship reaches its PWD or anchorage. Even with tug assist, the propulsion engines are still in operation. Hotelling (h/call) Hotelling is the time at PWD or anchorage when the vessel is operating auxiliary engines only or is cold ironing. Auxiliary engines are operating at some load conditions the entire time the vessel is manned, but peak loads will occur after the propulsion engines are shut down. The auxiliary engines are then responsible for all onboard power or are used to power off-loading equipment, or both. Cold ironing uses shore power to provide electricity to the ship instead of using the auxiliary engines. Hotelling needs to be divided into cold ironing and active to accurately account for reduced emissions from cold ironing. * Referred to as the transit zone in many inventory documents. Exhibit 3-37. Vessel movements and TIM descriptions within MEPA areas.

at a port. Hotelling times should be separated for those ships that use cold ironing at a port and those that do not. It is im- portant to also look for outliers (ships with extremely long hotelling times) to eliminate those in the average since they may represent ships at a PWD but not with auxiliary engines on. Miscalculation of hotelling time can directly affect emis- sion calculations. Hotelling emissions are generally a signifi- cant part of ship emissions near ports. Emission Factors. The current set of marine engine emis- sion factors come from ENTEC (111), which were derived from emissions data from 142 propulsion engines and 2 of the most recent research programs: Lloyd’s Register Engi- neering Services in 1995 and IVL Swedish Environmental Research Institute in 2002. ENTEC estimated uncertainties at the 95% confidence interval, presented in Exhibit 3-38 for the ENTEC emission factors. New work by IVL (112) shows major reductions in CO and HC emissions in comparisons with the previous ENTEC study. CO emissions for slow-speed diesel engines (SSDs) are about one-third of previous values, while the new study shows HC emissions at approximately half of prior values. In addition, PM emissions seem to vary significantly from ship to ship. Be- cause of these observed differences, it is likely that the actual uncertainty (within the given confidence intervals) on PM, CO, and HC emissions are much higher than those specified in Exhibit 3-38. Another assumption made in using emission factors is that they are constant down to about 20% load. Below that thresh- old, emission factors tend to increase as the load decreases. This trend results because diesel engines are less efficient at low loads and BSFC tends to increase. Thus, while mass emis- sions (grams per hour) decrease with low loads, the engine power tends to decrease more quickly, thereby increasing the emission factor (grams per engine power) as load decreases. Energy and Environmental Analysis Inc. (EEA) demonstrated this effect in a study prepared for EPA in 2000. (113) This study defined low-load adjustment factors that should be multi- plied by the propulsion engine emission factors when the load factor is below 20%. These factors can be large at very low loads. Although these low-load adjustment factors are used in most of the recent port inventory analyses and are recom- mended in the EPA guidance (114), they were derived mostly on distillate fuels, and much of the data came from Coast Guard cutters and ferries. Exhibit 3-39 shows the observa- tions and R2 values from the curve fits for the various emis- sions. As can be seen from Exhibit 3-39, the curve fits have relatively low R2 values. These low correlation coefficients and the small sample of ship types imply highly uncertain low-load adjustment fac- tors. It also should be noted that the PM adjustment factors are particularly suspect because they were only estimated based on smaller engines operating on distillate fuel. Although errors can occur in the determination of the low-load adjust- ment factor, the loads at which these adjustments are applied are very low, and the overall impact of these uncertainties is probably small. Summary of Strengths and Weaknesses. The analysis of strengths and weaknesses is included in Exhibit 3-40. Mid-Tier Methodology Some mid-size ports, or those preparing emission inven- tories with mid-sized resources, could prepare a simplified, mid-tier version of the inventory. This differs from the de- tailed methodology by averaging vessel characteristics and operational data by ship type. Even better resolution can be gained if the average information also is broken down by ship size (DWT range). Load factors and emission factors for each ship type and DWT range can be calculated using a method similar to that in the detailed methodology. Annual vessel calls for each ship type and DWT range should be determined at the port. Each call should be divided into the various modes 76 Pollutant Estimated Uncertainty NOx ± 20% SO2 ± 10% CO2 ± 10% HC ± 25% PM ± 25% BSFC ± 10% Exhibit 3-38. Estimated uncertainties at 95% confidence interval. Pollutant Observations R2 NOx 291 0.57 SO2 239 0.78 CO2 291 0.65 CO 291 0.52 HC 291 0.52 PM 31 0.95 Exhibit 3-39. Low-load adjustment factor derivation information.

of operation and each mode also should be averaged for the vessel type and DWT range. The mid-tier approach is detailed in Commercial Marine Port Inventory Development. (115) In this report, U.S. ACE entrances and clearances data are married with Lloyd’s data. Emissions for the modeled port were then determined by mode, ship type, engine type, and DWT range from similar categories at the paired typical port for which a detailed in- ventory was done. The same baseline errors and parameter uncertainties dis- cussed in the detailed methodology exist in this method. Ad- ditional uncertainties arise from the selection of the like port and the implications of that choice on the various activity modes. Like-Port Selection. This process involves determining a port for which a detailed inventory has been prepared (typ- ical port) that is similar to the port to be modeled (i.e., like or modeled port). The more similar the port chosen as the typ- ical port is to the modeled port, the more accurate the results. For large ports, the errors are probably small because most of the detailed inventories done to date were for large ports. However, if modeling a small port and using a large port as the typical port, the error margins could be large as different ship sizes service smaller ports and the port efficiency is usu- ally lower. Additional issues in port selection are discussed in each of the time-in-mode calculations discussed below. Cruise Mode. Cruise mode emissions are calculated by determining ratios of number of calls, average propulsion and auxiliary engine power, and vessel service speed between the modeled port and the typical port. Because this information is used, uncertainties in the cruise mode emissions for the mod- eled port are related to uncertainties in the detailed port analy- sis, the similarities between distances traveled at the two ports, and the vessel, engine, and fuel similarity at the two ports. The bias due to distance may be quantifiable and correctable. RSZ Mode. In the transit, or RSZ, mode, the average dis- tance and average speed is specified for both the typical and 77 Criteria Detailed Mid-Tier Streamlined Representation of physical processes Strength: Dominant physical processes included. Sensitivity to input parameters Strength: Method relies on detailed user inputs that may not be readily available, but should produce best results Weakness: General, overall uncertainty unknown Weakness: Method relies on surrogates for missing inputs; results highly sensitive to quality of inputs Flexibility Weakness: Requires detailed data collection Strength: Customizable to data limitations Ability to incorporate effects of emission reduction strategies Strength: Best information available; effects may be included in use of different EFs Strength: Highly customizable Representation of future emissions Strength: Projections available in the model and customizable to local information Consideration of alternative vehicle/fuel technologies Strength: May be achieved in methodology by using appropriate EFs Weakness: Does not consider alternative vehicle/fuel technologies Data quality Strength: Structured from best available information Weakness: Structured from available information Spatial variability Strength: Applicable to any location, but data requirements likely limit to smaller spatial scales Strength: Applicable to any location; data flexibility allows multiple spatial scales Temporal variability Weakness: Most likely limited to annual inventories Strength: Designed for annual inventories, but scalable with appropriate information Review process Strength: Documented in EPA Methodology Guidance Endorsements Strength: EPA endorsed Exhibit 3-40. Summary of strengths and weaknesses—comparison among methodologies.

modeled port. In addition to ratios of number of calls and propulsion and auxiliary engine power, ratios of propulsion load factors and TIM are also calculated and used in determin- ing emissions. This should provide results similar in accuracy to the detailed port analysis as long as the average speeds are fairly representative of the ships that call on both ports. If there is a disparity of speeds among the ships at the two ports, errors can result in the emission calculation. Maneuvering Mode. For the maneuvering mode, only ratios of number of calls and propulsion and auxiliary en- gine power are used. It is assumed that the modeled port and the typical port have the same maneuvering time and load factors. If the two ports are different in distances from the port entrance to the PWD or in the number of shifts that occur, errors in maneuvering emissions will result. How- ever, since maneuvering emissions are small compared to the other activity modes, the contribution to overall error will probably be small. Hotelling Mode. For the hotelling mode, only ratios of number of calls and auxiliary engine power are used. It is assumed that the average hotelling time for each ship type is the same between the typical port and the modeled port. This can lead to errors if the efficiency at the typical port is different than at the modeled port. Since hotelling emissions are significant, the resulting error could be sig- nificant as well. Summary of Strengths and Weaknesses. The analysis of strengths and weaknesses is included in Exhibit 3-40. Streamlined Methodology A streamlined methodology can be applied if those prepar- ing port inventories do not have sufficient resources to fol- low the mid-tier approach described. In this approach, those preparing port inventories should use an existing emission inventory from another similar port, scaling the emissions up or down based on the ratio of vessel operation data between the two ports. Two EPA activity guidance documents pro- vide details on estimating emission inventories from other ports. (116–117) These documents use U.S. ACE data to scale emissions based on the ratio of ship trips from a like port that has an existing inventory compared to the port in question. No adjustments are made, however, for average propulsion and auxiliary power or vessel speed. This can result in significant error if the typical port selected is differ- ent from the modeled port as discussed in the mid-tier methodology subsection. Summary of Strengths and Weaknesses. Exhibit 3-40 includes the analysis of strengths and weaknesses for the detailed, mid-tier, and streamlined methodologies. 3.5.3 Evaluation of Parameters Exhibit 3-41 summarizes all parameters relevant for calcu- lating emissions from OGVs calling at ports. Each of these has been detailed under the discussion of the appropriate model or method in Section 3.5.2. Also as discussed above, no quantita- tive assessments are provided, because the range of parameters is essentially unknown. 78 Parameter Methods/Models GeographicScale Pedigree Matrix Qualitative Assessment Quantitative Assessment Calls All All Engine Power Detailed and mid-tier All Load Factor Detailed and mid-tier All Activity Detailed All Emission Factors Detailed All Port Selection Mid-tier and streamlined All Fuel Type Secondary; used to determine emission factors All Growth Factor Optional and secondary; needed for future year projections All Engine Age Distribution Optional and secondary; needed to determine average emission factors All Key: indicates that a parameter is analyzed in the way denoted by the column: indicates that the parameter is not discussed in the way denoted by the column. Exhibit 3-41. Parameters.

Pedigree Matrix. Exhibit 3-42 shows the pedigree matrix for the six primary parameters for determining emissions from OGVs. Criteria to assign scores in the pedigree matrix are included in Appendix A. Calls. Emissions are linearly related to the number of calls. Call data should be determined for each ship type and DWT range. Thus, while accurate assessment of the number of ship calls is critical, in many cases there can be errors depending upon the source of the data and the geographic boundaries of the analysis. Engine Power. In the detailed and mid-tier approaches, propulsion power is determined directly from Lloyd’s data. Conversely, auxiliary power is estimated from surveys that produce ratios of auxiliary power to propulsion power by ship type. More accurate determination of auxiliary power would improve emission calculations. Load Factor. In the detailed approach, propulsion load factors are calculated using the Propeller Law. There are inherent errors in applying that law to all ships and speed ranges. Currently the Propeller Law is universally accepted as the method to use to determine propulsion load factors. It is doubtful that significant errors would result from these calculations. Auxiliary load factors, however, have been determined from surveys and tend to change with each new Starcrest inventory. More precise determination of auxiliary engine load factors, particularly during hotelling, would provide more accurate results. Emission Factors. Emission factors for ships were deter- mined for a small subset of engines. Although most ships use similar engines, this set does not represent a large enough sample to be accurate. This is particularly true of PM emis- sions. Measurement techniques of PM emissions vary and there is sensitivity to sampling methodology (e.g., tunnel length). PM emission factors need a more robust data set to determine them more accurately. In addition, current thinking is to estimate PM2.5 emission factors as 92% of PM10 emission factors for OGVs. Various studies have esti- mated PM2.5 emissions from 80% to 100% of PM10 emissions. Therefore, a more accurate determination of PM2.5 emission factors is needed. Low-load adjustment factors also need reviewing. The cur- rent methodology is based upon limited data and rough curve fits. Improvement of the low-load adjustment factors can result in more accurate emission calculations. Furthermore, the current emission factors were deter- mined for engines built before year 2000 when IMO set NOx emission standards on OGV engines. More testing is needed to determine the emission factors for engines built after 2000 as well as for future IMO Tier II and Tier III NOx emission standards. Port Selection. In the mid-tier and streamlined method- ologies, selecting a typical port that is like the port to be mod- eled is of utmost importance. EPA has provided some guidance on how to select the typical port and a list (118) based upon detailed inventories prepared at the time. As more ports pre- pare detailed inventories, this list should be expanded. 79 Parameter Im pa ct on R es ult Ac qu isi tio n M eth od In de pe nd en ce Re pr es en tat ive ne ss Te m po ra l C or re lat ion Ge og ra ph ic Co rre lat ion Te ch no lo gic al Co rre lat ion Ra ng e o f V ar iat ion Calls 4 1-2 1-2 1-2 1 1 Varies 3 Engine Power 4 2 1 1 1 Varies 2 2 Load Factor 4 3-4 3 2 1 Varies 3 4 Activity 4 2-4 3 3 1 Varies 1 3 Emission Factors 4 2-3 1-2 4-5 3 Varies 3 4 Port Selection 4 4 3 N/A N/A Varies N/A N/A Exhibit 3-42. Pedigree matrix—OGV parameters.

3.6 Waterborne/Harbor Craft A wide range of commercial harbor craft (H/C) is in oper- ation at or near ports, including assist tugboats, towboats/ pushboats/tugboats, ferries and excursion vessels, crew boats, work boats, government vessels, dredges and dredging sup- port vessels, commercial fishing vessels, and recreational vessels. These vessels serve many purposes other than just direct goods movement. From a freight perspective, it is worthwhile to focus only on those commercial H/C (SCC 2280002000) directly involved in goods movement, such as tug and towboat operations that move freight barges. Emis- sions and parameters relative to other commercial H/C are not considered here. There are no common models with the capability to esti- mate emissions from these vessels; neither CARB’s NON- ROAD nor EPA’s OFFROAD model considers commercial H/C. Instead, estimates of emissions for tug and towboats and other commercial H/C may be made through other method- ologies, such as the best practice or streamlined approaches discussed in EPA’s Current Methodologies document. (9) These general approaches rely on various sources for the necessary parameters and generally draw on the methodologies of the NONROAD or OFFROAD models, or other published stud- ies. They assemble parameters including a survey or estimate of the vessel and/or engine counts and engine activity and merge this information with emission and load factor data from the technical literature. For example, H/C emission inventories are commonly calculated using an equipment power method- ology, as shown in Equation 8. Where the sum is over the population of all main and auxiliary engines active in the fleet and the input parameters are as follow: EF = the emission factor for a given pollutant species and engine, HP = the engine horsepower, LF = the load factor, A = the annual activity, CF = the appropriate emission control factor. Any deterioration, low-load, transient, or other adjust- ment effects (if able to be characterized) are considered in the age- or tier-distributed EF. Both main and auxiliary engines are included. Differences in the best practices and stream- lined methodologies are chiefly dependent on the amount of data directly collected rather than derived through surro- gates. For the purposes of uncertainty assessment, they will be treated as the same methodology. EMIS EF Tier LF A HP CFPollutant Pollutant M = ( ) i i i i ain Auxiliary (Equation 8) + ∑ Two specific H/C methodologies are EPA’s national scale Regulatory Impact Analysis (RIA) done in support of the 2008 rulemaking (119) and CARB’s analysis of statewide H/C emis- sions analysis in support of its rulemaking. (120) Although constrained by the same limitations discussed previously, these analyses are both sufficiently developed and tailored to be discussed separately. A third, general methodology is dis- cussed for local scale analysis based on available guidance and previous project analyses. The following sections discuss these national, regional, and project-scale methodologies. Uncertainty in the resulting H/C emissions from these methodologies can then be attributed to either process uncer- tainty (that is, the degree to which Equation 8, or similar for- mulations, represent the actual processes causing emissions) or parameter uncertainty (that is, the uncertainty in the indi- vidual elements of Equation 8). Evaluation of process uncer- tainty is presented in the following three sections by domain; discussion of parameter uncertainty also appears for each methodology and is then summarized in Section 3.7.5. In both cases, any known biases should be corrected dur- ing the analysis. The effects of quantifiable residual uncer- tainty in input parameters on total calculated uncertainty may be made using standard error propagation methods, dis- cussed in Section 4.3.4. If no covariance is assumed for the parameters in Equation 8 the net error in total emissions would be given by Equation 9. Where σ2 indicates the variance. Note, however, that Equation 9 assumes the number of engines is sufficiently well known to complete the sum. More likely, the estimates and uncertainties are made by calcula- tions discussed in Section 3.6.2, which would allow inclusion of uncertainty in number as well. 3.6.1 Summary of Methods and Models As stated previously, the discussion here will focus on ele- ments of potential methods to estimate H/C emissions gen- erally, since few studies focus only on H/C directly involved in goods movement (i.e., ocean and line-haul tug and tow vessels). Since no models may be used to calculate H/C emis- sions directly, Exhibit 3-43 lists only methods. Two specific and one general method are listed, although the structure of each is very similar. The specific methods were developed by σ σ σ 2 2 2 2 2 Emis HP LF A EF EF LF A HP + EF HP = ( ) + ( ) i i i i i i A LF EF HP LF A Main ( ) + ( ) ⎢ ⎣ ⎢⎢⎢⎢⎢ ⎥ ⎦ ⎥⎥⎥⎥⎥ + 2 2 2 2 σ σi i Auxiliary (Equation 9)∑ 80

regulatory agencies to detail H/C emissions within a set geo- graphic range. The general method, which is labeled here as “the Local H/C Method,” is an aggregate of several studies that have been conducted at the project level. Neither of the specific methodologies, and most of the studies that form the basis of the general method, were applied solely to freight- moving H/C, although all could be modified to exclude other H/C types. 3.6.2 Evaluation of National Methods and Models The most current, national scale inventory of H/C emis- sions is related to EPA’s 2008 locomotive and marine engine rulemaking. (119) The Regulatory Impact Analysis (RIA) developed includes a baseline national emission inventory for Category 1 and 2 commercial marine vessels, including freight- related, commercial H/C. (89) EPA RIA Methodology In this case, separate inventories were developed for com- mercial marine diesel engines in the following three principal categories: • Category 1 propulsion engines, • Category 1 auxiliary engines, and • Category 2 propulsion engines. Propulsion and auxiliary engines less than 37 kW (50 hp) were also considered. Category 2 auxiliary engines were not considered, however, as these are only used on Category 3 ves- sels. These inventories include all commercial harbor craft, however, not only those directly involved in goods move- ment. Exhibit 3-44 shows the current definitions of marine compression-ignition engine categories. Exhibit 3-45 shows the strengths and weaknesses of the EPA RIA Methodology. Calculation Method. Commercial marine diesel en- gine inventories for HC, CO, NOx, and PM were estimated using spreadsheet calculations using the formula shown in Equation 10. Where E is the 50-state emission inventory (tons per year) for commercial marine vessels, N is engine population (units), P is the average rated power (kW), L is the load factor, A is the engine activity (operating hours/year), and EF is the emission factor (gram/kW-hr). Average rated power, load factor, and activity parameters are assumed constant across all simulation years but popula- tions and emission factors were considered to vary by year and age. Populations and the corresponding age distribution E N P L A EF (Equation 10)= × × × × 81 Method/Model Type Geographic Scale Pollutants Freight/Passenger EPA RIA Methodology Method National NOx, HC, PM, toxics Both ARB H/C Methodology Method Regional NOx, PM, ROG, CO Both Local H/C Method Method Local/Project All Both Exhibit 3-43. Harbor craft inventory methods. Category Specification Use Approximate Power Ratings 1 Gross engine power 37 kW*displacement < 5 liters per cylinder Small harbor craft and recreational propulsion < 1,000 kW 2 Displacement 5 and < 30 liters per cylinder OGV auxiliary engines, harbor craft, and smaller OGV propulsion 1,000–3,000 kW 3 Displacement 30 liters per cylinder OGV propulsion > 3,000 kW * EPA treats all engines with gross power below 37 kW (50 hp) separately. Exhibit 3-44. EPA marine compression ignition engine categories.

are calculated for the baseline year (generally 2002) and then projected. Emission factors vary with age to account for the effects of regulations and deterioration. PM emission factors also consider the in-use fuel sulfur level. Generally, the calculation methods are similar to those for CHE, including use of the NONROAD scrappage function, the linear deterioration factor, and sulfur PM adjustments. Inventory results are calculated in bins of power (in kW), en- gine displacement (L/cylinder), and power density (kW/L) to accommodate the form of the regulations, which differ from the standard break points used in the NONROAD model. Input Parameters. The population parameters were de- rived by displacement category, power density, and total power from historical sales estimates (provided by PSR [the Power Systems Research Database]), combined with scrappage, and then disaggregated into power and power density categories using the 2002 population and engine data. The average power values, load, and activity were population-weighted into ap- propriate bins to compute totals (see discussion under CHE in Section 3.7). Category 1 main engine load factor and activity estimates were determined from industry analysis and prior rulemak- ing as 0.45 and 943 h/year (engines less than 750 hp) and 0.79 and 4,503 h/year (greater than 750 hp). A median life of 13 years is used for all Category 1 main engines from indus- try estimates, with an annual growth rate of 1.009 (for domes- tic shipping from EIA). Baseline emission factors were taken from the 1999 Marine Diesel rulemaking, based on emissions data for uncontrolled engines. Tier I emission factors are esti- mated for NOx using 2006 certification data by displacement category; other pollutant factors equal the baseline values. Tier II PM, NOx, and HC emission factors are derived from 2006 certification data. Certification data relies on sales- weighted values from the E3 duty cycle. A parallel method was used for Category 1 auxiliary engines, but certification data from the D2 auxiliary cycle were used to derive load factors. Resulting load factor and activity estimates (from PSR) were 0.56 and 724 h/year for engines less than 750 hp and 0.65 and 2,500 h/year from the 1999 rulemaking for engines greater than 750 hp. A median life of 17 years is used for all Category 1 auxiliary engines. Category 2 main engine emissions also were calculated with a similar methodology, although here separate estimates were made for underway and idling activity. In this parameterization, an activity-based approach is substituted with a TIM approach. Accordingly, the activity parameter (in hours per year) is sub- stituted with the formula shown in Equation 11. In both cases, a “likely” load factor is used. Minimum, max- imum, likely load factors, and annual transit days are provided, as well as likely idle days. Activity estimates are discussed with a range of methods and resulting estimates, showing the uncer- Likely Annual Transit Days hours day f ( ) × ( )24 or underway emissions Likely Annual Idling Days hours day( ) × ( )24 E for idling emissions ( quation 11) 82 Criteria Strengths Weaknesses Representation of physical processes Overall average emissions processes included from all Category 1 and 2 H/C Variety of methods used to account for different input data Sensitivity to input parameters Method relies on documented inputs and discusses necessary choices Some inputs show significant differences from other studies; resulting overall uncertainty uncharacterized Flexibility Tailored methodology Not directly applicable to H/C subcategories or smaller spatial domains Ability to incorporate effects of emission reduction strategies Designed to model effects of future regulations Representation of future emissions Designed to model effects of future regulations Consideration of alternative vehicle/fuel technologies Data quality Information included and documented from testing and other authorities. Unknown uncertainty or bias Spatial variability No spatial analysis included Temporal variability Produces only annual inventories Review process Unclear from documentation Endorsements EPA Exhibit 3-45. Summary of strengths and weaknesses—EPA RIA methodology.

tainty inherent in this parameter via this analysis. In fact, one method relies on a Monte Carlo analysis, thus directly incorpo- rating uncertainty into the process. Additionally, for ferries (although not considered here as directly associated with goods movement), emissions are calculated using a total fuel con- sumption methodology. The median life for all Category 2 main engines is taken as 23 years. (121) Emission factors are taken from the 1999 commercial marine rulemaking (122) except for Tier I NOx, which was updated based on 2006 certification data. Uncertainty. Total uncertainty in this method is due to both process and parameter uncertainty. As discussed for CHE (Section 3.7), three potentially significant sources of process uncertainty are the 1. Appropriateness and representativeness of the characteri- zations, 2. Groupings used to categorize H/C, and 3. Potential for bias in inputs. The process used here is generally appropriate and tailored to its purposes. No spatial disaggregation is provided because this is a national-scale inventory, thus no uncertainty is asso- ciated with disaggregation or translation of values between regions, which is typical of a top-down inventory. Load and activity factors are based on industry characterization, binned, and averaged using power and population as weights since equivalent NONROAD factors are not applicable. Thus, uncer- tainty in the final emissions estimates is related to the number of engines in each bin and the estimates of other parameters by bin. The process used here is generally believed to rely on the best information available, minimizing grouping uncertainty and representativeness of the method. However, some parameters differ significantly from previ- ously published values, particularly load factors. This could either represent or correct significant bias. Reference is given to the duty cycles from which the load factors are derived, however without commonly accepted average harbor craft duty cycles, assessment of bias is impossible. The same is true for emissions and activity factors, which differ from those of other studies. (123) Another source of uncertainty in binning is the difference in Category 1 and 2 main engines, especially for tug and tow boats. In the rulemaking, EPA cites two different methods to separate values based on power, hull displacement, and other categories. The differences in these two methods implied that around 6% of tug vessels could not be clearly categorized in this method. Although this does not affect the total number of vessels directly, it does affect the total emissions as emis- sion factors, load factors, activity, and other parameters are dictated by the type of main engines equipped on the vessels. Also, the subdivision of values based on power, engine dis- placement, power density, and age is complex, although no known bias results from this method. Finally, it must be noted that the methodology here gener- ally does not distinguish between freight and non-freight movement. Thus, translation of the methods (and, particu- larly, parameters here) to freight-only calculations is likely to result in bias, due to the different engines used. Summary of Strengths and Weaknesses. Exhibit 3-45 includes the analysis of strengths and weaknesses for the EPA RIA methodology. 3.6.3 Evaluation of Regional Methods and Models As for CHE, the only regional analysis of emissions from commercial H/C has been prepared by CARB for its Novem- ber 2007 rulemaking. (124) This rule has special provisions that apply to tug, tow, and ferry vessels. CARB Harbor Craft Methodology. CARB developed a methodology to estimate emissions from all commercial H/C in California to support analysis of regulations to reduce commercial marine engine emissions. (125) Other goals of the inventory development included updating estimates to represent the current H/C fleet, showing effects of the various regulatory programs, and allowing allocation of the statewide emissions to local air pollution control districts (APCDs) and air basins. Particularly in this last goal, the CARB H/C method- ology differs from the EPA RIA methodology. The methodology is based on activity. It uses results from CARB’s 2004 Commercial Harbor Craft Survey (126) to estimate average emissions per engine per year for nine types of vessels: commercial fishing vessels, charter fishing vessels, crew and supply boats, ferry/excursion vessels, pilot vessels, tow boats, tug boats, work boats, and “others.” These regional emissions are then aggregated to statewide emissions by multiplying num- ber of engines in each engine category and in each region by av- erage emissions per engine. Among the findings are that tugs and tows (that is, vessels most directly involved in freight move- ment) account for 4% of the statewide vessel inventory, 7% of the statewide engine inventory, but about 25% of the statewide emissions inventory (i.e., between 21% and 25%, depending on the pollutant). Population. Base year populations are drawn principally from the CARB Harbor Craft Survey (126) and aggregated with data from the U.S. Coast Guard Vessel Documentation Program, the California Department of Fish and Game regis- tration data, and information from the Port of Los Angeles emissions inventory. Then, spatial distributions to the air dis- trict and county level were calculated. Future year populations are based on base year populations aggregated with fleet growth 83

rates from local air districts and scrappage rates based on the OFFROAD model. The CARB survey on which estimates are based collected information for about 900 vessels (i.e., about 1,900 engines), or about 20% of the statewide H/C population. Although the emission methodology assumes the results of the survey are representative of the overall California commercial H/C fleet and scales results up to statewide values, uncertainty is intro- duced in the parameters resulting from this relatively small sample size. Further, although the survey was distributed to approximately 5,000 potential owners and operators, only 704 surveys were returned. (127) Uncertainty and potential bias exist in how well these limited responses represent the average H/C fleet operating in California. Activity and Engine Parameters. Vessel activity parame- ters also were derived from the CARB survey, which included information on vessel use, age, annual fuel consumption, number of engines per vessel, engine make and model, age, horsepower, annual hours of operation, and other informa- tion. These data were aggregated into operating profiles by engine type by region. Number of engines per vessel by vessel type was also determined from the survey, as was engine life- time. In this study, total life was defined as the age when 90% of engines retire and useful life (UL) was defined as half of total life. These definitions both differ from the standard NONROAD formula used in many studies, although the shape of the scrappage curve is very similar to that of the NONROAD model. The uncertainty in this method is due to the defini- tions of the terms as employed. Annual activity was derived from the CARB survey. It is un- known if these values are biased, such as toward the activity at the state’s largest ports. However, the same uncertainty exists here as with other parameters derived from the survey. Auxiliary engine load factors were taken as 0.43, which is attributed to the NONROAD model, for all commercial H/C except tug boats, where a factor of 0.31 was used, based on the Port of Los Angeles’ study. (83) These values differ from the EPA RIA method values, and it is unclear whether the attri- bution of the 0.43 factor is appropriate, since NONROAD does not include commercial marine vessels. Thus, some un- certainty is associated with use of these parameter values. Main engine load factors are derived from results of CARB survey responses to fuel consumption, engine power, and an- nual operating hours as shown in Equation 12. Where LF is the vessel type specific propulsion engine load factor, BSFC is brake-specific fuel consumption (here taken as 0.058 gal/hp-hr from manufacturers’ marine engine data), HP is the rated engine power, LF BSFC HP Hr TF Equation 12= × × ( ) Hr is the number of annual operating hours of the engine, and TF is total, annual, per engine fuel consumption. Uncertainty in this approach comes from both parameters and the process. There is uncertainty in the method since it relies on survey results, which may be biased or inappropri- ately aggregated. There also seems to be no accounting for potential deterioration. Parameter uncertainty comes from the derivation of parameters from the survey, but particularly from the reliance on BSFC. NONROAD estimates BSFC as 0.367 lb/bhp-hr for engines larger than 100 hp, based on measured fuel consumption values during engine certifica- tion (which translates to 0.052 gal/hp-hr at 7.09 lb/gal for diesel fuel). Although only a 10% discrepancy exists between the two, there is uncertainty as to which, if either, is more ap- propriate, on the whole, to commercial marine vessels for goods movement. Ultimately, the load factors derived here are smaller than those from the EPA RIA method, although more in line with other analyses. In all cases, the uncertainties here are unquantifiable. Emission Factors. Emission factors were taken from the OFFROAD model, except for the following: • 1996–1999 model year engines use baseline/Tier 0 (1996) emission factors; • 2000 and later model-year engines use the smaller of EPA emission standards for marine engines or the NOx limits of the IMO MARPOL Annex VI; and • OFFROAD model emission factors were adjusted to reflect an E3 test cycle for main engines and D2 test cycle for aux- iliary engines. Uncertainty in this approach is due primarily to the choices made in the method, but also to underlying uncertainty in the emission factors of the OFFROAD model and baseline EPA emission factors, as well as in duty cycle characterizations. In particular, the lack of differentiation between 2-stroke and 4-stroke engine emissions may be a significant source of un- certainty in the emission factors applied. Fuel correction and engine deterioration factors employed are derived from the OFFROAD model. Section 3.7.3 discusses the uncertainty in this model. Calculation Methodology. Commercial H/C emissions per engine are estimated as shown in Equation 13. Where E is the amount of emissions inventory, EF0 is the model year, horsepower, and engine type (main or auxiliary) specific zero-hour emission factor, E EF F D A UL HP LF Hr= +( )0 1i i i i i (Equation 13) 84

F is the fuel correction factor, D is the (power and pollutant-specific) deterioration factor, A is the engine age, UL is the (vessel type and engine-use specific) engine useful life, HP is the engine-rated horsepower, LF is the load factor, and Hr is the annual engine activity (operating hours). Each of the parameters in Equation 13 has already been discussed in Section 3.6.3. CARB calculated statewide and regional emissions using this equation, the aforementioned parameters, a database model to estimate vessel type specific emission rates, and scaled up the emissions to statewide populations. Uncertainty in this methodology is due to process and parameter uncertainty. Uncertainty in each of the parameters has already been discussed in Section 3.6.3. Uncertainty in the process is due to any discrepancies between the analysis pre- sented here and the physical processes estimated. Although the process used here is believed to rely on the best informa- tion available and capture the dominant processes contribut- ing to commercial H/C emissions, three potentially significant sources of process uncertainty are as follow: 1. Appropriateness and representativeness of the characteri- zations, including those of the OFFROAD model, 2. Groupings used to categorize H/C, and 3. Potential for bias in the raw or extrapolated survey results. Until a comprehensive nonroad mobile emissions model is produced and validated, reliance on models such as NON- ROAD and OFFROAD will be required to estimate emissions parameters. Thus, any process uncertainty in the models and on assumptions involving use of these models—which are not designed to simulate commercial marine emissions—is propa- gated to total emissions calculation. Process uncertainty from groupings is due to the employed methodology, which relies on use of “vessel type specific emission rates . . . scaled up to the statewide population” (128) in the database construction. Because parameters are specific to engine, fuel, age, vessel type and/or power, process uncertainty will propagate due to the grouping and application of appropriately weighted central values in each bin. These uncertainties are due to choice and as- signment of values to equipment groupings. Additional process uncertainty—and potential bias—is due to the extrapolation of small sample set values to statewide H/C populations. Quantifi- cation of these uncertainties, however, generally is infeasible. Summary of Strengths and Weaknesses. The strengths and weaknesses of the CARB H/C methodology are shown in Exhibit 3-46. 3.6.4 Evaluation of Local/Project Methods and Models Several studies of port-related activity and emissions have been conducted that capture commercial H/C emissions at the local or project level. These are listed in Exhibit 3-47. 85 Criteria Strengths Weaknesses Representation of physical processes Overall average physical processes included Sensitivity to input parameters Method relies on best available inputs Method relies on OFFROAD model; uncharacterized overall uncertainty Flexibility Tailored methodology Ability to incorporate effects of emission reduction strategies Not included in base methodology, but could be applied if information provided Representation of future emissions Method projects populations and associated factors Consideration of alternative vehicle/fuel technologies Fuel effects included No apparent treatment for alternative fuels or technologies Data quality Information included from survey of fleet Unknown uncertainties from extrapolation scheme Spatial variability Emissions allocated to county and air basin, but not more finely Underlying data applicable only to CA Temporal variability Only produces annual inventories Review process Available for public review as part of rulemaking Endorsements ARB Exhibit 3-46. Summary of strengths and weaknesses—CARB H/C methodology.

A common theme shared by most of these studies is esti- mating emissions from limited information. In that sense, they are typically some variation of the streamlined method- ology discussed in EPA’s best practices document. (9) How- ever, the level of detailed information on H/C available to the studies varies. The similarity of these studies is driven both by the trend to similar methodologies and by the fact that the majority of studies are made by the same contractor. They are also very similar to the EPA RIA methodology or the CARB H/C methodology, albeit with a more limited spatial scope, where variation is made for the amount of information avail- able and the portion of the fleet considered and a method that is similar to that of NONROAD or OFFROAD models. Two of the inventories presented in Exhibit 3-47 discuss Great Lakes activity (those by LCA and ENVIRON) and only one discusses inland river activity (Bridgewater). However, the nation’s inland waterway system is a principal area of opera- tions for line-haul tug and tow vessels, as well as an area of interest in terms of marine emissions. One study that estimates emissions at various ports along the inland river system is by ARCADIS. (117) That study collected information on several principal ports and performed a detailed emission inventory, 86 Port Year Published Data Year Pollutants Contractor* Selected Alaska Ports ( 92 ) 2006 2002 SO 2 , NOx, PM 10 , PM 2.5 , CO, NH 3 , VOC Pechan Beaumont/Port Arthur ( 93 ) 2004 2000 NOx, CO, HC, PM 10 , SO 2 Starcrest Charleston ( 94 ) 2008 2005 NOx, TOG, CO, PM 10 , PM 2.5 , SO 2 Moffatt & Nichol Corpus Christi ( 95 ) 2003 1999 NOx, VOC, CO ACES Houston/Galveston ( 12 9 ) 2000 1997 NOx, VOC, CO, PM 10 Starcrest Houston ( 96 ) 2009 2007 NOx, VOC, CO, PM 10 , PM 2.5 , SO 2 , CO 2 Starcrest Great Lakes (Ports of Cleveland, OH, and Duluth, MN) ( 97 ) (Tugs only) 2006 2004 HC, NOx, CO, PM 10 , PM 2.5 , SO 2 Lake Carriers Assoc. (LCA) Lake Michigan Ports ( 98 ) 2007 2005 NOx, PM 10 , PM 2.5 , HC, CO, SOx Environ Los Angeles ( 110 ) 2005 2001 NOx, TOG, CO, PM 10 , PM 2.5 , SO 2 , DPM Starcrest Los Angeles ( 83 ) 2007 2005 NOx, TOG, CO, PM 10 , PM 2.5 , SO 2 , DPM Starcrest Los Angeles ( 99 ) 2008 2007 NOx, TOG, CO, PM 10 , PM 2.5 , SO 2 , DPM, CO 2 , CH 4 , N 2 O Starcrest Long Beach ( 13 0 ) 2007 2005 NOx, TOG, CO, PM 10 , PM 2.5 , SO 2 , DPM Starcrest Long Beach (100) 2009 2007 NOx, TOG, CO, PM 10 , PM 2.5 , SO 2 , DPM, CO 2 , CH 4 , N 2 O Starcrest New York/New Jersey ( 131 ) 2003 2000 NOx, VOC, CO, PM 10 , PM 2.5 , SO 2 Starcrest New York/New Jersey ( 101 ) (Tugs only) 2008 2006 NOx, VOC, CO, PM 10 , PM 2.5 , SO 2 , CO 2 , N 2 O, CH 4 Starcrest Oakland ( 10 2 ) 2008 2005 NOx, ROG, CO, PM, SOx Environ Portland ( 103 ) 2005 2004 NOx, HC, CO, SOx, PM10, PM 2.5 , CO 2 , 9 Air Toxics Bridgewater Consulting Puget Sound** ( 104 ) 2007 2005 NOx, TOG, CO, PM 10 , PM 2.5 , SO 2 , DPM, CO 2 , CH 4 , N 2 O Starcrest San Diego ( 10 5 ) 2008 2006 NOx, TOG, CO, PM 10 , PM 2.5 , SO 2 , DPM Starcrest Notes: * Starcrest = Starc rest Consulting Group LLC, ACES = Air Consulting and Engineering Solutions; Environ = Environ International Corp. ** Includes the Ports of Anacortes, Everett, Olympia, Port Angeles, Seattle, and Tacoma. Exhibit 3-47. Recently conducted port inventories containing H/C.

then used a principal port-like port analysis to scale activity and emissions to other harbor areas. The general method for producing a local/project scale commercial H/C emissions inventory—specifically targeted to goods movement—and its associated uncertainties are dis- cussed in the remainder of this section. Here we focus only on vessels directly moving goods, as follows: • Line-haul and short-haul tug and tow boats that make calls along the inland waterway systems, transporting barges and containerized goods, and • Ocean-service tug and tow boats. Specifics on the studies listed in Exhibit 3-47 are provided in the individual inventories cited. Local Harbor Craft Methodology Input Parameters. To calculate emissions from commer- cial H/C involved in goods movement, the following informa- tion needs to be collected from vessel owners and operators for the relevant types of harbor craft operating in the port area: • Hours of operation (annual and average daily, plus sched- ules if relevant and available); • Percentage of time-in-operational modes (e.g., idling, half power, full power); • Vessel characteristics; • Number, type, age, and rated power of main engine(s); • Number, type, age, and rated power of auxiliary engine(s); • Other operational parameters such as fuel consumption rates and fuel type; • Qualitative information regarding how the vessels are used in service, including operating domain; and • Any information on emissions-modifying methods applied to the vessels, such as exhaust after-treatment equipment installed or internal engine modifications. Ideally, average values of annual operating hours, number of main and auxiliary engines, engine power, and engine age should be determined from the information collected from the vessels operating at the specific port. This approach min- imizes parameter uncertainties because the calculations are made directly on the fleet in question. Process uncertainties remain on binning and methodology, and should be quanti- fied where possible. Inland river activity data often are taken from the ARCADIS study. (117) This provides detailed activity information includ- ing TIM and number of up- and down-river calls and passes for the 1995 base year segregated by HP bin for two principal inland river and two Great Lake ports. Although somewhat dated, the level of information contained is of high quality. Data may be updated to more current years by scaling, such as based on the calls or tonnage from other databases, although uncertainties would be associated with this scaling. In many other cases, too, the required level of information is not available to determine governing parameters and, instead, must be developed from surrogate data or translated from sim- ilar studies. It is likely that this approach will have inferior data quality and greater overall parameter uncertainty, even if the process is identical. For example, vessel counts by vessel type may be drawn from the USCG’s Merchant Vessels of the United States database as done in CARB’s harbor craft inventory. (132) However, this database includes no foreign vessels, may not be available for certain periods, suffers from much missing data, and has poor quality data for location of vessel activity. As discussed above, CARB was able to mitigate some of this uncertainty by focus- ing on larger domains and supplementing with locally specific information, however, this may not be available in all cases. Similar caveats apply to other databases, such as the U.S. Army Corps of Engineers’ comprehensive and current inventory for tug and towboats in the United States. (133) Although this database contains details on approximately 5,000 tow boats, the same caveats on operating domain may apply. In any case, it is likely that a vessel inventory may need to be estimated from a variety of databases for local inventories, which will exacer- bate uncertainty in the analysis. Uncertainty in the analysis can also arise from external databases if translation between vessel types is necessary. This process uncertainty can directly affect vessel population counts. Additional uncertainty may be caused by the need to distinguish Category 1 versus Category 2 engines for tug, tow, and push boats, as well as the lack of needed data in most databases. In the case of insufficiently detailed engine age distributions from direct surveys, a typical approach is to employ continu- ous age distribution profiles such as those commonly used in the NONROAD model for both main and auxiliary engines. (134) In many cases, reliance on median life, growth, and scrappage will be taken from other studies and age distribu- tions will be calculated for each vessel and engine type from the baseline year. Annual, linear growth in the population of harbor craft is commonly assumed, which may be taken from surrogate data, such as regional economic growth. Otherwise, default values for annual population growth, such as those used in the 2008 EPA RIA rulemaking, are employed. Process uncertainty is associated with the assumed shape of the age distribution. Parameter uncertainty in median age, growth, and other values assumed or translated from other studies is likely to be significantly larger than similar, directly observed parameters, although quantifying this uncertainty is infeasi- ble. Particularly, estimates derived following NONROAD guidance are known to produce unrealistic values for engine lifetime in marine applications. This can be mitigated by 87

forcing consistency between average model year predicted by the distribution and that drawn from surveys or translated from other studies. To minimize uncertainty, load, activity, emission, fuel cor- rection, and control factors also should be collected directly from the fleet being studied. This is not common. Rather, val- ues are commonly translated from other studies, such as the 2008 EPA rulemaking (119), the ports of Los Angeles (83, 99, 110) or Puget Sound (104) studies, the EPA best practices (9) document, the ARCADIS (116–117) studies, or EPA- or CARB- approved technology lists. As previously stated, parameter un- certainties are directly associated with these original values. Process uncertainties generally are introduced in the use of these parameters and in the translation of these parameters for a particular study. Quantification of these uncertainties is generally not possible. Emissions Calculation Methodology. Calculation of commercial H/C emissions in a local/project-scale inventory typically is done based on the parameters discussed in Section 3.6.3. As shown in Equation 14, emission estimates are gen- erated as the product of the following: Number of harbor craft vessels of a given type operating in the area (NH/C), Average number of main and auxiliary engines per vessel (NEng H/C main and NEng H/C aux), Load factor (LFH/C main and LFH/C aux), Average annual activity (ActivityH/C main and ActivityH/C aux), Average rated horsepower (HPH/C main and HPH/C aux), and Appropriate (pollutant, age, power, engine type, and, poten- tially, power density) emission factor (EFpollutant-H/C-main and EFpollutant-H/C-aux). In cases based on the ARCADIS methodology for inland river operations, emissions are calculated from a time-in- mode-based activity perspective instead of annual activity and average load factors. Other parameters are as shown in the list of variables for Equation 14. Transient adjustment and deterioration factors also may be considered and included in the emission factors parameteriza- tion for each engine. This approach parallels that for CHE dis- cussed in Section 3.7. As there, uncertainty in these emission estimates is due to both process and input parameters. Uncer- tainty may be included by the limited representation of the emission processes, especially the use of overall average param- eters. However, this total power approach is generally consid- Emissions N EF pollu t H C H C Pollu Main tan tant , , = i i N LF Activity HP EF Eng Main Main Main Main P , i i i( ) + ollu Aux Eng Aux Aux AuxN LF Activity HPtant, ,i i i i Aux( ) ⎧⎨⎩ ⎫⎬⎭ (Equation 14) ered to be adequate. More significant to the total uncertainty from the resulting emission calculations is the uncertainty in each of the input parameters, as discussed in the parameters sections, above. Summary of Strengths and Weaknesses. An analysis of local H/C methodology strengths and weaknesses is provided in Exhibit 3-48. 3.6.5 Evaluation of Parameters Exhibit 3-49 summarizes all parameters relevant for calcu- lating emissions from harbor craft. Each of these has been detailed under the discussion of the appropriate scale method in Sections 3.7.3 and 3.7.4. Only the primary parameters are discussed in detail here; the parameters that are used to derive these parameters may vary and are not listed here. The use of each is detailed in Section 3.6.4. Also as discussed above, no quantitative assessments are provided because the range of parameters is essentially unknown. Pedigree Matrix. Exhibit 3-50 shows the pedigree matrix for the five primary parameters determining emissions from harbor craft. Criteria to assign scores in the pedigree matrix are included in Appendix A. Note that both main and auxil- iary engine populations are ranked as “5” for Range of Vari- ation. This is because the variation in the variation of values between methods is wide, which is also considered a “5,” as documented in Appendix A. Population. Emissions are linearly related to engine populations. For commercial H/C, both main and auxiliary populations must be characterized, either directly or from vessel populations and average engines per vessel. Popula- tions may be characterized either by engine type, horsepower and age bin, or may only be listed by average values, depend- ing on the level of detail in the methodology. Thus, while accurate assessment of the engine inventory is critical, in many cases this parameter is uncertain, particularly for more streamlined approaches. For additional discussion, see Sec- tions 3.7.3 and 3.7.4. Load Factors. All methods require use of load factors for each engine and vessel type. This factor represents the aver- age load experienced by the engine over a period of use, typ- ically annually. This factor is ultimately derived from second order factors, such as the duty cycle. However, estimates for many specific types of equipment are not available and thus are aggregated from models, similar types of equipment, or similar studies. Because emissions are linearly related to the load factor, this can have a large impact on the uncertainty of the total emissions. More discussion is presented under Sec- tions 3.7.3 and 3.7.4. 88

Engine Power. Engine power represents the total rated power of each of the engine types installed on commercial H/C. Calculation of H/C emissions may require either disaggrega- tion into bins of specific type, age, and horsepower range or may just sum individual engines or even use overall averages, depending on the level of detail of the study. Because emissions are linearly related to total power, this can have a large impact on the uncertainty of total emissions. For additional discus- sion, see Sections 3.7.3 and 3.7.4. Activity. Commercial H/C engine activity determines the average operating hours of a given engine and vessel type in an annual period, and is typically described in hours per year. It may be broken down into bins of total power, power density, engine size, or left aggregated only at the H/C type level, depending on the methodology. Because emissions are linearly related to activity, uncertainty in this parameter can have a large impact on the uncertainty of total emissions. However, because activity also figures into the age distribu- tion of the NONROAD model, impact of its uncertainty may be somewhat mitigated if parameters are adjusted to ensure consistency in the age distribution. For additional discussion, see Sections 3.7.3 and 3.7.4. Emission Factors. All methods require the use of emission factors, although their source and quality level may vary. They may be defined for a given combination of engine power, density, size, and age, or vary only by equipment type and/ or age. As for other factors used to calculate emissions, the result is linearly proportional to this value, thus the impact of uncertainty in this parameter on that for the final calcu- lations can be significant. For additional discussion, see Sec- tions 3.7.3 and 3.7.4. 3.7 Cargo Handling Equipment Cargo handling equipment (CHE) is used to move or support movement of freight between modes at intermodal facilities, such as ports. Particularly at ports, a wide range of CHE is in use due to the diversity of cargo. Examples of types of CHE include • Cranes, • Forklifts, • Manlifts, • Sweepers, 89 Criteria Strengths Weaknesses Representation of physical processes Dominant physical processes typically included Structure of methodology is fluid; it must be ensured that adequate representation is included Sensitivity to input parameters High overall sensitivity to parameters, which are generally uncertain; some sensitivity mitigated by ensuring consistency between interim results (e.g., average age) Flexibility Extremely flexible—fluid method structure allows variation for available inputs and surrogates Ability to incorporate effects of emission reduction strategies Straightforward to include effects in calculations if use and effectiveness is known Representation of future emissions Future-year populations calculated may be projected if growth factors are known Consideration of alternative vehicle/fuel technologies Alternative technologies may be included by adjusting emission factors and populations Data quality No specific model on which to rely; information often comes from sources and surrogates of varying quality Spatial variability Tailored methodology allows application to range of domains, down to small/project scale Temporal variability Study may be designed for annual, daily, or seasonal inventories, depending on input data Review process Varies by application Endorsements Varies by application Exhibit 3-48. Summary of strengths and weaknesses—local H/C methodology.

• Container handlers, • Generators, • Specialized bulk handlers, • Nonroad vehicles, • Rail pushers, • Stackers, • Skid steer loaders, • Top handlers, • Tractors, • Excavators, • Welders, and • Yard tractors. Container terminals use CHE most extensively, while truck- to-rail equipment and dry bulk terminals also have high use of CHE. As examples, in 2007, the Port of Long Beach found that 81% of the CHE portwide was employed by its container ter- minals and that 8% of total NOx emissions were due to CHE (100); the Port of Houston found that 15% of its 2007 total NOx emissions came from CHE (96); New York/New Jersey found that 25% of their 2006 NOx emissions were due to CHE. (101) Thus, determining emissions from container terminal CHE is important in any landside emission inventory. Generally, CHE emissions from freight activities at ports are estimated using either the NONROAD or OFFROAD 90 Parameter Methods/Models Geographi c Scale Pedigree Matrix Qualitative Assessment Quantitative Assessment Main Engine Population EPA RIA method, CARB H/C method National, Regional Auxiliary Engine Population EPA RIA method, CARB H/C method National, Regional Harbor Craft Population Secondary: used to derive engine populations in local H/C method Local Number of Engines per Vessel Secondary: used to derive engine populations in local H/C method Local Load Factors All All Emission Factor All All Engine Power All All Activity All All Deterioration Factor Optional and secondary: used to derive in- use emission factors. All Growth Factor Optional and secondary: needed for future- year projections All Engine Age Optional and secondary: needed to determine average emission factors All Median Life Optional and secondary: needed to determine age distribution All Scrappage Intermediary: deriv ed from equipment age and median life All Duty Cycle Secondary: used to derive load and transient adjustment factors All Use of Retrofit Devices Optional, secondary. Used to calculate control factors on resulting emissions and/or correct modeled emission factors. All Fuel Type Secondary: used to determine emission factors All Key: indicates that a parameter is analyzed in the way denoted by the column: indicates that the parameter is not discussed in the way denoted by the column. Exhibit 3-49. Parameters.

emission models—or methods similar to those employed in these models—combined with parameters representing the CHE present, such as rated power, model year, type of fuel used, annual hours of operation, load data, use of retrofit de- vices or other emission mitigation measures, and fuel type. Uncertainty in each of these input parameters can lead to sig- nificant uncertainty in the final emissions estimated. Mod- els/methods and parameters are discussed separately in the following sections, however, the relationship between the two must be kept in mind. For example, the OFFROAD model generates emission inventories for a given type of equipment using an equipment- total power methodology as shown in Equation 15. Where f is the emission factor, P is the maximum rated equipment horsepower, L is the load factor, A is the annual activity, N is the equipment population, and In which f incorporates adjustments due to deterioration, transient use, and age-related effects. Uncertainty in the resulting CHE emissions can then be attributed to either the process uncertainty (that is, the degree to which Equation 15—or other OFFROAD algorithms— represents the actual emissions process) and parameter un- certainty (that is, the uncertainty in the individual elements Emissions f N P L A (Equation 15)=     of Equation 15). Evaluation of process uncertainty is presented in Sections 3.7.1 to 3.7.4. Evaluation of parameter uncertainty is presented in Section 3.7.5. In both cases, any known biases should be corrected. The effects of quantifiable uncertainty in input parameters on total calculated uncertainty may be made using standard error propagation methods, discussed in Sec- tion 4.3.4. If no covariance is assumed for the parameters in Equation 15, the net error in total emissions would be given by Equation 16, where σ2 indicates the variance. 3.7.1 Summary of Methods and Models Two general categories of methods are used to estimate CHE emissions. These are referred to as the best practice and streamlined methodologies. (10) Generally, these two differ only in the level of direct information collected and employed in the calculations. The best practice methodology dictates surveys of all equipment to establish correct parameters and then employs the NONROAD or OFFROAD models; the streamlined methodology allows for a greater degree of free- dom in collecting direct information by substituting surro- gate or otherwise derived information. It may then either use the models, or adjust the methodologies of the models them- selves for the available information. A special case, third methodology is used in CARB’s CHE inventory, which is essentially the best practice methodology without directly σ σ σ σ2 2 2 2 2 2 2emissions NPLA f fPLA N fNLA= ( ) + ( ) + ( ) P fNPA L fNPL A+ ( ) ( )2 2 2 2σ σ (Equation 16) 91 Parameter Im pa ct on R es ult Ac qu isi tio n M eth od In de pe nd en ce Re pr es en tat ive ne ss Te m po ra l C or re lat ion Ge og ra ph ic Co rre lat ion Te ch no lo gic al Co rre lat ion Ra ng e o f V ar iat ion Main Engine Population 4 Varies Varies Varies N/A Varies Varies 5 Auxiliary Engine Population 3 Varies Varies Varies N/A Varies Varies 5 Load Factor 4 2-3 1 2 N/A Varies 2 4 Emission Factor 4 2-3 1 2 N/A Varies 3 4 Engine Power 4 1 Varies Varies N/A Varies 1 1 Activity 4 Varies 1-2 3 N/A Varies 3 4 Exhibit 3-50. Pedigree matrix—harbor craft equipment parameters.

using the OFFROAD model. Exhibit 3-51 lists these three methods and two models. 3.7.2 Evaluation of National Methods and Models There are currently no national scale inventories of CHE emissions exclusively. The EPA prepares the NEI every three years, which includes emissions from nonroad sources, gener- ally broken out by SCC. Similarly, for the 2004 (Tier IV) Non- road Diesel Rulemaking, EPA prepared a baseline national emission inventory for nonroad engines with populations based on commercial inventories of equipment sales and calculations made via national-scale runs of the NONROAD model. (137) However, no details are given specifically to CHE, as results are reported only for “land-based nonroad engines.” Given the lack of a national-scale CHE emissions inventory, no uncertainties in such modeling are addressed here. 3.7.3 Evaluation of Regional Methods and Models California has conducted the only regional analysis of CHE emissions. To evaluate statewide emissions from CHE in sup- port of CARB’s Mobile Cargo Handling Equipment Regulation (adopted December 2005, effective December 2006), (138) CARB developed a statewide emission estimation methodol- ogy and corresponding emission inventory for CHE. (139) The regulation is in support of a statewide emission control strategy for CHE at ports and intermodal railyards. CARB CHE Methodology The CARB methodology, based on a survey conducted by CARB in early 2004 and the ports of Los Angeles (110) and Long Beach (84) emission inventories available at that time, es- timated population and activity data for CHE statewide by equipment type. The study developed emissions estimates at 16 ports and 14 intermodal railyards in the state for 8 equip- ment types. (CHE emissions also were estimated for the health risk studies for major rail yards in California.) Exhibit 3-52 shows these eight equipment types, the corresponding SCCs, and the SCC type. (Note that for most equipment types, multi- ple fuels are possible. The SCCs shown here are for off-highway diesel.) CARB (139) summarizes the methodology as follows: Briefly, the approach used to develop the cargo handling equipment emissions inventory estimates entailed determining 92 Method/Model Type Geographic Scale Pollutants Freight/Passenger NONROAD Emissions Model County or Larger* HC, NO X , CO, CO 2 , SO X , and PM; for exhaust and non- exhaust emissions Freight OFFROAD Emissions Model County, Air Basin, or Statewide (CA only) CO 2 and CH 4 ,** HC, CO, NO X , and PM; for exhaust, evaporative, and start. Freight Best Practices Methodology*** Method All All Freight Streamlined Methodology † Method All All Freight ARB Methodology †† Method County, Air Basin, or Statewide (CA only) All Freight Notes: * Model use is restricted to countywide definitions, but emission factors and methods may be extracted at scales down to equipment level. ** CO 2 and CH 4 emissions are produced by OFFROAD, however, these estimates are not currently used as the basis for CARB's official GHG inventory which is based on fuel usage information. (135) *** As documented in EPA’s draft best practice document. (10) This method includes locally specific information on fleets and use of models. † As documented in EPA’s draft best practice document. (10) This method includes use of surrogates for missing locally specific information. †† As documented by CARB. (136) The method used to derive the statewide CHE inventory is a slightly modified version of the best practices methodology, but without directly relying on the OFFROAD model, allowing a modified calculation of deterioration. Exhibit 3-51. List of cargo handling equipment methods and models.

the average annual emissions per engine for each equipment type and then multiplying that value by the total number of engines in that grouping. The majority of the inputs that went into developing the average annual emissions came from individual engine profiles developed using the information from a cargo handling equipment survey conducted by the [C]ARB in 2004 and cargo handling equipment population information pro- vided by the ports of Los Angeles and Long Beach. These inputs were then processed using a template based on the [C]ARB’s OFFROAD model to estimate annual emissions per engine for each equipment type. This data was then expanded to include the estimated statewide population of cargo handling equip- ment fitting a specific age and horsepower range. To estimate port specific emissions, the populations of cargo handling equipment were allocated based on the [C]ARB Survey and the port-specific data. Emission estimates were developed for the eight types of equipment described. . . . Estimates for NOx, HC, and PM were made. This methodology only differs from the best practice methodology by not relying directly on the OFFROAD (140) model and, instead, slightly modifying the calculation of deterioration. (141) Total uncertainty in this method is due to both process and parameter uncertainty. Although the process used here is generally believed to rely on the best information available at the time, three potentially significant sources of process un- certainty are (1) the appropriateness and representativeness of the OFFROAD model characterizations of CHE, (2) the groupings used to categorize CHE, and (3) the potential for bias in survey results. The OFFROAD model itself is discussed in the following subsection. The parameters used in this method are shown in Exhibit 3-53 and discussed in Section 3.7.5. Equipment Groupings The CARB CHE methodology states its choice to group equipment into eight categories (listed above) to make the analysis compatible with the OFFROAD model. (Note that this is different from the discussion in the Summary in that aerial lifts are grouped into general industrial equipment.) There is no particular bias or additional process uncertainty associated with groupings as long as the parameters within each group are appropriately weighted and applied, and re- sults are provided at the same resolution. That is, the result of total emissions calculations from more highly resolved cate- gories than those here should be consistent with the results of this study if values within each group are appropriately con- sidered. As in all similar cases, resolution must be balanced with accuracy; here the level of resolution was dictated by the use in the OFFROAD model. Specific discussion of uncertainty with parameters is given below. However, process uncertainty is associated with the as- signment of average parameters to bins. For example, in preparing emissions for cranes, the load factor used should be a number-weighted average of the load factors from each crane in the sample set. However, this value is not well known. The error in this parameter is the difference between the value used and the true average from all equipment in the bin. This uncertainty can be due to choice and assignment of values to equipment groupings. Potential Survey Bias. There is potential for bias in survey methods due to misreporting of equipment. This could be due 93 Aggregated CHE Type Estimated SCC SCC Type Cranes 2270002045 Construction Excavators 2270002036 Construction Forklifts 2270003020 Industrial Container handling equipment 2270003050 Industrial Other general industrial equipment 2270003040 Industrial Sweepers/scrubbers 2270003030 Industrial Tractors/loaders/backhoes 2270002063 Construction Yard trucks 2270003070 Industrial Exhibit 3-52. NONROAD cargo handling equipment types. Input Factor Source of Data (Gas and Diesel) Population (base year 2004) 2004 CARB Survey of Statewide Ports & Rail Yards; POLA & POLB data (2002) Useful life 2004 CARB Survey of Statewide Ports & Rail Yards Activity (h/yr) 2004 CARB Survey of Statewide Ports & Rail Yards Average horsepower 2004 CARB Survey of Statewide Ports & Rail Yards; POLA & POLB data (2002); Power Systems Research (1996) Load factor Power Systems Research (1996) Allocation factor 2004 CARB Survey of Statewide Ports & Rail Yards; POLA & POLB data (2002) Growth factor 2002 POLA Container TEUs data Survival rate Power Systems Research (1996) Source: CARB, Cargo Handling Equipment One Pager. Exhibit 3-53. Parameters from the CARB CHE inventory.

to a desire to underreport equipment or overstate control tech- nologies to underestimate emissions resulting from activities at a facility, omission of specific facilities due to a size cutoff, for example, or many other reasons. As noted earlier, any known bias should be removed from a sample set prior to analysis. Sampling was made by CARB for over 120 owner/operators statewide and results incorporated with detailed inventories from the 2001/2002 Port of Los Angeles (POLA) and Port of Long Beach (POLB) inventories. CARB corrected Los Ange- les and Long Beach inventories to a common year assuming a 3% annual growth factor. To adjust for limited information, CARB applied corrections to survey results for equipment populations where data were under-reported or not reported. Thus, no residual bias is likely for this study. Summary of Strengths and Weaknesses. CARB CHE methodology strengths and weaknesses are described in Exhibit 3-54. OFFROAD Model The OFFROAD2007 model is CARB’s current emissions and emission factor model designed to incorporate effects of proposed regulations, technology types, and seasonal condi- tions on emissions of nonroad equipment except ocean-going vessels, commercial harbor craft, locomotives, agricultural irrigation engines, and gas cans. The model consists of three main modules: population, activity, and emissions factor. Population is determined from a calendar year 2000 baseline equipment population, adjusted for growth and scrappage to produce model-year specific population distributions for years 1970 to 2040, allocated to geographic regions. Baseline emission factors are corrected for in-use and ambient condi- tions. Emission inventories are resolved to the county, air basin, or air district by fuel type, engine type, equipment cat- egory, and horsepower group. (140) Uncertainty in emission estimates by the OFFROAD model is driven by several aspects of the model, both in its structure and its input parameters. Calculation Method. The basic emissions calculations in the model are summarized by Equation 17. As noted above, this is essentially a total power approach to emissions calculations, rather than a TIM calculation or a fuel consumption approach. On average, a total power approach and a TIM approach should agree, if the more detailed activity profile and load in a given power setting agree with the average load factor employed by the power approach. (As noted by the lack of use of OFFROAD in creat- ing the California GHG inventory (10), a fuel consumption approach is not generally expected to agree.) However, uncer- tainty is inherent in this parameterization due to the physical representation of annual activity. Additional uncertainty due to best estimate parameters for average use conditions also exists in the model. This is dis- cussed in Section 3.7.5. However, OFFROAD model-specific discussion follows here. Emissions f N P L A (Equation 17)=     94 Criteria Strengths Weaknesses Representation of physical processes Dominant physical processes included Sensitivity to input parameters Method relies on well studied model inputs, and modifies when necessary Uncharacterized overall uncertainty Flexibility Tailored methodology Ability to incorporate effects of emission reduction strategies Information included from local authorities on reduction strategies implemented Representation of future emissions Method relies on well studied model inputs Consideration of alternative vehicle/fuel technologies Information included from local authorities on reduction strategies implemented Data quality Information included from local authorities; known biases corrected Spatial variability Applicable only to CA; emissions resolved only to county level Temporal variability Produces only annual inventories Review process Unclear from documentation Endorsements ARB Exhibit 3-54. Summary of strengths and weaknesses—CARB CHE methodology.

Population Parameters. Population in OFFROAD2007 is determined as the calendar year 2000 baseline equipment population adjusted for growth and scrappage. Growth factors are based on socioeconomic indicators such as housing units and manufacturing employment by category, by county, and with respect to year 1990 sales. Scrappage is fixed by equipment age and/or use and depends on engine type and horsepower group. For all CHE types, useful life is represented in years and is driven by the engine’s expected life (note that useful life for lawn and garden equipment and recreational vehicles is deter- mined by the equipment life). As for the NONROAD model, the equipment useful life is defined by the sample median; total lifetime is twice the useful life. Since emissions estimates are linearly proportional to pop- ulation, significant uncertainties may result from uncertainties in population, as discussed in Section 3.7.5. For OFFROAD, particularly, many of these uncertainties are driven by the pop- ulation projections to specific calendar years. These uncertain- ties may be mitigated by using observed counts of CHE instead, as in the CARB CHE methodology. Uncertainties also exist in the methods used to allocate populations to smaller domains, such as counties or air basins. Similarly, the shape of the age-to- median age curve could be inappropriate for a given equipment type. Neither of these uncertainties is generally quantifiable, but could lead to uncertainty in resulting inventories. Activity Parameters. Activity estimates in OFFROAD 2007 include annual average usage, load factors, brake-specific fuel consumption (BSFC), and number of starts per year. Values are included for each equipment category by fuel and engine types and horsepower group. Activity profiles also include seasonal and temporal variations by industrial category. Uncertainty exists in these parameters on the appropriate cate- gory binning and application across categories. Particularly, this is true for equipment that could have uses in multiple indus- tries or placed in a more general category. There also are issues with attributing usage fractions to freight activity only. For example, an average (no peak) usage pattern is exhibited by airport ground service and TRUs while construction and in- dustrial equipment is assigned primarily a weekday profile. However, much CHE is likely to be considered industrial equipment, although having a profile more similar to air GSE. Similarly, a skid steer loader used in a mining applica- tion is not likely to represent the activity profile of one used at a bulk cargo terminal. Emission Factors. Exhaust emission factors are engine- specific and vary by fuel type, horsepower group, and model year. Equipment-specific emission rates are based on the combination of engine emission factors and equipment duty cycles. Deterioration rates are generally based on on-road emissions data. Use of on-road deterioration rates, application of esti- mated duty cycles, and assumed zero-hour emission factors all may add uncertainty into model results. Nonroad CHE active at ports is likely to have a different duty cycle than sim- ilar nonroad equipment used in other industrial applications. Further, the use of on-road deterioration factors from a 1990 study (142) seems unlikely to represent a current fleet of non- road engines. Sources of zero-hour emission factors also are unclear. Each of these leads to an unquantified uncertainty in the model results. Note that the CARB CHE inventory did not rely on deterioration rates in the OFFROAD model. Summary of Strengths and Weaknesses. Strengths and weaknesses of the OFFROAD model are described in Exhibit 3-55. 3.7.4 Evaluation of Local/Project-Level Methods and Models Several studies of CHE emissions have been conducted at the local/project level. Principally, these include studies at ports throughout the United States, as detailed by Exhibit 3-56. Other studies of note include CHE active at intermodal rail- yards throughout California (143) and NEPA and CEQA studies that have characterized impacts from CHE. (144) Typically, these studies either rely on the best practices methodology directly or a variation of it, where calculations are made externally, but in a similar method to that of NONROAD or OFFROAD models. In some cases, particularly for the less detailed studies, a streamlined approach is used. These methods and models are discussed in the following subsections. Best Practice Methodology Best practices in developing an emissions inventory from CHE activity dictate that one should gather detailed informa- tion on all CHE present at the port in question (within the study boundaries) and make simulations using the NON- ROAD (outside of California) or OFFROAD (in California) model. This methodology is rooted in observations of all active CHE, including information on the following: • Equipment type, • Rated horsepower, • Model year, • Type of fuel used, including fuel sulfur level for diesel, • Annual hours of operation, • Equipment load data, and • Retrofit devices or other emission mitigation measures employed. Using the data collected on equipment numbers, types, horsepower, model year, hours of operation and load data, in- puts can be generated for the various NONROAD (OFFROAD) 95

96 Criteria Strengths Weaknesses Representation of physical processes Dominant physical processes included Sensitivity to input parameters Model relies on user-customizable inputs; sensitivity to these inputs varies Uncharacterized overall uncertainty Flexibility Moderately flexible; customization requires familiarity with model, or replication of calculations Ability to incorporate effects of emission reduction strategies May be included after model runs, using CARB- certified reductions; unclear how to include in simulations Representation of future emissions Projections available in the model Consideration of alternative vehicle/fuel technologies Unclear Data quality Generally structured from best available information Spatial variability Applicable only to CA; emissions resolved only to county level Temporal variability Produces only annual inventories Review process Unclear from documentation Endorsements ARB Exhibit 3-55. Summary of strengths and weaknesses—OFFROAD model. Port Year Published Data Year Pollutants * Contractor* Charleston ( 94 ) 2008 2005 NO x , TOG, CO, PM 10 , PM 2.5 , SO 2 Moffatt & Nichol Houston/Galveston ( 145 ) 2003 2001 NO x , VOC, CO Starcrest Houston ( 96 ) 2009 2007 NO x , VOC, CO, PM 10 , PM 2.5 , SO 2 , CO 2 Starcrest Los Angeles ( 110 ) 2005 2001 NO x , TOG, CO, PM 10 , PM 2.5 , SO 2 , DPM Starcrest Los Angeles ( 83 ) 2007 2005 NO x , TOG, CO, PM 10 , PM 2.5 , SO 2 , DPM Starcrest Los Angeles ( 99 ) 2008 2007 NO x , TOG, CO, PM 10 , PM 2.5 , SO 2 , DPM, CO 2 , CH 4 , N 2 O Starcrest Long Beach ( 14 6 ) 2004 2002 NO x , TOG, CO, PM 10 , PM 2.5 , SO 2 , DPM Starcrest Long Beach ( 13 0 ) 2007 2005 NO x , TOG, CO, PM 10 , PM 2.5 , SO 2 , DPM Starcrest Long Beach (100) 2009 2007 NO x , TOG, CO, PM 10 , PM 2.5 , SO 2 , DPM, CO 2 , CH 4 , N 2 O Starcrest New York/New Jersey ( 147 ) 2003 2002 NO x , VOC, CO, PM 10 , PM 2.5 , SO 2 Starcrest New York/New Jersey ( 148 ) 2005 2004 NO x , VOC, CO, PM 10 , PM 2.5 , SO 2 Starcrest New York/New Jersey ( 101 ) 2008 2006 NO x , VOC, CO, PM 10 , PM 2.5 , SO 2 , CO 2 , N 2 O, CH 4 Starcrest Oakland** ( 10 2 ) 2008 2005 NO x x, ROG, CO, PM, SO Environ Portland ( 103 ) 2005 2004 NO x x, HC, CO, SO , PM10 , PM 2.5 , CO 2 , 9 Air Toxics Bridgewater Consulting Puget Sound*** ( 10 4 ) 2007 2005 NO x , TOG, CO, PM 10 , PM 2.5 , SO 2 , DPM, CO 2 , CH 4 , N 2 O Starcrest San Diego ( 10 5 ) 2008 2006 NO x , TOG, CO, PM 10 , PM 2.5 , SO 2 , DPM Starcrest Notes: * Starcrest = Starcrest Consulting Group LLC, Environ = Environ International Corp. ** Definitive results are not included for cargo handling equipment in this inv entory. *** Includes the ports of Anacortes, Everett, Olympia, Port Angeles, Seattle, and Tacoma. Exhibit 3-56. Recently conducted port inventories containing CHE.

equipment types to determine emissions for CHE at the port. Use of retrofit or emission control devices must be treated outside the model. In these cases, emission factors may be de- termined using the NONROAD (OFFROAD) models for diesel equipment and then appropriate emission reduction per- centages applied. For retrofit devices such as diesel oxidation catalysts, diesel particulate filters, or other technologies, re- ductions specified in the following sources should be applied: EPA’s Verified Retrofit Technology website; (149) EPA’s Diesel Emission Quantifier; (150) or CARB’s list of currently veri- fied technologies. (151) Other sources may be relied upon, but may be considered more uncertain. Specific discussion of these models and their associated un- certainties is given in Section 3.7.3. Total uncertainty in this method is due to both process and parameter uncertainty. The process described here is generally structured to rely on the best information available for a given project. However, at the time, three potentially significant sources of process uncertainty are (1) the appropriateness and representativeness of the NONROAD (OFFROAD) model characterizations, (2) the groupings used to categorize CHE in analysis, and (3) the po- tential for bias or error in equipment inventory counts. The appropriateness and uncertainty of the models is dis- cussed in their respective sections. Equipment Groupings. The best practice methodology should minimize uncertainty associated with grouping CHE into categories by following the categories already provided by each model to make the analysis compatible with the model being employed. As in all similar cases, resolution must be bal- anced with accuracy; here the level of resolution will be dic- tated by the emissions model. There is no particular bias or additional process uncertainty associated with groupings as long as the parameters within each group are appropriately weighted and applied, and results are provided at the same level of resolution. That is, the result of total emissions calculations from more highly resolved cat- egories should be consistent with the total emissions from a coarser study if values within each group are appropriately considered. Specific discussion of uncertainty with input pa- rameters for the OFFROAD model is given above; discussion of the NONROAD model is below. However, process uncer- tainty is associated with the assignment of average parameters to bins. This uncertainty can be due to choice and assignment of values to equipment groupings. Because the bins are deter- mined by model designations and the CHE sample size is expected to be moderate to small for local/project-scale analy- ses, fleet characterization should not cause much uncertainty in the emissions analysis. Survey Error. There is potential for bias and error in sur- vey methods due to miscounting of equipment at the facility for a variety of reasons such as inappropriate boundary con- ditions, accidental omission of facilities or equipment, a desire to misrepresent activity, or incorrect survey methodology or results processing, for example. As noted earlier, any known bias should be removed from a sample set prior to analysis. If an appropriate survey is conducted following the best practice guidelines, these uncertainties should be small. Summary of Strengths and Weaknesses. Best practice methodology strengths and weaknesses are provided in Exhibit 3-57. Streamlined Methodology In cases where all necessary information is not available, resulting emissions from CHE activity may be approximated using a more streamlined approach than that of the best prac- tice approach, allowing emission estimations without directly observed equipment inventories and other parameters. Recently, a variety of detailed, local/project-scale CHE emission inventories have become available (see Exhibit 3-56). Unlike vessel emissions, there is no standardized methodol- ogy for developing estimates of port CHE emissions. Devel- oping a detailed CHE inventory may require extensive time and resources to survey tenants within the study boundaries re- garding their equipment. As an alternative to this level of effort, CHE emissions are sometimes estimated based on inputs developed for CHE inventories prepared by other sources. The essence of a streamlined CHE evaluation is to estimate any missing values in a local survey of equipment types, counts, and/or parameters from other published studies—commonly by applying ratios of known parameters, such as cargo tonnage throughput—to other detailed ports, followed by calculations using the NONROAD (OFFROAD) model or methodology. Uncertainty in this method can be significant, although general quantification of this uncertainty is difficult. Uncer- tainty is propagated into the analysis via the parameters input to the model, such as in the number inventory and properties of CHE. For example, one might use tonnage throughput ra- tios between two projects to determine the number of cranes at a second project from that at the first, but translate all other parameters for those cranes (e.g., power, load, activity) di- rectly from the values at the known port. The net uncertainty on resulting emissions could be tracked from the uncertainty resulting from the scaled input parameters, but the source of this uncertainty is the process used to translate the parameters. Specifically, it is due to the assumptions used and choices made. Bias can be minimized by selecting projects that are sim- ilar, both in scope (by using methodologies such as the princi- pal port-like analysis of the ARCADIS guidance, (116–117) for example) and in equipment age, activity, and other parame- ters. Regardless, uncertainty in this methodology is likely to be significant. 97

Summary of Strengths and Weaknesses. Strengths and weaknesses of the streamlined methodology are shown in Exhibit 3-58. NONROAD Model In April 2009, EPA released the current version of its nonroad, mobile emissions and emission factor model, NONROAD2008. The NONROAD model (152) predicts emissions for recreational land and marine vehicles as well as logging, agricultural, construction, industrial, and lawn and garden equipment. It includes over 80 basic and 260 spe- cific types of nonroad equipment stratified by horsepower rating, and considers equipment fueled by gasoline, diesel, compressed natural gas (CNG), and liquefied petroleum gas (LPG). NONROAD2008 also includes emission reductions 98 Criteria Strengths Weaknesses Representation of physical processes Dominant physical processes included Sensitivity to input parameters Method relies on detailed user inputs that may not be readily available, but should produce best results General, overall uncertainty unknown Flexibility Low Flexibility; requires detailed data collection Ability to incorporate effects of emission reduction strategies Best information available; effects may be included after model runs Representation of future emissions Projections available in the model and customizable to local information Consideration of alternative vehicle/fuel technologies May be achieved in methodology with suitable model runs Data quality Structured from best available information Spatial variability Applicable to any location, but data requirements likely limit to smaller spatial scales Temporal variability Most likely limited to annual inventories Review process Documented in EPA Methodology Guidance Endorsements EPA Exhibit 3-57. Summary of strengths and weaknesses—best practice methodology. Criteria Strengths Weaknesses Representation of physical processes Dominant physical processes included Sensitivity to input parameters Method relies on surrogates for missing inputs; results highly sensitive to quality of inputs Flexibility Highly flexible; customizable to data limitations Ability to incorporate effects of emission reduction strategies Highly customizable. Representation of future emissions Projections available in the model and customizable to local information Consideration of alternative vehicle/fuel technologies May be achieved in methodology with suitable model runs Data quality Structured from available information Spatial variability Applicable to any location. Data flexibility allows multiple spatial scales Temporal variability Designed for annual inventories, but scalable with appropriate information Review process Documented in EPA Methodology Guidance Endorsements Exhibit 3-58. Summary of strengths and weaknesses—streamlined methodology.

associated with the 2008 diesel recreational marine standards from the locomotive/marine and small spark-ignition (SI) and SI recreational marine final rules. The model is capable of estimating subcounty emissions with specific inputs. How- ever, the practical geographic domains vary between county and national extents. NONROAD is intended to eventually be replaced by a version of the MOVES model that will incor- porate nonroad modeling capability. EPA has indicated that it intends to include this capability in the release of the final version of MOVES2010 (focused on on-road vehicles and scheduled to be released by the end of 2009), however, that version would not be expected to yield substantially different results compared to NONROAD2008. (153) Exhibit 3-59 provides an example of equipment types and the corresponding SCC used in the NONROAD model to estimate emissions from CHE. The majority of CHE can be classified into one of these equipment types. (Note that for most equip- ment types, multiple fuels are possible. The SCCs shown here are for off-highway diesel.) Uncertainty in emission estimates by the NONROAD model is driven by several aspects of the model, both in its structure and its input parameters. Population Parameters. NONROAD maintains 1996, 1998, and 1999 baseline populations and determines future year populations by assigning an average growth rate to esti- mate emissions in subsequent years. (154) To produce emis- sions for a given calendar year, growth can be set to zero so that the emissions will not increase over time and the results will be accurate for the analysis year. For future forecasts, updated inputs for population and activity are required. This is due to the NONROAD method- ology, which calculates both population and age distribution in which the model uses a population growth rate to project equipment populations from a base year to an evaluation year. (155) For all base years (projected or current) the model fits the population numbers to a predetermined form as a func- tion of growth and scrappage. The number of units of each model year (or, equivalently, age) is determined for each age for 50 years back. Populations with ages greater than twice the median life are assumed scrapped. (156) Significant uncertainty in emissions may arise from this for- mulation of age distribution, due to the assignment of engine tiers to specific ages (and power bins). The driving parameters here are the growth rates, shape of the population distribution curve, and median lifetime of equipment. Any event that leads to a difference in real world age distribution from that assumed by the model will lead to different average emission factors, and thus different emissions. This bias could result from a mischaracterization of equipment median life or growth rates, both of which shift the overall curve of popula- tion versus age. The resulting uncertainty could bias the results in either direction, as an under- (over-) estimated median life 99 Aggregated CHE Type Estimated SCC SCC Type Compressor 2270006015 Commercial Crane 2270002045 Construction Forklift 2270003020 Industrial Manlift 2270003010 Industrial Sweeper 2270003030 Industrial Car loader 2270003050 Industrial Chassis rotator 2270003040 Industrial Empty container handler 2270003050 Industrial Generator 2270006005 Commercial Light tower 2270002027 Construction Specialized bulk handler 2270003050 Industrial Nonroad vehicle 2270002051 Construction Gantry Crane 2270002060 Industrial Rail pusher 2270003040 Industrial Reach stacker 2270003050 Industrial Roller 2270002015 Construction Side handler 2270003050 Industrial Skid steer loader 2270002072 Construction Top handler 2270003050 Industrial Tractor 2270002063 Construction Excavator 2270002036 Construction Welder 2270006025 Commercial Yard Tractor 2270003070 Industrial Exhibit 3-59. NONROAD cargo handling equipment types.

would lead to relatively more (fewer) newer engines and lower (higher) overall emissions. Similarly, a difference in the general shape of the real world age distribution from that parameter- ized in the model could lead to bias in either direction. This bias is more difficult to quantify without explicitly knowing the full age distribution of the sample population, but one exam- ple could be in cases where a type of equipment, engine, or technology is newly introduced and the older tail of the age distribution is not yet populated. In that case, the bias would be to higher emissions estimates by over-predicting the num- ber of older, more polluting engines. Uncertainty in the results may also be attributed to correct estimation of growth factors by equipment type. This param- eter may be set in model inputs, however, and is directly con- trollable by the user. Mischaracterization, however, could lead to significant bias in resulting populations. A similar sce- nario exists with equipment population. Default NONROAD equipment populations by geographic areas are determined from national-level estimates using economic factors, such as construction expenditures, farm acreage, and building square footage. (157) Reliance on these, rather than directly observed current year population counts, may lead to bias in resulting emissions. As emissions estimates are linearly proportional to popu- lation, significant uncertainties may result from uncertainties in population, as discussed in Section 3.7.5. In all cases, these uncertainties may be mitigated by using observed counts of CHE of each age for the given project. Usage Parameters. Engine median lifetime shapes the population distribution, as discussed previously. Annual activity and load determine the engine usage. All are dis- cussed together by EPA. (158) The parameters are related because NONROAD uses annual activity and load factor values to calculate emissions by engine type and uses activ- ity, load factor, and median life together to calculate fleet age distributions. See Equation 18. NONROAD assumes equipment lifetime equals engine life; engine life is determined based on the expected lifetime of highway diesel engines operated continuously at full load and adjusted to in-use values by dividing by the average load factor and annual activity. The NONROAD methodology assumes that nonroad engines are not rebuilt and that equipment never fails before the engine is worn out. These underlying assump- tions in the model may lead to significant resulting uncertainty in calculated emissions. However, engine rebuilds would lead the model to underestimate the equipment fleet, while engine Median Life years Median Life At Full Load h( ) = ours Activity hours year Load Factor ( ) ( ) Equa( tion 18) wear out would lead to overestimation of the fleet, thus the net bias due to in-use lifetime is expected to be small. (157) Uncer- tainty due to the representativeness of on-road engine lifetimes for off-road applications is unknown, although EPA did con- sider the data underlying these estimates in NONROAD devel- opment. (158) Load factors in NONROAD are based on seven operational duty cycles for agricultural tractors, backhoe-loaders, crawler tractors, skid-steer loaders, arc welders, wheel loaders, and excavators. Extrapolation of the seven duty cycles to every type of equipment was done by grouping the seven cycles into three categories—transient cycles with high loads (average of 0.59 with range 0.48–0.78), transient cycles with low loads (average of 0.21 with range 0.19–0.23), and steady-state cycles (average of 0.43)—and assigning all equipment to one of these three categories. (157) Uncertainty in the measured range of high- and low-cycle values is about 10% to 30%, although that for the steady cycles is uncharacterized, but could be as high as about 80%. Uncertainty due to assignment of measured emission factors to equipment groups is unknown. However, the assumed load factor is likely to be a significant source of uncertainty in NONROAD modeling, both in directly calcu- lating equipment emissions and in determining population age distribution. EPA claims that the effects of load and life- time in determining emissions and population are offsetting when computing total emissions, such that uncertainties in these parameters should have little effect on total emission uncertainty. (157) Activity values in NONROAD are based on surveys of equipment users by a private company using proprietary methods that estimate annual activity by equipment type but not by engine size, age, or model year. The uncertainty in average values and the actual sensitivity of activity to equip- ment size and age are all unknown. (157) Thus, the effects of these on overall emissions estimates is unknown. NONROAD estimates brake-specific fuel consumption (BSFC) as 0.408 lb/bhp-hr for engines smaller than (or equal to) 100 hp and 0.367 lb/bhp-hr for engines larger than 100 hp, based on measured fuel consumption values during engine certification. (157) Uncertainty in these estimates is unknown. Emission Factors. Emission factors in NONROAD (159) consist of zero-hour, steady-state emission factors, transient adjustment factors, and deterioration factors; fuel sulfur im- pacts on emission rates are included. Zero-hour, steady-state emission factors (EFs) are a function of model year and power, which defines the technology type. Transient adjustment fac- tors (TAFs) vary by equipment type. Deterioration factors (DFs) are functions of the technology type and engine age. See Equation 19. EF EF TAF DFInUse SteadyState=   (Equation 19) 100

In addition to exhaust emissions, crankcase HC emissions are computed as a simple 2% fraction of exhaust HC emis- sions for Tier 0 to III engines and are zero for Tier IV engines. Zero-hour, steady-state emission factors are drawn from a variety of sources, including NEVES (baseline engines > 50 hp), CARB’s OFFROAD values (Tier 0 engines less than 50 hp), emission rate tests (Tier 0 engines greater than 50 hp), EPA en- gine certification data (all Tier I engines and Tier II engines 300–600 hp), methods for the remaining Tier II and all Tier III engines (including compliance margins on emission standards, certification results, CARB engine test data, and engineering judgment). All Tier IV emission factors are based on compli- ance margins from emission standards. Since each element is chosen based on the best available information, all bias is as- sumed to be minimized. However, significant but unquantified uncertainty persists in most factors. Factors based on standards are likely to be less uncertain, since engines must be designed to meet specific thresholds, but a range of values is still likely. TAFs (159) are applied to the emission factors of all engines except Tier IV, where transient control is expected to be part of all engine design. TAFs in NONROAD were calculated by averaging tests for each engine, pollutant, and test cycle, and comparing these measured emission factors for off-road equip- ment duty cycles to the zero-hour steady state emission factors. Thus, in-use emission factors should have reduced uncer- tainty relative to using zero-hour steady state emission rates as emission factors. Deterioration factors (159) in NONROAD increase with en- gine age up to its median life, at which point it is held constant, under the assumption that increased deterioration is offset by maintenance. For compression ignition engines, deterioration is linear. In all cases, due to a lack of data for nonroad engines, the factors are based on data derived from highway engines. Uncertainty in these factors is unknown, particularly any addi- tional effects due to deterioration, mal maintenance, tamper- ing, or the effects from use of fuel with various sulfur levels. Calculation Method. The basic emissions calculations in the model are summarized by Equation 20. Where Popi,j is the population of engines of equipment type i within power bin j, Powerj is the average power (hp) of bin j, LFi is the load factor (fraction of available power) of equip- ment type i, Ai is the annual activity (hours/year) of equipment type i, and EF is the emission factor (g/hp-hr). Emissions Pop Power LF A EFi j j i i i jji= ∑∑ , , (     Equation 20) As noted, this is essentially a total power approach to emissions calculations, rather than a TIM calculation or a fuel consumption approach. On average, a total power ap- proach and a TIM approach should agree, if the more de- tailed activity profile and load in a given power setting agree with the average load factor employed by the power approach. However, uncertainty is inherent in the model due to the physical representation of annual activity. Addi- tional uncertainty due to best estimate parameters for aver- age use conditions also exists in the model. This is discussed in Section 3.7.5. Summary of Strengths and Weaknesses. NONROAD model strengths and weaknesses are shown in Exhibit 3-60. OFFROAD Model The OFFROAD model was discussed in Section 3.7.3. Since the model is appropriate at the project/local scale as well as the regional scale, the discussion is not repeated here. 3.7.5 Evaluation of Parameters Exhibit 3-61 summarizes all parameters relevant for calcu- lating emissions from CHE. Each of these has been detailed under the discussion of the appropriate model or method in Section 3.7.3 and 3.7.4. Only the primary parameters are discussed in detail here. That is, many of the parameters are used to derive the parameters in Equations 15 and 20, but not discussed here. The use of each is detailed above. Also as discussed previously, no quantitative assessments are provided, because the range of parameters is essentially unknown. Pedigree Matrix. Exhibit 3-62 shows the pedigree matrix for the five primary parameters determining emissions from CHE. Criteria to assign scores in the pedigree matrix are included in Appendix A. Note that population is ranked as a “5” for the range of values. This is actually because the varia- tion in the variation of values between methods is wide, which is also considered a “5” in Appendix A. Population. Emissions are linearly related to the equip- ment population, as shown by the previously provided equa- tions. Populations should be determined for each type of equipment, for each horsepower and age bin employed. Thus, although accurate assessment of the equipment inventory is critical, in many cases this parameter is uncertain, particularly for projected years or streamlined methods. More discussion has been presented under Sections 3.7.3 and 3.7.4. Note that population is shown as “varies” in Exhibit 3-62 because the range of values varies too widely to be ranked, depending on the methodology employed. 101

102 Criteria Strengths Weaknesses Representation of physical processes Dominant physical processes included Sensitivity to input parameters Sensitivity to some parameters mitigated by model structure (load, activity); overall sensitivity depends on the parameters Flexibility Moderately flexible; most inputs adjustable in input files Ability to incorporate effects of emission reduction strategies Unclear Representation of future emissions Future year populations calculated in the model Consideration of alternative vehicle/fuel technologies Unclear Data quality Model relies on best available information at time of development, with public review Spatial variability Applicable to domains from countywide to national Temporal variability Designed for annual inventories Review process Publicly reviewed Endorsements EPA Exhibit 3-60. Summary of strengths and weaknesses—NONROAD model. Parameter Methods/Models GeographicScale Pedigree Matrix Qualitative Assessment Quantitative Assessment Population All All Load Factor All All Emission Factor All All Engine Power All All Activity All All Deterioration Factor Optional and secondary: used to derive in-use emission factors All Growth Factor Optional and secondary: needed for future-year projections All Engine Age Optional and secondary: needed to determine average emission factors All Median Life Optional and secondary: needed to determine age distribution All Scrappage Intermediary: derived from equipment age and median life All Duty Cycle Secondary: used to derive load and transient adjustment factors. All Use of Retrofit Devices Optional, secondary: used to calculate control factors on resulting emissions and/or correct modeled emission factors All Fuel Type Secondary: used to determine emission factors All Key: indicates that a parameter is analyzed in the way denoted by the column: indicates that the parameter is not discussed in the way denoted by the column. Exhibit 3-61. Parameters.

Load Factors. Most models either require input of, or use default values for, load factors for a piece of equipment. This factor represents the average load experienced by an engine over a period of use, typically annually. This factor is ultimately derived from second-order factors, such as the duty cycle. However, estimates for many specific types of equipment are not available and are aggregated from average values of similar equipment types. Because emissions are lin- early related to load factor, this can have a large impact on the uncertainty of total emissions. More discussion has been pre- sented in Sections 3.7.3 and 3.7.4. Engine Power. Engine power represents the total rated power of CHE engines. Calculation of CHE emissions gener- ally requires disaggregation of equipment into bins of specific horsepower range since the power and age typically deter- mine the engine category for regulatory purposes. Databases of CHE within these bins are incorporated into the NONROAD and OFFROAD models, or should be collected through sur- veys. Because emissions are linearly related to total power, this can have a large impact on the uncertainty of total emis- sions. More discussion has been presented in Sections 3.7.3 and 3.7.4. Activity. Engine activity determines the average operating hours a given piece or group of equipment types have in an annual period, typically described in hours per year. It is not commonly broken down into power bins, but left at the CHE type level. Because emissions are linearly related to activity, uncertainty in this parameter can have a large impact on the uncertainty of total emissions. However, because activity also figures into the age distribution of the NONROAD model, im- pact of its uncertainty is somewhat mitigated. More discus- sion has been presented in Sections 3.7.3 and 3.7.4. Emission Factors. Most models either require input of, or use default values for, emission factors. Typically, these are defined for a given combination of engine power and age. As for other factors used to calculate emissions, the result is lin- early proportional to this value, thus the impact of uncer- tainty in this parameter on that for the final calculations can be significant. Emission factors are determined from a range of activities, including measurements, certification databases, and engineering judgment. More discussion has been pre- sented in Sections 3.7.3 and 3.7.4. 3.8 Air Transportation For aviation-related emissions the following two modeling approaches are reviewed and evaluated: • Version 1.5 of the FAA’s System for Assessing Aviation’s Global Emissions (SAGE) modeling system is the primary method for national or regional emission analysis in the United States. Other national or regional aircraft models are under development but these are focused on non-U.S. regions (e.g., AEM [EUROCONTROL], AERO2k [UK/ QinetiQ], and FAST [UK/MMU]). The emphasis of these models is on global-scale emission inventories with regional emphasis on european issues. (160) • Version 5.1 of the Emissions and Dispersion Modeling Sys- tem (EDMS), released September 19, 2008, was developed by FAA specifically to address the impacts of airport emission 103 Parameter Im pa ct on R es ult Ac qu isi tio n M eth od In de pe nd en ce Re pr es en tat ive ne ss Te m po ra l C or re lat ion Ge og ra ph ic Co rre lat ion Te ch no lo gic al Co rre lat ion Ra ng e o f V ar iat ion Population 4 Varies Varies Varies N/A Varies Varies 5 Load Factor 4 2-3 1 2 N/A Varies 2 4 Emission Factor 4 2-3 1 2 N/A Varies 3 4 Engine Power 4 1 Varies Varies N/A Varies 1 1 Activity 4 Varies 1-2 3 N/A Varies 3 4 Exhibit 3-62. Pedigree matrix—cargo handling equipment parameters.

sources, including ground-level sources and associated support activity. FAA requires the use of the model in per- forming air quality analyses for aviation sources. Recent improvements to the model include speciated air toxic emissions, CO2 emissions from aircraft, improved method- ology for PM emission estimates, and the addition of 63 new engines and 40 aircraft. FAA also is sponsoring ongoing research through the Partnership for Air Transportation Noise and Emission Reduction, to understand and evalu- ate the potential role of aviation emissions in local and re- gional air quality. The main objective of the project is to quantify the potential incremental contribution of aviation emissions to local and regional air quality though their chemical interaction with the background air. A summary of air freight methods and models is shown in Exhibit 3-63. 3.8.1 Evaluation of National and Regional Models The FAA has been working on the development of a na- tional-to-global version of SAGE since 2001. The most current version has been used to develop annual emission inventories for commercial (civil) aircraft fuel consumption for CO, NOx, SO2, HC, H2O and CO2. (37) Because the model operates at the level of individual flight by airport it can potentially be applied to a limited regional analysis as well as at the national level. This version of SAGE dynamically models aircraft per- formance, fuel consumption, and emissions, and includes such factors as capacity and delay at airports. The model does not have the current capability to separate freight-only travel from freight and passenger operations nor does the model include military air cargo activity. The model is driven primarily by a set of databases that are used to develop the emission inventory. The key databases include the following: • International Civil Aviation Organization (ICAO) emis- sions databank with information on certification emissions and fuel flow rates for a wide variety of jet engines that have entered service; • Base of Aircraft Data (BADA), which is a collection of air- craft performance and operation parameters, includes co- efficients in the data that allow calculation of lift and drag forces; • Information on airport location and altitude; • Official Airline Guide (OAG) database contains information on trip origination, trip length, type of aircraft, destination, and aircraft type for all commercial activity; • Enhanced Traffic Management System (ETMS), a database on the electronic recording of flight position and flight plan used for air traffic management, captures every flight within coverage of FAA radars; • ICAO’s Forecasting and Economics Sub-Group Forecast contains forecasts of the number of aircraft, number of air- craft seats, number of flights, capacity, and average seating numbers per aircraft by region; • FAA’s Terminal Area Forecast provides information on passenger boarding and aircraft operations for each U.S. airport; • USDOT’s Bureau of Transportation Statistics (BTS) data- base provides on-time performance for the 10 largest U.S. air carriers; • FAA’s annual runway capacity “benchmarking” report of U.S. airports provides a basis for delay data; and • Aircraft retirement parameters (data that categorize the survivor curves [i.e., polynomial equations] for the aircraft fleet population that survived the retirement process) by aircraft category type and age. Model Overview The fundamental modeling unit in SAGE is a single flight. All data, including those related to flight schedules, trajecto- ries, performance, and emissions, are represented at a level of detail sufficient to support the modeling of a single flight. This allows high resolution modeling of emission inventories. Each flight is modeled from gate to gate. Although a single flight in SAGE is the modeling unit, the simulation is con- 104 Method/Model Type GeographicScale Pollutants Freight/Passenger FAA SAGE (version 1.5) Model Global, National to Regional CO, hydrocarbons NOx, CO2, H2O,* and SOx Commercial freight and passenger (no military) EDMS (version 5.1) Model Local Criteria pollutants, NMHC, CO2 and 44 air toxics Freight and passenger * Water (H2O) is included here because when emitted at cruising altitude into the lower stratosphere/upper troposphere, it acts as a greenhouse gas via contrail development. Exhibit 3-63. Air freight methods and models.

ducted at a more detailed level (i.e., each individual segment of flight—referred to as a flight chord—is calculated by the model). Typical flights in SAGE are represented by 40 to 50 chords, depending on the stage length and availability of detailed radar trajectory data. The flight chords allow the abil- ity to express outputs in a variety of different formats (e.g., gridded and per flight mode) and allow for dynamic aircraft performance modeling in SAGE. Such modeling provides an opportunity for improvements in accuracy relative to those based on aggregated TIMs or simplified performance lookup tables. To accomplish the detailed flight-by-flight modeling, SAGE includes information on a variety of aircraft fleet, op- erations, and performance data, as well as the modules to process the information and perform computations. The model reports information both on a vertical and horizontal distribution. The SAGE model was last updated in September 2005. The model’s primary purpose is to provide FAA and, indirectly, the international aviation community, with a tool to evaluate the effects of various policies, technology, and operational scenarios on aircraft fuel use and emissions. The current ver- sion of the model is not considered a standalone model; it is used primarily as an FAA research tool. The information presented on SAGE is principally based on the analysis, review, and discussion of the SAGE Version 1.5 Technical Manual, (161) SAGE: Validation Assessment, Model Assumptions, and Uncertainties, (162) “System for assessing Aviation’s Global Emissions (SAGE), Part 1: Model Descrip- tion and Inventory Results,” (163) and “System for Assessing Aviation’s Global Emissions (SAGE), Part 2: Uncertainty Assessment.” (164) Summary of Strengths and Weaknesses. An analysis of the strengths and weaknesses of FAA’s SAGE is included in Exhibit 3-64. Analysis of Process Uncertainty. To estimate emissions, SAGE uses Boeing Fuel Flow Method 2 (BFFM2), which is a method developed based on engine performance and emis- sions data obtained from ground-level engine tests. BFFM2 uses ICAO certification fuel flow and emissions data taken at 7%, 30%, 85%, and 100% rated outputs at sea level pressure as the basis for correcting emissions indices for installation ef- fects, ambient conditions, and flight speed. At the four certifi- cation points, BFFM2 provides an agreement between meas- ured and calculated emissions indices that is within ±10% for most jet engine types. Increased uncertainty occurs in estimat- ing idle emissions below 7%, particularly for HC, and these errors may be large and tend to be an underprediction. (165) The interpolation method (curve fitting), used between certification emission portions, is another source of uncer- tainty. A comparison undertaken by ICAO found agreement between direct measurement and fuel flow correlations using curve-fitted ICAO data to within a standard deviation of 6% and a maximum error of 13%. (166) Other sources of uncer- tainty in most emissions data, including certification data, consist of the variability in emissions inherent among engines in the fleet and aging of the engine. (11, 165, 167) Overall the emission indices for NOx have been estimated to have a standard deviation for an approximate normal dis- tribution of ±24% based on the aggregation of 16% uncer- tainty incorporated in the engine certification process, adding (using sum of squares) the uncertainty in curve fitting and BFFM2 (6% and 10%, respectively) and then accounting for the bias error due to aircraft engine degradation (4%). (11) The uncertainties implied in the certification process for HC and CO emission indices are 54% and ±23%, respectively, and also have been aggregated with those of curve fitting and BFFM2 with a resulting estimated standard deviation in the uncertainty of ±55% HC and ±26% for CO. However, these estimated un- certainties need to be confirmed by comparing SAGE emissions results to measured emissions over a wide range of emission points, power settings, and engine types that are not readily available at present. Below the mixing height (i.e., around 3,000 ft) where air- craft emissions have the greatest impact on local air quality, the landing and takeoff (LTO) procedures are an important source of uncertainty. LTO procedures mainly consist of en- gine throttle setting, rate of climb/descent, and flight speed. The analysis of a major carrier’s computer flight data recorder results showed that the throttle setting and resulting change in the emission index for NOx, HC, and CO were the most important parameters, accounting for 30% to as much as 70% of the total variance of the emissions. Other LTO proce- dures such as the rate of climb, descent plus flight speed, and aerodynamic drag explained most of the remaining variance in the emissions estimates below 3,000 ft. Finally, although individual uncertainties for specific air- craft may be large, it is likely that the current version of SAGE can distinguish between the emissions associated with the typical policy options that are directed across all aircraft and engine types. However, it would be important to analyze the uncertainties and account for them when interpreting any type of policy scenario analyses. 3.8.2 Evaluation of Local/Project-Level Models Starting in the mid-1980s, the FAA developed the Emissions and Dispersion Modeling System (EDMS) to assess local air quality impacts in the near airport vicinity for a single airport. The current version of EDMS incorporates EPA-approved emissions inventory methodologies and dispersion models to ensure that analyses performed are consistent with EPA guidelines. EDMS is used primarily in complying with local 105

106 Criteria Strengths Weaknesses Representation of physical processe s The model details actual flight path and trajectories with flight-by-flight modeling. SAGE includes specific aircraft fleet information and operations, combined with the use of engine performance data. Such detailed modeling represents an improvement in accuracy over other methods based on aggregated TIMs or simplified performance lookup tables . The model only includes emissions from commercial aircraft—no general aviation or military flights. Studies have suggested that military aviation in the U.S. is responsible for up to 15% of aircraft emissions . Requires an extensive database to make the emission calculations. Relies on emissions indices as a function of fuel consumption to estimate emission s. Model sensitivity to input parameters No formal sensitivity analysis has been conducted with the model but the model is highly dependent upon the emission indices, which are a function of the fuel burn which, in turn, are sensitive to aerodynamic and engine performance, aircraft take-off weight, and flight speed. Individual flights not using winds aloft information also use standard day ambient temperature. The fuel burn rate is based on engine performance and emissions data obtained from ground-level, full-scale engine tests. These are the ICAO certification fuel flow and emissions data taken at 7% , 30%, 85%, and 10 0% rated outputs at sea level with corrections for installation effects, ambient conditions, and flight speed. NO x emission indices are the best developed, having been created for a broad range of engine types, and power settings and measured fuel flow rates. Lack of a model-s pecific sensitivity analysis makes it difficult to quantify the model’s sensitivity. Emission indices are best developed for NO x , followed by HC, CO, and water vapor. Ability to incorporate effects of emission reduction strategies This design of SAGE allows a user to quantify the effects of communication, navigation, and surveillance/air traffic management (CNS/ATM) initiatives, determine the benefits of reduced vertical separation minimum (RVSM), investigate trajectory optimizations, and compute potential emissions benefits from the use of a continuous descent approach (CDA). Requires detailed knowledge of the model databases and can only be performed by FAA or their supporting organizations . Representation of future emissions The forecasting module uses flight forecasts from the FAA’s Terminal Area Forecast (TAF) for U.S. flights and ICAO’s Forecasting and Economics Sub Group (FESG) for the rest of the world. The TAF method involves cr eating a week’s worth of official airline guide scheduled flights to represent the growth in demand for a future year. The week was a balance between accuracy and computational efficiency. Also inc luded are the effects of aircraft retirements and replacements. The result is a future schedule of flights reflecting the effects of fleet growth and retirements with replacements The forecast is created from airport-based projections. Model can be updated through database for new aircraft and engines but it has been almost 4 years since last public update. Data quality Use of radar-based flight trajectories and speed are highly accurate in most cases. Although specific flights may be in error by up to 40 %, average fleet emissions showed less than 10% error. The most important default assumptions in determining the modeled emission rates are the use of the International Standard Atmosphere temperature, not correcting for winds aloft, use of Base of Aircraft Data (BADA) aerodynamic performance, and aircraft take-off weight and flight speed. Other important concerns are the OAG-based flight trajectories and the use of the Boeing Fuel Flow method to estimate the emission indic es . Separation of air cargo from passenger travel In general, commercial aircraft fly the same airframe design and engine technology whether the intended load is passengers or air freight. For airports that can separate aircraft operations performed exclusively for air freight transport, the current version of SAGE can be used to assess air freight emissions. The model does not have a method for separating air ca rgo from passenger transport activity, nor does the approach separately identify those aircraft used exclusively for air cargo transport. Spatial variability The model’s accuracy for spatial representativeness is directly tied to the quality of the TAF and FESG databases. It is anticipated that the accuracy of results from the model is dependant upon the depth of activity and load information in these databases. Vertical distributions are anticipated to be more accurate because of the use of actual flight path and trajectories with flight-by-flight modeling. Most aircraft activity is in the U.S. and Europe, although substantial growth is projected over Asia. Greenhouse gas emissions Unlike criteria pollutants, the locations of emissions are irrelevant, but all of the emissions from aircraft need to be determined. The Intergov ernmental Panel on Climate Change (IPCC) protocol recommends that each flight’s emissions be attributed to the departure airport. Also, the IPCC prefers that the method be based on aircraft performance and operating data rather than fuel sales. This is the approach employed in SAGE for CO 2 emissions using extensive information on aircraft fleet, flight schedules, trajectories, and aircraft performance. Based on results from applying SAGE, FAA is expected to begin releasing these results publicly in the near future for airport operators to use in determining GHG emissions. This information would be reported as fuel burned and CO 2 emissions for ground level (taxi/idle mode), above ground to below 3,000 ft (takeoff, climb-out, and approach modes), and abov e 3,000 ft (reflecting cruise). The fuel consumption data must be used with generic emissions factors for jet fuel to calculate emissions of CH4 and N2O. The FAA SAGE dataset will not include future emission projections. Emission rates are not separated into cargo and passenger modes. Review process Review process has been limited to peer review publications of model results and meeting presentation on findings and methodology . Only user guide documentation available, code and databases are not publicly available. This limits full and open comparison. The large database sizes have limited the model’s distribution. The model has been made available to support various International Civil Aviation Organization/Committee on Aviation Environmental Protection (ICAO/CAEP) activities but with FAA running the model. Endorsements FAA recommends that the model may be used in developing policy, technology, and operational scenarios on aircraft fuel use and emissions. Neither ICAO nor EPA have made statements about the model. FAA continues to support the model, but it has been almost 4 years since the last public update to the model. Model comparison/evaluation studies Comparison with research-oriented methodologies such as the NASA/Boeing scheduled inventory and SAGE totals show a 30% difference, which may partly be explained through differences in trajectory modeling (Great Circle used by NASA/Boeing versus track distributions used in SAGE) and the inclusion of unscheduled flights in SAGE (unaccounted for in NASA/Boeing studies). However, estimates for any given flight may be off by ±50% or more. An assessment of the aircraft performance module showed that when comparing point-by-point fuel flows from SAGE against data from a major U.S. airline and NASA, the overall agreement was good with mean errors of 6.95 % and 0.24%, respectively. Similarly, system-level (aggregated flight-level) comparisons of fuel burn against data from one major U.S. airline and two major Japanese airlines also showed good agreement with mean errors of 2.62% and 0.42%, respectiv el y Lack of a public av ailability of the model has hampered external review by outside agencies and the international community . Exhibit 3-64. Analysis of strengths and weaknesses—FAA SAGE.

air quality requirements (e.g., NEPA documents, EIS/EIR, air toxic risk assessments, and general conformity). The model uses a comprehensive database of aircraft engines and emis- sion factors in different modes, ground support equipment, auxiliary power units, and vehicular and stationary source emission factor data. The model includes emissions for CO2, CO, THC, NMHC, NOx, SOx, PM2.5, PM10, and 395 speciated hydrocarbons for use in air toxic assessments. The CO2 emis- sions are calculated only for aircraft. Aircraft PM emissions are only available for aircraft with ICAO-certified engines. The model offers two approaches for estimating emissions: one based on an ICAO EPA TIM approach and another that uses an aircraft performance module that dynamically models the flight of an individual aircraft based on its flight profile. The air quality dispersion analysis uses EPA’s AERMOD dispersion modeling system. Concentrations of the pollutants are output for comparison with the NAAQS. The model does allow the specification of user-created aircraft modes when operated with the aircraft performance module in which specifications can be made for cargo-only aircraft. However, the model does not determine the freight fraction for aircraft that move both cargo and passengers. The model has no future forecasting capabilities. (168) The model is driven primarily by a set of databases that are used to develop the emission inventory. The key databases include the following: • ICAO emissions databank containing information on cer- tification emissions, default TIM, and fuel flow rates for a wide variety of jet engines that have entered service. • Aircraft performance-based database using system air- craft-engine (SAE AIR 1845) TIM as an alternative to the ICAO emissions databank. These data were developed for, and adapted from, the Integrated Noise Model (INM). EUROCONTROL Base of Aircraft Data [BADA] is used in the aircraft performance modeling. (169) • Various ground support equipment emission factors for use in EDMS are based on EPA’s NONROAD2005 model using the fuel type, brake horsepower, and load factor variables. In addition, a deterioration factor is applied based on the age of the engine. A national default fleet average age may be used for a particular equipment type or a facility-specific age of an individual piece of equipment may be specified. • Motor vehicle activity can be incorporated into the model using information on the number of vehicle trips and av- erage speed while traveling on roadways with emission fac- tors based on EPA’s MOBILE 6.2. National default age dis- tribution can use a base year assignment up to 2025. Model Overview The EDMS is both an emissions inventory development model and an air dispersion model. It is used to assess air quality impacts in the near airport vicinity by developing an emission inventory from the emissions from aircraft, auxil- iary power units, ground support equipment, and stationary sources. Emissions are developed based on a combination of EPA models and best-available models from other sources such as an aircraft performance module for calculating air- craft emissions, on-road (MOBILE6.2) and off-road vehicles (NONROAD2005). EDMS has an extensive database with in- formation from engine manufacturers, FAA, and EPA on the aircraft flight performance arrivals (approach and taxi-idle) and departures (takeoff and climb out). This information is indexed by aircraft types, which is cross-referenced with nominal takeoff weight and glide slope angle. The dispersion- modeling module uses EPA’s AERMOD (version 07026) and its supporting weather and terrain processors to determine air concentrations. EDMS offers the flexibility of allowing the user to perform an emissions inventory only or to perform air dispersion modeling as well. Results are reported as ground- level concentrations. (170) EDMS was last updated in September 2008 (version 5.1). EDMS is used primarily to assess the local air quality impact in the vicinity of individual airports as part of an environ- mental impact assessment under NEPA or general conform- ity requirements. Recent integration efforts are underway to integrate EDMS as part of an Aviation Environmental Design Tool (AEDT) that will result in the ability to model noise and emissions interdependencies in the same model- ing platform. A research version of EDMS, funded by FAA and NASA, is underway to develop a 3D representation of aviation emissions at the regional scale, coupled with a re- gional-scale air quality model (i.e., community multiscale air quality [CMAQ] model) to assess air quality impacts for regional-scale air pollutants of fine particulate matter, ozone, and air toxics. The primary information on EDMS comes from the analy- sis, review, and discussion of the EDMS 5.1 User’s Manual (171) and the EDMS 4.2 Technical Manual. (172) Summary of Strengths and Weaknesses. An analysis of the strengths and weaknesses of EDMS is provided in Exhibit 3-65. Analysis of Process Uncertainty. Ideally, a comprehen- sive validation of the EDMS model would be conducted using field data to scientifically determine the accuracy of the model and ensure the model results are defensible. This effort would include a multi-year measurement plan and analysis follow- ing EPA protocols so that EDMS could be fully evaluated using several detailed steps as follows: 1. Identification and collection of previously collected field data that potentially could be used in validation, 2. Assessment of the quality/applicability of collected data, 107

3. Collection of additional field-measured data, 4. Rigorous exercising of EDMS for comparison with col- lected data, and 5. Comparison of EDMS performance with that of other similar models (e.g., the ADMS-Urban or Eurocontrol’s ALAQ). The results of this effort would be used to identify EDMS limitations, correct major deficiencies, as well as determine the overall accuracy and sensitivity of EDMS. However, cur- rent FAA priorities are focused on the development and eval- uation of ADET, the replacement model for EDMS. FAA’s priority is that this model has a complete and informed process analysis so that a comprehensive understanding of the model’s uncertainty, inputs, and assumptions is devel- oped. As part of the development of AEDT, FAA plans to conduct a formal parametric sensitivity and uncertainty analysis. This analysis would be completed on individual components of AEDT as well as on the whole tool. The analy- sis would consist of quantifying uncertainties of AEDT and rank ordering of the most important assumptions and limi- tations. Gaps in functionality potentially would be identified that significantly impact AEDT, leading to the identification of priority areas for further research and development. In addition, the evaluation would examine the modeling factors that contribute to model output uncertainty. 3.8.3 Freight Disaggregation EDMS does not distinguish between freight and passenger movements. To solve this problem, the study team developed an approach to allocate each airport’s total commercial air- craft emissions to the freight and non-freight sectors. (67) For each airport of interest, the team used the BTS Air Carrier Statistics Database to obtain aircraft departure records with the following data fields: 108 Criteria Strengths Weaknesses Representation of physical processes The model details aircraft activity bas ed on dynamic aircraft performance- based modeling. The model includes operational profiles in 15-minute bins of aircraft delay and sequencing. Model only examines primary emissions; does not allocate aircraft emissions in a full 3D environment. Model sensitivity to input parameters Aircraft emissions in EDMS are dependent upon two main parameters: the emission factors obtained from the aircraft/engine combination and the vertical flight profile. Within the flight profile the least well-established parameter is the TIM. FAA has announc ed plans to conduct a robust sensitiv ity analysis on EDMS, but has not published results from any studies to date. This is because the EDMS system has undergone a tremendous number of changes over the past 6 years with the release of 2 major changes (EDMS 4 and EDMS 5) and 6 extensive model changes. Ability to incorporate effects of emission reduction strategie s By providing user-specified aircraft emission factors and performance data, emission reductions for aircraft engines can be assessed. No capabilities for testing operational changes such as aircraft approach and descent changes . Representation of future emissions The model has the capability for assessing future emission changes for on- road and nonroad vehicles. The Voluntary Airport Low Emissions (VALE) Program can be evaluated for selected airports. Lacks a capability for assessing aircraft emission reductions. Latest version of NONROAD model is now NONROAD2008. Similarly, MOVES is to be released in 2009 and current version of EDMS incorporates MOBILE6.2. Data quality Comparison tests with other local-scale airport models suggest overall emission strength is reasonable, but that a large variability exists in aircraft grouping, TIMs, and emission factors. Large variability exists for the TIM between aircraft grouping (e.g., business jet vs. small jet) and TIM default varies widely from airport to airport. Als o, emission factors vary widely from one grouping to another. Separation of air cargo from passenger travel The model allows the specification of user-created aircraft modes when operated with the aircraft performance module in whic h specifications can be made for usage as cargo-only aircraft. The model does not determine freight fraction for aircraft that move both cargo and passengers. Spatial variability User specifies horizontal locations of emissions. Vertical profiles in 2D are specified for the aircraft approach and takeoff grids from which the emissions are released. Does not represent aircraft emissions on a full 3D grid. Greenhouse gas emissions Includes CO 2 emissions for aircraft only. Uses same approach as criteria pollutants. Emissions reporting can be reported by aircraft mode. No emission estimates for N 2 O or CH 4 . Emission rates are not separated into air cargo and passenger modes. Review process Validation and uncertainty studies have not been made publicly available to date. Although, FAA has announced plans to actively pursue a better understanding of the uncertainties of the modeling components for the new Aviation Environmental Design Tool (AEDT)/EDMS with plans for formal parametric sensitivity and uncertainty analyses. No external peer review has been performed although DOT’s Volpe Center reports that a program is underway for modeling validation and uncertainty assessment. FAA continues efforts to improve model but moving towards development of AEDT. Source code not publicly available. Technical User Guide was last releas ed with version 4.2 of EDMS now at version 5.1. Executable of model available but requires licensing with EUROCONTROL Base Aircraft Data. Endorsements FAA requires that EDMS be used in air quality impacts of airport emission sources for purpos es of complying with NEPA and general conformity. In addition, FAA and EPA have co-written a recommended best practice document with an accompanying tec hnical support document for issuanc e by each agency scheduled for the summer of 2009. Onl y User’s Manua l and Technical Manual available. To date, little documentation and/or model verification testing has been reported or made publicly available. FAA has plans to improve this as they move toward release of AEDT in December 2011. Model comparison/evaluation studies Intercomparison of three different loc al airport emission inventory tools— ALAQ, LASPORT, and EDMS—showed that all models have similar global results for aircraft, but that aircraft emissions were dependent mainly on the engine emission factors and climb-out profiles. ( 16 0 ) Limited comparison or model validation has occurred by outside agencies or from the international community. Lack of a performance evaluation dataset has hampered this type of evaluation. Exhibit 3-65. Analysis of strengths and weaknesses—FAA EDMS.

• Carrier, • Origin, • Number of departures performed, • Tonnage of freight, tonnage of mail, • Number of passengers, • Seating capacity, and • Payload. Using this database, it is possible to estimate the number of aircraft departures attributable to freight transport and the number attributable to passenger transport, then to use the freight fraction to split each airport’s aircraft emission total into a freight and non-freight component. This allocation process can be summarized as follows: • Air cargo aircraft – Aircraft departures that do not have any passengers or seats but have payload capacity of 18,000 lbs or larger are assumed to be air cargo commercial aircraft. This definition is consistent with that used by FAA for cate- gorizing the aircraft type when reporting emissions. – Aircraft departures for which no freight tonnage and no passengers are reported are assumed to be non-freight (passenger) movements if departure was reported as having a seating capacity greater than zero, otherwise it was assumed to be a freight movement. • Passenger aircraft – For flights with passengers, it is assumed the flight is a commercial flight if the plane has 60 or more seats. This definition is consistent with that used by FAA for report- ing aircraft emissions. – For those aircraft that are commercial and that carry both freight and passengers, the number of departures is allo- cated to freight activity and non-freight activity based on weight fractions. The freight weight fraction is the com- bined weight of the freight plus mail divided by the sum of all weight—passengers, mail, and freight (average pas- senger weight of 240 lbs was used based on a March 21, 2003 FAA-sponsored weight survey of more than 6,000 passengers that included an average adult passenger weight of 196 lbs, 16 lbs of carry-on items, and 29 lbs of checked baggage). Similarly, the passenger weight frac- tion is the weight of all passengers divided by the sum of all weight. These fractions are then multiplied by the number of departures for each record. – The weighted freight and non-freight departures are summed for all flights departing from the airport in 2002, using the ratio of freight departures to total departures to apportion the airport’s emission total to a freight com- ponent. This approach assumes that all departures have a corresponding arrival, so the freight departure frac- tion is equivalent to the freight LTO fraction. 3.9 Air Quality Air quality refers to the level of contaminants in ambient air. It is either determined through measurement techniques and/or estimated through applications of models—numerical techniques to predict ambient levels of pollutants from atmos- pheric releases. Most air quality impacts from goods move- ment activities are assessed either by modeling studies alone or coupled with measurements. This section discusses air qual- ity modeling assessments and associated uncertainties. 3.9.1 Summary of Methods and Models To characterize ambient concentrations, all air quality mod- els require some level of input information for meteorology (e.g., winds, stability, atmospheric structure) and source infor- mation (e.g., emission rate, stack height, initial plume size, temporal profile). Numerous other inputs may be required, depending on the complexity of the model and application. Most commonly applied air quality model formulations are deterministic and include Gaussian plume, puff, and box models. These models approximate the physical (e.g., trans- port, dispersion, and removal) and chemical (e.g., scavenging, secondary formation) processes that operate on pollutants re- leased into the atmosphere. These models work by parameter- izing the controlling processes that occur at emission source(s) and between the source(s) and receptor(s) at discrete time steps. Other special modeling cases include approaches based on computational fluid dynamics (CFD)—used particularly to characterize source-induced and downwind turbulence effects on the flow, and stochastic approaches that approximate air quality distributions from data sets of controlling variables— including regression, Monte-Carlo, and extreme-value theory- based approaches, (173) as well as those that incorporate sto- chastic properties in a deterministic setting such as combined puff-particle models (e.g., Puff-Particle Model inclusions in CALPUFF). (174) EPA (175) distinguishes air quality models into the follow- ing three categories: • Dispersion models typically are used for small spatial scales and to estimate impacts from individual source(s). These models contain either no or limited chemistry and may be plume or puff formulations. EPA recommended/guideline dispersion models include the following: – AERMOD and – CALPUFF. – Other specialized preferred/recommended models in this category include  BLP,  CALINE3,  CAL3QHC(R), 109

 CTDMPLUS, and  OCD. – Other models in this category include  ADAM,  ADMS-3,  AFTOX,  ASPEN,  Canyon-Plume-Box Model (not a regulatory model but a research-grade FHWA model to demonstrate nonlinear effects of vortex separation and resulting dispersion from roads within cut sections),  EDMS,  HOTMAC/RAPTAD,  HYROAD,  ISC3 (ISC-PRIME),  Panache,  PLUVUEII,  SCIPUFF, and  SDM. • Photochemical grid models typically are used to assess cumulative impacts or interactions of a range of sources over large spatial scales. These are box models but typically also contain plume or puff formulations. This group of models includes the following: – Community Multiscale Air Quality (CMAQ), – Comprehensive Air quality Model with extensions (CAMx), and – Regional Modeling System for Aerosols and Deposition (REMSAD). • Receptor Models that relate observed concentrations to source types and contributions. The focus of this analysis is not a review of the models commonly used to estimate the ambient concentrations associated with goods movement, but rather the methodolo- gies and inputs used by these models. That is, how the emission outputs discussed in Sections 3.2 to 3.8 are used to predict downwind pollutant concentrations. As such, this section does not review the uncertainties in any given model or the uncertainties in any other parameter input to these models. The focus here is only on the emissions-relevant model parameters and processes and does not include other neces- sary model inputs (such as meteorological data, surface and terrain characteristics, biogenic or coincidental emissions data, chemical schema, etc.). Generally, one of two methods will be employed in air qual- ity modeling, depending on the domain size, the physical and chemical processes to be included, and the desired output res- olution. Note that there is significant overlap between these criteria. The two methods we consider here are grid modeling for national and regional scales (typically applied to citywide and larger analyses) and dispersion modeling for local/project scales (facility to citywide analyses). Note that at some scales, either method could be appropriate. Exhibit 3-66 shows these methods. 3.9.2 Evaluation of National and Regional Methods and Models National or regional simulations of air quality will most likely be made with a photochemical grid model. In many cases, these are limited in time to episodic simulations, although annual or even multi-annual simulations are capa- ble in some models. In all cases, input preparation and model executions are resource intensive. Photochemical Grid Model Methodology Photochemical grid models (PGMs) rely on gridded model domains and simulate all processes that influence concentra- tion (chemistry, diffusion, advection) in each grid cell during a time step. However, this approach is physically limited for small spatial scale applications due to artificial dilution of emis- sions, unrealistic near-source concentrations, and spatially un- resolved receptors for sizes smaller than an individual grid cell. Most current models allow for plume in grid (PiG) or other subgrid scale treatment of gas, aqueous, and aerosol chemistry, at least for major or elevated point sources. Other parameters are (horizontally) resolved only at the grid-cell level (typically 2 km to 36 km), including emissions and meteorology. Some 110 Method/Model Type Geographic Scale Pollutants Freight/Passenger Grid modeling methodology Method National and regional Primary and secondary criteria, toxics All* Dispersion modeling methodology Method Local/project Primary criteria and toxics** Both * Typically requires simulation of all sources, but specialized techniques may be used to identify impacts from individual elements. ** Some limited chemistry may be included for primary reactions and secondary species. Exhibit 3-66. Air quality modeling methods.

current models extend subgrid cell treatment to nonpoint sources, such as resolution of individual roads and receptors within a grid cell, such as the use of PiG for near roadway con- centrations of mobile source air toxics (MSATs). (176–177) This formulation highly parallels that of the dispersion mod- eling methodology discussed in the following section, but of- fers the additional ability of full chemistry simulations. It is, however, highly computationally expensive. Inputs to PGMs are typically in the form of detailed input files describing the emissions, meteorology, initial and bound- ary conditions, underlying surface geographical and topolog- ical characteristics, appropriate chemical reactions and rates, as well as domain and simulation period. Each of these must be derived from other sources by a series of typically complex processes and carries inherent uncertainty. For example, meteorological inputs are commonly derived from a diagnostic application of a different model that simulates the meteorolog- ical environment during the period. As discussed, however, the only uncertainty in a photochemical grid modeling method- ology related to goods movement activities is that of the emis- sions parameters. Emissions Parameters. Emission parameters are typically detailed for the PGMs using emissions input files. These describe both low-level and elevated emissions. Low-level emissions are those released within the lowest atmospheric layer (surface layer—typically tens of meters), and are com- prised of area, mobile, low-level point, and biogenic sources. Area sources are representations of groups of point sources that are either spatially distributed or poorly spatially charac- terized, but collectively important. They include, for example, various industrial and agricultural processes, dry cleaners, etc. Elevated emissions include releases from tall point sources, such as power plant stacks. Emission inputs are generally pre- pared for PGMs using external tools, such as SMOKE or EPS. Goods movement activities are included in PGMs either as mobile (e.g., trucking) or area (e.g., cargo handling equip- ment) sources. In air quality modeling, the strength, location, and profile of emissions are all influential. Because goods movement emission inventories, such as those described pre- viously, are typically annual totals, temporal profiles must be assigned. This may be an additional source of uncertainty. The spatial distribution of mobile and area sources is not typically critical within a given grid cell, since that is the minimum res- olution of PGMs in most contexts. However, uncertainty in lo- cations that lead to source placement across cell borders may lead to biased predictions of concentration. This, too, is an ad- ditional source of uncertainty for emissions not characterized previously. Also, PGMs require simulation of all sources in the model domain for correct chemical analysis, not just those of a given project or those from freight transport. This additional burden may introduce uncertainties or lead to bias. Total uncertainty in predicted concentrations from PGMs is due to uncertainty in the emission inputs as well as the un- certainties in all other inputs (meteorology, chemistry, model formulation, etc.) and model formulations. This value is gen- erally unquantifiable. It is possible, however, to characterize the sensitivity of predicted concentrations to the representa- tion of emissions, particularly emission strength. Because PGMs involve nonlinear processes, this is typically done nu- merically by performing multiple PGM simulations of vary- ing emission levels while other parameters are kept fixed to estimate the relative change about a default state (linear error term). However, this sensitivity would be context specific and, in general, could not be generalized to overall model sensitivity (or uncertainty). Summary of Strengths and Weaknesses. Strengths and weaknesses of the photochemical grid modeling methodol- ogy are summarized in Exhibit 3-67. 3.9.3 Evaluation of Local/Project-Level Methods and Models Evaluation of the impacts of emissions from project-specific scale applications, such as individual ports, intramodal yards, freeways, or intersections, are typically done with a dispersion modeling method. Dispersion Modeling Methodology Dispersion models simulate the effects of atmospheric tur- bulence, mixing depth, and wind flow that drives the advec- tion and diffusion of pollutants following their release into the atmosphere. Dispersion models simulate these processes as either a straight line Gaussian plume or as an advecting puff. Both formulations have advantages and disadvantages. Most Gaussian plume models—including AERMOD, which is the EPA’s preferred model for near-field regulatory applications—have either no or highly simplified chemistry. Furthermore, guideline models such as AERMOD were designed to predict peak concentration distributions, not to accurately assess temporally and spatially varying concentra- tions. (178) Although skill is being improved, the limitations of these model formulations (i.e., assessing source contribu- tions to all receptors at each simulated hour) must be consid- ered. These models are relatively straightforward, however, and have shown reasonable predictive skill in their operating range (50 km for AERMOD). (179) They also have several advanced or specialized treatments that make their appli- cation for specific projects advantageous. For example, AERMOD has state-of-the-science boundary layer physics, plume rise, deposition, and building downwash methods. (180) The CALINE series of models is designed to characterize 111

enhanced turbulence from vehicle motions and hot-exhaust rise near the emission sources on roadways. (181) The OCD model is designed to simulate pollutants on-shore after being dispersed in the over-water boundary layer. (182) Gaussian plume models are relatively straightforward to apply, how- ever, these models cannot predict more complex impacts from air circulation, stagnation, or other non-steady-state condi- tions. Exhibit 3-68 shows a schematic illustration of Gaussian dispersion, and Equation 21 provides the general equation for Gaussian dispersion. Where C is the concentration at point (x, y, z), Q is the emission rate, u is the wind speed, σy and σz are the horizontal and vertical dispersion coeffi- cients (at a downwind distance), and h is the effective stack height. Advecting puff models, such as CALPUFF, simulate non- continuous plumes. CALPUFF is a non-steady-state La- grangian puff model that can include the effects of a three- dimensional wind field on the puff as it migrates through complex terrain. CALPUFF is EPA’s preferred model for long- range transport applications (greater than 50 km, and prima- rily for Class I increment studies) or for near field applications involving complex winds, although complete verification of current versions is still being undertaken by EPA. (183) In C x y z Q u z h y z z , , exp( ) = − −( )⎡ ⎣⎢⎢ ⎤ ⎦⎥⎥ ⎧⎨⎪⎩2 2 2 2π σ σ σ⎪ + − +( )⎡ ⎣⎢⎢ ⎤ ⎦⎥⎥ ⎫⎬⎪⎭⎪ − ⎡ ⎣⎢exp exp z h y z z 2 2 2 22 2σ σ ⎤ ⎦⎥ ⎧⎨⎩ ⎫⎬⎭ ( )Equation 21 112 Criteria Strengths Weaknesses Representation of physical processes Complex physical and chemical processes parameterized Limited to model spatial (and temporal) resolution Sensitivity to input parameters A number of parameters may affect model results Highly susceptible to uncertain, complex inputs Flexibility Ability to incorporate effects of emission reduction strategies Yields most realistic air quality impacts since model explicitly treats nearly all of the important chemical and dispersion processes Indirect: incorporated via emission characterization Representation of future emissions Incorporated via emission characterization. Consideration of alternative vehicle/fuel technologies Indirect: incorporated via emission characterization Data quality Models are typically verified against observed data for some air pollutants, lending confidence to other air concentration predictions Relies on numerous inputs of varying quality and uncertainty Spatial variability Generally limited to grid cell resolution (typically 2 km or more) Temporal variability Current concentration is a function of all of the previous hour’s emissions Generally limited to hourly time steps Review process Models and methods have undergone continuous revisions since the 1970s. Endorsements EPA and other federal, state, and local agencies Exhibit 3-67. Summary of strengths and weaknesses—photochemical grid modeling methodology. Source: http://upload.wikimedia.org/wikipedia/commons/1/10/Gaussian_Plume.png Exhibit 3-68. Gaussian dispersion.

these models, a “puff” of pollutants is followed from emission source, through the atmosphere, and to a receptor. During this transport, simple chemical changes, effects of wind shear, effects of terrain, and wind circulations are simulated. This allows the models to more completely parameterize atmo- spheric effects than simple straight-line, steady-state Gaus- sian models. However, the setup, execution, and model form- ulation are all more complex. In many circumstances, their performance is not sufficiently enhanced over Gaussian plume models to justify their use. Emission Parameters. Although the atmospheric effects on pollutants are parameterized differently using the two types of dispersion models discussed in this section, the para- meterization of emissions is similar. Most dispersion models treat emission sources as either point, area, volume, or line sources. In all cases, the locations of the emission releases do not change during simulations. Point sources typically represent emissions from station- ary stacks and are generally buoyant. They could be used to represent exhaust stacks of hotelling vessels, for example. Input parameters required include location, instantaneous or average emission rate, release height, exit temperature, exit velocity, and stack inside diameter (or flow rate). For stacks where building downwash is important, additional parame- ters also must be included to simulate these effects. Uncer- tainty in any of these parameters will lead to uncertainty in output concentrations. The relationship between most of these parameters and concentrations may be complex, due to interactions with input meteorology as formulated in the model. Concentration is linearly proportional to emission rate in all cases; standard propagation of uncertainty can be used to show uncertainty in concentration from a known emission uncertainty. Uncertainty due to other (nonlinear) parameters may be derived between a specific source and re- ceptor due to propagation of uncertainty and Equation 21. Line, area, and volume sources are one-, two-, and three- dimensional source types commonly used to describe emissions where the spatial distribution of emissions within a particular boundary is not fully known (e.g., an industrial complex) or within which the emissions occur more or less uniformly (e.g., a freeway link). Their governing equations are a varia- tion on the equation for point sources found in Equation 21. Area and volume sources may be non-buoyant (AERMOD) or buoyant (CALPUFF, area sources only). Further, some models do not contain the ability to model line sources ex- plicitly (e.g. AERMOD); instead, modeling sources such as roads, rail lines, or shipping channels may be achieved by as- sembling adjacent groups of volume or area sources. Emission inputs for these source types include emission rate, location, orientation, release height, and initial plume size (lateral and vertical dimensions). If buoyancy is considered, exit temper- ature must also be included. As for point sources, uncertainty in any of these parameters will lead to uncertainty in output concentrations. Here, too, the relationships between most of these param- eters and concentrations are complex, except for the linear re- lationship between concentration and emission rate. If uncer- tainty in any input parameter is known, standard propagation of uncertainty can be used to show uncertainty in concentra- tion. Commonly, uncertainty is not known, especially for methodological or choice issues to fit model requirements. For example, when modeling freight emissions from HDVs, some line source models (e.g., CALINE) may only require a single release height for both light- and heavy-duty vehicles. Selection of an appropriate value is sometimes discussed in modeling guidance documents. Total uncertainty in predicted concentrations from goods movement represented using a dispersion methodology is due to uncertainty in the emission input parameters, uncertainties in all other input parameters (e.g., meteorology), as well as uncertainties in methodology (e.g., model formulation and choice). This value is generally unquantifiable without comparison to observed concentrations. Those uncertainties due to calculated emission rate, however, may be characterized directly from the input uncertainty. Other emission param- eters due to the methodology by which the emitting process is represented—such as spatial scale of activity—generally can not be characterized, but could be assessed for any par- ticular scenario. Summary of Strengths and Weaknesses. Strengths and weaknesses of the dispersion modeling methodology are summarized in Exhibit 3-69. 3.9.4 Evaluation of Parameters Exhibit 3-70 summarizes all emissions-related parameters relevant for calculating concentrations from goods movement activities. Each of these has been discussed. Other parameters, such as initial and boundary chemical conditions, meteorol- ogy, and selection of appropriate models and methods, are not included here. Pedigree Matrix. Exhibit 3-71 shows the pedigree matrix for the seven general parameters that relate goods movement emissions to pollutant concentration through the use of air quality modeling methodologies. Criteria to assign scores in the pedigree matrix are included in Appendix A. Note that all entries here are ranked as “5” for “Range of Variation.” This is because the variation in the variation of values between methods, models, and applications is wide, which is also con- sidered a “5.” See documentation in Appendix A for further explanation. 113

Emission Rate. Concentrations are always directly pro- portional to emissions. Uncertainty in characterizing the total emissions from any given source (and from all modeled sources, if chemistry is included) leads directly to uncertainty in concentrations. In this case, emission rate refers to the emis- sions, usually in grams/second, emitted by a given source at a given time, which is usually determined from the (annual) emission inventory and the emission temporal profile. The relationship of concentration to emissions becomes more complicated as the modeling becomes more complex, but accurate emission inventories for given sources are the key- stone of reasonable model predictions. Uncertainty in emis- sions has been discussed in all previous sections. Source Location. Uncertainty in geographic placement of sources leads to uncertainty in concentration at a given re- ceptor site due to the uncertainty in transit distance between the two locations. In plume or puff modeling, this distance al- lows the pollutants to be more (less) diffuse and have greater (less) time for chemical transformation reactions, settling, 114 Criteria Strengths Weaknesses Representation of physical processes Dominant processes generally parameterized, as long as operated within model limitations (e.g., spatial scale) Model formulations are generally simplistic Sensitivity to input parameters Generally rely on readily available inputs Susceptible to uncertain inputs Flexibility Generally adaptable to a variety of scenarios and available information Gaussian plume models operate on an underlying assumption of a steady-state Ability to incorporate effects of emission reduction strategies Indirect: incorporated via emission characterization Representation of future emissions Indirect: incorporated via emission characterization Consideration of alternative vehicle/fuel technologies Indirect: incorporated via emission characterization Data quality Varies: relies on input data quality and model formulations; is particularly susceptible to inappropriate model choice or input variables Spatial variability Can model concentrations in close proximity to source and with arbitrarily high spatial resolution Gaussian plume model formulations may not represent variability well in complex terrain or wind flow regimes Temporal variability Limited only by input data resolution Review process Models and methods continuously updated and expanded. Result is model-specific Endorsements EPA and other federal, state, and local agencies Result is model-specific Exhibit 3-69. Summary of strengths and weaknesses—dispersion modeling methodology. Parameter Methods/Models GeographicScale Pedigree Matrix Qualitative Assessment Quantitative Assessment Emission rate All All Source location All All Emission temporal profile All All Release height All All Initial plume size and shape All All Source orientation, size, and shape All All Exhaust temperature and other buoyancy parameters All (if plume rise is considered) All Exhibit 3-70. Parameters.

and other removal processes to act if the source is placed fur- ther (closer) to the receptor. In a grid modeling application, the locations of the sources (within the model resolution) are less important, as long as they are assigned to the correct grid cell. The general relationship between location and concen- tration is unquantifiable, but uncertainties in location will impact simulated concentrations and are likely to change the spatial distribution of concentrations. Emission Temporal Profile. Concentration estimates are highly sensitive to the temporal profile imposed on the total, annual emission rate determined from an inventory of goods movement activities. The temporal profile assigns emissions to specific hours of the year where the model pairs them with corresponding meteorological and other parameters. If the di- urnal, weekly, or other cycles are mischaracterized, the dis- persion will be, too. Values of the profiles are often taken from published studies of activity of specific equipment types (184) based on SCCs. More accurate representation would require knowledge of activity profiles throughout the inven- tory period, which are often unavailable. The impact of emis- sion temporal profile on total concentration uncertainty is not generally quantifiable, but may be determined for specific scenarios. Release Height. Release height is the vertical component of source location. The relationship between release and re- ceptor height, in combination with terrain, stability, initial dispersion, building downwash, and other parameters can greatly influence modeled concentrations. Because of these complex relationships, concentration uncertainty caused by uncertainty in release height can not generally be quantified. Initial Plume Size and Shape. Dispersion generally in- creases plume/puff size, and therefore dilutes concentrations. Thus, the concentration observed at a particular receptor lo- cation will be due to both the processes acting on the pollu- tants after emission and on the initial state of the emissions. As the size and shape of the initial plume influences the downwind concentration at a given location, uncertainty in initial shape will lead to uncertainty in resulting concentra- tions. This uncertainty can be mitigated by following pub- lished modeling guidance and characterizing the sources in as realistic a method as possible. Source Orientation, Size, and Shape. Particularly for nonpoint sources, the initial size, shape, and orientation of the source can dictate the dispersion characteristics. Orienta- tion can change the size of the source relative to a given wind direction, and therefore influence the downwind concentra- tion. Generally, the initial plume size is related to the source size; thus, the uncertainties discussed for initial plume size relate here, too. Buoyancy Parameters. Buoyancy and rise of the emitted pollutants is related to the initial exhaust temperature relative to the ambient temperature and exhaust flow rate. This has an effect similar to raising the release height. Thus, uncertain- ties here propagate to concentration in a method similar to that discussed for release height. 115 Parameter Im pa ct on R es ult Ac qu isi tio n M eth od In de pe nd en ce Re pr es en tat ive ne ss Te m po ra l C or re lat ion Ge og ra ph ic Co rre lat ion Te ch no lo gic al Co rre lat ion Ra ng e o f V ar iat ion Emission rate 4 Varies Varies Varies N/A N/A Varies 5 Source location 3 Varies Varies Varies N/A N/A Varies 5 Emission temporal profile 3 Varies Varies Varies N/A N/A Varies 5 Release height 3 Varies Varies Varies N/A N/A Varies 5 Initial plume size and shape 3 Varies Varies Varies N/A N/A Varies 5 Source orientation, size, and shape 2 Varies Varies Varies N/A N/A Varies 5 Exhaust temperature and other buoyancy parameters 3 Varies Varies Varies N/A N/A Varies 5 Exhibit 3-71. Pedigree matrix—harbor craft equipment parameters.

Next: Chapter 4 - Conceptual Model »
Representing Freight in Air Quality and Greenhouse Gas Models Get This Book
×
MyNAP members save 10% online.
Login or Register to save!
Download Free PDF

TRB’s National Freight Cooperative Research Program (NFCRP) Report 4: Representing Freight in Air Quality and Greenhouse Gas Models explores the current methods used to generate air emissions information from all freight transportation activities and their suitability for purposes such as health and climate risk assessments, prioritization of emission reduction activities, and public education.

The report highlights the state of the practice, and potential gaps, strengths, and limitations of current emissions data estimates and methods. The report also examines a conceptual model that offers a comprehensive representation of freight activity by all transportation modes and relationships between modes.

  1. ×

    Welcome to OpenBook!

    You're looking at OpenBook, NAP.edu's online reading room since 1999. Based on feedback from you, our users, we've made some improvements that make it easier than ever to read thousands of publications on our website.

    Do you want to take a quick tour of the OpenBook's features?

    No Thanks Take a Tour »
  2. ×

    Show this book's table of contents, where you can jump to any chapter by name.

    « Back Next »
  3. ×

    ...or use these buttons to go back to the previous chapter or skip to the next one.

    « Back Next »
  4. ×

    Jump up to the previous page or down to the next one. Also, you can type in a page number and press Enter to go directly to that page in the book.

    « Back Next »
  5. ×

    To search the entire text of this book, type in your search term here and press Enter.

    « Back Next »
  6. ×

    Share a link to this book page on your preferred social network or via email.

    « Back Next »
  7. ×

    View our suggested citation for this chapter.

    « Back Next »
  8. ×

    Ready to take your reading offline? Click here to buy this book in print or download it as a free PDF, if available.

    « Back Next »
Stay Connected!