National Academies Press: OpenBook

Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design (2004)

Chapter: Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design

« Previous: Front Matter
Page 1
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 1
Page 2
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 2
Page 3
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 3
Page 4
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 4
Page 5
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 5
Page 6
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 6
Page 7
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 7
Page 8
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 8
Page 9
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 9
Page 10
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 10
Page 11
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 11
Page 12
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 12
Page 13
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 13
Page 14
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 14
Page 15
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 15
Page 16
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 16
Page 17
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 17
Page 18
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 18
Page 19
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 19
Page 20
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 20
Page 21
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 21
Page 22
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 22
Page 23
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 23
Page 24
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 24
Page 25
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 25
Page 26
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 26
Page 27
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 27
Page 28
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 28
Page 29
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 29
Page 30
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 30
Page 31
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 31
Page 32
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 32
Page 33
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 33
Page 34
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 34
Page 35
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 35
Page 36
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 36
Page 37
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 37
Page 38
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 38
Page 39
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 39
Page 40
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 40
Page 41
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 41
Page 42
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 42
Page 43
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 43
Page 44
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 44
Page 45
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 45
Page 46
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 46
Page 47
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 47
Page 48
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 48
Page 49
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 49
Page 50
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 50
Page 51
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 51
Page 52
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 52
Page 53
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 53
Page 54
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 54
Page 55
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 55
Page 56
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 56
Page 57
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 57
Page 58
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 58
Page 59
Suggested Citation:"Part 1 - Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design ." National Academies of Sciences, Engineering, and Medicine. 2004. Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design. Washington, DC: The National Academies Press. doi: 10.17226/13781.
×
Page 59

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.

Part 1 Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design

1.0 Introduction The AASHTO Joint Technical Committee on Pavements has undertaken an effort to develop an improved guide for the design of pavement. This effort, undertaken under NCHRP Proj- ect 1-37A, Development of the 2002 Guide for the Design of New and Rehabilitated Pavement Struc- tures1 (the “Pavement Design Guide”), will provide engineers with practical and realistic pave- ment design procedures and software that use existing mechanistic-empirical principles. The mechanistic-based distress prediction models used in the Pavement Design Guide will require the input of specific data for each axle type and axle-load group. The goal of NCHRP Project 1-39 was to develop procedures and software for collecting and processing traffic data required by the Pavement Design Guide procedures. In addressing that goal, the Project 1-39 Team has produced the following: • Guidelines for collecting traffic data to be used in pavement design. These guidelines are presented in Part 2 of this report. • Software for analyzing traffic data and producing the traffic data inputs required by the Pavement Design Guide software (the “Design Guide software”). The software developed under Project 1-39 has been named “TrafLoad.” TrafLoad is contained online at http:// trb.org/news/blurb_detail.asp?id=4403. The use of TrafLoad is described in a user’s man- ual, and algorithms used by TrafLoad are documented. • A guide for choosing equipment for collecting classification counts and weigh-in-motion (WIM) data. This guide has been published as NCHRP Report 509.2 Recommended steps to be taken in the course of implementing the data-collection procedures are presented in Chapter 2.0 of Part 1. The following chapter presents the results of analyses of the effect of the length of the data-collection period on the accuracy of pavement damage factors developed from short-duration WIM data collection. The fourth chapter of Part 1 discusses three technical issues relating to the design of software for analyzing traffic and presents recommended solutions that have been incorporated into TrafLoad. In Section 4.1, it is recommended that short-duration traffic counts be converted to estimates of annual average daily traffic (AADT) by dividing by “traffic ratios” rather than by multiplying by “traffic factors.” In Section 4.2, it is recommended that, if use is made of partial- day classification counts, “truck traffic distribution factors” (TTDFs) not be used for convert- ing these counts to estimates of 24-hour traffic volume by vehicle class; instead, it is recom- mended that “hourly traffic ratios” (or “hourly fractions”) be used. Section 4.3 discusses some issues relating to the use of simple versus weighted averaging of traffic data. 1-1 1 http://www.2002designguide.com/ 2 Hallenbeck, Mark, and Weinblatt, Herbert, NCHRP Report 509: Equipment for Collecting Traffic Load Data, Transportation Research Board of the National Academies, Washington, D.C., 2004.

The final chapter of Part 1 presents recommendations for potential improvements to TrafLoad to be made in the future as well as recommended areas for future research. Part 1 also contains two appendixes. Appendix A discusses procedures for forecasting traffic volumes, including several relatively complex alternatives to the procedure that is recom- mended in Section 3.6 of Part 2. Appendix B presents a procedure that had been proposed for estimating coefficients of variation (CVs) for estimates of AADT by vehicle class.3 1-2 3 The original design for the Design Guide software would have required users to provide these CVs. However, the final version of this software makes no use of these CVs.

2.0 Implementation Needs Successful implementation of the TrafLoad software by state highway agencies will require several actions at both the state and the national level. The most important of these actions is an increased level of communication and cooperation between the traffic data-collection staff and the pavement design staff. Successful communication between these groups hinges, in part, upon better knowledge between both groups as to what traffic data are needed for pave- ment design, how variations in traffic loads affect pavement design, and how to account for those variations. Consequently, the implementation requirements at both the state and the national level involve the following: • Development and execution of training programs to improve the knowledge of both groups, including both specific training in the TrafLoad software and more general instruction as to how traffic loads affect pavement design; and • Removal of institutional barriers that limit the interaction between pavement design and traffic data-collection and analysis staff.  2.1 National-Level Implementation Actions Three key actions are required at the national level: • Development of training material, • Financial support of training programs, and • Agency support of the institutional changes required at the state level that encourage greater interaction between the pavement design and traffic data-collection groups in sup- port of the implementation of the Pavement Design Guide procedures. Development of Training Material Because of the national scope of the Pavement Design Guide’s implementation effort, it is most appropriate for training material to be developed at the national level. The training material should cover the following topics: • Variations in traffic loads with time and location, • The effect of those variations on pavement design, 1-3

• The effect of those variations on traffic data-collection procedures, • The analytical steps required to convert the raw data into inputs for the pavement design process, • Instruction in the operation of the TrafLoad software, and • The type and timing of interactions between the pavement design and traffic data-collection groups that are required to produce cost-effective traffic data-collection programs. The training program should teach pavement design engineers about the need for traffic data and difficulties in its collection and analysis. And it should teach the traffic data-collection staff about how traffic data (and particularly variations in traffic data) affect pavement design. Financial Support for Training National financial support for training is required in two areas. The first area is the creation of the primary training material described above. The second area is assistance in actually teach- ing the course material. This second level of federal funding includes both subsidizing the cost to state agencies of providing the courses and funding multi-state workshops that allow state personnel to learn from each other. Multi-state training workshops provide an excellent mech- anism for exploring potential institutional and organizational changes that break down the communication barriers that exist between state agency personnel working in different offices. Support for State Agency Organizational/Institutional Change Finally, the Federal Highway Administration (FHWA) should actively support and encour- age the institutional changes that are needed to incorporate accurate traffic load estimates in the pavement-design procedures. Changes in state highway agency culture are difficult to achieve without “top-down” direction. The FHWA is the appropriate agency for providing this direction and for encouraging top-down direction within the highway agency itself.  2.2 State-Level Implementation Actions The implementation tasks needed at the state highway agency level relate to the following: • Training personnel, • Changing internal work processes to institutionalize the communication needed to ensure the collection and use of accurate traffic load data, and • Refining the current traffic data-collection and summarization process to improve the quality of the load estimates available for use in the Pavement Design software. 1-4

Training State highway agency staff will require training in the collection, summarization, and entry of traffic data into the TrafLoad software. The project team recommends the cross training of pavement design and traffic data-collection and analysis personnel. Members of both groups need to understand how traffic load and variations in that load affect pavement design and how traffic load varies with time and location. Only when both groups understand this inter- action can cost-effective decisions be made as to how much traffic data to collect, where and when to collect the data, and how the data should be summarized and processed by the TrafLoad software for use in the Pavement Design Guide software. Institutional Changes to Improve Communication Traditionally, interaction between the pavement design group and the traffic data-collection and analysis group is limited to the responses of one group to routine requests issued by the other. This lack of interaction is exacerbated by the fact that these groups are often separated physically (working in different buildings or on different floors of the same building) and report within different branches of the organization (design/construction versus planning). To implement improvements in traffic-load data needed for successful use of the Pavement Design Guide, frequent and effective communication between the two groups is required. State highway agencies should adopt procedures that allow the following communications to take place between the pavement design and traffic data-collection groups: • The pavement design group must communicate to the traffic data-collection group what and where data are needed for design purposes. This communication must take place in a timeframe that allows the traffic data-collection group to collect any required site-specific data cost-effectively. • Not all traffic data that could be used by the Pavement Design Guide can be collected cost- effectively. Hence, the two groups must consider the resources available for traffic data collection and agree to compromises on when, where, and how much data are collected for pavement design purposes. (These compromises might include the use of design funds to supplement the routine data-collection budget.) • Procedures must be developed that allow either group to communicate data quality con- cerns to the other and receive feedback about how those concerns are addressed. • Both groups need to be involved in developing the procedures that allow a smooth flow of traffic data into the TrafLoad software and a smooth flow of data from that software into the Pavement Design Guide software. • Annual reviews should be conducted of the traffic data being used for pavement design and of the processes used for collecting and summarizing the data and loading it into the TrafLoad and 2002 Pavement Design Guide software. These reviews should identify lim- itations in the processes (e.g., lack of data on specific highways or in specific parts of the state), so that weaknesses are identified and eliminated over time. 1-5

Many of these communications will require changes in the way that procedures are per- formed. Consequently, it will be necessary for upper-level management of the state highway agency to provide guidance and institutional support for changing current procedures to sup- port these efforts. Improvements in communication should also include other groups within the state govern- ment. For example, forecasting of future truck volumes and/or loads can be improved if the groups charged with support of statewide economic development are consulted about expected changes in statewide economic activity that might affect truck volumes. (Are there expected rail-line abandonments that would add substantial truck traffic to specific roads? Are there new factories being built in a specific area that will increase truck traffic?) Similarly, highway maintenance workers and others often can provide insight into the commodities being carried on specific roads and the presence of heavily loaded or overloaded trucks on those roads. Thus, implementation requires a thorough review of the many opportunities for improved communication. Changes in Collection and Summarization of Data While the 2001 edition of the FHWA’s Traffic Monitoring Guide (TMG) lists steps needed to sup- ply the traffic load data needed for pavement design, most states are still in the early stages of implementing the procedures described in the TMG and in the NCHRP 1-39 reports. Support for continued improvements in the data-collection process will be needed. In particular, a careful review of the data-collection procedures described in Part 2 should be performed, and required changes to the state’s current data-collection program should be identified. Improving the process will result in improving the quality of the truck volume and axle-load values used as input to the design process and will affect the volume of data collected, the sites that are mon- itored, and the handling of the data that are collected. The refinement process should be an ongoing effort. Much remains to be learned about truck volume and weight, and data-collection resources are not sufficient to fill all the knowledge gaps quickly. Therefore, states should expect their ongoing review of truck volumes and weights, combined with their need for pavement designs, to result in periodic shifts in data- collection resources as priorities change, as new needs become apparent, and as weaknesses in the current system are identified and eliminated. These refinements will be most effectively performed if they involve significant input from the pavement design community. Pavement designers should indicate where their needs are greatest and the relative priorities that they place on the various components of the data-collection process. 1-6

3.0 The Effect of Length of Collection Period for WIM Data Several analyses were performed to estimate the accuracy of estimates of annual pavement loads that are developed from short periods of WIM data collection. The data-collection peri- ods consisted of either 7 consecutive days or 2 consecutive weekdays. The analyses of 7-day periods were conducted as part of NCHRP Project 139, while those of 48-hour periods were conducted under a recently completed project performed for FHWA.1 This chapter presents the results of both sets of analyses. The analyses require the use of unidimensional measures of pavement load. For this purpose, the pavement load created by any given set of axle loads was measured in terms of 18,000- pound equivalent single-axle loads (ESALs) for flexible pavement, and the analyses focused on the resulting values of average ESALs per vehicle (AEPV) and annual AEPV (AAEPV). The methodology used for the analyses is described in the first section of this chapter, and the results of the analyses are described in the second section.  3.1 Methodology All analyses were conducted using WIM data collected in 2000 by the state of California’s traf- fic monitoring program and stored by the University of California’s Pavement Research Cen- ter. The analyses used data for 55 sites for which there were at least 8 months of available data that met the checks on consistency of calibration.2 Estimates of AAEPV were developed for each of the 55 sites using a procedure that attempts to minimize the effects of missing days of the week and missing months.3 For each site, separate estimates of AAEPV were developed for each of three groups of FHWA vehicle classes (VCs): Class 5; Classes 6 and 7; and Classes 8–13. The effectiveness of using short-duration WIM data to estimate annual conditions was evalu- ated by comparing the estimates of AAEPV derived from “annual” data with estimates derived from data collected over a 48-hour or 7-day period. For each of the 55 sites, estimates 1-7 1 Cambridge Systematics, Inc., Accuracy of Traffic Load Monitoring and Projections, Volume II: The Accu- racy of ESALs Estimates, February 2003. 2 Ibid., Chapter 2.0. 3 Ibid., Appendix.

of AAEPV were obtained using data from pairs of consecutive weekdays (or from periods of 7 consecutive days) that exclude federal holidays, Christmas Eve, and the days before and after Thanksgiving. The use of 48-hour periods was analyzed using all possible pairs of weekdays, with a maxi- mum of four pairs per week. The use of 7-day periods was analyzed using all periods of 7 con- secutive days for which data were available that did not include any of the above holidays. For most sites, the database contains data for a maximum of 12 weeks (1 week per month). Hence, with a few exceptions, data were available for a maximum of 48 estimates for each site and VC group when 48-hour data were used and a maximum of 12 estimates when 7-day data were used. Each of the resulting estimates was compared to the “true” value of the AAEPV for the site, producing one estimate of the resulting error. The absolute values of the errors were obtained and were converted to percentages and averaged, producing values of mean absolute percent error (MAPE). Since MAPE uses absolute percent errors (rather than signed values of percent errors), large positive errors are not offset by large negative errors. Separate statistics were obtained for each of three VC groups (5, 6 and 7, and 8–13). One complete set of AAEPV estimates was developed using data for each 48-hour period with- out adjustment to produce estimates of AAEPV for each VC group; a second set was similarly developed using data for each 7-day period. In addition, several sets of AAEPV estimates were produced using ESAL ratios to “factor” the data. For 48-hour data, this factoring process involved three major steps performed separately for each VC group distinguished: 1. For each of the 55 sites for which good estimates of AAEPV have been obtained, a set of seven day-of-week (DOW) ESAL ratios is developed, as well as a set of 12 or fewer monthly ESAL ratios, as described subsequently. 2. For any site of interest, a second set of DOW and monthly ESAL ratios is obtained as an unweighted average of the Step 1 ESAL ratios obtained at some or all of the other 54 sites. In this step, Monday ratios are obtained as averages of Monday ratios from Step 1, Tues- day ratios as averages of Tuesday ratios from Step 1, etc. 3. For a particular site, all raw (i.e., unfactored) AEPV values for pairs of consecutive week- days are converted to factored estimates by a) Using the raw data to obtain separate AEPV estimates by DOW and, if necessary, by month; b) Dividing each AEPV estimate by the corresponding monthly and DOW ESAL ratios obtained for the site in Step 2; and c) Taking a weighted average of the factored AEPV estimates produced in Step 3(b). The weights used in Step 3(c) are the sums of the monthly average day-of-week (MADW) vol- umes for the site and VC group. 1-8

For each site, VC group, and day of the week, the DOW ESAL ratio developed in Step 1 is obtained as follows: • For each month for which a MADW ESAL value exists, divide this value by the corre- sponding value of monthly AEPV (MAEPV) and • Average the resulting ratios. Similarly, for each site, VC group, and month for which MAEPV exist, a monthly ESAL ratio is obtained in Step 1 by dividing MAEPV by AAEPV. For 7-day data, DOW factoring is unnecessary. Accordingly, for 7-day data, the above factor- ing process was used without any DOW factoring. The results reported here were developed using “statewide factoring” and simple (as opposed to weighted) averages. For this purpose, simple averages are produced in Step 2 using ratios obtained from all sites except the site from which the data to be factored were obtained.4  3.2 Results The results of the above analyses are summarized in Table 3.1. Although the errors for VCs 6 and 7 are appreciably larger than they are for the other two VC groups, the results are other- wise similar for the three VC groups. The use of 48 hours of unfactored data produces moderate MAPEs in the estimates of AAEPV, ranging from 7.3 percent for VCs 8–13 to 13.0 percent for VCs 6 and 7. The table indicates that moderate reductions can be obtained by factoring and/or extending the data collection period to 7 days. For combinations (VCs 8–13), factoring produces a slightly larger improvement than does extending the period to 7 days. However, for single-unit trucks (SUTs) (VC 5 and VCs 6 and 7), extending the collection period produces an appreciably greater improvement than does factoring. The use of 7 days of factored data produces MAPEs that range from 5.5 per- cent (for VC 5) to 9.9 percent (for VCs 6 and 7). The switch from 48 hours to 7 days produces greater improvements in the estimates for SUTs than in the estimates for combinations, particularly when unfactored data are used. This dif- ference is due to differences in the extent of DOW variation in AEPV for SUTs and combina- tions. The daily ESAL ratios shown in Figure 3.1 reflect statewide average values of AEPV for each day of the week that have been normalized by dividing by annual AEPV. As can be seen 1-9 4 The FHWA report also presents results for several variants of the procedure. The use of weighted aver- ages produced very slight reductions in the MAPEs. The use of several factor groups (instead of statewide factoring) did not produce any consistent improvement, but the factor groups used in the tests were created for another purpose and not specifically designed to be used as factor groups. It is likely that more carefully designed factor groups could produce small reductions in the MAPEs.

1-10 Table 3.1 Errors Produced by Using Short-Duration WIM Data to Estimate Average ESALs per Vehicle Figure 3.1 Daily ESAL Ratios 0.70 0.80 0.90 1.00 1.10 1 2 3 4 5 6 7 Sunday Monday Tuesday Wednesday Thursday Friday Saturday Day of Week ESAL Ratio VC 5 VC 6-7 VC 8-13 Source: Cambridge Systematics, Inc., Accuracy of Traffic Load Monitoring and Projections, Volume II: The Accuracy of ESALs Estimates, prepared for FHWA, February 2003, Figure 5.1. Mean Absolute Percentage Error Vehicle Classes 5 6 and 7 8–13 Unfactored Data 48 hours 8.1% 13.0% 7.3% 7 days 5.7% 10.1% 6.6% Factored Data 48 hours 7.0% 12.7% 6.4% 7 days 5.5% 9.9% 5.7%

from the figure, AEPV for SUTs drops sharply on weekends, but AEPV for combinations varies only slightly by DOW, reaching its highest level on Sunday. The effectiveness of factoring depends on the consistency of the DOW and month-of-year pat- terns in AEPV. When using 48-hour data, factoring produces a 1.1-percent reduction in MAPE for VC 5 and smaller reductions for the other VC groups. However, when using 7-day data, factoring produces only a 0.2-percent reduction in MAPE for VC 5. These results imply that there is a substantial degree of similarity in the DOW patterns of AEPV for VC 5 for different sites but much less similarity in the month-of-year patterns. On the other hand, for VCs 8–13, factoring produces the same 0.9-percent reductions in MAPE when using either 48 hours of data or 7 days of data. This result implies that, for VCs 8–13, there is a substantial degree of similarity in the month-of-year patterns in AEPV at different sites but relatively little similarity in the DOW patterns. For VCs 6 and 7, factoring produces relatively small (0.2- to 0.3-percent) error reductions, implying that there is relatively little site-to-site consistency in the DOW and month-of-year patterns in AEPV for these vehicles. Indeed, for VCs 6 and 7, AEPV at most sites is likely to vary with the percentages of these vehicles contributed by construction-related dump trucks. These percentages, in turn, vary with the level and type of activity at nearby construction sites. Since there is little relationship between the levels of activity at different construction sites, there is likely to be little correlation between the DOW and month-of-year AEPV patterns at different WIM sites; therefore, for VCs 6 and 7, factoring produces relatively small reductions in the error. 1-11

1-12 4.0 Three Technical Issues This chapter addresses three technical issues that arise in the development of software, such as TrafLoad, that is designed for analyzing traffic data. Short-duration counts of total traffic (or traffic by VC) can be converted into estimates of AADT (or AADT by VC) by multiplying the counts by appropriate “factors” or by dividing them by “traffic ratios.” Section 4.1 discusses these two options and explains why the use of traffic ratios is likely to produce slightly better estimates of AADT. For this reason, TrafLoad uses monthly and day-of-week traffic ratios (rather than factors) for converting short-duration classification counts into estimates of AADT by VC. TrafLoad also uses “hourly traffic ratios” (rather than “truck traffic distribution factors,” or TTDFs) for converting partial-day classification counts into estimates of 24-hour traffic vol- umes by VC. Section 4.2 explains why TTDFs produce upwardly biased estimates of truck vol- umes at most urban sites. Finally, there is the issue of how to average data collected at several sites that have been assigned to a specific group (such as a “factor group”). Averaging can be performed by tak- ing simple averages of values (e.g., traffic ratios) obtained at each site in the group; alterna- tively, weighted averages can be obtained, most frequently using traffic volumes at the sites as weights. With two exceptions, TrafLoad uses simple averages. Section 4.3 summarizes the use of simple and weighted averages by TrafLoad, discusses the two cases in which weighted averages are used, and explains why simple averages (rather than weighted averages) are used for developing load spectra for “Truck Weight Road Groups.”  4.1 Traffic Ratios versus Traffic Factors A 24-hour traffic count can be converted to an estimate of annual average daily traffic (AADT) by multiplying the count by an appropriate factor or by dividing it by an appro- priate traffic ratio. The two processes are both commonly referred to as “factoring,” and there are several variants of both processes. One common variant (used by TrafLoad for counts from Level 2 classification sites) uses a day-of-week (DOW) traffic ratio (or factor) to convert the count to an estimate of monthly average daily traffic (MADT) and a monthly (or seasonal) traffic ratio (or factor) to convert the MADT to an estimate of AADT.1 When esti- mating AADT by VC, the counts used are classification counts, and separate traffic ratios 1 Another variant (used by TrafLoad for counts from Level 1B classification sites) uses a combined monthly and DOW traffic ratio (or factor) to convert the count directly to an estimate of AADT.

1-13 (or factors) should be used for major groups of VCs (e.g., personal-use vehicles, single-unit trucks, and combinations).2 Factoring procedures that use traffic ratios produce AADT estimates that generally differ only slightly from those produced by the corresponding procedures that use traffic factors. How- ever, to the extent that the estimates differ, those produced using traffic ratios are likely to be the better ones. For this reason, TrafLoad uses traffic ratios rather than traffic factors. This section presents information about why traffic ratios produced very slightly better esti- mates than traffic factors. To simplify the presentation, the research team focuses on the use of monthly traffic ratios and monthly traffic factors for converting MADT estimates to AADT estimates, and the research team focuses on estimates of total traffic volume. The cases of developing estimates for individual VCs, or for VC groups, are completely analogous. Traffic Factors For any month of the year, m, a monthly factor, MFm, can be developed using data from a sin- gle continuous-count site, i: (4.1) Alternatively, MFm can be developed using data from a specified group of continuous-count sites: (4.2) where the average is taken over all sites, i, in the group. To convert an estimate of MADT for any short-duration count site, i′, to an estimate of AADT, the former estimate, MADTmi ′, is multiplied by the appropriate monthly factor: (4.3) Traffic Ratios The process for developing and using traffic ratios is similar to that of developing and using traffic factors. For any month of the year, m, a monthly traffic ratio, MTRm, can be developed using data from a single continuous-count site, i: (4.4)MTR MADT AADTm mi i = AADT MF MADT= ′ ′ =i m mi MF Avg AADT MADTm i i mi =     MF AADT MADTm i mi = 2 The formation of VC groups to be used for factoring classification counts is discussed in Part 2, Sec- tion 3.3.

1-14 Alternatively, MTRm can be developed using data from a specified group of continuous- count sites: (4.5) where the average is taken over all sites, i, in the group. To convert an estimate of MADT for any short-duration count site, i′, to an estimate of AADT, the former estimate, MADTmi ′, is divided by the appropriate monthly traffic ratio: (4.6) Comparing the Two Processes A comparison of Equations 4.3 and 4.6 indicates that the two processes will produce identical estimates of AADTi ′ whenever (4.7) If the factors and traffic ratios are derived using data from a single continuous-count site (i.e., if they are derived using Equations 4.1 and 4.4), Equation 4.7 always holds. However, if they are derived using data from a group of continuous sites, Equation 4.7 generally does not hold. When using data from a group of sites, one actually gets (4.8) or (4.9) with equality holding only for months for which the ratios AADTi/MADTmi are the same for all continuous-count sites in the group. To shed light on this inequality, consider a pair of sites, both of which have the same value of AADT, say 10,000. Assume that, for month m, the values of the MADT are different, say 9,000 and 11,000. The individual MTRs for these sites are 0.9 and 1.1, and the average MTR for this group of sites is 1.0. The individual MFs for these sites are 1.11 and 0.91, and the average MF is 1.01. Hence, when using short-duration counts for this month for sites in this group, the AADT estimates produced using MFs will be about 1 percent higher than those produced using MTRs. The information presented also suggests that a “neutral value” of 1.0 for the traf- fic ratio or factor is likely to be slightly preferable to the actual value of 1.01 for the MF. It is useful to view the averaging process used in the development of traffic ratios and factors as attempts to determine the extent to which these quantities should differ from the neutral Avg AADT MADT Avg MADT AADTi i mi i mi i        ≥ MF MTRm m ≥ 1 MF MTRm m = 1 AADT MADT MTR′ ′ =i mi m MTR Avg MADT AADTm i mi i =    

value of 1.0. The averaging process used for deriving the MF applies extra weight to differ- ences from 1.0 for any site for which these differences are positive (producing, in the above example, the value of 1.0 + 0.11 = 1.11 for Site 1), with this weight increasing nonlinearly as these differences grow. Similarly, the process applies reduced weight to differences from 1.0 for any site for which the differences are negative (producing the value of 1.0 − 0.09 = 0.91 for Site 2). From this perspective, the averaging process used for deriving traffic factors produces a (generally small) upward bias in the values produced for these factors. This upward bias is a mildly undesirable characteristic of traffic factors produced using Equation 4.2. The bias that exists in traffic factors does not exist in the development of traffic ratios. Thus, in the above example, the MTRs for the two sites are 1.0 − 0.1 = 0.9 and 1.0 + 0.1 = 1.1, pro- ducing an average of 1.0 for this group of two sites. The research team concludes that traffic ratios are likely to work better than traffic factors. For this reason, all factoring performed by TrafLoad is performed using traffic ratios.  4.2 Partial-Day Classification Counts and Truck Traffic Distribution Factors In order to classify vehicles reliably on the basis of axle-spacing criteria, automatic vehicle clas- sifiers must be located where vehicles are neither accelerating nor decelerating and where the spacing between vehicles is sufficient to allow consecutive vehicles to be readily distinguished. Because these conditions are difficult to meet in urban areas, urban classification counts fre- quently are collected manually. (Alternatively, classification on urban streets and roads may be limited to length classification.) Manual classification counts are usually collected only dur- ing daylight hours, usually for a period of 6 to 12 consecutive hours. The first step in using the resulting partial-day classification counts is to convert each of these counts to estimates of volume by VC for the day in which the count was collected. For this pur- pose, TrafLoad uses a time-of-day factoring procedure that uses hourly traffic ratios (or, as they are called in Part 2, “hourly fractions”) that are analogous to the monthly and day-of- week traffic ratios used by other TrafLoad factoring procedures.3 This factoring procedure was chosen over the more commonly used truck traffic distribution factor (TTDF) procedure because the latter procedure produces upward biases in truck volume estimates for sites at which most truck traffic is “business-day” truck traffic—a characteristic of most urban sites. This bias is discussed below. The TTDF procedure uses a partial-day classification count as the basis for distributing a machine count of total daily traffic among VCs. The procedure implicitly assumes that this distribution is approximately the same during the daytime hours when manual counting is performed as it is during the remainder of the day. However, this assumption does not hold for most urban sites. At these sites, most truck traffic occurs on weekdays between 6:00 a.m. and 6:00 p.m. Figure 4.1a shows how weekday truck and automobile volumes are distributed over a 24-hour day on a typical urban street; and Figure 4.1b shows how the percent trucks (i.e., the percentage of vehicles that are in FHWA Classes 4–13) varies by hour of day. 1-15 3 See Part 2, Section 3.4.

1-16 Figure 4.1a Typical Time-of-Day Patterns for Urban Sites at which Most Trucks are Business Day Trucks Figure 4.1b Percent Trucks by Hour of Day 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour of Day 1 2 3 4 5 6 7 8 9 Trucks Autos Percent 1 2 Derived from data for urban other principal arterials (Functional System 14) in Mark Hallenbeck et al., Vehicle Volume Distributions by Classification, Chaparral Systems Corporation and Washington State Transportation Center, June 1997, for FHWA, FHWA-PL-97-025, pp. 79-80. 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hour of Day 0 Percent 2 4 6 8 10 12 14 Percent Trucks Daily Average

1-17 Table 4.1 Truck Traffic Percentages Derived from Partial-Day Weekday Classification Counts Collected for Various Periods of Time Hours Counted Period Counted Percent Trucks 6 6:00 a.m. - Noon 8.7 8:00 a.m. - 2:00 p.m. 10.1 10:00 a.m. - 4:00 p.m. 9.3 Noon - 6:00 p.m. 7.35 8 8:00 a.m. - 4:00 p.m. 9.4 10:00 a.m. - 6:00 p.m. 8.1 10 8:00 a.m. - 6:00 p.m. 8.4 12 6:00 a.m. - 6:00 p.m. 8.0 24 7.0 Source: See Figure 4.1. TTDFs derived from counts collected during periods when truck percentages are high (par- ticularly between 9:00 a.m. and 3:00 p.m.) and excluding periods when these percentages are low (particularly after 5:00 p.m.) will produce overestimates of daily truck traffic. Table 4.1 shows how the choice of a time period for collecting partial-day classification counts affects the resulting estimates of the overall percentage of trucks. The table shows the expected value of the estimated percent trucks that will be obtained if classification counts are allocated for various daytime periods. All periods shown result in upwardly biased estimates. The small- est upward bias shown in the table (5 percent) occurs for counts taken between noon and 6:00 p.m.; the largest (44 percent) occurs for counts collected between 8:00 a.m. and 2:00 p.m.  4.3 Simple versus Weighted Averages In several situations, traffic data analysis software, such as TrafLoad, is required to average data obtained from several sites in a group. An example is the averaging of factors or traffic ratios obtained for continuous-count sites belonging to a factor group. Any of these averages can be obtained as • A simple average of the values obtained for each site in the group; • A weighted average of these values, using as weights the volume of relevant vehicles observed during a specified time period; or • A weighted average of these values, using as weights the total number of relevant vehi- cles observed at each site.

The first option has the advantage of computational simplicity.4 The second and third options have the advantage of applying low weights to values obtained from sites at which few vehi- cles belonging to the relevant VC(s) were observed. The second and third options thus mini- mize the effects of potentially misleading values derived from small sample sizes. The second option has the further advantage of weighting the values obtained for each site in approxi- mate proportion to the relative volume of vehicles in the relevant class(es) at each site. Consider a group that consists of several sites for which some type of data (such as load spec- tra or monthly and annual traffic volumes) are available and a larger number of sites for which these data are not available. The goal of the averaging process is to use data from the first set of sites to produce values (for load spectra, monthly traffic ratios, etc.) that are reasonably representative of the (unknown) values that exist at the second set of sites. In the case of TWRGs, the goal is to use load spectra developed from site-specific WIM data collected at several sites in order to produce a set of “average” load spectra that produces pavement loads that are rea- sonably representative of the loads that exist at sites that are in the group and for which site- specific data are not available. In the case of TWRGs, the research team originally hypothesized that weighted averages would work somewhat better than simple averages. However, a study for FHWA5 found that simple averages produced somewhat better results than weighted averages.6 This result appears to be caused by a significant positive correlation between the volume of trucks in a particular VC operating at a site and the average loads of these trucks. Because of this correlation, weighted averages generally produce higher estimates of average pavement load per vehicle than do simple averages. But the highest traffic volumes and truck loads exist only at a few sites, leading to estimates of average pavement load that overestimate loads at most sites.7 For this reason, TrafLoad was designed to use simple averages when developing a set of load spec- tra for a TWRG. 1-18 4 Although the derivation of weighted averages usually is more complex than the derivation of simple averages, the difference in complexity is sometimes quite small. For example, traffic ratios for a group of n sites are obtained as 5 Cambridge Systematics, Inc. Accuracy of Traffic Load Monitoring and Projections, Volume II, February 2003, pp. 16-18. 6 The measure of pavement stress used in this study was equivalent single-axle loads (ESALs). How- ever, the result is likely to hold for load spectra as well. 7 The analysis only used data from continuous WIM sites. Since these sites are likely to have higher- than-average truck volumes, the actual tendency for weighted averages to overestimate average pave- ment load per truck is likely to be even stronger than indicated by the analysis. 1 1 1 1 n j mj n j j n j j n MADT AADT using simple averages, and as MADT AADT using weighted averages.- - = ∑ ∑ ∑

Most other averaging performed by TrafLoad is also performed using simple averages. In par- ticular, the factor-group averages of monthly, day-of-week, and hourly traffic ratios are obtained as simple averages. Similarly, TrafLoad uses monthly and day-of-week adjustment factors to modify the load spectra obtained for 1 month (or day of week) to be more repre- sentative of another month (or day of week);8 when the resulting “load spectra adjustment fac- tors” are averaged, simple averaging is used. There are, however, two situations in which TrafLoad generally obtains weighted averages. One is the development of a set of statewide default load spectra to be used by the Design Guide software when designing pavement for any site that has not been assigned to a TWRG and for which site-specific load spectra are not available. These statewide default load spec- tra are obtained by TrafLoad as averages of the TWRG load spectra. For this purpose, the TrafLoad user is allowed to assign weights to the TWRGs so that, in this averaging process, the load spectra for TWRGs that are relatively large or are believed to be particularly repre- sentative of statewide conditions can be weighted more heavily than the load spectra for other TWRGs. The second situation in which TrafLoad uses weighted averages is the development of day- of-week adjustment factors to be applied to the load spectra. The procedure used for devel- oping these factors uses truck volumes by day of week as weights in the averaging process.9 The process used is designed to develop average monthly load spectra that are not unduly influenced by load characteristics observed on days (usually weekend days) when truck vol- umes are low. 1-19 8 See Part 2, Section 2.2. 9 Part 4, Section 3.3, Step WB, available online at http://trb.org/news/blurb_detail.asp?id=4403.

1-20 5.0 Areas for Future Work The first section of this chapter presents brief discussions of some areas for future research relating to the collection and analysis of traffic data. The second section discusses several potential improvements to TrafLoad.  5.1 Areas for Future Research The following sections briefly discuss some areas for future research relating to the collection and analysis of traffic data to be used in pavement design. The discussion excludes most research into improvements in WIM equipment and WIM technologies, although continued improvement in those areas is extremely important for improving the quality of data used in pavement design. The one exception to this exclusion is research into the accuracy and relia- bility benefits of multi-sensor WIM systems discussed immediately below. Calibration of WIM Equipment There is a clear need for research into better procedures for calibrating WIM equipment and for maintaining calibration over time. In particular, both the Pavement Design Guide proce- dure and TrafLoad’s load spectra factoring procedure place substantial reliance on maintain- ing WIM calibration over time. In the case of the Design Guide procedure, temporal variations in WIM calibration may result in monthly load distribution factors that reflect the effects of variations in calibration rather than variations in axle load, thus compromising the use of these distribution factors in pavement design. In the case of TrafLoad, these temporal variations have the further effect of producing seasonal adjustments to the load spectra for an individ- ual site that may be more influenced by variations in calibration than by actual variations in axle loads, compromising the effect of the seasonal adjustments in estimating annual average load spectra. For these reasons, further research into improving WIM calibration is strongly recommended. In particular, research should assess the costs and benefits of using multiple sensors, both as a means of calibrating a WIM installation in a manner that is not overly influenced by the dynamic characteristics of one or two test trucks and as a means of collecting more accurate axle-load information. A clear set of standards should be developed to ensure that, if auto- calibration procedures are used for a particular WIM site, they are used only after the initial cal- ibration of the WIM equipment and only after the autocalibration procedures themselves have been calibrated to reflect site-specific characteristics (such as front-axle weight) of vehicles observed at the site. Finally, there should be testing, validation, and refinement of the process being developed by the Long-Term Pavement Performance Project for using the results of his- torically collected data as a means of identifying WIM scales that are in need of calibration.

1-21 Factoring Procedures for WIM Data TrafLoad uses traffic ratios to convert classification counts into estimates of AADT by VC in a manner that attempts to minimize the effects of monthly and day-of-week (DOW) variations in the volume of vehicles in several VC groups. Similarly, TrafLoad develops sets of load spec- tra adjustment factors that are used to modify load spectra collected on certain days of the week and/or certain months so that they better reflect the overall pavement-damaging effects of loads traversing a site on an annual basis or during a particular month. This load spectra factoring procedure is a generalized version of the ESAL factoring procedure described in Chapter 2.0 that was developed and tested as part of a recent study for FHWA.1 These two procedures are the first procedures developed for adjusting WIM data for the effects of sea- sonal and DOW variations in axle loads. Tests of the ESAL factoring procedure (see Table 3.1 of Part 1) indicate that the procedure pro- duces moderate improvements in the resulting estimates of pavement damage, and the research team believes that the same is true of the load spectra factoring procedure that is incorporated into TrafLoad. However, the latter procedure has not been subjected to system- atic testing. Also, the benefits of both procedures are reduced if the seasonal adjustments are derived from data collected at sites at which the WIM equipment has not been consistently calibrated over the 12-month collection period. (In the research team’s previous research, the team was able to identify and eliminate a few sites at which the calibration changed signifi- cantly, and the research team assumed that calibration at the remaining sites was reasonably consistent over the year.) TrafLoad’s load spectra factoring procedure also warrants further review. In developing this latter procedure, the research team did not have the time or resources needed to consider pos- sible variants of the procedure. In particular, the TrafLoad factoring procedure is based on the earlier ESAL factoring procedure. Accordingly, as the basis for all factoring, it uses the fourth root of average ESALs per vehicle by vehicle-class group. A possible alternative to the current TrafLoad procedure would use average weight by axle- group type and vehicle-class group. This alternative would be somewhat easier to understand than the current procedure, and it would likely allow for a moderate reduction in computa- tion time. However, implementation of this alternative procedure would require a moderate programming effort as well as development of more complete specifications of the procedure. Additional development and testing of procedures for factoring load spectra (and other WIM data) is warranted. One possible line of research would involve implementation of the average- weight factoring procedure and comparing its performance to the current TrafLoad procedure. Development of TWRGs Both the Pavement Design Guide procedure and the earlier ESAL-based procedure presume a good understanding of the axle loads that will be incurred by the new pavement. This 1 Cambridge Systematics, Inc., Accuracy of Traffic Load Monitoring and Projections, Volume I, Chapter 5, and Volume II, Section 5.1, prepared for FHWA, February 2003.

1-22 understanding can best be obtained from site-specific data collected by well-calibrated WIM equipment using in-pavement sensors—an expensive and relatively time-consuming option. A less expensive option is to use as defaults data from permanent WIM installations at other sites at which the axle loads (ESALs or load spectra) are believed to be similar to those occur- ring at the site for which pavement is being designed. For this purpose, the research team’s procedures (as well as those recommended by the 2001 TMG) require the states to develop a set of TWRGs and the research team provides some simple recommendations for forming these TWRGs (see Part 2, Section 2.4). However, testing of these procedures2 indicates that the resulting default values obtained from the resulting TWRGs are only moderate improvements over those that would be obtained from statewide data and they are appreciably poorer than values obtained from site-specific WIM, even if the site-specific data are collected for only a short period of time. Clearly, further research would be desirable into the development and use of TWRGs. Seasonal Variations in Truck Weights Analysis3 indicates that in California the weights of combination trucks generally are higher during the spring and summer than during the fall and winter. However, there is relatively little published information on the weight patterns in other states or on how these patterns may vary with functional system or location. Such information would provide an improved understanding of the value of TrafLoad’s seasonal adjustments for load spectra and would be helpful in the review and interpretation of WIM data. More information on day-of-week vari- ations in weights (which can be quite significant for single-unit trucks) would also be useful. Sensitivity Analyses of Designs Produced by the Pavement Design Guide The Pavement Design Guide requires an extensive amount of environmental and traffic data, but there is little published information on the relative importance of the many variables to the resulting pavement designs. A good set of sensitivity analyses would provide a better understanding of the extent to which the collection of good traffic data is warranted. These analyses could also provide the basis for modifying procedures for collecting traffic data so as to improve the resulting estimates of the input variables to which the Pavement Design Guide procedures are most sensitive. Averaging Procedures In several situations, systems for analyzing traffic data create averages of data obtained from a specified group of sites, such as a factor group. These averages may be developed as simple averages of the values obtained for each monitored site in the group, or they may be obtained as weighted averages of these values, using the volume of relevant vehicles observed at each site 2 Ibid., Volume II, Section 4.4. 3 Ibid., Volume II, Section 5.2.

as weights. The latter option is preferable if it is believed that values obtained for high-volume monitored sites are more representative of values for unmonitored sites in the group than are values obtained for low-volume monitored sites. Otherwise, the former option is preferable. In one specific case, it has been determined that values from high-volume monitored sites probably are less representative of values from unmonitored sites than are values from low- volume monitored sites. This is the case of TWRGs, in which ESALs or axle-load data from WIM sites in a TWRG are used to provide default values for other sites in the TWRG.4 For most TWRGs, site-specific values of ESALs and axle loads per vehicle for combination trucks tend to be positively correlated with truck volumes, so that weighted averages will produce higher values for ESALs and load spectra than will unweighted averages. However, there is also a tendency to install WIM equipment at sites with relatively high truck volumes. Therefore, unmonitored sites are likely to have somewhat lower-than-average values of ESALs and axle loads per vehicles, and weighted averages are likely to produce poorer load spectra values for a TWRG than are unweighted averages. For this reason, TrafLoad uses unweighted averages for developing a set of average load spectra for a TWRG. TrafLoad also uses unweighted averages in several other cases, including developing sets of average monthly, DOW, or time-of-day traffic ratios for a factor group and averaging the monthly and DOW load spectra adjustments developed from different seasonal load spectra datasets. The use of unweighted averages is simpler, and, for these purposes, unweighted averages are likely to produce results that are at least as good as weighted averages. However, this last assertion has not been tested. Some further evaluation of the effectiveness of weighted and unweighted averages for these purposes may be warranted. Classifying Trucks in Urban Areas Most automatic vehicle classifiers require vehicles to travel at relatively constant speed with clear gaps between vehicles. In urban areas, there generally are few locations at which these conditions exist, particularly during peak periods. Development of classifiers that work bet- ter in urban conditions would allow substantial improvement in the estimates of the numbers and types of trucks operating on urban streets.  5.2 Potential Improvements to TrafLoad TrafLoad is a software system that is designed to analyze traffic data and load spectra and to produce outputs to be used by the Design Guide software. Although several TrafLoad analy- ses are based on analyses performed by existing systems for analyzing traffic data,5 other TrafLoad analyses have been designed to meet the specialized requirements of the Design Guide procedures. Much of the effort expended in the design of TrafLoad was consumed in the design of procedures to perform the latter analyses, and more could have been expended if additional time and resources had been available. 1-23 4 Ibid., Volume II, p. 18. 5 E.g., the TRADAS traffic data analysis system developed by Chaparral Systems.

This section identifies several ways in which the implementation of TrafLoad can be improved in the future. This section is divided into three subsections. The first identifies some steps that can be taken (including some improvements to the user interface) to improve the learning experience for new users. The second subsection lists some recommended improvements to TrafLoad’s functional capabilities that have clear value. These improvements should be imple- mented in the near future. The final subsection lists potential improvements whose value is less clear or whose precise design may be better determined after some experience is gained in using TrafLoad. It is assumed that additional useful improvements will be identified by TrafLoad users in the course of using the system. Making TrafLoad Easier to Learn TrafLoad beta testers have offered some suggestions for making TrafLoad easier to learn: • Develop a Tutorial. The initial steps in using TrafLoad (the Setup and Loading phases described in Sections 2.2 and 2.3) are somewhat involved. A tutorial presentation of these steps would be helpful. • Re-Sequence Menu Items. Some re-sequencing of items on TrafLoad’s menus would be helpful to first-time users. • Assign Sites to Groups. Currently, the Maintain Site Information screen is used for pro- viding TrafLoad with a complete set of characteristics for each site, including its assign- ment to factor groups and to a TWRG. Thus, assignments to groups are performed for each site separately. An alternative would be to use a set of matrices for this purpose, one matrix for each type of group. Each matrix would list all sites in the system and have columns corresponding to all groups of a particular type, e.g., all TWRGs. The user could then identify the sites that should be assigned to a particular TWRG and enter checks in the appropriate columns. This alternative would allow the user to see, on a single screen, how the sites have been grouped and provide visual cues for assigning the remaining sites. A capability for copy- ing the resulting matrices into a spreadsheet (or, possibly, for printing the matrices directly) should also be provided. • Group Test. A group testing session could be attended by potential users from several states as well as by software developers. This would permit the developers to identify sys- tem characteristics that users find confusing. Such a session is likely to enable the devel- opers to identify additional improvements to the interface and/or documentation that would be helpful to subsequent users. Highly Recommended Improvements to Functional Capabilities The following improvements to TrafLoad’s functional capabilities are recommended for implementation in the near future: 1-24

• User-Defined VCs. The initial version of TrafLoad estimates the AADT and load spectra only for FHWA VCs 4–13 and aggregates of these VCs. Several states collect data for more refined sets of VCs. In particular, some states subdivide Class 9 vehicles into tractor semi- trailers with conventional tandem rear axles and semi-trailers with spread tandem rear axles. And some states that allow the operation of longer combination vehicles collect sep- arate data for seven-axle (Rocky Mountain) doubles, nine-axle (“turnpike”) doubles, and triples. Use of this additional information will allow the development of improved esti- mates, for any site, of the numbers of axles of each type and the weights on these axles. • TWRGs and VC Groups. The initial version of TrafLoad requires that the assignment of sites to a TWRG be independent of VC group, i.e., all sites assigned to a particular TWRG for one VC group (e.g., for combinations) are automatically assigned to that TWRG for all other VC groups. This requirement creates an undesirable restriction on the way TWRGs are formed.6 Therefore, it would be desirable to eliminate this requirement. The same undesirable restriction also applies to the assignment of sites to load spectra factor groups. • Load Spectra and “Direction.” Consider a site for which load spectra are available for two design lanes, one for each direction of travel. TrafLoad currently allows the two sets of load spectra to be processed separately, but it does not allow the two lanes to be assigned to separate factor groups or to separate TWRGs. The latter limitation can be significant, since there are many sites at which vehicles are loaded more heavily in one direction than in the other. It would be desirable for TrafLoad to allow two design lanes at a site to be assigned to TWRGs and seasonal load spectra factor groups independently of each other.7 • Quality Control Checks. The software assumes that the vehicle classification and weight records have been through a quality-control process. This assumption may be restrictive in many cases. Some legacy systems may do little, if any, checking of the incoming data. Addition of quality-control checks would be an economical enhancement for users who have such systems. The checks to be implemented should include at least the type and range of checks used in the FHWA Vehicle Traffic Information System (VTIS) (http://www.fhwa.dot.gov/ohim/ohimvtis.htm). In addition, some of the checks from the traffic data editing pooled fund study might be a useful addition. The quality checks should produce a statement of the number of data records read into the system and the number rejected because of the quality checks. With some additional design and coding, an option could be provided to reject an entire data file if the number of rejected records exceeds a user-specified threshold. • Traffic Growth Between Base Year and Year in Which Pavement Will Be Improved. It would be desirable to provide TrafLoad users with a simple facility for telling TrafLoad the number of years that will pass between the last year for which traffic data have been 1-25 6 For example, consider the sites at which the most heavily loaded combinations operate. When look- ing at combination trucks, these sites are appropriately assigned to a single TWRG. However, the same degree of uniformity does not necessarily exist for the weights of single-unit trucks (SUTs) operating at these sites. If SUTs at some of these sites are heavy and SUTs at other sites are light, it would be desirable to assign these sites to separate TWRGs when analyzing SUTs. 7 In the absence of this improvement, it is possible to treat the two design lanes as belonging to sepa- rate sites (with separate site identification numbers).

collected (e.g., Design Guide) and the year in which redesigned pavement is expected to go into service (e.g., 2005). With this information, TrafLoad would be able to generate fore- casts that automatically take traffic growth during this period into account, eliminating the need for a work-around for this limitation. Other Potential Improvements Other potential improvements to be considered for future implementation include the following: • User-Defined VC Groups for Forecasting. TrafLoad currently allows users to specify one set of forecast information for VCs 4–7 and a separate set for VCs 8–13. It would be desir- able to allow users to define their own VC groups for this purpose. • Allow Form of Forecast to Vary by VC Group. TrafLoad currently allows users to spec- ify exponential growth for all classification VC groups or linear growth for all such groups, but not exponential growth for one VC group and linear growth for another. It would be desirable to remove this restriction. • AADT by VC for Level 3A Sites. TrafLoad currently estimates AADT by VC and direc- tion at Level 3A sites by using estimates of AADT by VC and direction at a nearby “asso- ciated” site on the same road. A somewhat better procedure for developing estimates for Level 3A sites is described in Part 2, Section 3.5. • Monthly Distribution Factors (MDFs) for Level 1 Sites. TrafLoad could be modified to produce separate MDFs by lane (or by direction) for Level 1 sites (but not for Level 2 sites). Such MDFs might be marginally better than the non-directional MDFs that are currently developed for these sites. • “Direct Scaling” by Direction. TrafLoad has an optional “direct scaling” procedure that may be used (at the user’s request) for estimating AADT by VC at Level 1B sites that have only one lane per direction.8 The scale factors used by this procedure are developed by lane. A variant of this procedure would develop scale factors by direction and could be applied at multi-lane Level 1B sites. • DOW ESAL Ratios. TrafLoad’s procedure for adjusting load spectra to reflect DOW vari- ations in axle weights (for a given VC group) uses a set of DOW load spectra ESAL ratios.9 For any site, these ratios are developed for any month for which WIM data exist for each day of the week. There is at least one relatively unusual situation (when data are available for only 1 week and that week contains a holiday) when the resulting ratios might produce relatively unreliable adjustment ratios. It would desirable for TrafLoad to identify this sit- uation and to ignore the resulting ESAL ratios. Another alternative would be for the annual average load spectra to be developed as a weighted average of the adjusted monthly load 1-26 8 See Part 4, Section 2.2, Procedure CE, available online at http://trb.org/news/blurb_detail.asp?id=4403. 9 See Part 4, Section 3.3, Procedure WB, available online at http://trb.org/news/blurb_detail.asp?id=4403.

spectra using weights that vary with the number of observations for the individual months. Also, rules could be developed that would enable TrafLoad to develop DOW ESAL ratios using data for months for which data exist for some, but not all, of the five weekdays. • Sorting Input Files. TrafLoad’s procedure for loading a set of weight or vehicle- classification records is most efficient when the set is sorted by site. (This is because of problems that arise when a pair of consecutive records contains data for two different sites.) An optional procedure for sorting these records before loading could be helpful to users whose weight and vehicle-classification files are not already sorted. 1-27

1-28 Glossary AADT Annual average daily traffic. AAEPV Annual average ESALs per vehicle. AEPV Average ESALs per vehicle. CV Coefficient of variation. DOW Day-of-week. ESALs (18,000-pound) equivalent single-axle loads. MADT Monthly average daily traffic. MADW Monthly average day-of-week. MAEPV Monthly average ESALs per vehicle. MAPE Mean absolute percent error. SUT Single-unit truck (or bus). TMG Traffic Monitoring Guide. TrafLoad The traffic-data analysis software developed under this project. TTDFs Truck traffic distribution factors. TWRG Truck weight road group. VC Vehicle class. WIM Weigh-in-motion.

Levels of Classification Site 1A Site for which AVC data are available for periods of at least 1 week for at least 12 consecutive months. 1B AVC site that is reasonably near a Level 1A site on the same road. 2A Site for which an AVC count is available for a period of at least 48 hours. 2B Site for which a manual classification count for a minimum of 6 weekday hours is available. 3A Any other site for which volume counts are available and that is on the same road as a Level 1 or 2 site. 3B Any other volume-count site. 1-29

1-30 Levels of WIM Site 1 Site for which site-specific WIM data are available. 2 Non-Level 1 WIM sites that have been assigned to a TWRG. 3 All other WIM sites.

1-31 Appendix A: Forecasting The Pavement Design Guide software requires forecasts of linear or exponential rates of change in truck volumes over the design life of the pavement. For this purpose, TrafLoad allows the user to provide these rates of change separately for single-unit trucks (SUTs) and for combination trucks (CTs). A simple procedure for estimating these rates is presented in Section 3.6 of Part 2. This procedure is recommended for use by most users of TrafLoad. This appendix presents a more extensive discussion of potential forecasting procedures that was developed in Phase I of the current project. The simplest approach to forecasting truck traffic for a particular road is to estimate the aver- age percentage rate of past growth in truck vehicle-miles of travel (VMT) on a set of similar roads and to assume that truck traffic on the road in question will grow at this rate in the future. A minor variant of this procedure is to adjust the estimated rate of growth to reflect information that suggests that truck traffic on the road is likely to grow faster or slower than implied by the estimated growth rate (e.g., because of the expected opening or closing of a major generator of truck traffic). For most pavement-design efforts, this approach is likely to be the most cost-effective to use. Because of differing influences on the use of SUTs and com- binations, the procedure should be applied separately to SUTs and combinations (or to small and large trucks), but the procedure can also be applied to total truck volumes to produce a single growth rate for all trucks. Section A.1 describes the above approach in some detail. Also described in that section is a related approach in which a linear trend in truck VMT is estimated and the road in question is assumed to receive a proportional share of this linear growth. Section A.2 discusses some more sophisticated procedures that can be used for forecasting truck traffic. These procedures may be used by highway planners to estimate traffic on a pro- posed new road or to evaluate the likely effects of other significant changes in the highway network. When forecasts have been developed for such purposes using these procedures, they may also be used for pavement design. Section A.3 discusses issues relating to forecasting changes in the distribution of vehicles in use, as characterized by their axle configuration, and the load spectra of these vehicles. TrafLoad accepts separate forecasts of SUTs and CTs. Nearly all SUTs are used primarily in local service, and changes in SUT volumes tend to closely relate to changes in the local econ- omy. On the other hand, many CTs are used for longer hauls, so changes in CT volumes tend to be influenced by a mix of local and non-local economic factors. Accordingly, in many areas, SUT traffic and CT traffic exhibit different growth rates. For this reason, it usually will be bet- ter to develop separate growth rates for SUTs and CTs. However, for conciseness, the fore- casting procedures presented in this appendix frequently refer simply to truck volumes and truck AADT. It should be borne in mind that these procedures can, and usually should, be applied separately to SUT volumes and CT volumes.

1-32  A.1 Simple Trend Analysis Trend analysis is a methodology for forecasting future truck volumes that relies solely on his- torical estimates of truck volume or truck AADT. While it is theoretically possible to develop trends of volumes from only 2 years of data, more data points are desirable so that the impacts of non-representative data points can be minimized. Trend analysis formulas are very simple to develop and apply and can be contained in a single spreadsheet. A.1.1 The Procedures The simplest form of trend analysis is a linear regression method that forecasts future truck volumes based solely on historical truck volumes, developing a trend line of volumes into the future. This method can be used to estimate either the annual change in truck volume (linear trend) or the annual percentage change in truck volume (exponential trend). Exponential trends incorporate the compounding effect of growth and generally are the more appropriate assumption for truck growth. Because of the compounding effect, exponential trends gener- ally produce higher forecasts. For roads that have experienced high growth rates, the differ- ences in forecasts between the two methods can be significant. A linear trend analysis uses the formula (A.1) where: Vi is the i th observation of the dependent variable (to be predicted); Yi is the i th observation of the independent variable (explanatory); and a and b are parameters to be estimated by linear regression in a manner that minimizes εi (the error term). In this analysis, Vi could be annual VMT of SUTs or CTs in the state or in the region of inter- est, and Yi could be the corresponding calendar year of the observation. An exponential trend analysis incorporates a compound annual growth rate and follows the formula (A.2) where r is the annual rate of growth and the other variables are defined above. Equation A.2 is estimated by first taking logarithms of both sides of the equation to produce (A.3) and then using linear regression. Log Log LogV a Y ri i( ) = + ∗ +( ) ( )1 V a ri Yi = +( )1 V a b Yi i i= + ⋅ + ε

1-33 As implied above, truck forecasts generally are developed using data on total truck VMT on a set of roads, rather than truck AADT on the road in question. Although the latter option may seem appropriate, its use poses several problems. In particular, data on truck volumes on a given road usually are not collected annually, and even if they are, the factoring process used to convert short-duration truck counts to estimates of truck AADT introduces artifact into the resulting time series that adversely affects the regression results. When an exponential trend analysis is performed using truck VMT, the growth rate that is estimated for VMT on an entire set of roads usually is assumed to be valid for any road in the set. However, as discussed below, if there is a good reason to believe that the future growth rate for truck traffic will differ from the past rate, or that the growth rate for the road of inter- est differs from that of the entire set of roads, it may be desirable to adjust the rate judgmen- tally before applying it. When an estimate of linear growth in VMT on a set of roads is developed, this estimate has to be scaled before it can be used to forecast truck AADT on a particular road. This scaling can be performed using the equation (A.4) where: To is AADT of SUTs or CTs in the last year for which historical data are available (usually also used as the base year for forecasts), Vo is the corresponding value of VMT in that year, b is the linear growth rate for VMT estimated in Equation A.1, and g is the resulting estimate of annual growth in AADT of SUTs or CTs on the road in question. As observed above, there are some circumstances in which a small judgmental adjustment to an estimated growth rate may be appropriate. Some circumstances that may warrant such an adjustment are the expected diversion of existing truck traffic to a new road that is being built or the expectation that new industrial facilities being built on the road will accelerate the growth of truck traffic on the road. In making such adjustments, analysts should bear in mind that upward adjustments will produce more conservative (and expensive) pavement designs with a reduced likelihood of premature failure, while downward adjustments will produce less conservative designs with an increased likelihood of premature failure. For this reason, downward adjustments should be made only with great care. In performing a trend analysis, the historical data used should be selected to provide a rea- sonable indicator of likely future growth on the road in question. The beginning and end years of the historical data should be selected to be in corresponding phases of the business cycle. Starting the time series in a recession year and ending it in a boom year will result in over- estimating likely future growth, while starting in a boom year and ending in a recession will have the opposite effect. g T V bo o =

1-34 The historical data used in a trend analysis should be plotted and examined to ensure that they exhibit a relatively steady growth rate over time. If the year-to-year changes appear erratic, then the assumption underlying the simple procedure—a relatively constant growth rate over time—is called into question. In addition, the analyst should examine the plotted data to deter- mine outlier observations that differ significantly from the trend of other observations. A.1.2 Examples Data for two numeric examples of trend analyses are presented in Table A.1. The third column of the table shows assumed estimates of statewide VMT of CTs for 1992 through 1999, and the fourth column shows the natural logs of the VMT estimates. (Logs to the base 10 could also be used.) No VMT estimate is shown for 1996, reflecting an assumption that data for this year are unavailable or have been found to be unreliable. The following shows first how the above procedures can be used to forecast statewide VMT of CTs and then how these results can be used to estimate the AADT of CTs (AADTc) on a par- ticular road. Most commercial spreadsheet programs incorporate a linear regression function. Applying such a function to the second and third columns of Table A.1 produces the equation where: V is estimated VMT of CTs, in billions; and Y is the number of years since 1992, the year of the first observation. V Y= +25 38 0 4085. . Table A.1 Historical Truck Volumes Year of Observation Number of Years Since First Observation Truck VMT (Billions) ln (Truck VMT) 1992 0 25.61 23.97 1993 1 25.24 23.95 1994 2 26.23 23.99 1995 3 27.42 24.03 1997 5 26.76 24.01 1998 6 27.33 24.03 1999 7 28.85 24.09

1-35 The regression output generally also contains statistical variables that indicate how well the equation explains the observed data. For the example, in addition to the values of the coeffi- cients, the output indicates that the R2 (which indicates the portion of the total variation in the observation that is explained by the equation) is 0.76 and that the t-statistics are 60.4 for the constant term and 4.09 for the variable term. (A t-statistic greater than 1.96 indicates that the coefficient is statistically significant to a confidence level of 95 percent.) Using this result, fore- cast VMT in 2022 is estimated as The same data can also be used to estimate an exponential trend or growth rate. Here the regression is performed using the natural log of truck VMT as the dependent variable and the number of years from 1992 as the independent variable. In this case, the resulting coefficients in the output are also natural logs of the desired variables and are transformed to the desired terms by using the exponential function, ex or exp(). The estimated equation is Exponentiating both sides of the equation produces which can then be simplified to where Y is the number of years from 1992, the first observation. The estimated annual growth rate is 1.52 percent (1.0152 − 1.0). The R2 for this regression is 0.77, and the t-statistics are 1,548 for the constant term and 4.13 for the variable term. For the purpose of forecasting the AADT of combination trucks (AADTc) on a single road, the exponential growth rate (1.52 percent per year) can be applied directly to AADTc for any base year for which AADTc is available. Thus, if AADTc for a given road is estimated to be 1,000 in 1999, then the forecast for any subsequent year is where n is the number of years between 1999 and the year of interest. For 2022, the forecast is When using TrafLoad, it is only necessary to specify that 1.0152 is the estimated exponential growth ratio for the road. On the other hand, if linear growth is assumed, Equation A.4 must be used to convert the above estimate of linear growth in statewide VMT (408.5 million VMT per year) to an estimate of the linear growth rate for the road in question. Substituting this value in Equation A.4, along with the 1999 estimates of AADTc and statewide VMT of combinations, produces a linear growth rate of AADTc( ) , . ,23 1 000 1 0152 1 41523= × = AADTc nn( ) , .= ×1 000 1 0152 V Y= ×25 4 1 0152. . V Y= exp( . )exp( . )23 96 0 0151 ln(V) 23.96 0.0151Y= + 25 38 30 0 4085 37 64. . .+ × = billion

1-36 The forecast for any subsequent year is where n is the number of years between 1999 and the year of interest. For 2022, the forecast is as follows: This value is appreciably lower than the 1,415 produced using an exponential growth rate. When using a linear growth rate with TrafLoad, the software requires only this growth rate, 14.17 CTs per year.  A.2 More Sophisticated Approaches This section describes three more sophisticated approaches for forecasting truck volumes: • Multivariate linear regression, • Growth-factors methods, and • Travel demand models. The first of these is a somewhat more sophisticated alternative to the univariate regression analysis procedure presented in the preceding section. Multivariate linear regression may be a useful alternative to the earlier procedure when forecasting truck traffic in areas where there are identifiable factors other than time that influence truck volumes. The second and third approaches, and particularly the third, are appreciably more complex and generally not warranted for forecasts that will be used only for designing pavement. How- ever, these approaches may be of interest to highway planners who are developing forecasts to be used for multiple purposes. In particular, the third approach (the development of travel demand models, or TDMs) is commonly used for estimating the traffic diversion effects of the construction of a new road or of other significant changes in an area’s highway network. Pave- ment designers frequently will be able to use either existing forecasts developed using these approaches or others. In the case of a planned new road, a TDM may be the best source of a forecast. The three approaches are described in the first three subsections below. A fourth subsection then describes several potential sources of data that may be used with these approaches. Additional AADTc( ) , . ,23 1 000 14 17 23 1 326= + × = AADTc n n( ) , .= +1 000 14 17 1 000 0 4085 10 28 85 10 14 17 9 9 , . . .× × × = CTs/year

1-37 information about potential data sources is contained in Appendix A of NCHRP Report 3881 and in Appendixes G–L of the Quick Response Freight Manual.2 The approaches presented in this section generally produce forecasts for a specified future year. For use by TrafLoad, these forecasts must be converted to estimates of either annual growth rates or constant annual growth. The conversion is described in the last subsection below. A.2.1 Multivariate Linear Regression Multivariate linear regression allows the use of multiple independent variables. This approach is useful if there are multiple independent factors that are believed to cause growth or fluctu- ations in truck traffic. In agricultural areas, one such factor would be the size of the harvest. In general, independent variables should be chosen that represent a set of significant and dis- tinct influences on truck traffic for which historical data exist and for which reasonable fore- casts are available. If influences are not distinct, the variables will be correlated, the regression analysis will not be able to distinguish the influences, and the procedure will produce unreli- able coefficients. Population, employment, personal income, and time are potential indepen- dent variables that tend to be correlated with each other and that generally should not be used as separate independent variables in an ordinary least-squares regression (though it may be possible to use them together when more sophisticated techniques, such as two-stage least- squares regression, are used3). For this regression, R2 is 0.85, and the t-statistics are 2.68 for the year, 1.54 for grain, and 10.8 for the constant term. An example of a simple multiple variable regression using a trend term and an additional independent variable, grain production, uses the data in Table A.2. Using a spreadsheet regression function, one obtains the estimated equation where: V is estimated VMT of CTs, in billions; Y is the number of years since 1992, the year of the first observation; and G is grain production in billions of bushels. V Y G= + +22 27 0 3018 2 15. . . 1 Cambridge Systematics, Inc., et al., NCHRP Report 388: A Guidebook for Forecasting Freight Transporta- tion Demand, Transportation Research Board, National Research Council, 1997. 2 Cambridge Systematics, Inc., COMSIS Corporation, and University of Wisconsin-Milwaukee, Quick Response Freight Manual, prepared for the U.S. DOT and U.S. EPA, September 1996. 3 E.g., see Peter Kennedy, A Guide to Econometrics, Fourth Edition, MIT Press, 1998. Additional discussion of econometric techniques for forecasting truck traffic is also contained in NCHRP Report 388.

1-38 If there is a forecast for grain production in 2022 of 2.1 billion bushels, the forecast of CT VMT in that year will be The use of a separate grain-production term in this example makes it possible to distinguish the effects of growth and fluctuations in grain production from other influences that cause truck VMT to trend upward. The assumed data indicate that the 1993 harvest was poor and the 1999 harvest was particularly good. The indicated sharp growth in grain production over the 7-year period appears to be responsible for a significant share of the observed VMT growth. With a separate grain term, the estimated coefficient of Y (0.3018) is appreciably lower than the 0.4085 value obtained in the earlier example without a term for grain. With a fairly modest forecast for future growth in grain production, the resulting VMT forecast for 2022 (35.8 billion) is somewhat lower than the forecast that was produced using a simple linear trend (37.6 billion). Conversion of this forecast to an estimate of either linear annual growth or an exponential annual growth rate is discussed in Section A.2.5. A.2.2 Growth-Factor Methods Growth-factor methods use forecasts of growth in specific economic sectors as indicators of corresponding growth in related truck traffic. For this purpose, the most desirable indicator variables are those that measure goods output or demand in physical units (tons, cubic feet, etc.). However, forecasts of such variables frequently are not available. More commonly avail- able indicator variables are constant-dollar measures of output or demand; employment; or, for certain categories of truck traffic, population or real personal income. 22 27 30 0 3018 2 15 2 1 35 8. . . . .+ × + × = billion Table A.2 Historical Truck Volumes and Grain Production Year of Observation Number of Years Since First Observation Truck VMT Billions Grain Production (Billion Bushels) 1992 0 25.61 1.365 1993 1 25.24 1.643 1994 2 26.23 1.456 1995 3 27.42 1.756 1997 5 26.76 1.462 1998 6 27.33 1.626 1999 7 28.85 1.986

1-39 The Procedure There is considerable flexibility in determining how to determine growth factors, but the basic procedure is as follows: 1. Select the commodity or industry groups that will be used as the economic indicators in the analysis. Obtain base-year and forecast-year estimates for the selected indicators. The groups should be selected to distinguish the most important commodities carried on the road, but should be aggregated to reflect a manageable number of groups for analysis purposes. Generally 10 or fewer commodity/industry groups are appropriate. Forecasts of dollar-valued output or commodity production can be obtained from the U.S. Depart- ment of Commerce’s Bureau of Economic Analysis (BEA), from input/output models, from state agencies or other public forecasting groups, and from private vendors. Fore- casts of employment can often be obtained from state labor departments and from private vendors.4 2. Allocate the base-year truck volume to commodity/industry group. A key to growth-factor forecasting is determining how the truck traffic is allocated among the different commodity or industry groups in the base year. One method for allocating truck traffic to different commodity or industry groups is to estimate VMT for each group. The principal source of information for this is the Vehicle Inventory and Use Survey (VIUS) dataset.5 However, VIUS is an imperfect source of information about truck opera- tions in a particular state, because state-level VIUS data reflect all operations of trucks based in that state (including out-of-state operations) and excludes all in-state operations of trucks based elsewhere. An alternative is to obtain this information from a roadside survey. In addition to provid- ing data on the commodity or industry group for each vehicle, surveys can provide infor- mation on trip origin and destination, vehicle routing, trip frequency, etc. These surveys are not routinely available for roadway sections, and the expense of conducting these stud- ies would not be justified for pavement design projects. The conduct of such surveys is dis- cussed briefly in Section A.2.4 (below) and discussed further in the Quick Response Freight Manual.6 3. Determine the growth factor for each commodity or industry group. For each commodity/industry group, an annual growth factor, gi, is obtained from the equation (A.5)g X Xi if oif io Y Y =     −     1 4 These data sources are described further in Section A.2.4. 5 VIUS is described further in Section A.2.4. 6 Chapter 6.

1-40 where Xio and Xif are the values of the ith indicator variable in the base year, Yo, and in some future year, Yif, respectively, for which a forecast for this variable is available. (Note that the Yif may vary by indicator variable.) 4. Apply the growth factor to the base-year truck volume for each commodity/industry group. The growth factors for each commodity are applied to the truck volumes allocated to each commodity group in the base year to estimate future volumes. The general formula is (A.6) where: Vio is the share of base-year truck volume allocated to indicator variable i; Vif is the forecast of corresponding truck volume in the forecast year; gi is the annual growth factor for this indicator variable (obtained from Equation A.4); and n is the number of years in the forecast period. 5. Aggregate the forecasts across the commodity/industry groups. The forecast-year truck volumes are obtained by summing the truck volume forecasts across all commodity/industry groups. A Numeric Example An illustrative example is shown in Table A.3. The total base-year volume is the 1999 average daily truck volume from the previous examples. The distribution of the base-year traffic was obtained from 1997 VIUS data for six-tire trucks. The annual growth factors for each industry sector are assumed to have been obtained from employment forecasts produced by a local eco- nomic development agency. The growth factors are applied to the base-year truck volumes using Equation A.6, with n, the number of years in the forecast period, set to 23 (2022 − 1999). The annual growth factors in Table A.3 can be converted to annual growth rates by subtract- ing 1.0. Most of the resulting growth rates are higher than the 1.98 percent obtained in the sec- ond example in Section A.1.2; accordingly, they produce a higher forecast of truck traffic: 2,879 trucks per day. Conversion of this forecast to an estimate of either linear annual growth or an exponential annual growth rate is discussed in Section A.2.5. Recent Use by States Brief descriptions of three states’ use of growth factors for truck forecasting are presented below. Section A.2.4 contains additional information about the data sources mentioned. V V gif io in=

1-41 Texas As part of the Texas Strategic Plan,7 the state used growth factors to forecast VMT generated by trucks for several multi-county districts. The TRANSEARCH commodity-flow database was used to develop growth rates of economic activity by industry sector. Growth rates were applied to base-year truck average annual daily traffic (AADT) to estimate the future-year truck AADT by functional class and district. West Virginia West Virginia employs a growth-factor approach to forecast truck traffic. Growth factors are applied to baseline truck traffic data in order to project future truck activities. Growth factors are developed for each county based on socioeconomic forecasts, if the data are available, or historical traffic growth trends, otherwise. Colorado During 1998 and 1999, the Colorado Freight Infrastructure Study8 developed detailed infor- mation on the state’s freight infrastructure at both the state and transportation planning region (TPR) level. This project developed regional- and county-level growth rates for different Table A.3 Growth-Factor Example Commodity/Industrial Category Daily Truck Traffic (Base Year 1999) Annual Growth Factor Daily Truck Traffic (Forecast Year 2022) Agriculture (Farming) 105 1.017 150 Forestry and Lumbering 37 1.059 123 Mining and Quarrying 21 1.031 40 Construction 190 1.016 265 Manufacturing 119 1.031 226 Wholesale Trade 175 1.03 326 Retail Trade 121 1.033 239 Transportation and Public Utilities 723 1.03 1,345 Services 96 1.026 165 TOTAL 1,587 2,879 7 Texas Department of Transportation, Texas DOT Strategic Plan FY 2001–2005. 8 HNTB, Inc., Colorado Freight Infrastructure Study, prepared for the Colorado Department of Transportation, Division of Transportation Development, 2000.

1-42 commodities that could be used for growth-factor forecasting. The 1993 Commodity Flow Sur- vey (CFS) was used to identify statewide commodity flow origins and destinations; the 1992 Truck Inventory and Use Survey (a predecessor to VIUS) was used to develop factors to convert origin and destination commodity flows to truck trips; and the 1995 County Business Patterns (CBP) data (from the U.S. Bureau of Census) were used to allocate the commodity flows and truck trips for both origins and destinations from the state to county level. County-level commodity flows and truck trips were then aggregated to represent TPR activities. Statewide forecasts of employment by industry group were used as the basis for predicting county-level growth factors. A.2.3 Travel Demand Models A well-established and relatively sophisticated technique for forecasting traffic volumes for an entire region is through the use of travel demand models (TDMs). These models are most commonly used for forecasting total traffic volumes in metropolitan areas, but some TDMs have been developed for forecasting truck volumes for a state or region. If separate forecasts of truck volume are not produced by a TDM, future truck volumes can be estimated by mul- tiplying the forecasts of total volume by current percentages of trucks in the traffic stream and adjusting the result upward to reflect the tendency of truck volumes to grow faster than vol- umes of other vehicles. Truck forecasts produced by existing TDMs may be used for the pur- pose of pavement design, but the development of new models is too complex to be justified solely on the basis of their potential use in the pavement-design process. TDMs estimate demand through the use of one or more economic-indicator variables (employment, economic production, etc.) associated with specific units of geography (traffic analysis zones, or TAZs) together with a computerized network representing individual sec- tions of road connecting the zones. In sequence, these models estimate 1. How many truck trips begin and end in each zone (trip generation), 2. How many of those trips travel between each pair of zones (trip distribution), and 3. What specific sections of road are used to travel between zones (assignment). As network models, TDMs are capable of estimating the traffic to be carried by a proposed new road as well as the effects on other roads of diversion to the new road. Users of TDMs should be familiar with several issues: • Travel demand models produce forecasts for various scenarios. When obtaining informa- tion from TDMs, care should be taken to determine exactly which economic and trans- portation scenarios are represented in each forecast. • Travel demand models are validated to groups of highway sections, referred to as screen- lines, not to individual highway sections. When obtaining a forecast, the analyst should obtain the base-year observed truck count, the base-year model volume, and the future- year model volume. Common practice is to assume that the ratio of base-year observed volume to base-year model volume can be used to adjust future-year model volume to pro- duce an appropriate estimate of future-year volume.

1-43 • Travel demand models load truck traffic from a TAZ to the highway network through a centroid connector. The loading of all traffic from an area to the highway network through these discrete points causes high volumes near these points that are an artifact of the model process and do not reflect reality. The use of truck model volumes near these loading points should be done with the assistance of the staff responsible for the model. • While some agencies’ models have specific truck models, many TDMs include trucks only as a percentage of all vehicle trips. The analyst should obtain all relevant information from the model concerning the methodology used for trucks and make any post-processing adjustments that appear necessary. Additional information about TDMs can be obtained from the U.S. DOT’s Travel Model Improvement Program web site (http://tmip.fhwa.dot.gov). Recently Developed TDMs Brief descriptions of several TDMs that have recently been developed by the states are pre- sented below. Section A.2.4 contains additional information about the data sources mentioned. Arizona As part of the Arizona Long-Range Transportation Plan, the state is developing passenger and freight travel demand forecasting capabilities. Base and future estimates of VMT are being developed using available state and regional population, employment, and traffic data. The TRANSEARCH database is being used to identify base-year freight flows, and the VIUS dataset is being used to convert these flows to truck trips. These truck trips are then allocated to the highway network using data from the Highway Performance Monitoring System. Sources of socioeconomic forecasts are being identified as part of the project and are being used to develop forecasts of freight movements. Colorado The Colorado DOT is sponsoring the Eastern Colorado Mobility Study to develop passenger and freight modeling capabilities in the eastern part of the State. The study is using a hybrid modeling approach for trucks, using commodity flow data to develop interregional freight- trip tables to be disaggregated to the zonal level and a more traditional truck-trip generation and distribution process to capture truck traffic generated locally by activities such as service and utility businesses, construction, and local parcel delivery. New Jersey New Jersey includes a vehicle-based truck model as part of the statewide travel demand model.9 The truck model develops truck trips for individual zones using trip rates for zonal 9 URS Greiner Woodward Clyde, Statewide Model Truck Trip Table Update Project, prepared for the New Jersey Department of Transportation, January 1999.

1-44 households and employment by category; estimates of truck trips generated by special gen- erators, such as truck terminals and intermodal facilities; and observed volumes at border crossings. Gravity models are used to distribute medium and heavy trucks separately. Fore- casts for truck trips at major intermodal facilities were developed using forecasts from termi- nal operators, and external station forecasts were developed using trend analysis. Indiana The Indiana statewide freight model10 was developed by Professor William Black at Indiana University in the 1990s. The model was originally developed to provide forecasts of both truck and rail shipments of freight throughout the state and includes zones for each county and 53 external zones representing other states. The model is based on publicly available data from the 1993 Commodity Flow Survey, supplemented with proprietary county-level data from Woods and Poole and the Federal Railroad Administration (FRA) Rail Waybill Sample. Future growth in commodity flows depends on estimates of growth in population and employment within each county or state. Michigan The Michigan statewide truck model11 was developed in the mid-1990s as part of a statewide model for all highway travel. Portions of the truck model were developed at various levels of geographic detail ranging from entire states outside Michigan to commodity analysis zones and TAZs within the state. The model uses REMI model forecasts to project growth in employ- ment for 14 industry groups. Wisconsin The Wisconsin statewide truck model12 was developed in the early 1990s to provide truck vol- ume predictions on state highways and to explore the potential for shifts in the future from truck-only to intermodal (truck/rail) flows for a portion of the present commodity move- ments. The model uses TRANSEARCH as the basis for commodity distribution. Future pro- jections of freight flows were obtained by applying growth factors to individual data elements based on expected changes within Wisconsin in industrial output measures, in employment, and in productivity forecasts by county and industry and on BEA forecasts for other states. Forecasts of future productivity levels were obtained from REMI model forecasts. 10 W. R. Black, Transport Flows in the State of Indiana: Commodity Database Development and Traffic Assign- ment, Phase 2, Transportation Research Center, Indiana University, Bloomington, July 1997. 11 Parsons Brinckerhoff Quade and Douglas, Inc., The Michigan Statewide Truck Travel Forecasting Model, Final Report, February 1996; Parsons Brinckerhoff Quade and Douglas, Inc., Analysis of the 1994 Michi- gan Truck Survey Data, Technical Report, March 1995; Parsons Brinckerhoff Quade and Douglas, Inc., Statewide Travel Demand Model Update and Calibration, Phase II, February 1996, Chapter 3. 12 Wisconsin Department of Transportation, Translinks 21 Technical Report Series: Multimodal Freight Fore- casts for Wisconsin, Draft No. 2, 1995.

1-45 A.2.4 Data Sources The three procedures presented in this section all require some form of economic forecasts; growth-factor methods also require data for allocating base-year truck volumes to industry/ commodity groups; and TDMs have additional data requirements. This section discusses sev- eral useful sources of data. Additional information about potential data sources is contained in Appendix A of NCHRP Report 388. Economic Forecasts The procedures presented in this section all require some form of economic forecasts. Indeed, the availability of appropriate forecasts is an important consideration in the selection of inde- pendent variables to be used in multivariate regressions and economic-indicator variables to be used in growth-factor methods. All states have data centers that maintain economic data, many states also produce statewide employment forecasts,13 and some also have groups (frequently based at universities) that develop more detailed forecasts of employment and/or production by industry. Economic forecasts can also be purchased from several private vendors, including • Global Insight,14 which provides an extensive variety of economic data and forecasts; • Woods and Poole,15 which develops long-term economic forecasts by county; • Regional Economic Models, Inc.,16 which develops regional economic and demographic forecasting models; and • The Minnesota IMPLAN Group,17 which also produces software for economic forecasting. Until recently, the Bureau of Labor Statistics and the Bureau of Economic Analysis produced forecasts of several economic variables by state and industry at 2.5- and 5-year intervals.18 However, these forecasts are no longer available. Allocation of Truck Volumes to Commodities and Industries Vehicle Inventory and Use Survey The Vehicle Inventory and Use Survey (VIUS) is a U.S. Census Bureau dataset collected every 5 years that provides extensive information on truck activity. The VIUS data file includes 13 Links to state websites that provide employment forecasts can be found at http://www.projections central.com 14 http://www.globalinsight.com 15 http://www.woodsandpoole.com 16 http://www.remi.com 17 http://www.implan.com 18 NCHRP Report 388, p. 14.

1-46 information on annual truck VMT by axle configuration, commodities carried, and state in which the trucks are based (but not the states in which they operate). The industries to which the trucks’ owners belong are also identified; however, for many combination trucks, the industry is identified simply as “for-hire trucking.” For these trucks, information about com- modities carried generally is more useful. VIUS data are not reported below the statewide level. Thus, commodity or industry-group esti- mates of VMT that are derived from VIUS could reflect statewide averages, averages for a group of states, or nationwide estimates. If a road carries a significant percentage of long-distance, non- local commodity movements, it may be appropriate to use nationwide data or data for a group of states for estimating VMT. TRANSEARCH TRANSEARCH is a proprietary transportation-related database developed by Reebie Associ- ates (in Greenwich, Connecticut) by combining data from a variety of public sources and sup- plementing these with information from their own surveys, including an annual survey of motor carriers. The data include estimates of tonnage of manufactured goods transported by transport mode and commodity group, between multi-county regions, and, with somewhat less detail and accuracy, between individual counties. Surveys More specific information about trucks operating on a given road can potentially be obtained via a roadside survey. Such surveys are unlikely to be warranted solely for use in the pavement- design process. However, the results of previously conducted surveys may be available for this purpose, and, in some cases, there could be other reasons for conducting a new survey. The conduct of roadside surveys entails both public-sector and private-sector costs. Surveys usually require the active participation of police officers to flag trucks to be surveyed and to ensure the safety of motorists and surveyors alike. Also, surveys can be conducted only at locations that have sufficient space to allow trucks to stop safely to be surveyed. There are also issues related to the sampling of vehicles. Most roadside surveys are collected during daylight hours only, while many long-haul commodity-carrying trucks travel at night. So a daylight survey may not report local truck traffic characteristics accurately. There may also be significant seasonal or day-of-week variation in truck traffic that would not be cap- tured by a single sample. If not all trucks are to be surveyed, then a sampling plan should be developed that does not introduce bias into the results. For roads whose truck traffic is all locally generated, or nearly so, another option is to survey the local businesses that operate these trucks. This type of survey is appreciably less expen- sive than a roadside survey. However, it generally is difficult to select a truly representative sample of truck operators, so it may be necessary to extend such a survey to cover all opera- tors of trucks operating to or from the area. Other Sources Other sources of data that may be used for allocating truck volumes to commodity/industry groups are the following:

1-47 • The Commodity Flow Survey (CFS),19 a U.S. Census Bureau dataset that contains infor- mation on freight flows by origin, destination, transportation mode, commodity group, distance shipped, and shipment size. Data are collected from a quinquennial survey of manufacturing, mining, and wholesale establishments and selected types of retail and ser- vice establishments. No data are collected on imports received directly from foreign sources or on shipments from farms. • County Business Patterns,20 published annually by the U.S. Census Bureau, contains data on employment and wages by county and industry. The data are obtained from Federal Insurance Contributions Act (FICA) reports and exclude data on employment that are exempt from FICA. For employees in the construction, mining, transportation, communi- cations, and utilities industries, place of employment generally refers to a firm’s relatively permanent offices rather than to actual worksites. Truck-Trip Generation TDMs require estimates of truck trips generated (i.e., originating or terminating) in each TAZ. Extensive information about truck-trip generation is contained in the Quick Response Freight Manual, which includes a compilation of truck-trip generation rates from a number of studies. Tables are provided listing trip generation rates by employee, by square feet of office space, by developed acre, and by linear regression formulas. Rates are listed by land-use category and vehicle size where available. Because of the availability of regional economic forecasts expressed in terms of number of jobs, employment growth is the most common basis for fore- casting truck volume growth. However, localized forecasts (such as site development pro- posals and regional economic development plans) may instead use units of developed acres or square feet of developed space. Two additional sources of trip-generation data are published by the Institute of Transporta- tion Engineers.21 The latest editions of these publications contain information obtained from studies released since the Quick Response Freight Manual was published. Recent information is available in NCHRP Synthesis of Highway Practice 298: Truck Trip Generation Data. A.2.5 Conversion to Linear or Exponential Annual Growth Rates The procedures presented above generally produce forecasts of truck volume for a single future year. For use by the Pavement Design Guide software, these forecasts must be converted to 19 http://www.census.gov/econ/www/se0700.html 20 http://www.census.gov/epcd/cbp/view/cbpview.html 21 Institute of Transportation Engineers, Trip Generation, 7th ed., 2003, and Trip Generation Handbook, 2nd edition, 2004.

1-48 forecasts of either exponential annual growth rates or linear annual growth rates. This con- version will be performed by TrafLoad or by using the formulas presented below. Exponential Growth Rate The formula for deriving an exponential annual growth rate using estimates of AADT of SUTs or CTs for any 2 years is (A.7) where: Yo is the base year, Yf is the future year, To and Tf are the corresponding estimates of AADT of SUTs or CTs (or any other variable of interest), and r is the estimated exponential growth rate. In the example of Section A.2.2 (shown in Table A.3), AADT was estimated to be 1,587 in 1999 and forecast to be 2,879 in 2022. With these values, Equation A.7 becomes In percentage terms, the resulting exponential growth rate is 2.88 percent. Linear Growth Rate The corresponding formula for linear growth is (A.8) where b is the estimated linear growth rate and the other variables are defined as in Equa- tion A.7. Substituting data from the example of Section A.2.2 into Equation A.8 produces the following estimate of linear annual growth rate: b = − − = 2 879 1 587 2022 1999 61 52, , . trucks per year b T T Y Y f o f o = − − r = − =      2 879 1 587 1 0 0288 1 23, , . r T T f o Y Yf o = −     −     1 1

1-49  A.3 Changes in Vehicle Use The previous sections of this appendix present procedures for forecasting changes in the num- bers of SUTs and CTs operating on a given road. Such changes can be expected to have cor- responding effects on the stresses on a pavement over its lifetime—increases in the daily num- ber of trucks operating on the road generally result in increases in the daily stresses on the pavement. However, these changes are not the only ones that may affect these stresses. Daily (or annual) stresses on pavement may also be affected by changes in the following: 1. Commodity mix. A shift in the mix of commodities carried on the road toward (or away from) dense commodities will tend to increase (or decrease) pavement stresses. 2. Payload density. For an appreciable number of manufactured products, increasing use of protective packaging has caused shipment density to decline. If this trend continues, it will tend to reduce pavement stresses produced by trucks carrying these products. 3. Size and weight limits. Changes in truck size and weight limits can have significant effects, in either direction, on truck configurations used and the load spectra of these configurations. Each of these three types of change in vehicle use will affect the load spectra of affected vehi- cle classes, and changes in size and weight limits also can have significant effects on the num- ber of trucks in affected vehicle classes. Because the current version of the Pavement Design Guide software is not designed to use forecast changes in load spectra, TrafLoad has no capa- bility for developing such forecasts. However, the effects of these potential changes in vehicle use do warrant some further discussion. Of the three types of change, the last is potentially the most significant and also is the least pre- dictable. Possible changes in size and weight limits include the following: • Increases in size limits for existing configurations (with no change in weight limits). For cube-limited shipments, such increases would allow increasing shipment sizes, resulting in heavier axle loads and increasing pavement stresses. For vehicles whose gross vehicle weight (GVW) is limited by the bridge formula, increased length limits may also allow increasing GVWs for weight-limited shipments, also resulting in heavier axle loads and increasing pavement stresses. • Changes in axle-weight limits. Increases in axle-weight limits would increase pavement stresses; decreases would decrease these stresses. • Elimination of the 80,000-pound cap on gross vehicle weight (GVW) that currently exists on most roads in the Interstate system. The effects of such a change would depend on the limits that would control GVW in the absence of the 80,000-pound cap. However, most proposals for eliminating this cap (and the limits that currently apply on most roads that do not have this cap) would result in converting some traffic from five-axle semi-trailer combinations to longer combination vehicles (LCVs) with seven to nine axles. On a per- vehicle basis, this change would result in increasing both payloads and pavement stresses, with the increase in payloads generally being greater than the increase in pavement stresses

per vehicle. The result generally would be a modest shift in VMT from Class 9 to Class 13 vehicles and a small reduction in total VMT of CTs that would more than balance the increase in pavement stresses per vehicle. We observe that the Pavement Design Guide software will be capable of analyzing some of the effects of changes in size and weight limits. For example, forecasts of AADT for Class 9 and Class 13 trucks could be adjusted to reflect the effects of a shift in usage that is expected to result from possible lifting of the 80,000-pound cap. However, such an analysis would be incomplete. The change in weight limits would have a significant effect on the load spectra of Class 13 trucks (and, most likely, a small effect on the load spectra of Class 9 trucks). Because the results of such a partial analysis could be misleading, it should not be used. More gener- ally, for the purpose of pavement design, forecasts of AADT by vehicle class should not be adjusted for expected shifts between vehicle classes. Partly for this reason, TrafLoad has not been designed to allow users to specify separate growth rates for each vehicle class (though it is designed to allow separate rates for SUTs and combinations). A more significant issue relating to size and weight limits stems from the difficulty that exists in forecasting likely changes in these limits and in determining whether such future changes are likely to increase or decrease total pavement stresses. Except in the special case of changes in size and weight limits that have already been enacted but that were not in effect during the most recent year for which historical data are available, it is not possible to predict with confi- dence whether such changes will increase or decrease pavement stresses. Accordingly, the inability of the current version of the Pavement Design Guide software to use forecast changes in load spectra is of little consequence, and addressing this limitation should be a low priority. 1-50

1-51 Appendix B: Coefficients of Variation The coefficient of variation (CV) is a statistic that is designed to measure the likely error in an estimate in percentage terms. The CV of a variable is obtained by dividing its standard devi- ation (s) by the estimate. As originally designed,1 the Pavement Design Guide software would have required estimated CVs for each value of annual average daily traffic (AADT) by vehicle class. Accordingly, a pre- liminary set of procedures for estimating the required CVs was developed; it was subse- quently redesigned so that no use was made of CVs. Accordingly, there is no current need to estimate CVs, and TrafLoad does not produce CVs. However, there is some possibility that a future version of the Pavement Design Guide software will require CVs. For this reason, a slightly edited version of the preliminary procedures for estimating the CVs is presented in this appendix. In the case of the AADT, there are several sources of error. One source, the use of factor groups in the factoring of Level 2 classification counts, generates enough data to permit estimation of the standard deviation and the associated CV from this source. However, other sources (dis- cussed below) do not provide such information. In order to avoid underestimating the over- all error in the AADT estimates, the proposed procedure combines a statistical estimate of the CV resulting from the use of factor groups (if used) with subjective estimates of the additional error due to other sources. The application of this procedure to Level 2 sites is described, in some detail, in the first section of this appendix. The estimation of CVs for Level 1 and Level 3 sites is discussed in the second and third sections. And a concluding section discusses an issue relating to how the CVs should be applied by the pavement-design software.  B.1 Level 2 Sites Consider a Level 2 classification site for which AADT estimates are developed for each vehi- cle class (VC) using combined seasonal/day-of-week (DOW) factoring, and assume that the factor group to which it is assigned consists of n Level 1A sites. If factors are developed sep- arately for each of the Level 1A sites, they can be applied to a classification count from the Level 2 site to produce n sets of AADT estimates. These estimates, in turn, can be averaged to produce mean values of AADT for each VC. By construction, these mean values of AADT are identical to the AADT estimates produced by a combined seasonal/DOW factoring procedure. 1 ERES Consultants and FUGRO-BRE, Draft Report, prepared for NCHRP Project 1-37A, 2000, pp. 414–415.

1-52 For each VC, differences between the original n estimates of AADT and the overall mean can be used to infer information about the major source of error in the overall mean, the “within- group” variance among the seasonal and DOW patterns occurring at different sites in the fac- tor group. The first subsection below presents a procedure for estimating the CV due to the major source of error. TrafLoad can be readily modified to perform all computations required by this pro- cedure. The second and third subsections present more qualitative discussions of other poten- tial sources of error. And the fourth subsection proposes a procedure that could be incorpo- rated into a future version of TrafLoad that would estimate overall CVs for AADT for all Level 2 sites. Within-Group Variance Consider a combined seasonal/DOW factor group2 consisting of n Level 1A sites, and con- sider one or more short-duration classification counts obtained at a Level 2 site for a particu- lar vehicle class. Let xi = the estimate of AADT for this vehicle class produced by applying factors obtained from site i (i = 1, …, n) to the count(s) in question and x = the mean of the xi (= the estimate produced by applying the seasonal/DOW factors obtained from the entire group of Level 1A sites). Then the differences between the xi and x can be used to compute the standard deviation, s, of x: (B.1) The CV resulting from the use of factors that are derived using data from an entire factor group can be estimated as follows: (B.2) Similarly, assume that separate seasonal and DOW factors are being used, with the seasonal factors derived using data from n Level 1A sites and the DOW factors derived using data from m Level 1A sites (which may or may not overlap the first n sites). Then the above derivation yields a similar formula for estimating the CV due to the use of seasonal and DOW factors from these two groups: CV ( 1) 2 =1 2= −∑ − ( )x x n x ij i n s x x n i i = − − ( )∑ 2 1 2 The current version of TrafLoad uses separate seasonal and DOW factor groups. The development of CVs for these factor groups is presented at the end of this subsection. (See Equation B.3.)

1-53 (B.3) It may be noted that the CVs produced by Equations B.2 and B.3 depend, in part, on the days and months on which the short-duration classification counts are obtained. For example, it is likely that, for many factor groups, there will be more variance in the Friday factors than in the Wednesday factors. This difference in variances will result in higher CVs for counts obtained on Fridays than for those obtained on Wednesdays. Other Sources of Error There are several sources of error in the AADT estimates developed for Level 2 sites in addition to those due to within-group variance. The additional sources of error include the following: 1. Equipment errors at the site in question or at the Level 1A sites used as a source of factors. 2. Random variation in the seasonal and DOW patterns of truck volumes at a Level 1A site that is atypical of other sites belonging to the same factor group. 3. The use of a single set of factors developed using data for several vehicle classes (e.g., Classes 8–13) to produce separate AADT estimates for each of the vehicle classes. 4. Some other minor approximations used in the factoring process. (For example, in an area in which the harvest season ends in early November, counts collected early in the month are likely to be higher than counts collected later in the month. Accordingly, use of a November/Tuesday factor is likely to produce an upward bias in AADT estimates obtained from counts collected during the first Tuesday in November.) 5. Deficiencies in the process of associating Level 2 sites with factor groups (discussed below). Of the above sources of error, the least significant is the third. This limitation in the factoring procedure does not affect the estimates of AADT of combination trucks as a group, but only the distribution of AADT among the individual classes of combinations. Thus, the errors (as in Classes 8–13) tend to cancel each other, though they may influence pavement design. There are two types of site at which the second source of error, random variation, may be an issue. One type consists of sites with relatively low truck volumes. Random variation in these volumes can produce patterns that are atypical of other sites belonging to the same factor group. For this reason, factoring generally is performed using data that come only from sites with relatively high truck volumes. Random variation can also be a problem for Level 1A sites at which a significant portion of total truck traffic is influenced by a small number of decision-makers. For example, if a sig- nificant portion of truck traffic at a particular site serves a nearby construction site, the sea- sonal volume pattern at the site may be very different from the patterns at other sites in the CV ( 1) 2 11 2= −∑∑ − ( ) == x x nm x ij j m i n

1-54 state and also from the pattern at the site in the following year (when the construction project may be completed). The last source of error requires additional discussion. There are two issues: the degree of ambiguity in the assignment of Level 2 sites to factor groups and the representativeness of Level 1A sites in the factor group of the Level 2 site to be factored. The representativeness issue arises for Level 2 sites in the lower functional systems. As an example, assume there is a single “Rural Other” (RO) factor group that is used for all rural non-Interstate sites. Most or all of the Level 1A sites in this factor group are likely to be princi- pal arterials. Truck traffic at these sites is likely to include some traffic that both originates and terminates at relatively distant locations, as well as traffic that originates or terminates nearby— two types of traffic that can have somewhat different seasonal and DOW patterns. Almost all truck traffic at sites in the lower functional systems, on the other hand, are likely to be locally generated. Accordingly, if the two types of truck traffic actually do have different seasonal or DOW patterns, the RO factors (obtained from principal arterials) will tend to produce some unknown bias in the AADT estimates produced for Level 2 sites on the lower systems.3 The issue of ambiguity in the assignment of Level 2 sites to factor groups is one that only arises for certain types of factor groups. If a factor group is defined to apply to all sites in one or more functional systems, there is no ambiguity in the process. On the other hand, it may be desir- able to use other information in developing the factor groups. Factor groups developed in this way may have lower variances than ones that are based entirely on a functional system, and so Equations B.2 and B.3 will produce lower CVs. However, some additional error may be cre- ated in the AADT estimates for individual Level 2 sites because of ambiguity in the definitions of the factor groups. As an example, consider a state with truck activity on the RO that has distinctly different sea- sonal patterns in the eastern and western parts of the state, with a blend of the two patterns in the middle. If the RO group is divided into an eastern RO group and a western RO group, some decision will be required as to where to place the boundary between the two groups. Because seasonal patterns at all Level 1A sites are known, it generally will be easy to make sure that all Level 1A sites are placed in the appropriate group. However, seasonal patterns at Level 2 sites generally are unknown, so it is likely that some of these sites will not be placed in the appropriate group. The resulting misassignment will produce a (generally small) addi- tional error in the AADT estimates that is not captured by Equations B.2 and B.3. Level 2B Sites Level 2B sites are Level 2 sites at which only a partial-day manual classification count is avail- able. Procedures for using a partial-day classification count to estimate full-day truck volumes 3 The alternative of creating a separate factor group for the lower functional systems may be considered. However, this alternative would increase data collection costs and, because of the random variation in truck volumes at sites with relatively low truck volumes, could produce even poorer estimates of AADT.

1-55 (by class) for the day of the count are presented in Part 2, Section 3.4. In addition to the errors discussed in the preceding subsection, AADT estimates developed for Level 2B sites incorpo- rate errors that result from the conversion of partial-day counts to estimates of full-day truck volumes. There are two types of site: 1. Sites that are dominated by business-day trucking and for which truck traffic distribution factors (TTDFs) are derived directly from the partial-day counts and 2. Other sites. As discussed in Part 2, the estimates of full-day truck volumes that were developed for the first type of site will almost always be conservative; for these sites, the probability of under- estimating truck volumes for the day in question is negligible. Thus, for these sites, the prob- ability that errors in the partial-day/full-day conversion process will contribute to underesti- mating AADT will be negligible. However, the Pavement Design Guide software requires error estimates for the AADT primarily as an indicator of the extent to which the AADT may have been underestimated. Because the errors in this conversion process are almost certain not to contribute to underestimates of AADT, they need not be reflected in CVs that are produced by the Pavement Design Guide software. The same cannot be said for the other type of site. For these sites, the conversion process may result in either underestimating or overestimating full-day truck volumes on the day in ques- tion, and the extent of any underestimation could be fairly significant. Hence, for these sites, this source of error cannot be ignored. CVs for Level 2 Sites The preceding discussion identified several reasons why the AADT estimates are likely to incorporate some errors that are not reflected in the CVs produced by Equations B.2 and B.3; the actual CVs generally will be slightly larger than those indicated by Equations B.2 and B.3. Accordingly, it is recommended that CVs for Level 2 sites incorporate small upward adjust- ments to the value produced by Equation B.2 (for combined seasonal/DOW factoring) or B.3 (for separate seasonal and DOW factoring). The suggested adjustments are • 0.01 for all sites, • An additional 0.03 for sites with ambiguous assignments to factor groups, • An additional 0.02 for sites in functional systems that are unrepresented or significantly underrepresented among the Level 1A sites from which the seasonal and DOW factors are developed, and • An additional 0.05 for Level 2B sites that states believe fit into some state-defined com- posite of the business-day and through-truck TOD patterns. These adjustments are all added to CVs produced by Equation B.2 or Equation B.3.

1-56  B.2 Level 1 Sites Level 1A Sites Level 1A sites are sites at which base-year values of AADT are obtained from classification counts collected at the site during all of, or most of, a 12-month period. The only errors in these values are those resulting from equipment malfunction and from the approximation proce- dures that are used to estimate truck volumes during any periods of time when the equip- ment was malfunctioning or out of service. For such sites, 0.01 is a suggested conservative value for the CVs. Level 1B Sites Level 1B sites are sites on the same road as an associated Level 1A site. AADT at Level 1B sites is estimated by obtaining a set of short-duration classification counts at the site and applying factors obtained from the associated Level 1A site. Errors in the AADT estimates at Level 1B sites can be caused by the following: equipment malfunction at either the site or the associated Level 1A site, some of the inherent limitations of the factoring process (such as the use of an average “November/Tuesday” factor for any Tuesday in November), and differences in the seasonal and DOW patterns at the two sites. The CVs at these sites are clearly higher than those at Level 1A sites. If the associated Level 1A site were reasonably close to the Level 1B site in question, it would appear reasonable to assume a CV of 0.02. If the two sites are more distant, with some intersections/interchanges with significant truck routes occurring between the two sites, the CV is likely to be somewhat larger. Accordingly, for Level 1B sites, 0.02 is a suggested default value for the CV, and the user should increase this value (typically to 0.03 or 0.04) when the associated Level 1A site is relatively distant or lies beyond intersections or inter- changes with one or more major truck routes.  B.3 Level 3 Sites Level 3A Sites Level 3A sites are sites that are on the same road as an associated Level 1 or 2 site and for which a volume count exists (for a time period during which traffic was also being counted at the associated site) but for which classification counts do not exist. Traffic volumes at Level 3A sites are sufficiently different from those at their associated sites to warrant separate analyses of the Level 3A sites, but a Level 3A site and its associated site are presumed to have fairly similar percentages of vehicles in the major truck classes. Two sources of error affect AADT estimates developed for a Level 3A site but do not affect the estimates developed for the associated site: • The ratio of the short-duration volume counts obtained at the two sites may not accurately reflect the ratio of AADT at the two sites and

1-57 • The distribution of total traffic among vehicle classes is likely to differ somewhat at the two sites. Because of these differences, and particularly because of the second difference, CVs for Level 3A sites should exceed the CVs at the associated sites by 0.10 for associated Level 1 sites and by 0.08 for associated Level 2 sites.4 Level 3B Sites Level 3B sites are sites for which TrafLoad produces only an estimate of annual average daily truck traffic (AADTT) and identification of the site’s Truck Traffic Classification (TTC) group. There are two kinds of Level 3B site. The first kind of Level 3B site is one for which a recent traffic count exists. For such sites, AADT is estimated by factoring the traffic count, and AADTT is estimated by multiplying AADT by a non–site-specific estimate of percent trucks. As stated in Part 2, Section 3.5, this procedure should only be applied at sites with very low volumes of trucks and buses. Errors in the fac- toring process and in the estimate of percent trucks are likely to produce fairly high CVs. CVs of 0.30 should be used for these sites. The second kind of Level 3B site is on a planned new road. For these sites, estimates of total traffic and truck volumes are, at best, developed from travel demand models. Errors in the resulting estimates of AADTT are likely to be quite high. For these sites, use CVs of 1.0.  B.4 Applying the CVs The factoring procedure presented in Part 2, Section 3.3 uses one set of factors for each Type 1 VC group. For purposes of discussion, assume that there is one VC group for all SUTs and a second group for all combination trucks (CTs). As a result, for any Level 2 site, the procedures for estimating CVs will produce the same value for the CV for the four SUT vehicle classes and a different value for the CV for the six CT classes. The same will be true for Level 3A sites whose associated site is a Level 2 site. For any other Level 3 sites, and for any Level 1 site, the above procedures produce a single value for the CV for all 10 vehicle classes. Consider the AADT estimates produced for the six CT classes for a Level 2 site. These errors will be correlated with each other (because of imperfections in the factors applied to the six classification counts), but the correlations will not be perfect (in part, because of differences in the seasonal and DOW patterns exhibited by the six classes). The Pavement Design Guide soft- 4 A lower increment is proposed for associated Level 2 sites because of the tendency of errors from inde- pendent sources to partially cancel each other. This tendency has very little effect for associated Level 1 sites (because Level 1 sites have very small CVs), but it is likely to have a moderate effect for associ- ated Level 2 sites.

1-58 ware, on the other hand, can be designed to assume that the errors are totally uncorrelated or that they are perfectly correlated. Because uncorrelated errors tend to cancel each other, the former option will tend to understate the effects of the partially correlated errors while the latter option will tend to overstate these effects. The goal of producing conservative pavement designs suggests that it would be preferable for the Pavement Design Guide software to adopt the latter option, i.e., to assume that the errors for all CT classes are perfectly correlated. Though the discussion here focused on Level 2 sites, this assumption appears to be appropriate for Level 1 and 3 sites as well. In the case of the SUT classes, the conclusion is somewhat different. In many areas, there is likely to be substantial correlation between the AADT errors for Vehicle Classes 6 and 7 (three- axle trucks and trucks with four or more axles). However, these two classes are likely to exhibit seasonal and DOW patterns that differ greatly from those exhibited by buses (Class 4) and by two-axle, six-tire trucks (Class 5). Hence, it would be preferable for the Pavement Design Guide software to use three independent estimates of AADT error for SUTs: one for Class 4, one for Class 5, and one for Classes 6 and 7.

Next: Part 2 - Guidelines for Collecting Traffic Data to be Used in Pavement Design »
Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design Get This Book
×
MyNAP members save 10% online.
Login or Register to save!
Download Free PDF

TRB's National Cooperative Highway Research Project (NCHRP) Report 538: Traffic Data Collection, Analysis, and Forecasting for Mechanistic Pavement Design provides guidance for collecting traffic data to be used in pavement design and includes software—designated TrafLoad—for analyzing traffic data and producing traffic data inputs required for mechanistic pavement analysis and design. The TrafLoad software is designed to produce traffic data for input to the 2002 AASHTO pavement design software. TrafLoad is based on a new mechanistic-empirical approach to pavement design, it relies on axle load spectra rather than equivalent single axle loads. For each of four axle types, the load spectra specify the percentages of axles falling into each of several load ranges.

Help on Burning an .ISO CD-ROM Image

Download the TrafLoad .ISO CD-ROM Image

TrafLoad User’s Manual

TrafLoad Procedures Manual

  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!