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

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 2 Guidelines for Collecting Traffic Data to Be Used in Pavement Design

1.0 Introduction The AASHTO Technical Committee on Pavements has undertaken an effort to develop an improved guide for the design of pavements. This effort, undertaken under NCHRP Project 1-37A, Development of the 2002 Guide for the Design of New and Rehabilitated Pavement Structures1 (the “Pavement Design Guide”), will provide engineers with practical and realistic pavement design procedures and software that use existing mechanistic-empirical principles. The mech- anistic-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. Under NCHRP Project 1-39, a software system, TrafLoad, has been developed for generating the traffic data required by the Pavement Design Guide software. This part of the report pre- sents guidelines for collecting traffic data required by TrafLoad and provides brief descrip- tions of some of the analyses that TrafLoad performs in the course of converting these data into inputs required by the Pavement Design Guide software. The presentation is designed to complement the more extensive discussion of the collection of traffic data that is contained in the FHWA’s Traffic Monitoring Guide (2001).2 A user’s manual for TrafLoad is presented as Part 3, which is available online at http://trb.org/news/blurb_detail.asp?id=4403. This introduction presents a brief summary of the data required by the Pavement Design Guide software, followed by an outline of the remainder of Part 2.
1.1 Traffic Data Requirements for the Pavement Design Guide Software The Pavement Design Guide distinguishes three broad levels of input data that vary with the quality of information that the designer has about the loads to be applied at the site. These input levels are described as sites for which the designer has 1) good, 2) fair, or 3) poor infor- mation about the truck volumes and axle loads to be applied, with Level 3 further divided into Levels 3A and 3B. Corresponding to each input level, the Pavement Design Guide software requires site-specific, region-specific, or default values for several types of traffic data. In addition, there are other types of optional traffic data for which the software can use state-supplied values, or the software can use default values that have been developed in NCHRP Project 1-37A from national data. 2-1 1 http://www.2002designguide.com/ 2 http://www.fhwa.dot.gov/////ohim/tmguide/index.htm

TrafLoad has been designed to be the principal source of traffic data for the Pavement Design Guide software. A summary of the traffic data elements produced by TrafLoad for use by the Pavement Design Guide software is presented in Table 1.1. TrafLoad currently allows the use of the 13 standard FHWA vehicle classes (VCs) (see Table 1.2) and/or an aggregation of these classes into a smaller set of user-defined classes.3 The latter option makes it possible to develop pavement designs for sites for which vehicle classification data are available only by length class.
1.2 The Remainder of Part 2 Chapter 2.0 of Part 2 presents in some detail the development of a program for collecting the weight data that TrafLoad requires for generating axle-load distribution factors and estimates 2-2 3 The Pavement Design Guide software currently allows the use of up to 13 classes. Table 1.1 Traffic Data Elements to be Produced by TrafLoad AADTi a by direction for up to 13 VCs (i) (1, 2, and 3A) b Annual Average Daily Truck Traffic (all classes, combined) (3B) Truck Traffic Classification Groupc (3B) Monthly Traffic Distribution Factors by VC (1, 2, and 3A) Axle-Load Distribution Factors – Site Specific (1) Axle-Load Distribution Factors – Regional (2) Axle-Load Distribution Factors – Statewide (3) Linear or Exponential Growth Rates Directional Distribution Factorsd (1, 2, and 3A) Axle Groups per Vehicle (for each VC) Hourly Distribution Factors (1, 2, and 3A) a Annual average daily traffic by VC (i). b Numbers in parentheses identify the input levels for which the data are used. c Each Level 3B site is assigned by the user to a truck traffic clas- sification group. This assignment is passed by TrafLoad to the Pavement Design Guide software without modification. d Because all AADTi estimates will be provided separately by direc- tion of travel, all corresponding directional distribution factors for Level 1, 2, and 3A sites will be set to 1.0.

of axle groups per vehicle for each VC. TrafLoad is capable of using data provided by the state to generate these values for every site for which such data are requested, though the quality of the estimates will be higher for Level 1 sites (for which site-specific data are provided) than for Level 2 or 3 sites. Chapter 3.0 presents a similar discussion of the development of a program for collecting the classification counts TrafLoad requires for generating most of the other traffic data required by TrafLoad. As indicated in a footnote to Table 1.1, all estimates of annual average daily traf- fic by VC (AADTi) for Level 1, 2, and 3B sites are produced by direction, so the corresponding directional distribution factors are set to 1.0. For Level 3B sites, no directional information is supplied to TrafLoad, so it does not produce directional factors for these sites. Instead, the Pavement Design Guide software will use its own default values. The final chapter discusses the collection and handling of the traffic data required to create the necessary datasets. 2-3 Table 1.2 FHWA Vehicle Classes 1. Motorcycles 2. Passenger cars 3. 4-tire trucks 4. Buses 5. 2-axle 6-tire trucks 6. 3-axle trucks 7. 4+ axle trucks 8. 3-4 axle single-trailer combinations 9. 5-axle single-trailer combinations 10. 6+ axle single-trailer combinations 11. 5-axle multi-trailer combinations 12. 6-axle multi-trailer combinations 13. 7+ axle multi-trailer combinations

2-4 2.0 Weight Data The Pavement Design Guide procedure, developed under NCHRP Project 1-37A, performs a detailed analysis of pavement deterioration caused by single, tandem, tridem, and quad axles of varying weights. Accordingly, instead of estimates of 18,000-pound equivalent single-axle loads (ESALs), the Pavement Design Guide software requires estimates of the numbers and weights of four types of axles: single, tandem, tridem, and quad. For each VC and axle type, it is necessary to estimate the percentages of axles falling into each of several specified load ranges. These load ranges are listed in Table 2.1. For single axles, there are thirty-eight 1,000-pound load ranges covering the 3,000- to 41,000-pound weight range plus one range for lighter axles. For tandem axles, there are thirty-eight 2,000-pound ranges covering the 6,000- to 82,000-pound weight range, etc. For each VC, the Pavement Design Guide software requires (and the Project 139 software, TrafLoad, produces) between one and four of these load distributions, also known as load spectra. The Pavement Design Guide software treats the load spectra for each VC as being con- stant over time. The load spectra are obtained by aggregating vehicle weight data collected at one or more weigh-in-motion (WIM) sites. The WIM data used for developing the load spectra also produce the average number of axles of each type associated with each class of vehicle. For example, a standard five-axle tractor semi-trailer has one single axle and two tandem axles. Thus, to compute the total load applied by 1,000 of these trucks, it is necessary to multiply the single-axle load spectrum by 1,000 and the tandem-axle spectrum by 2,000. The expected number of axles of each type for a given number of vehicles in a particular VC is developed from WIM data in conjunction with the development of the load spectra. For any pavement project, the required load spectra are developed by TrafLoad using data col- lected from WIM equipment either on the same road at a site reasonably near the pavement proj- ect (Pavement Design Guide Level 1 data), or from WIM data collected elsewhere (Levels 2 and 3). From these same data, TrafLoad develops the number of axles of each type associated with each class of vehicle. The first section of this chapter discusses potential sources of error in the load spectra so that readers can understand what is important to ensure accurate load estimates, why some data- collection plans are better than others, and why specific data are requested for developing accurate load estimates. Section 2.2 provides brief descriptions of the three levels of WIM data collection that TrafLoad supports. Sections 2.3–2.5 discuss in more detail the three levels of data collection and provide guidelines for collecting the required data. And the final section of this chapter presents additional information on the collection of weight data.

2-5 Table 2.1 Load Ranges Used for Load Spectra Upper Limit of Load Range (kipsa) by Type of Axle Group Load Range Single Tandem Tridem Quad 1 3 6 12 12 2 4 8 15 15 3 5 10 18 18 12 21 21 14 24 24 16 27 27 18 30 30 20 33 33 22 36 36 24 39 39 26 42 42 28 45 45 30 48 48 32 51 51 34 54 54 36 57 57 38 60 60 40 63 63 42 66 66 44 69 69 46 72 72 48 75 75 50 78 78 52 81 81 54 84 84 56 87 87 58 90 90 60 93 93 62 96 96 64 99 99 4 6 5 7 6 8 7 9 8 10 9 11 10 12 11 13 12 14 13 15 14 16 15 17 16 18 17 19 18 20 19 21 20 22 21 23 22 24 23 25 24 26 25 27 26 28 27 29 28 30 29 31 30 32 31 33 66 102 102 32 34 68 33 35 70 34 36 72 35 37 74 36 38 76 37 39 78 38 40 80 39 41 82 a One kip = 1,000 pounds = 4.448 kN.

2-6
2.1 Sources of Estimation Error To create a data-collection and analysis process that produces accurate axle-load distributions for pavement design, errors resulting from the following sources must be minimized: 1. The calibration of the data-collection equipment, 2. Differences in axle-weight distributions among different VCs, 3. Differences in vehicle loading rates between one road and another, 4. Differences in vehicle load by direction, 5. Variation in axle weights caused by changes in loading conditions by time of day, 6. Variation in axle weights caused by changes in loading conditions by day of the week, 7. Variation in axle weights caused by changes in loading conditions by time of year, and 8. Future changes in vehicle loading conditions. The first of these sources of error, calibration error, is a function of equipment operation. It is important that highway agencies calibrate their data-collection equipment carefully and use that equipment in locations where it can operate correctly. WIM equipment is particularly sen- sitive to calibration error, and calibration error is affected by a variety of factors, including sen- sor installation and condition, pavement roughness and condition, environmental conditions, and even roadway geometrics. Individual WIM sensor technologies are affected differently by these factors. Additional information about the selection, calibration, and operation of WIM equipment is contained in NCHRP Report 509.1 The second source of error, differences in weight among different VCs, is handled by sepa- rating the axle-weight data of one class of vehicles from those of other classes. Thus, separate axle-load distribution tables are required for each class of vehicles. VC may be defined by each highway agency. TrafLoad currently allows the use of the 13 standard FHWA VCs and/or an aggregation of these classes into a smaller set of user-defined classes. The latter option makes it possible to develop pavement designs for sites for which vehicle classification data are avail- able only by length class. To compute accurate load estimates, a highway agency must also collect vehicle classification volume data using VCs that can be correlated directly to the weight data that the agency col- lects. Chapter 3.0 provides instructions on how to collect and manipulate the data needed to determine the number of vehicles, by class, that will operate on the pavement being designed. The third and fourth sources of error arise because different roads can serve similar trucks that have very different sets of loading characteristics. This is also true for different direc- 1 Hallenbeck, Mark, and Herbert Weinblatt, NCHRP Report 509: Equipment for Collecting Traffic Load Data, Transportation Research Board of the National Academies, Washington, D.C., 2004.

2-7 tions of travel on the same road. These variations in loading characteristics are caused by dif- ferences in the commodities being carried and by differences in the percentages of trucks that are loaded. For example, on a road leading to a gravel pit, the lane of travel leading away from the pit will experience heavier trucks than the lane of travel leading to that pit, because most trucks leave the pit full but return to the pit empty. For this example, the same number of trucks and the same type of trucks operate in both directions, but the load experienced in each direction is very different. This same effect occurs when two parallel county arterials of simi- lar size carry very different heavy-vehicle loads. The loads are determined not by the size of the arterials but by the nature of the economic activity associated with each road. (For exam- ple, one arterial may lead through land zoned for residential use, while another leads through land zoned for warehouse use.) The next three sources of error, variation by time of day, day of week, and time of year, also occur because the kinds of trucks using a road can change during these time periods. Week- day and weekend traffic can be very different because of the presence or absence of truck traf- fic generated by businesses that are open only during weekdays. Similarly, vehicle weights can change by time of day. Lastly, a number of phenomena can cause changes in vehicle weights over the course of a year. For example, in agricultural areas trucks may be very heavy during some times of the year (e.g., harvest time) and lighter during others. In some states, legal weight limits change over the course of the year; loads can be legally increased during the winter months when pavement is strong but must be reduced during the spring-thaw months when the subgrade strength is low and heavy loads cause serious pavement damage. (Note that spring load restrictions are also a source of variation from one road to another, as some states only apply load restrictions to minor roads or to roads in specific geographic areas of the state.) Finally, loading rates can change over the years as weight laws change or as economic activ- ity affecting a roadway changes. The Pavement Design Guide software assumes that the shapes of the normalized axle-load distribution curves do not change from year to year, and the research team makes the same assumption in TrafLoad. The research team adopts this assumption both for consistency with the Pavement Design Guide software and because the research team cannot accurately forecast these types of changes. While it is possible to fore- cast changes in truck volumes on the basis of expected changes in economic activity, it is extremely difficult to forecast the effects such changes will have on axle-load distributions. Load distributions are a function of • Truck size and weight laws (will changes in regulations encourage heavier gross vehicle weights but cause those weights to be carried on more axles?), • The commodity characteristics of specific routes (do the commodities “cube out,” or are loads limited by weight laws?), and • The fraction of loaded and unloaded trucks on the roadway (which is a function of both trucking fleet efficiency, types of haul, and the nature of truck ownership and use, all of which are difficult to forecast). This chapter deals exclusively with creating a data-collection program that, in conjunction with the TrafLoad software, helps an agency account for variations in vehicle loads by location, time

2-8 of day, day of week, and time of year. Estimating the number of vehicles (by class) is addressed in the next chapter, and equipment calibration is covered in a companion report.2 Seasonal variation in axle weights are of particular interest in pavement design because the pavement damage caused by axle loads is significantly affected by seasonal variations in soil conditions such as wetness, freezing, and thawing. As a result, the Pavement Design Guide software is designed to use separate sets of load spectra for each month of the year and to incorporate the effects of seasonal variations in the load spectra in the resulting pavement designs, and TrafLoad has been designed to produce such seasonally varying load spectra. Because WIM data collection tends to be difficult and expensive, highway agencies cannot afford to collect all the data needed to precisely measure and account for each source of vari- ation for all pavement design efforts. Consequently, a series of data-collection options are pro- vided to help states optimize the amount of data they collect, given the accuracy of the load estimate they require and the funding available for data collection. In general, the recommended program for collecting WIM data is stratified into three levels of data collection. Each level corresponds to how well an agency understands the location com- ponent of truck weight variation. Level 1 design is for sites where site-specific WIM data are available, and thus errors associated with locational differences are negligible. Level 2 design is for sites where some general knowledge of loading rates can be applied but actual WIM data have not been collected. Level 3 design is for sites where knowledge of loading rates is lim- ited enough that statewide average loading rates are the best available. Each of these three general conditions is discussed further in the remainder of this chapter.
2.2 Alternative Data-Collection Programs There are three levels of axle-load distribution (or load spectra) data in the data collection and analysis: • Site specific, • Truck weight road group (TWRG), and • Statewide averages. These different levels of data collection and application are introduced below and discussed in more detail in subsequent sections of this chapter. Site-Specific Load Distribution Data Site-specific data collection means that the state is able to accurately weigh trucks on the road on which the new pavement will be laid. The axle-weight data-collection site must be located 2 Ibid.

2-9 so that the traffic measured by the WIM scale is basically the same as the traffic that operates in the design lane of the roadway segment being designed. The intent of the definition of “site specific” is to allow a state to collect data on the same road- way as, but possibly at a location somewhat removed from, the pavement project’s segment limits. For example, the WIM scale used to provide data on I80 in Wyoming might be located at the Utah-Wyoming border on I80. The corresponding pavement project could easily be 50 (or more) miles away on I80 in Wyoming. Because little change in trucking activity occurs between those locations, data from the border-crossing WIM would be considered site spe- cific. However, data collected on I80 20 miles east of Salt Lake City would not be considered site specific for a pavement 20 miles west of Salt Lake City on I80 because considerable change in loading rates can occur within major metropolitan areas. In addition, Level 1 data collection assumes that the highway agency will use the WIM data collected to calculate the axle-load distribution tables directly. This means that the agency must be satisfied with the performance of the WIM scale being used. The scales used must be properly calibrated, and quality control checks must indicate that the data are valid. (If the agency uses a device that is not adequately calibrated, the site-specific data should be used only to iden- tify which TWRG dataset is most appropriate for that pavement project; and a Level 2 design, as discussed in Section 2.4, should be performed.) TrafLoad accepts WIM data collected over a 12- to 24-month period as well as WIM data col- lected over shorter periods of time. The software uses seasonal load spectra datasets that con- tain data for the 12 months of the year as the basis for imputing seasonal adjustments to data collected for periods of less than 12 months. The process for performing these adjustments (pre- sented in Part 4, which is available online at http://trb.org/news/blurb_detail.asp?id=4403) uses seasonal load spectra factor groups, which are discussed in Section 2.3. These factor groups are similar to automatic traffic recorder (ATR) factor groups except that sites are assigned to a group on the basis of seasonal variation in axle weights instead of seasonal vari- ation in the traffic volume. Errors in the axle distributions at site-specific locations usually are due to equipment and cal- ibration errors and errors caused by sampling of the traffic stream. In general, these site-specific estimates are relatively good. TWRG Data TWRG axle-load distributions are summary load distributions that represent axle loads found on roads with similar truck weight characteristics. These groups are similar to ATR factor groups, in which all roads in a group have similar seasonal volume patterns. In the case of TWRGs, however, all roads in a group have similar axle-load distributions. Because TWRGs are designed to have similar axle-load distributions rather than similar volume patterns, TWRGs generally will differ from the seasonal and day-of-week (DOW) factor groups. The FHWA’s Traffic Monitoring Guide recommends TWRGs as a way for highway agencies to collect, summarize, and report summary statistics for groups of roads. The intent is to group roads by their trucking characteristics so that the load spectra on all the roads in a group are fairly similar. Road characteristics that can be used to define road groups include the region of the state, particularly where the economic activity in a state differs from region to region

2-10 (e.g., agricultural areas versus mining areas); the nature of the commodities being carried (e.g., roads leading to a port versus roads in other parts of an urban area); and sometimes the type and location of the facility (e.g., urban freeway versus suburban arterial). The Traffic Monitoring Guide expects TWRGs to be state specific, but multiple states can work together to create regional load distribution tables if the states involved in the regional effort have similar truck weight laws. (Different truck size and weight laws would invariably lead to different truck weight characteristics and thus different axle-load distribution tables.) TWRG axle-loading tables are needed because most states do not have (and cannot afford to col- lect) site-specific WIM data for the majority of pavements they design each year. However, these tables produce poorer estimates of pavement stresses than tables derived using site-specific data. Recent analyses of California data indicate that, for combination trucks, TWRGs produce mean absolute percentage errors (MAPEs) for pavement stresses (as measured in 18,000- pound equivalent single-axle loads) of 17–20 percent,3 varying with the degree of disaggre- gation of the TWRGs and the care with which they are constructed. In contrast, the analyses indicate that site-specific WIM data collected over a 48-hour period produces a MAPE of only 7 percent (exclusive of equipment and calibration error). Because load distributions vary significantly, it is important that the pavement designer understand the approximate range of loads being applied. Using such knowledge greatly improves the reliability of the pavement design. Figure 2.1 shows the tandem-axle distribu- tions found at three different WIM scales. Figure 2.1(a) represents a site where a large per- centage of trucks are operating empty or in a partially loaded condition. Figure 2.1(b) repre- sents a moderate loading condition, while Figure 2.1(c) illustrates a site with very heavy (but predominantly legal) loading. Ideally, each of these three sites should be in a different TWRG. If each of the three roads carries the same number of trucks, the different loading conditions should result in three very different pavement designs. The challenge for each state is to deter- mine which roads (and directions of travel in some cases) are typified by which of these (or other) basic loading conditions. This grouping process requires analysis of a state’s existing weight data and trucking patterns, and it results in the creation of appropriate TWRGs. (Note that states may easily have more than three loading conditions. Also note that the TWRGs gen- erally will not correspond to the groups used for factoring classification counts. The number of TWRGs distinguished in the analyses of California data varied between 3 and 10, with the best results obtained using a set of 10 TWRGs that distinguished functional system, region, and direction of travel.) For each TWRG, a set of tables for the axle-load spectra will be created by the software. These tables summarize the distribution of all axle loads measured for trucks weighed at scales within each group of roads. In addition, for each TWRG, the average number of axles for each VC will be computed. All or most WIM scales in a state or multi-state region should be assigned to a TWRG. For each truck class, the axle weights of all trucks in the class weighed at the scales in a given TWRG are then used by the software to compute a corresponding set of load spectra. 3 Cambridge Systematics, Inc., Accuracy of Traffic Load Monitoring and Projections, Volume II, The Accu- racy of ESALs Estimates, prepared for FHWA, February 2003, Tables 4.5 and 5.1.

2-11 Figure 2.1 Load Distributions for Tandem Axles of FHWA Class 9 Trucks at Three Different Sites Figure 2.1(a) Lightly Loaded Trucks Fraction of Tandem Axles in Weight Group 0 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.16 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 Figure 2.1(b) Moderately Loaded Trucks Fraction of Tandem Axles in Weight Group 0 0.02 0.04 0.06 0.08 0.10 0.12 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 Figure 2.1(c) Heavily Loaded Trucks Fraction of Tandem Axles in Weight Group 0 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.16 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 Axle Weight Groupa a Each group is identified by the maximum weight, in thousands of pounds, for the group.

2-12 The error associated with applying the load spectra calculated using TWRG data is caused by the error sources that affect site-specific load distributions (i.e., the equipment error and the sampling error associated with less than complete annual data being collected at each of the TWRG WIM sites) and by two additional error sources: the error associated with using an average condition for several roads to represent a specific road and the error associated with assigning a given road to a specific TWRG. These two errors are considerably larger than the first error. Statewide Load Distribution Data The last type of axle-load distribution, statewide, should be used only to provide load esti- mates when the state highway agency has little knowledge of the loads trucks will carry on the roadway being designed. This means that the agency has little confidence in its ability to predict the TWRG for the pavement section. The above-cited analyses indicate that use of con- temporaneous statewide load data produces MAPEs of about 25 percent for combination trucks.4 Statewide load distributions are obtained (for each VC) by combining the data collected from all WIM sites in a state. These distributions (most likely moderate distributions) then serve to represent average conditions that can be used whenever something better is not available. Fig- ure 2.2 illustrates a possible statewide distribution for tandem axles of Class 9 trucks. (This distribution was obtained by averaging the three load distributions shown in Figure 2.1.) Statewide distributions are the least reliable load spectra for pavement design. Their use is an acknowledgment that the state has little idea about the vehicle weights being carried on the pavement in question. However, because these distributions represent average load condi- tions, they will be poor representations for pavements that experience very heavy or very light loading conditions. Thus, the pavement design will be reasonable if not optimal.
2.3 Level 1 WIM Sites Level 1 WIM sites are sites at which design-lane WIM data have been collected for some period of time. These include sites at which WIM data have been collected continuously for a year or more, as well as sites at which data have only been collected for one or more short periods of time. Data collected at these sites are used by TrafLoad to produce one or more sets of load spectra for each site. All data should be quality checked and collected using accurately cali- brated WIM equipment. The following subsections describe the way in which TrafLoad uses WIM data and other infor- mation required by TrafLoad. The discussions include recommendations for developing a WIM data-collection system that will produce data that can be used effectively by TrafLoad. 4 Ibid., Table 4.5.

2-13 WIM Inputs to TrafLoad For each Level 1 site, TrafLoad requires a current load spectra dataset. For some sites, TrafLoad should also be provided with a seasonal load spectra dataset. Each of these datasets consists of separate sets of load spectra by month and DOW, and there is a maximum of 84 sets of load spectra for each dataset. TrafLoad takes advantage of the resulting disaggregate sets of load spectra in two ways. First, for any month for which all seven DOW sets of load spectra are available, the monthly load spectra are developed by averaging the seven sets of load spectra in a way that avoids overweighting or underweighting the load spectra collected on any particular day of the week.5 For all months for which all seven DOW load spectra are available, this form of averaging Figure 2.2 A Typical Statewide Load Distribution for FHWA Class 9 Trucks 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 Fraction of Axles in Weight Range Axle Weight Groupa a Each group is identified by the maximum weight, in thousands of pounds, for the group. 5 For each VC, the averaging procedure uses the monthly average DOW traffic volumes for that class to obtain weighted averages of the seven DOW load spectra for the class. The procedure is described in Section 3.3 of Part 4, which is available online at http://trb.org/news/blurb_detail.asp?id=4403, and is based on the AASHTO procedure for summarizing traffic data (American Association of State Highway and Transportation Officials, AASHTO Guidelines for Traffic Data Programs, 1992, Chapter 5: Summarizing Traffic Data).

2-14 produces monthly load spectra that have no DOW bias. To improve the probability of obtain- ing complete sets of seven DOW load spectra, it is recommended that, if practical, WIM data from continuously monitored sites be collected and stored for at least 2 weeks of each month or, even better, for the entire month. (Collecting data for the entire month has the additional benefit of avoiding time-of-month bias, though this type of bias is usually much smaller than DOW bias.) Secondly, for a site and month for which only a partial set of DOW load spectra are available (i.e., load spectra for 1 to 6 days of the week), TrafLoad uses information about how pavement damage factors for each VC vary by DOW to produce estimated load spectra for the month that combine the available DOW load spectra with information about the extent to which the load spectra for the other days of the week are likely to be more or less damaging than the available load spectra. (See Part 4, Section 3.3, available online at http://trb.org/news/blurb_ detail.asp?id=4403.) The seasonal and current load spectra datasets are discussed further in the following subsections. Continuous WIM Sites and Seasonal Load Spectra Datasets Seasonal load spectra datasets are obtained from data collected at some or all of the continuous WIM sites in the states. The most significant use of data from these datasets is the develop- ment of monthly adjustment factors. These factors are used to adjust the load spectra obtained at other WIM sites for a given month so that they are reasonably representative of the load spectra for another month for which reliable WIM data are unavailable. If load spectra are available by month and DOW, data from seasonal load spectra datasets are also used to produce DOW adjustment ratios that are used in developing a set of monthly load spectra from a set of DOW load spectra for a month for which one or more DOW load spectra are missing. Each seasonal load spectra dataset must contain data for each month of the year, with a full week of data for at least one of these months (and preferably for all months). Ideally, this con- dition is met by defining a seasonal load spectra dataset to consist of 12 consecutive months of data. However, to allow for outages of WIM equipment, the data-collection period represented by a seasonal load spectra dataset may be extended to cover a period of up to 24 months. If, for any month of the year, the resulting dataset contains 2 months of data, only the data for the later month are used. Seasonal load spectra datasets are developed using data collected from a single lane (or design lane) using WIM equipment that has been consistently calibrated over the entire collection period. Each such dataset should be collected over a period during which seasonal changes in axle weights are believed to be reasonably representative of those that currently occur at the site. Month-to-month variations in calibration should be minimized, as should any secular upward or downward drift in calibration. TrafLoad will interpret any such variations or drift as representing actual changes in axle weights, and, as a result, it will assume that similar vari- ations in axle weights also occur at sites at which only 1 or 2 months of WIM data have been collected.

2-15 Seasonal Load Spectra Factor Groups To allow the application of seasonal adjustments to load spectra obtained from various WIM sites, each state must establish a set of seasonal load spectra factor groups and assign every WIM site in the state to one of these groups. Furthermore, each of these groups must contain at least one WIM site for which a seasonal load spectra dataset exists. For each of these groups, the software will use the seasonal load spectra datasets to develop a set of monthly adjustment factors. The assignment of WIM sites to seasonal load spectra factor groups is based on how average pavement damage per vehicle varies over the course of a year. A convenient unidimensional measure of this last quantity (and one that is used by TrafLoad) is average 18,000-pound ESALs per vehicle (average ESALs per vehicle, or AEPV). The WIM sites that are assigned to a given load spectra factor group should have values of AEPV that exhibit similar seasonal variations; i.e., the peaks and valleys in the AEPV should occur at the same time of year, and the ratios between the maximum and minimum values of AEPV should be similar. A reasonable starting point for the development of seasonal load spectra factor groups would be to distinguish three sets of functional systems: urban (U); rural Interstate (RI); and rural other (RO). In addition, any set of roads that has its own set of seasonally varying weight limits (such as spring-thaw restrictions or higher weights during winter freeze) should be assigned to a separate factor group or to separate factor groups. In establishing these factor groups, consideration might also be given to the way in which the month of the heaviest truck loadings varies regionally (e.g., as a result of differences in harvest season). As observed above, each seasonal load spectra factor group must include at least one WIM site for which a seasonal load spectra dataset exists. When establishing seasonal load spectra fac- tor groups, a state is likely to use its own WIM sites for this purpose. However, this is not a requirement. Any WIM site at which the seasonal variation in truck weights is believed to be typical of the seasonal load spectra factor group can be used. WIM VC Groups The seasonal adjustments to the load spectra are performed separately for each of several user- defined VC groups. These VC groups are referred to as WIM VC groups to distinguish them from the VC groups (Type 1 VC groups) that are used for factoring classification counts. (See Section 3.3.) The principal purpose of allowing the use of WIM VC groups is to make sure that, for any VC, the load spectra adjustments are based on a reasonable amount of data. For this purpose, any uncommon VC should be grouped with a more common class. In particular, any VC that is not commonly observed on a daily basis at all WIM sites that are used for seasonal load spec- tra datasets should be grouped with a more common class. However, because of a limitation in the current adjustment procedure, it is recommended that all common VCs be assigned to separate WIM VC groups.

2-16 Current Load Spectra Datasets Current load spectra datasets can be obtained from most or all WIM sites. There are two prin- cipal uses for a current load spectra dataset obtained for the design lane at a particular site: • It can be used in designing pavement for that site; and • It can be used in the development of a set of load spectra for a TWRG to which the site is assigned. (The development and use of load spectra for TWRGs is discussed in Section 2.4.) Since the current load spectra dataset for a particular site may be used for designing pavement for that site, it is important that this dataset provide as good a set of estimates of the current load spectra at the site as practical. Accordingly, a current load spectra dataset should contain data collected using calibrated WIM equipment over a recent period when the axle loads are believed to be representative of the current axle loads observed at the site. The data-collection period may be up to 24 months long. If a seasonal load spectra dataset exists for a particular site, it may also be used as the current load spectra dataset for the site. Alternatively, a separate dataset that is believed to better reflect current conditions may be used. If a current load spectra dataset contains less than 12 months of data, the software uses monthly adjustment factors to impute load spectra for the missing month. For a particular site, the monthly adjustment factors that are used normally6 are the ones developed for the sea- sonal load spectra factor group to which the site belongs. In developing a current load spectra dataset for a particular site, it is important to minimize calibration error. To the extent practical, use of portable sensors should be minimized, since their effects on the road profile adversely affect data accuracy. There is no minimum on the number of days of data incorporated into a current load spectra dataset. However, increasing the amount of data collected decreases the amount of imputa- tion required to produce a complete set of monthly load spectra and increases the reliability of the resulting values. Imputation is unnecessary when data are collected continuously over the course of a year, and relatively little imputation is required when data are collected over 1-week periods in 3 or 4 months uniformly spread over the year. Using data that are collected in only 1 month places appreciably greater reliance on the imputation procedure. (However, even in this case, the software’s imputation procedure is likely to produce a better represen- tation of the seasonally varying pavement loads than would the simple alternative of using the load spectra that are observed in 1 month to represent the load spectra that occur in the other 11 months of the year.) 6 In the special case of a site for which a separate seasonal load spectra dataset exists, the monthly adjust- ment factors used are the ones developed directly from that seasonal load spectra dataset. TrafLoad also allows the user to associate any site for which a seasonal load spectra dataset does not exist with a site for which such a dataset does exist and at which seasonal and DOW variations in load are believed to be very similar to those for the site in question. If such an association is made, daily and monthly adjustment factors developed using data from the associated site will be applied to data from the site in question in the absence of factors developed from a seasonal load spectra dataset for that site.

2-17 There is also the question of how much data to collect in a given month. The California data summarized in Figure 2.3 indicate that there is substantial variation in axle weights during weekends for single-unit trucks. Hence, monthly load spectra developed from a full week of data generally will be more accurate than load spectra developed from smaller amounts of data. However, this difference in accuracy is somewhat mitigated by TrafLoad’s procedure (discussed earlier) for adjusting load spectra to account for the approximate effect of any miss- ing days of the week.
2.4 Level 2 WIM Sites and TWRGs Level 2 and 3 WIM sites are sites for which site-specific WIM data are not available. Pave- ment designs for such a site are developed using either a set of default load spectra devel- oped on a statewide basis or a default set developed for a particular TWRG to which the site has been assigned. The goal of assigning sites to TWRGs is to permit the use of load spectra that better describe the axle loads at member sites than would a statewide set of load spec- tra. The following subsections provide guidance for developing TWRGs to be used for this purpose. 0.70 0.80 0.90 1.00 1.10 1 2 3 4 5 6 7 Day of Week Sunday Monday Tuesday Wednesday Thursday Friday Saturday 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. Figure 2.3 Daily ESAL Ratios

2-18 Background The basic goals in forming TWRGs for default values are to group roads at which axle loads are likely to be reasonably similar (e.g., axle loads are likely to be higher than average) and to assign roads whose axle loads are not expected to be similar to different TWRGs. The TWRGs should be defined so that every road or road segment for which load spectra may be of inter- est is unambiguously assigned to a specific TWRG. A key principle in forming TWRGs is that all weight limits should be essentially the same on all roads in the group. Thus, if some combinations are allowed to operate routinely at weights above 80,000 pounds on a set of designated roads, then the designated roads should be assigned to one or more TWRGs that are separate from the TWRGs to which other roads are assigned. Similarly, roads on which axle-weight limits vary seasonally (e.g., during spring thaw or during winter freeze) should be assigned to different TWRGs than roads on which axle-weight limits do not vary seasonally. A corollary to the above principle is that the WIM sites whose data are used for deriving load spectra for a given TWRG need not all be in the same state, provided that they are all subject to the same weight limits and that the trucks operating at the out-of-state WIM site(s) are believed to carry loads that are similar to those carried at other sites in the TWRG. One of the major influences on the load spectra for any site is the mix of empty, partial, and full loads of vehicles at the site. This influence is a significant factor affecting load spectra for combination trucks. Combination trucks on long trips (i.e., trips of more than 200 miles) are likely to be fully loaded. Past studies have indicated that only about 15 percent of combination trucks operat- ing on the rural Interstate system are empty. However, for short trips, most trucks operate loaded in one direction and empty in the other, while other trucks are used for pickup-and- delivery service, carrying partial loads for much of their trips. Since the percentage of combi- nation trucks making short trips rises in urban areas (particularly in large urban areas), the percentage of fully loaded trucks usually declines, and so do axle weights. Similarly, non- Interstate roads in rural areas also carry a mix of long-haul and short-haul traffic, so axle loads for combination trucks are also likely to be lower on these roads than on the rural Interstate system. The mix of empty and full loads can also result in significant differences in load spectra by direction. Directional differences are likely to be greatest on roads where most trucks are trav- eling to or from a particular site. For many such sites, nearly all these trucks are likely to oper- ate empty in one direction and full in the opposite direction. On the other hand, trucks oper- ating to or from a containerport usually operate loaded (and frequently very heavily loaded) in both directions. Roads that function primarily as access roads to a containerport are likely to warrant a TWRG of their own. Another major influence on the load spectra on any road is the weight of fully loaded vehicles operating on the road. If most of the trucks that use a particular set of roads carry commodi- ties that are likely to produce axle loads that are atypically high or low, that set of roads should be assigned to a separate TWRG. However, there is little reason to split a state into regional TWRGs if there are no regional variations in the commodities carried.

2-19 Guidelines The above discussion leads to the following guidelines for developing a set of TWRGs to be used for load spectra defaults: 1. If there are any significant differences in the size and weight limits applied to vehicles on different roads in the state, partition all roads into two or more sets, each with uniform size and weight limits. 2. For each of these sets of roads, develop a separate set of TWRGs, and assign the roads to these TWRGs on the basis of – Functional class, – Region, and/or – Direction. The second step should be performed using judgment, local knowledge of the trucks operat- ing in various parts of the state, and available WIM data.7 Some observations that may be use- ful in carrying out this step are the following: • There is almost certainly value in distinguishing roads by functional system: urban, rural Interstate system, and rural other. • If there are significant regional differences in the density of commodities carried (particu- larly on rural other roads), these differences may warrant either using a combination of regions and functional systems or using regions instead of functional systems. • Similarly, if, within any region, there are significant differences between the density of com- modities carried on East-West roads and that carried on North-South roads, these differences may warrant using combinations of regions, functional systems, and road orientation. • In the case of any TWRG that consists primarily or entirely of divided roads, if heavy (i.e., loaded) and light (i.e., empty) directions can be readily distinguished without using any WIM data,8 it is likely to be desirable to divide the TWRG into heavy and light directions. If practical, there should be between three and eight WIM sites in a TWRG. However, one or two WIM sites may be used for some small TWRGs. Three sites is the minimum number nec- essary to provide some confidence that all sites in the TWRG have reasonably similar load spectra. On the other hand, as the number of WIM sites in a TWRG grows, opportunities also grow for splitting the TWRG to produce smaller TWRGs, each with more uniform sets of load spectra. 7 Additional discussion of the formation of TWRGs is contained in FHWA’s Traffic Monitoring Guide (May 2001, Section 5, Chapter 3). 8 The heavy/light distinction can be useful only if, for any project site for which there are no WIM data, local knowledge can be used to distinguish the heavy direction from the light direction.

2-20 Special Purpose TWRGs Consider the case in which pavement is being designed for a site at which truck weights are believed to combine the weight characteristics of trucks operating at two Level 1 WIM sites but that is not itself a Level 1 site. One option is to assign such a site to one of the standard TWRGs (discussed above) and to use the load spectra developed for that TWRG as the load spectra for the site in question. Another option is to develop a special-purpose TWRG that is used just for this pavement design. If the only Level 1 WIM sites assigned to this TWRG are the two sites mentioned above, the load spectra developed for this TWRG will be formed as unweighted averages of those developed for the two sites. (The current version of TrafLoad does not allow for using weighted averages in developing load spectra for TWRGs.) In general, development and use of such a special-purpose TWRG will require a separate run of TrafLoad, with the two Level 1 WIM sites assigned to the special-purpose TWRG instead of the TWRG(s) to which they are normally assigned. (The current version of TrafLoad allows sites to be assigned to only one TWRG.)
2.5 Level 3 WIM Sites Level 3 design can be used when little is understood about the axle-weight distributions of the road segment being designed. Level 3 designs use the state’s average axle-load distribution curves. They can be applied to any design project in the state, but their use entails a relatively high likelihood of at least modest load estimation error because few roads are truly average. Thus, the pavement design is likely to be either over- or under-designed. Consequently, states are encouraged to use Level 2 load spectra whenever the combination of available data and engineering knowledge permits accurate assignment of a road to a TWRG with a reasonable degree of confidence. Pavement designs for Level 3 sites are developed using a set of statewide load spectra defaults that are formed as a weighted average of the sets of load spectra developed for the TWRGs. The weights used in this process are user specified. For averaging, relatively high weights should be assigned to the more important TWRGs (in terms of size and/or degree of represen- tativeness of conditions at Level 3 WIM sites) and lower weights to the less important TWRGs. Because of the wide variety of loading conditions that occur from one road to another, the error associated with statewide load spectra defaults will be large. To allow for the possibility of large errors in the load spectra for Level 3 sites, the Pavement Design Guide software pro- duces a very conservative pavement design. For roads with heavy loading conditions, such a design will result in a small probability of premature failure. However, for other roads, pave- ment will be over-designed (significantly so for roads with light loading conditions), causing state funds to be used inefficiently.

2-21
2.6 Weight Data Collection The truck weight data-collection program is somewhat different from the classification data- collection effort. Truck weight data are primarily collected using WIM technology, although a few state highway agencies also use weight data collected from static scales. In an ideal world, vehicle (and axle) weights would be collected continuously at a limited number of per- manent locations as well as at a large number of sites where only short weighing sessions would be performed. Unfortunately, both the cost of WIM data-collection and functional limitations in WIM sensor technology restrict the number and location of data-collection points at which state highway agencies can collect WIM data. WIM equipment only works accurately on flat, smooth pavements that are in good condition.9 In addition, each time a WIM scale is placed in or on a pavement, the effects of road profile and roughness on vehicle dynamics mean that the scale must be recali- brated in order to collect data accurate enough to be used as input to the pavement design process. Because calibration of a WIM scale is both time consuming and expensive, the cost of using accurate portable equipment is very high. Similarly, the cost of permanently installing WIM sensors is also high. The result is that state highway agencies generally operate relatively few WIM sites. The design of the WIM data-collection program is intended to obtain the best weight data for traffic load estimation, given these limitations. The first recommendation of the Traffic Monitoring Guide is to make sure that the data being collected are accurate. This often means that the number of WIM sites must be reduced in order to ensure that the sites that are used are supplying accurate data. The NCHRP 139 proj- ect team fully endorses this recommendation. Use of weight data from poorly calibrated WIM scales can create significant biases in the pavement design process,10 leading to unreliable pavement performance. Given a limited number of WIM locations within each state, it is recommended that those sites be distributed across the state in such a way as to discover and measure truck weight patterns that differ by geographic region and/or by type of road. Thus, in a state such as Kentucky, some scales should be placed on roads that carry significant volumes of coal trucks, while other scales should be placed on roads that carry little or no coal traffic. Data from these dif- ferent locations are then summarized to create regional load estimates that can be used within the TrafLoad and Pavement Design Guide design software. When deteriorating pavement conditions at a WIM site cause the data to become unreliable, the WIM equipment should be moved to a new location where little is known about truck weights. This allows a state to slowly expand the geographic coverage of its weight data-collection effort. Slowly expanding the geographic coverage of the WIM program helps an agency learn about 9 See the equipment descriptions in Cambridge Systematics, Inc., and Washington State Transportation Center, op. cit. 10 WIM Calibration, a Vital Activity, FHWA Publication Number FHWA-RD-98-104, July 1998.

2-22 the variations in trucking characteristics that occur in the state, while staying within available data-collection budgets. Where resources exist to collect and analyze the data collected, WIM sites should operate as permanent, continuous data-collection sites. (Note that these sites also provide continuous classification data as well as continuous volume data and thus take the place of ATRs and per- manent vehicle classifiers.) Analyses of these continuous data sources allow states to learn if truck weights are changing over time or if they change by season of year or even by time of day. These data can also be used to direct the timing of enforcement actions intended to pre- vent overloaded trucks from using the roadway system. Where resources for analyzing WIM data are extremely limited, discontinuous data may be col- lected, even from permanently mounted sensors. Limiting the data collected from permanently mounted sensors simply allows the state highway agency to focus its available resources on productive data-collection efforts. Where possible, even in cases of limited resources, sufficient data should be collected to measure possible changes in vehicle and axle weights over time.

3.0 Vehicle Classification Data Chapter 2.0 discussed the collection of weigh-in-motion (WIM) data and the capabilities of the TrafLoad software for using these data to estimate the pavement loads produced by vehicles in various VCs. This chapter addresses the complementary topic of the data required by TrafLoad to estimate the number of vehicles in each class operating at a particular site. All data provided to TrafLoad are assumed to be quality checked. As in the case of WIM sites, in this chapter different levels of classification-count sites are iden- tified, reflecting the amount and quality of the counts collected at the sites. These levels are discussed in the first section of this chapter. The second section provides a brief summary of the traffic data produced by TrafLoad for use by the Pavement Design Guide software. Sec- tions 3.3–3.5 discuss in some detail the classification data that are required by TrafLoad in order to produce these outputs. And the final section presents a recommended procedure for forecasting volumes of heavy vehicles over the design life of the pavement.
3.1 Levels of Classification Site TrafLoad provides the user with substantial flexibility in the amount of vehicle classification data to be collected for any site. The options have been grouped into three levels: 1. Sites for which continuous data from an automatic vehicle classifier (AVC) are available for periods of at least 1 week for at least 12 consecutive months. A distinction is made between 1A. Data collected at the site in question and 1B. Data collected at a reasonably nearby site on the same road. 2. All other sites for which vehicle classification counts are available. A distinction is made between 2A. Sites for which at least one AVC count is available for a period of at least 48 weekday hours and 2B. Sites for which only a manual classification count is available for a period of at least 6 weekday hours. 3. Sites for which volume counts are available but not classification counts. A distinction is made between 2-23

3A. Sites on the same road as a Level 1 or 2 site and 3B. Other sites. The assignment of sites to classification levels is independent of their assignment on the basis of WIM data. Thus, it may be practical to collect Level 1 classification data at some sites for which only Level 2 weight data are available and vice versa.
3.2 Data Produced for the Pavement Design Guide Software TrafLoad produces a moderate amount of traffic data for Level 1, 2, and 3A classification sites and a very limited amount of data for Level 3B classification sites. Brief summaries of the data produced for these two categories of sites are presented below, and more extensive discus- sions of the data required by TrafLoad for each of the levels are presented in subsequent sec- tions of this chapter. Level 1, 2, and 3A Classification Sites The most important data produced by TrafLoad’s analyses of classification counts are esti- mates of annual average daily traffic by VC, or AADTi, where the subscript i denotes VC. The VCs may be the standard FHWA truck and bus classes (Classes 4–13) or any smaller set of user-defined classes into which the FHWA classes can be unambiguously mapped. (For exam- ple, the user may define “short,” “long,” and “very long” length classes, mapping FHWA Classes 4–7, 8–12, and 13 into the three length classes.) For Level 2 and 3A sites, TrafLoad produces estimates of AADTi by direction. However, for Level 1 sites, at the user’s option, TrafLoad will produce estimates of AADTi by lane or by direction. When provided with AADTi by direction, the Pavement Design Guide software applies a lane distribution factor (LDF), which is a function of the number of lanes, to estimate AADTi for the design lane. Users of the Pavement Design Guide software may either provide the LDFs or allow the software to use its own default values. If the user requests estimates of AADTi by lane, the user must identify the design lane (or lanes) and pass the appropriate data to the Pavement Design Guide software. If the Pavement Design Guide software is told that it is receiving AADTi for a single lane, then it uses these values with- out further adjustment. This alternative will enable the Pavement Design Guide software to use better estimates of design lane AADTi, provided that the user is able to readily recognize the design lane (as is frequently the case) and provided that the actual lane distribution of heavy vehicles is not expected to change as a result of the highway improvement. (Reasons for a change in lane distribution include adding lanes or eliminating badly deteriorated pavement that is affecting the pre-improvement lane distribution). When these conditions do not hold, the user should request that TrafLoad produce estimates of AADTi by direction. 2-24

In addition to the AADTi, TrafLoad uses classification counts collected at Level 1A sites to derive or infer monthly traffic distribution factors to be used at all Level 1, 2, and 3A sites. Sim- ilarly, TrafLoad uses classification counts collected at Level 1A and 2A sites to derive hourly distribution factors (HDFs) for these sites and to infer HDFs for many other sites.1 The monthly and hourly distribution factors provide the Pavement Design Guide software with the ability to estimate how the pavement load varies by time of day and time of year. This information, in turn, is used by the software in analyzing the effects of diurnal and seasonally varying envi- ronmental factors that affect the pavement’s susceptibility to damage. Since all estimates of AADTi for Level 1, 2, and 3A sites are developed by direction or lane, TrafLoad sets an accompanying set of directional distribution factors (DDFs) to 1.0, indicating to the Pavement Design Guide software that the AADTi represents traffic in one direction only. (If not provided with these DDFs, the Pavement Design Guide software would assume that the estimates represent two-way traffic, and an appropriate default DDF would be used.) Additional information about the traffic data that TrafLoad produces for Level 1, 2, and 3A classification sites is presented in Sections 3.3–3.5. Level 3B Classification Sites The Pavement Design Guide software requires, and TrafLoad produces, just two pieces of information for Level 3B sites: • Total (two-way) annual average daily truck traffic and • The “Truck Traffic Classification” group to which the site is assigned. This information is discussed further in the second part of Section 3.5.
3.3 Level 1 Classification Sites Level 1A sites are sites at which continuously operating AVCs have been used to collect a min- imum of 1 week of classification counts for 12 consecutive months. The first two subsections below discuss these data and the several uses that are made of them. Level 1B sites are classi- fication sites that are on the same road as a Level 1A site and reasonably near that site; these sites are discussed in the third subsection below. Continuous Classification Counts The principal goal of the continuous classification-count program is the creation of factors needed to estimate annual average daily truck volumes from short-duration classification 2-25 1 For sites for which HDFs are not provided to the Pavement Design Guide software, the software uses its own set of HDF default values.

counts. The same information is also used to estimate seasonal fluctuations in truck volumes so that these changes can be accounted for in the design process. To accomplish this goal, it is necessary to measure day-of-week and seasonal variation in truck traffic and to develop factors that can be applied to short-duration counts. As illustrated by Figures 3.1 and 3.2, truck volumes vary significantly by time of day (TOD) and day of week (DOW), and different patterns exist for local (predominantly business-day) trucks and for long-distance trucks. A sufficient number of continuous-count locations are needed to measure each of the differ- ent truck volume patterns found in a state or region. This means that continuous counters should be placed on different functional classes of roads and in different geographic locations within each state. It is especially important to be able to measure the differences in truck vol- ume patterns between roads that carry primarily local truck traffic and those that serve through traffic. A good rule of thumb is that the continuous classification-count program should be roughly the same size as the traditional continuous volume count program. (The latter program, con- ducted with automatic traffic recorders, is frequently called the ATR program.) In fact, the design of the continuous classification-count program is very similar to the design of the ATR 2-26 Figure 3.1 Typical TOD Patterns 0 1 2 3 4. 5 6 7 8 9 Percent of Daily Traffic Hour of Day 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Through Trucks Business-Day Trucks Rural Cars Urban Cars

program. While the recommended continuous-count program requires a significant number of count locations, it is important to note that continuous classifiers also serve as ATRs. Thus, it is possible to use the classification counters in place of ATRs at the same time as they are used to supply continuous classification data. Such a step significantly reduces the number of contin- uous counters an agency needs and reduces unnecessary duplication of the data-collection effort. Permanent, continuous counts also provide an excellent source of information on truck vol- ume trends. In particular, all highway agencies should monitor the total volume of heavy trucks. The trend in truck volume should be examined both for each individual roadway on which a permanent data-collection device is located and for each of the geographic areas in the state. (Truck traffic tends to vary with both the economic activity taking place in a geo- graphic region and the amount of through traffic passing through the area.) Also of interest are changes in the mix of trucks. Changes in truck size and weight laws can have significant effects on the total number (and percentage of) large trucks of specific designs. FHWA’s Traffic Monitoring Guide provides wide latitude in the selection of locations where permanent classifiers are placed. For the purposes of both general monitoring and pavement design, permanent classifiers should be placed on a variety of roads throughout the state. Thus, some classifiers should be on Interstate highways and other major routes that carry 2-27 Figure 3.2 Typical DOW Patterns 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2 1.3 Daily Traffic Ratio Day of Week Sunday Monday Tuesday Wednesday Thursday Friday Saturday Recreational Car Traditional Car Business-Day Truck Through Truck

heavy through-traffic volumes. Others should be placed so that trucking patterns specific to within-state movements of freight can be monitored. Lastly, where possible, urban locations should also be monitored, so that urban truck volumes can be measured. Uses of Data from Continuous Classification-Count Sites The highest quality estimates of AADTi are those that are developed using 12 months of data from a continuous classification-count site. Such a site is referred to as a Level 1A classifica- tion site. For such a site, estimates of AADTi, monthly traffic distribution factors by VC, and hourly distribution factors are developed entirely from the classification counts obtained for the site. In addition, classification counts obtained at these sites are used for developing several types of traffic ratio.2 Two of these types of traffic ratio are monthly and DOW traffic ratios that are used for seasonal and DOW factoring of short-duration classification counts obtained at Level 2 classification sites. The use of separate sets of monthly and DOW traffic ratios for this purpose makes it possible to adopt independent definitions of the seasonal and DOW factor groups that are used for this purpose. The development and use of the seasonal and DOW fac- tor groups are discussed in the first and third subsections below, and a related concept, VC groups, is discussed in the second subsection. Monthly traffic ratios that are used for adjusting any short-duration classification count should (if pos- sible) be current year traffic ratios; that is, they should be developed from data that are collected over a 12-month period that includes the months during which the short-duration counts are collected. The resulting estimates of AADTi generally will be better estimates of AADTi for that 12-month period than estimates developed using traffic ratios developed using data from earlier 12-month periods. In particular, current-year traffic ratios will provide a better adjust- ment for any unusual conditions affecting truck volumes at the time that a short-duration count is collected. (Such conditions include unusual weather, a poor harvest, or the beginning of a sharp recession or of an economic recovery.) A third type of traffic ratio developed from classification counts obtained at Level 1A classifi- cation sites is TOD traffic ratios. These traffic ratios are combined with partial-day classification counts obtained at manual-count sites to estimate 24-hour traffic volumes, by VC, at these sites. The development of TOD factor groups is discussed in the fourth subsection below. A fourth type of traffic ratio developed from classification counts obtained at Level 1A classi- fication sites is applied to data from Level 1B classification sites and is described in a subse- quent section discussing such sites. 2-28 2 TrafLoad users should think of the “traffic ratios” used by TrafLoad as “factors.” The technical dis- tinction between “traffic ratios” and “factors” is presented in Section 4.1 of Part 1, along with an expla- nation of the advantage of using traffic ratios.

DOW Factor Groups As observed earlier (see Figure 3.2), “through trucks” and “business-day trucks” have very different DOW volume patterns. The volume of through trucks varies only slightly from day to day, while the volume of business-day trucks drops substantially on Saturdays and Sun- days. For purposes of the present discussion, we shall replace the term “business-day” with “business-week” to emphasize that our interest is in the drop-off in activity during the week- end (and not the drop-off at night). The volume pattern shown in Figure 3.2 for business-week trucks provides a good example of the importance of DOW factoring. For a site where nearly all trucks are business-week trucks, estimates of AADTi derived from weekday classification counts without DOW factor- ing will tend to be overestimates. (The plot in Figure 3.2 indicates that, on average, the over- estimates will be in the 15- to 25-percent range.) DOW factoring is designed to correct for the overestimates that are reflected in weekday classification counts. In order to use DOW factoring effectively, it is necessary to distinguish VCs and sites where business-week trucking predominates from VCs and sites where through trucking predomi- nates. Since nearly all single-unit trucks are used primarily for local service, business-week trucking is likely to be dominant for FHWA Classes 5–7 at nearly all sites. In the case of buses and combination trucks (Classes 4 and 8–13), the situation is more com- plex. At sites on the Interstate system that are more than 200 miles from a major urban area, most vehicles in these classes are likely to be through vehicles. On the other hand, in major urban areas, vehicles in these classes are more likely to be providing service of a more local nature, especially on roads that carry little or no through traffic. In these areas, the volume of vehicles in these classes is likely to be appreciably lower on weekends than on weekdays. A schematic summary of the above observations is presented in Table 3.1. The table indicates that Class 5–7 vehicles are likely to exhibit a business-week volume pattern at nearly all sites at which significant numbers of these vehicles operate, while other truck and bus classes are likely to exhibit a business-week pattern at some sites and a through pattern at other sites. For this 2-29 Table 3.1 Commonly Observed DOW Volume Patterns, by VC VCs DOW Volume Pattern 5 - 7 4 and 8 - 13 Business-Week Pattern X X Through Pattern - X Key: X Pattern is likely to exist at many sites. – Pattern occurs only under unusual circumstances.

2-30 reason, when developing DOW factor groups, attention should be focused on the bus and combination-truck classes, particularly on the most important of these classes (usually Class 9). The general approach to developing DOW factor groups is to start by distinguishing two or three such groups: • One consisting of sites at which buses and combination trucks exhibit a business-week pattern; • One consisting of sites at which they exhibit a through pattern; and • Perhaps, one consisting of sites at which they exhibit an intermediate pattern. For the purpose of developing DOW traffic ratios for each of these factor groups, there should be a minimum of three continuous classification-count sites in each group, with a larger num- ber (five to eight) used wherever possible. For many states, the two or three DOW factor groups described above will suffice. However, states with large numbers of continuous classification-count sites may wish to consider estab- lishing additional DOW factor groups. One possibility that might be considered is increasing the number of groups corresponding to intermediate “through”/“business-week” patterns. Another possibility for increasing the number of factor groups involves identifying and dis- tinguishing different DOW patterns for through trucks. In particular, the dip in through-truck volume that, in Figure 3.2, is shown as occurring on Monday is actually affected by distance from the trucks’ origins and destinations. As a result, these DOW patterns may vary by road orientation (North-South versus East-West) or by direction of travel. These differences are likely to be most significant in the mountain and western plains states, where many trucks are traveling to and from relatively distant origins and destinations. VC Groups In concept, it would be desirable to develop separate sets of DOW traffic ratios for each VC. However, attempting to do so may produce zero values for some DOW traffic ratios for uncommon classes, resulting in division by zero when traffic counts are subsequently divided by these ratios. To avoid this problem, a set of user-defined vehicle-class (VC) groups are established. In TrafLoad, VC groups used in the factoring of classification counts are called “Type 1 VC groups” to distinguish them from the WIM VC groups discussed in the preceding chapter. The Type 1 VC groups are also used in the development and application of seasonal and TOD traf- fic ratios. When defining Type 1 VC groups, a general rule is that VCs that have appreciably different DOW or seasonal volume patterns should be assigned to separate VC groups. If TOD factor- ing is to be used, this rule also applies to TOD volume patterns. As observed previously, FHWA VCs 5–7 tend to exhibit business-week and business-day volume patterns that are

2-31 stronger than those exhibited by the other VCs, suggesting that these three VCs generally should be assigned to a separate Type 1 VC group from the other VCs. When first setting up Type 1 VC groups, two VC groups may be found to be sufficient. How- ever, TrafLoad allows users to define a larger number of VC groups, and some users may wish to take advantage of this capability to divide the VC groups further. The one limitation in this process is that VCs that are rarely used should be assigned to VC groups that include one or more VCs that are frequently used. If this is not done, there is a small possibility that the TrafLoad factoring procedure will be forced to terminate abnormally in order to avoid divid- ing by zero.3 Seasonal Factor Groups As discussed above, the DOW factor groups should be designed to group sites with similar DOW patterns of truck volume. Similarly, the seasonal factor groups should be designed to group sites with similar seasonal (or month-of-year) patterns of truck volume. Toward this end, the research team makes several observations about seasonal variations in truck volumes: • As in the case of automobiles, seasonal variations in truck volumes tend to be weaker in urban areas than in rural areas. • In many areas, the highest truck volumes occur during the May–October period, and the lowest volumes occur in January. • Local influences (commodities carried, harvest season, etc.) can produce substantial site- to-site variation in the timing and intensity of the seasonal peak in truck volumes on rural non-Interstate roads. • The greater diversity of trucks using the Interstate system mutes the effects of local influ- ences on seasonal variations in truck volumes, producing more consistent seasonal patterns. The above observations suggest that, for many states, the development of seasonal factor groups might begin with the creation of an urban group and a rural Interstate group. Some consideration might also be given to creating a third group consisting of sites whose seasonal variations are in between those of the first two groups. This group might include urban IS sites with relatively high volumes of through trucks. 3 As an example, consider a DOW factor group that contains no Level 1A site at which any vehicle in a specific VC group was observed on a Sunday. In this case, the Sunday traffic ratio for this DOW fac- tor group and this VC group will be zero. Assume that a 7-day classification count is obtained for a Level 2 site that has been assigned to this DOW factor group and that one or more vehicles in that VC group are observed on Sunday at this site. Then TrafLoad will be unable to factor the Sunday count for this VC group at this site. Similar problems can also be constructed for seasonal and TOD factoring, and they are even more likely to occur if combined monthly/DOW traffic ratios are used for factoring counts obtained at Level 1B sites.

The remaining issue is how to develop seasonal factors to be applied to classification counts obtained at rural Level 2 sites that are not on the Interstate system. A simple alternative is to create a single rural non-Interstate factor group for this purpose.4 Issues relating to the devel- opment of seasonal factor groups for classification counting are discussed further in the Traf- fic Monitoring Guide (pp. 4-22 through 4-32). As in the case of DOW factors, TrafLoad develops separate sets of seasonal factors for each seasonal factor group and each Type 1 VC group. TOD Factor Groups TrafLoad uses TOD factoring to convert partial-day classification counts (collected at Level 2B sites) to estimates of 24-hour traffic volumes by VC. As in the case of seasonal and DOW traf- fic ratios, TrafLoad uses classification counts from Level 1A sites to develop several sets of TOD traffic ratios. In particular, for each of the user-defined Type 1 VC groups, TrafLoad develops a separate set of 24 TOD traffic ratios (or “hourly fractions”) for each user-defined TOD factor group. Since partial-day classification counts are almost always collected on a weekday, only weekday classification counts are used in developing TOD traffic ratios. At a minimum, the TOD factor groups should be designed to distinguish sites at which business- day trucking predominates from sites at which through trucking predominates. Additional TOD factor groups may also be created to represent intermediate situations and/or more extreme cases of business-day or through-trucking patterns. It is likely that the TOD factor groups frequently will be identical to the DOW factor groups (discussed earlier), but the soft- ware allows the user to identify differences where appropriate. (For example, sites on a road that is used primarily to access a truck terminal or warehouse that operate 24 hours per day, 5 days per week, might be treated as “through-trucking” sites for the purpose of TOD factor- ing but not for the purpose of DOW factoring.) FHWA VCs 5–7 almost always exhibit a business-day volume pattern (just as they almost always exhibit a business-week volume pattern). For this reason, when developing TOD fac- tor groups, attention should be focused on the bus and combination-truck classes, just as in the case of DOW factor groups. TOD factoring is performed only if there are sites at which partial-day classification counts are collected. If no such sites exist, it is not necessary to define TOD factor groups. Level 1B Sites Consider a classification site that is not a Level 1A site but that is on the same road as a Level 1A site. If it is believed that the two sites are sufficiently close that most trucks that pass 2-32 4 A more ambitious alternative would be to examine the seasonal patterns of all Level 1A rural non- Interstate sites, and, on the basis of this review, create two or more separate rural non-Interstate fac- tor groups. If this alternative is adopted, it will then be necessary to determine how to assign rural non-Interstate Level 2 sites to factor groups. However, if this assignment is performed well, the result- ing estimates of AADTi for these sites are likely to be better than those that would result from using a single rural non-Interstate factor group.

one of the sites pass both sites, then the site in question qualifies as a Level 1B site that is “asso- ciated” with the Level 1A site, and this information should be provided to the TrafLoad soft- ware. TrafLoad is capable of producing high-quality AADTi estimates for Level 1B sites, with the quality of these estimates depending on the similarity of the truck traffic at the two sites. TrafLoad has two procedures for producing AADTi estimates for Level 1B sites. The choice as to which procedure to use for a particular site is made automatically by TrafLoad. Brief, somewhat technical descriptions of the two procedures and how the choice is made are pre- sented below. The monthly traffic distribution factors and hourly distribution factors for any Level 1B site are assumed to be the same as those for the associated Level 1A site. Direct Scaling The simpler of the two procedures for estimating AADTi at a Level 1B site is “direct scaling.” TrafLoad uses direct scaling whenever a) Classification counts at a Level 1A site associated with a Level 1B site have been obtained for the same hours and dates as the classification counts that were obtained at the Level 1B site and b) Both sites have no more than one lane in each direction. Under these circumstances, the ratios of the counts at the two sites are used to scale the AADTi at the Level 1A site to produce estimates of the AADTi at the Level 1B site.5 Separate scale fac- tors are used for each Type 1 VC group. Factored Counts The second procedure for estimating AADTi at a Level 1B site is a factoring procedure. For this purpose, for each Level 1A site, a set of combined monthly/DOW traffic ratios is developed. For each Level 1A site, each direction, and each Type 1 VC group, 84 such ratios are devel- oped, corresponding to all combinations of the 12 months and 7 days of the week.6 Each of these ratios is developed by TrafLoad by obtaining monthly average day-of-week traffic (MADW) for a given direction and VC group and dividing by AADT for that direction and VC 2-33 5 If the user has requested that AADTi be estimated by direction (rather than by lane), then direct scal- ing would be appropriate even if Condition (b) does not hold. However, the current version of TrafLoad does not perform direct scaling in this case. 6 The use of 84 combined monthly/DOW traffic ratios allows the factoring procedure to reflect the com- bination of monthly and DOW variations in volume better than can be done with separate monthly and DOW traffic ratios (12 monthly ratios and 7 DOW ratios). However, combined traffic ratios can- not be used in conjunction with monthly and DOW factor groups that are developed independently of each other. Hence, TrafLoad uses combined traffic ratios for factoring counts from Level 1B sites and separate monthly and DOW traffic ratios, developed using data from groups of Level 1A sites, for factoring counts from Level 2 sites.

group. TrafLoad uses the traffic ratios obtained at a Level 1A site to convert short-duration clas- sification counts obtained at any associated Level 1B site to estimates of AADTi by lane. (See Part 4, Step CF, available online at http://trb.org/news/blurb_detail.asp?id=4403.) As in the case of monthly traffic ratios, the monthly/DOW traffic ratios applied to short-duration counts obtained at any Level 1B site should be “current year” traffic ratios; i.e., they should be developed from data that are collected over a 12-month period that includes the month dur- ing which the short-duration count is collected. The research team makes two observations about this factoring procedure. The first is that the conversion process involves dividing by the MADW for each Type 1 VC group. Hence, the accuracy of the resulting estimates of AADTi is affected by the similarity of the seasonal and DOW volume patterns for the VCs within each VC group but not by the similarity (or dis- similarity) of these patterns between VC groups. The second observation is that the adjust- ment procedure uses traffic ratios obtained from a single Level 1A site, a site that should have seasonal and DOW volume patterns that are very similar to those at the corresponding Level 1B site(s). For this reason, the resulting estimates of AADTi should be substantially bet- ter than those that can be produced at Level 2 sites.
3.4 Level 2 Classification Sites Level 2 classification sites are classification sites that do not qualify as Level 1 classification sites. That is, less than 12 months of current data are available for these sites, and they are not associated with another site on the same road for which 12 months of current data are avail- able. A few of these sites are continuous classification sites at which data are missing for one or more months. However, most of these sites are ones at which classification counts are col- lected as part of a state’s short-duration classification-count program. Counts collected as part of this program fall into three categories: 1. Coverage counts that are collected periodically (e.g., once every third year) at a relatively fixed set of sites to provide general information about truck volumes and how these vol- umes are changing over time; 2. Expected project counts, collected at sites at which highway projects are anticipated, to pro- vide data for use in the planning and design process; and 3. Project-specific counts, collected either to provide additional information about sites at which expected project counts have already been collected or to provide data to be used for projects that had not been anticipated. Most short-duration counts collected for pavement design projects are likely to fall into the second of these categories. That is, they are likely to be collected to support pavement design projects that are anticipated to occur in the near future. These potential projects should be 2-34

identified by highway planners as soon as practical. Planning and programming tools avail- able for this purpose include pavement management systems. The early identification of pavement projects and scheduling of classification counts requires coordination among the data-collection staff, the pavement design staff, and other agency staff involved in the programming and prioritization process. While this level of communication is not easy, it has several advantages. It allows the traffic engineering office to schedule needed counts so that they can be collected efficiently. It ensures that data are available to designers when needed, thus speeding up the design process. And finally, it provides an opportunity to collect extra counts to be used in the design of major projects. As discussed below, estimates of AADTi developed from three or four 7-day counts collected over the course of a year are likely to produce appreciably better estimates of AADTi than similar estimates developed from a single 48-hour classification count. A key to this approach is to be generous when estimating possible pavement design locations. Traffic volume and classification counts at most locations are considered to be reliable for at least 2 years. Thus, even if an expected pavement design project does not make this year’s design list, it will likely make next year’s list, and traffic data will already be collected and available for that location. Even with good communication between pavement and traffic engineering staff, it may not be possible to collect all the traffic data required as part of the routine data-collection effort. Accordingly, allowance should always be made for a possible need for project-specific counts to supplement the expected project counts. Short-duration classification counts are usually collected using automatic vehicle classifiers (AVCs). However, at some urban sites, manual classification may be preferred. Because man- ual counts usually cover only part of a day, estimates of AADTi derived from manual counts are not likely to be as good as estimates derived from accurate classification counts obtained with AVCs for periods of 48 hours or more. Accordingly, these two types of short-duration counts are distinguished from each other by calling AVC sites Level 2A sites and calling man- ual classification sites Level 2B sites. The two following subsections contain brief discussions of the collection and analysis of counts at these two types of classification sites. AVC Sites (Level 2A) Level 2A classification sites are sites at which automatic vehicle classifiers (AVCs) are used to obtain one or more classification counts over the course of a year. Accurate AVC counts usu- ally require that vehicles are traveling at constant speed with adequate spacing between vehi- cles, conditions that may be difficult to meet in urban areas. Each AVC count should cover a period of at least 48 weekday hours (though TrafLoad’s Level 2A procedure is capable of esti- mating AADTi from 24-hour classification counts). Improved estimates of AADTi will be pro- duced if count duration is extended or if multiple classification counts are collected over the course of a year. 2-35

Section 3.3 included a discussion of TrafLoad’s use of data from continuous classification- count sites (Level 1A sites) to develop sets of monthly and DOW traffic ratios. These traffic ratios are used by TrafLoad to convert the short-duration counts collected at Level 2A sites to estimates of AADTi. For any Level 2A site, the traffic ratios used are those developed for the seasonal and DOW factor groups to which the site belongs. As observed in Section 3.3, monthly traffic ratios that are derived from current year data work better for this purpose than monthly traffic ratios derived from historic data. For this reason, when providing TrafLoad with a set of Level 2 classification counts to be factored, the user should (if possible) also provide TrafLoad with Level 1A counts for a 12-month period that includes the month(s) during which the Level 2 counts were collected. The quality of the estimates of AADTi that are produced will tend to vary with the degree to which the seasonal and DOW patterns in truck volumes at the Level 2A site match the vol- ume patterns in the seasonal and DOW factor groups to which the site has been assigned. Use of 7-day counts reduces or eliminates the need for DOW factoring, and use of multiple counts over the course of a year reduces the need for seasonal factoring. Thus, 7-day counts and mul- tiple counts are strategies for increasing the amount of data collected in order to improve the quality of AADTi estimates. Hourly distribution factors for each Level 2A site are developed by TrafLoad from the counts collected at the site. TrafLoad sets the monthly traffic distribution factors for each Level 2A site equal to the monthly traffic ratios for the seasonal factor group to which the site has been assigned. Manual Classification-Count Sites (Level 2B) In order to classify vehicles reliably on the basis of axle-spacing criteria, AVCs must be located where vehicles are neither accelerating nor decelerating and where the spacing between vehi- cles is sufficient to allow consecutive vehicles to be readily distinguished. Because these con- ditions are difficult to meet in urban areas, urban classification counts frequently are collected manually. (Alternatively, classification on urban streets and roads may be limited to length classification.) Manual classification counts are usually collected only during daylight hours, usually for a period of 6 to 12 consecutive hours. Conversion of these partial-day counts to estimates of AADTi is a two-step process: 1. Each set of partial-day classification counts is converted to a set of estimates of volume by VC for the day on which the counts were collected. 2. The procedures discussed in the preceding section are used to convert these estimates of 24-hour volume by VC to estimates of AADTi. Procedures for performing the first of these two steps are discussed below. This step adds some additional error to the resulting estimates of AADTi (over and above the error intro- duced by the factoring procedures discussed above). Accordingly, sites at which classification counts are obtained manually are described as Level 2B sites. 2-36

TrafLoad uses TOD traffic ratios to convert partial-day classification counts to estimates of vol- ume by VC for the day on which the count is collected. These traffic ratios generally are devel- oped from hourly classification counts obtained for weekdays at Level 1A sites, as discussed in Section 3.3. For this purpose, each Level 2B site must be assigned to a TOD factor group that is believed to have a TOD pattern for truck volume (particularly for combination trucks) that is similar to the pattern that is believed to exist at the site in question. A separate set of TOD traffic ratios is used for each Type 1 VC group. For Level 2B sites, TrafLoad also produces a set of monthly traffic distribution factors and, at user option, it may produce a set of hourly distribution factors. As in the case of Level 2A sites, the monthly traffic distribution factors for each Level 2B site are equal to the monthly traffic ratios for the seasonal factor group to which the site has been assigned. For sites identified by the user as ones at which a business-day truck pattern exists, TrafLoad produces a set of hourly distribution factors that have the values shown in Table 3.2. For all other Level 2B sites, TrafLoad does not produce any hourly distribution factors. Instead, the Pavement Design Guide software provides its default TOD distribution.7
3.5 Level 3 Classification Sites Level 3 classification sites are sites for which volume counts exist but classification counts do not exist. There are two types of Level 3 classification site: • Level 3A sites are sites on the same road as an associated Level 1 or Level 2 site and suffi- ciently close to that site to carry a traffic mix that is similar to the mix at the associated site. • Level 3B sites are all other sites. For both types of Level 3 site, TrafLoad requires either an estimate of overall (two-way) annual average daily truck traffic (AADTT) or estimates of overall AADT and overall percent trucks from which AADTT can be derived. Level 3A Sites Level 3A classification sites are sites that are on the same road as an associated Level 1 or Level 2 site and sufficiently close to that site to carry a traffic mix that is similar to the mix at the associated site. For such sites, TrafLoad uses the estimates of AADTi by direction at the associated site to distribute the user-supplied estimate of AADTT over the various truck 2-37 7 There is a relationship between the hourly distribution factors (HDFs) required by the Pavement Design Guide software and the TOD traffic ratios (TODTRs) developed and used by TrafLoad. How- ever, the Pavement Design Guide software requires a single set of HDFs for all VCs combined, while TrafLoad generates separate sets of TODTRs for each VC group. The current version of TrafLoad does not contain a procedure for converting the TODTRs into HDFs.

2-38 Table 3.2 Hourly Distribution Factors Used by TrafLoad for Level 2B Sites with Business-Day Trucking Hour Hourly Distribution Factor 0 0.6% 1 0.4% 2 0.4% 3 0.4% 4 0.9% 5 2.8% 6 4.8% 7 6.1% 8 7.4% 9 7.8% 10 7.7% 11 7.6% 12 7.5% 13 7.9% 14 8.0% 15 7.0% 16 5.9% 17 4.7% 18 3.6% 19 2.6% 20 1.9% 21 1.6% 22 1.3% 23 1.1% Derived from data for urban other principal arte- rials (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.

2-39 classes and over the two directions of travel.8 TrafLoad sets the monthly distribution factors for the Level 3A site equal to the corresponding factors for the associated site. Similarly, the hourly distribution factors for the Level 3A site are set equal to the ones for the associated site, if they exist.9 Level 3B Sites Level 3B sites are similar to Level 3A sites except that they have no associated Level 1 or Level 2 site. Instead, users are required to assign each Level 3B site to one of 17 Truck Traffic Clas- sification (TTC) groups that have been defined by the Pavement Design Guide team.10 Table 3.3 lists the 17 TTCs along with the criteria used to distinguish among them. The Pavement Design Guide software uses the TTCs as the basis for disaggregating estimated AADTT into the standard FHWA VCs. Pavement designs developed by the Pavement Design Guide software for Level 3B classifica- tion sites require the use of load spectra for the standard FHWA VCs. These load spectra nor- mally will be developed by TrafLoad. If these are not supplied by TrafLoad, the Pavement Design Guide software will use a set of default load spectra that have been developed from national data.
3.6 Forecasts The Pavement Design Guide software requires forecasts of linear or exponential rates of change in the AADTi over the design life of the pavement. A simple procedure for estimating 8 There are two potential improvements to the current TrafLoad procedure for handling Level 3A sites. One improvement would require TrafLoad to be modified to produce estimates of overall AADT for Level 1 and Level 2 sites (instead of just AADTi for truck classes). If TrafLoad has an overall AADT value for each associated site, then, for Level 3A sites, TrafLoad would require estimates only of AADT (but not percent trucks), since the percentage of trucks could be assumed to be the same as at the asso- ciated site. An alternative improvement would entail implementing a somewhat more sophisticated (and more demanding) algorithm for analyzing Level 3A sites. This algorithm would require that total traffic be counted at the 3A site at the same time as it is being counted at the associated site, with the counts at the 3A site being obtained by direction and, if practical, for a period of at least 48 hours. TrafLoad would then estimate AADTi for the Level 3A site, by direction, by using the volume counts at the two sites as the basis for scaling the corresponding estimates of AADTi for the associated site. 9 As discussed above, TrafLoad does not create hourly distribution factors for all Level 2B sites. Accord- ingly, TrafLoad does not create hourly distribution factors for some Level 3A sites that are associated with Level 2B sites. 10 ERES Consultants and FUGRO-BRE, Determination of Traffic Information and Data for Pavement Struc- tural Design and Evaluation, NCHRP Project 1-37A, Interim Report, December 1999, pp. 39–53.

2-40 these rates of change is presented below. Some more sophisticated forecasting procedures are discussed in Part 1, Appendix A. A Simple Procedure A simple procedure for forecasting the rates of change in traffic volumes for the design lane or design direction at any particular project site is presented below. In the procedure, the rates of change are referred to as “growth rates” to emphasize that, for the purpose of pavement design, traffic growth is of primary interest. However, the procedure may also be applied to sites at which traffic is expected to decline. The procedure consists of six steps: Table 3.3 Truck Traffic Classification Groups Percentage of AADTT in Key VCs TTC Description VC 9 VC 5 VC 13 VC 4 1 Major single-trailer truck route (type I) > 70 < 15 < 3 - 2 Major single-trailer truck route (type II) 60 - 70 < 25 < 3 - 3 Major single- and multi-trailer truck route (type I) 60 - 70 5- 30 3 - 12 - 4 Major single-trailer truck route (type III) 50 - 60 8- 30 0 - 7.5 - 5 Major single- and multi-trailer truck route (type II) 50 - 60 8 - 30 > 7.5 - 6 Intermediate light and single-trailer truck route (type I) 40 - 50 15 - 40 < 6 - 7 Major mixed truck route (type I) 40 - 50 15 - 35 6 - 11 - 8 Major multi-trailer truck route (type I) 40 - 50 9 - 25 > 11 - 9 Intermediate light and single-trailer truck route (type II) 30 - 40 20 - 45 < 3 - 10 Major mixed truck route (type II) 30 - 40 25 - 40 3 - 8 - 11 Major multi-trailer truck route (type II) 30 - 40 20 - 45 > 8 - 12 Intermediate light and single-trailer truck route (type III) 20 - 30 25 - 50 0 - 8 - 13 Major mixed truck route (type III) 20 - 30 30 - 40 > 8 - 14 Major light truck route (type I) < 20 40 - 70 < 3 - 15 Major light truck route (type II) < 20 45 - 65 3 - 7 - 16 Major light and multi-trailer truck route < 20 50 - 55 > 7 - 17 Major bus route - - - > 35% Source: ERES Consultants and FUGRO-BRE, Determination of Traffic Information and Data for Pavement Structural Design and Evaluation, NCHRP Project 1-37A, Interim Report, December 1999, pp. 40 and 42.

2-41 1. Distinguish two groups of VCs: single-unit trucks and buses (FHWA Classes 4–7); and combina- tions (Classes 8–13).11 The distinction between the two VC groups permits the development of separate growth rates for single-unit trucks (which are used almost exclusively to serve the local economy) and combinations (whose usage responds to a much wider range of influences). 2. Identify all Level 1A sites for which estimates of AADTi have been developed for at least 4 years and that are believed to have historic rates of growth in the volume of heavy vehicles that are similar to those at the project site. 3. Associate the project site with one or more Level 1A sites identified in Step 2. Only Level 1A sites are used for this purpose because the AADTi estimates developed for these sites are likely to achieve a much greater level of consistency over time than estimates developed for other sites. 4. Use regression to estimate either linear growth rates or exponential growth rates for each Level 1A site for each VC group.12 In choosing between the two types of growth, a simple option is to choose the type that is believed to best describe expected future growth in truck traffic at the project site—linear growth if it is believed that the annual increase in this traffic is not likely to grow and exponential growth if this annual increase is expected to grow. (The Pavement Design Guide software has no provision for sites at which the annual increase is expected to decline over time. For such sites, linear growth should be assumed.) If this option is used, the same type of growth should be assumed for both VC groups (single- unit trucks and combinations). A slightly more complex option is to choose the type of growth that best fits the historic data at the Level 1A sites and then to modify the type of growth in Step 6. This option is discussed further in the latter part of the next subsection. If this option is used, the type of growth used in the regressions need not be limited to linear or exponential, and the regres- sions for single-unit trucks can be one type of growth and those for combinations can use a different type of growth. 5. For each VC group, average the growth rates obtained in Step 4 for the associated Level 1A sites. 6. Judgmentally adjust the growth rates on the basis of a review of macroeconomic and site-specific factors. The Step 6 review should consider any identifiable factors suggesting that future growth in heavy-vehicle traffic at the target site is likely to differ from past growth at the Level 1A sites. Factors to be considered include • Expected changes in macroeconomic trends, 11 The vehicle-class groupings recommended here are the only ones handled by the current version of TrafLoad. 12 Regression capabilities are available in most computer spreadsheets as well as in many other types of software.

2-42 • Planned and recently completed facilities that may affect the generation of truck trips, and • Planned and recently completed highway projects that may affect truck routings. This last category includes both new and upgraded feeder routes and new and upgraded par- allel facilities. An interesting example of the effect of upgrading is the recent conversion of New Mexico SR-44 from two lanes to a four-lane divided highway (and its redesignation as US-550). The upgraded facility has attracted a significant amount of truck traffic heading northwest from Albuquerque that formerly used several other Interstate and U.S. highways. The facility is now feeding an increased number of trucks onto roads heading north and west from the Farmington area. Procedures that may be used for making the required adjustments are discussed in the next subsection. Adjusting the Forecast The adjustments made in Step 6 may be made directly to the (linear or exponential) growth rates developed in Step 5. Alternatively, it may be helpful to plot the Step 5 results in a spread- sheet and to use the plot as an aid in making the adjustments. As an example, assume that, for a site of interest, AADT of combination trucks in the base year is estimated to be 1,000 and the forecast growth rate is estimated to be 3 percent per year. The solid line in Figure 3.3 represents this forecast over a 20-year period. Forecast AADT for com- bination trucks at the end of this period is 1,754. Assume that the analyst believes that some downward adjustment of this forecast to a 2-percent annual growth rate may be appropriate. The dotted line in Figure 3.3 shows this alternative fore- Figure 3.3 Three Alternative Forecasts 0 200 400 600 800 1,000 1,200 1,400 1,600 1,800 2,000 AADTCT Year 3% per Year 2% per Year to+20to Linear

2-43 cast, producing a forecast-year value of AADT for combination trucks of 1,457. The forecast-year volume is 17 percent lower, but the total volume over the entire 20-year design period has only been reduced by 9.6 percent (from 9.81 million to 8.87 million). A third forecast assumes the same truck volumes in the base year and in the forecast year as those of the second forecast, but the third forecast assumes linear growth. This third fore- cast is also shown in Figure 3.3. This forecast assumes somewhat larger increases in traffic vol- umes than does the second forecast in the early years but smaller increases in the later years. The total volume over the 20-year design period is 8.97 million, slightly higher than the total volume produced by the second forecast. In the above example, the initial forecast (Steps 4 and 5) was developed under the assumption of exponential growth, but consideration was given to substituting linear growth. Such a pos- sible substitution is always an option, subject to two restrictions on the selected form for the forecast: • It must be either linear or exponential and • It must be the same for both VC groups.13 For the purpose of pavement design, the most important traffic volume estimates are the total numbers of trucks and buses, by VC, expected over the design period. For the three forecasts discussed above, these estimates are represented graphically by the areas under the three curves in Figure 3.3. In developing these estimates, it should be recognized that the most important information consists of the base-year estimates of existing AADTi. Forecasts of over- all growth in traffic volume (which are necessarily more speculative) are less significant—in the example, a 65-percent increase in overall growth (from an increase of 457 combinations per day for the two lower curves to an increase of 754 for the highest curve) increases the total number of combinations over 20 years by only about 10 percent. Using TrafLoad Although TrafLoad does not develop traffic forecasts, it does provide the user with substan- tial flexibility for entering forecasts. These options are summarized in Table 3.4. It allows the user to provide a linear or exponential growth rate developed in Step 5 of the above proce- dure. Alternatively, modified forecasts produced in Step 6 can be described in terms of the overall change in traffic volume over the forecast period or in terms of the volume forecast for the end of the period. For Level 1, 2, and 3A sites, TrafLoad allows separate specification of forecasts for two VC groups: single-unit vehicles and combinations. However, it does not currently allow separate forecasts for individual VCs. Exponential growth rates specified for one of the two VC groups are applied to each VC in the group. Linear growth rates specified for one of the VC groups are distributed among the corresponding VCs in proportion to their base-year volumes (AADTi). For Level 1, 2, and 3A sites, if TrafLoad has been requested to estimate AADTi for 13 The second restriction applies only if forecasts are entered into the system via TrafLoad. See the next subsection.

the design lane, then linear forecasts of growth are interpreted as being for this lane; if TrafLoad has been requested to estimate AADTi for a given direction, then linear forecasts of growth are interpreted as being growth in traffic for that direction. For Level 1, 2, and 3A sites, TrafLoad also requires that the same input option and the same type of growth (linear or expo- nential) be used for both VC groups. For Level 3B sites, TrafLoad accepts only a single forecast of growth. This forecast is applied to total two-way truck volume, with no distinctions by lane, direction, or VC. 2-44 Table 3.4 TrafLoad Input Options for Forecasts User Inputs Input Option Linear Growth Exponential Growth Annual Annual change Annual percentage change Overall change Total change over period* Percentage change over period* Forecast AADT Forecast AADT for VC group* Forecast AADT for VC group* * Also requires specification of the base year and forecast year.

4.0 Data Handling This chapter discusses the data collection and handling needed to create the datasets used by TrafLoad and the relationship of this data handling to a state’s traffic data-collection program. Highway agencies currently collect, manipulate, store, and report traffic data. No fundamen- tal change in this existing data flow is required to meet the traffic data requirements of the Pavement Design Guide. • Data must still be collected from the field, preferably using modern, calibrated, automated data-collection equipment. • Data are downloaded from the devices used to collect data from the field and analyzed in the office. This analysis process includes checks for data quality, a summarization step, and a storage process that allows for later use. • These summaries are then extracted and manipulated as needed to produce the traffic load estimates required by the Pavement Design Guide. These activities are discussed in the first two sections of this chapter, and related administrative and institutional issues are discussed in Section 4.3. Extensive information on data-collection equipment is presented in a companion volume.1
4.1 Data Collection The Pavement Design Guide mechanistic design software does not require collection of new types of traffic load data. The TrafLoad data analysis system that feeds the Pavement Design Guide software uses the traditional measures of volume, vehicle classification, and truck axle weights to compute the traffic loading inputs needed. All of these measures are currently col- lected to one degree or another by every state highway agency and are discussed in FHWA’s Traffic Monitoring Guide. While all state highway agencies already collect data, it is likely that the number, timing, and location of counts that highway agencies perform will change in order to produce better traffic loading estimates. In addition, some state highway agencies may have to create new summary 2-45 1 Cambridge Systematics, Inc., and Washington State Transportation Center, Equipment for Collecting Traffic Load Data, prepared under NCHRP Project 1-39, June 2003 available online at http://trb.org/ news/blurb_detail.asp?id=4403.

output reports and data files from the data they are already collecting in order to input traffic loading estimates into the new pavement design software. These changes in count location and duration are purely voluntary. However, the availability and quality of data collected by each state will have a direct impact on the accuracy of traffic load inputs to the pavement design process and consequently on the reliability of the pave- ment designs developed with the new software. The basic data-collection design for providing traffic load data fits within the general traffic data-collection guidance provided by the FHWA in the 2001 Traffic Monitoring Guide. A key point is that a large portion of the traffic data collection required for estimating traffic loads should be collected as part of the routine traffic data-collection program. Thus, pavement design engineers need to work closely with those engineers who select, schedule, perform, and analyze the traffic data being collected. This increased level of communication will ensure that traffic load estimates can be collected cost effectively and that the summary statistics needed by the pavement designers are readily available and easily loaded into the pavement design software. Each state highway agency should have a traffic count program that, at a minimum, collects • Short-duration volume counts, • Continuous volume counts, • Short-duration classification counts, • Continuous classification counts, and • Weigh-in-motion (WIM) measurements (i.e., truck weighing). Because pavement depth is not significantly impacted by the volume of light-duty vehicles, mechanistic design is primarily concerned with the number and weight of trucks and buses using the roadway in question. Volume data collection is not discussed in this report. Uses of classification and weight data are discussed in earlier chapters, and the collection and han- dling of these data are discussed in this chapter. Short-duration classification count program elements are designed to provide site-specific vol- ume (by VC) measurements that determine the total number of axle loads on a given roadway segment. Continuous-count elements provide measures of temporal variation needed to con- vert short-duration counts into unbiased measures of average annual conditions. WIM mea- surements provide data on the weights of each axle group. These data-collection program elements provide all of the information needed for producing the traffic loading estimates required by TrafLoad. Table 4.1 summarizes the data require- ments of TrafLoad and identifies the traffic data-collection elements that provide the raw data needed to meet these requirements. Table 4.2 describes the data that each of these elements contributes to the pavement design process. Both tables identify distinctions between the three levels of classification data discussed in Chapter 3.0 and between the three levels of weight data discussed in Chapter 2.0. 2-46

4.2 Data Analysis Once data are collected from the field, the data must be analyzed. This process consists of • Quality control review of the collected data (to ensure that the equipment operated correctly), • Summarization of the data into statistics and record formats that can be readily used by others inside and outside the state highway department, and 2-47 Table 4.1 Data Required by the Pavement Design Guide Software Required Data Source for Data AADTi a for up to 13 VCs (1, 2, and 3A)b Continuous classification counts, or Short-duration classification counts adjusted for day of week and season AADT and Percent Trucks (3B) Short-duration volume counts, adjusted for day of week and season and State estimates of truck percentages (from a combination of short and continuous classification counts) Truck Traffic Classification Group (3B) Judgment Monthly Traffic Distribution Factors by VC Continuous classification counts Axle-Load Distribution Factors– Site Specific (1) Weigh-in-motion data collection Axle-Load Distribution Factors– Regional (2) Weigh-in-motion data collection–statewide program Axle-Load Distribution Factors– Statewide (3) Weigh-in-motion data collection–statewide program Linear or Exponential Growth Rate Various sources Directional Distribution Factor Set to 1.0, except for Level 3B analyses Axle Groups per Vehicle (for each VC) Weigh-in-motion data collection Hourly Distribution Factors Continuous classification counts or Short-duration classification counts a AADTi is AADT by VC. b Numbers in parentheses identify the input levels for which the data are used.

• Storage of summary statistics in a form that permits ready retrieval and use by other analy- sis tools. Data that are not reviewed, summarized, and stored for easy use simply waste the available data-collection resources. Mechanistic pavement design does not require that state highway agencies perform these tasks in a particular manner. It does require that specific output reports be made available 2-48 Table 4.2 Data-Collection Elements for TrafLoad Type of Traffic Data Collection Data Produced for TrafLoad Short-Duration Volume Counts Provides a “counted” measure of average daily traffic (ADT), which serves as an input to the computation of AADT (Class Level 3) Continuous Traffic Counts Used to compute the seasonal and day-of-week adjustment factors necessary to compute AADT from ADT values Short-Duration Vehicle Classification Counts Actual truck volumes (by type of truck) on the road for which the measurement was made (Level 1 or 2 class data) TOD distribution factors by VC Continuous Vehicle Classification Counts Day-of-week and seasonal adjustment factors for trucks Actual truck volumes for Level 1 (class) sites Monthly traffic distribution factors by VC Trend measurements used when forecasting future truck volumes Short-Duration WIM Measurements Current load spectra datasets (Weight Level 1) (if a well- calibrated site) Used in the computation of Level 2 (weight) regional axle-load spectra by Truck Weight Road Group (TWRG) and Level 3 statewide axle load spectra Used to correctly assign a specific roadway to a specific TWRG Continuous WIM Measurements Seasonal and current load spectra datasets (Weight Level 1) Day-of-week and seasonal adjustments for load spectra datasets developed from short-duration WIM measurements Used in the computation of Level 2 (weight) regional axle-load spectra by Truck Weight Road Group (TWRG) and Level 3 statewide average axle load spectra Also used for continuous classification data

from the collected data. It also requires that effective quality assurance procedures be adopted and followed in order to maintain the quality of the data being used as input to the design process. The key components of this process are discussed below. Quality Control Data-collection equipment does not always work as intended. Sensors fail, come loose, or are improperly installed. Settings can be inappropriate. The equipment may not be properly cal- ibrated, or the calibration may drift over time as environmental conditions change. In some cases, operating conditions may not allow the equipment to function as designed. Data from equipment that is not operating correctly yield inaccurate measurements of traffic loads that in turn result in poor design of pavement depths. Quality control programs are intended to identify malfunctioning or poorly calibrated equipment and to remove data col- lected by that equipment from the analysis process. In some cases, this means that additional data must be collected to replace the invalid data. In other cases, alternative data may be avail- able (e.g., loss of 2 weeks of data from a continuous-count location is not serious). Performing quality checks quickly allows repair or recalibration efforts to be undertaken quickly, which in turn prevents loss of a large volume of data. Quality control is particularly important for weigh-in-motion data, as many WIM scales are subject to calibration drift. Calibration drift of as little as 10 percent can result in errors of up to 40 percent in the estimates of pavement damage.2 For these reasons, each data-collection agency should have a quality assurance process that checks incoming data for errors. This can be a significant task, depending on the type of data collection being performed, the volume of data being collected, and the amount of automa- tion present in the traffic data processing system operated by the state highway agency. A pooled-fund study led by the Minnesota DOT developed a knowledge-based system for performing data quality checks for volume, classification, and weight data.3 Other projects, such as FHWA’s Long-Term Pavement Performance project, have also developed and pub- lished basic quality assurance procedures.4 A summary of the most common data quality checks is provided in Section 5.5 of a companion report.5 All quality check procedures compare measured traffic characteristics with a set of known val- ues. Known values are drawn either from previous data-collection experience for that location 2-49 2 WIM Calibration, a Vital Activity, FHWA Publication Number FHWA-RD-98-104, July 1998. 3 Intelligent Decision Technologies, Ltd., Traffic Data Quality Procedures, Pooled-Fund Study, Expert Knowledge Base, Interim Task A3 Report, prepared for Minnesota DOT, November 1997. 4 FHWA, LTPP Division, Data Collection Guide for Long-Term Pavement Performance Studies, Operational Guide No. SHRP-LTPP-OG-001, Revised October 1993. 5 Cambridge Systematics, Inc., and Washington State Transportation Center, Equipment for Collecting Traffic Load Data, prepared under NCHRP Project 1-39, June 2003, available online at http://trb.org/ news/blurb_detail.asp?id=4403.

or from independently measured sources. (For example, to determine if the clock on a data- collection device correctly distinguishes daytime from nighttime, 1:00 a.m. and 1:00 p.m. vol- umes might be compared using the known fact that 1:00 p.m. volumes normally exceed 1:00 a.m. volumes.) A key to the quality assurance effort is to make sure the known values against which collected data are compared are accurate measures of the expected traffic patterns. For example, traffic volume on the freeway connecting Los Angeles and Las Vegas often has 1:00 a.m. traffic vol- umes that are large enough to exceed 1:00 p.m. volumes. Thus, the check described above is not an appropriate quality control check for this location, even though it is quite applicable to most other roadways in the nation. This same key point is important when known values are used for automatically adjusting the calibration of data-collection equipment such as WIM scales. Such algorithms can work, but only when the known values are correct and when a sufficient number of vehicles cross the scale in the time period observed. If any of the key assumptions used for auto-calibration are incorrect, the auto-calibration system will not work effectively and can actually decrease the accuracy of the data collected. Auto-calibration problems may exist if the average axle weights of either passenger cars or Class 9 truck steering axles are not known, if either of these aver- ages varies over time, or if adequate samples of these two vehicle types are not observed dur- ing any calibration period. Periodic collection of independent data is required to confirm that the values used for quality assurance checks are correct. These independent tests include (1) the calibration of WIM and classifier systems when they are first installed and used at a site and (2) visual confirmation that portable classifiers are correctly functioning when they are placed on a roadway. Once the initial equipment operation can be verified, datasets can be collected and used for deter- mining the known traffic patterns against which new data are compared. This type of quality control procedure is designed to identify suspect data (i.e., data that do not fit expected patterns). If unexpected patterns are observed, additional forensic work is required. In some cases, it is readily apparent that equipment or sensors have failed. For per- manent data-collection sites, such failures indicate that repairs are needed as quickly as prac- tical. In the case of short-duration data collection, the affected data must be discarded and, usually, replaced by new data. In other cases, the unusual data are plausible but unexpected (for example the Los Angeles/ Las Vegas TOD patterns mentioned above). In these cases, additional data should be collected to confirm or invalidate the unusual data. For these second-chance data-collection efforts, par- ticular attention should be paid to setting up and calibrating the equipment to ensure that the confirmation dataset is accurate. If the new data support the unexpected traffic pattern, then the known value for this site must be updated to reflect the new information. Data Summarization Once the collected traffic data have successfully passed through the quality assurance process, an efficient mechanism is needed for storing and summarizing the data so that they can be used when needed for pavement design. Most states have existing programs that collect and store both volume and classification data on a section-by-section or count-by-count basis. In 2-50

many states, these data can be retrieved by section through the state highway agency’s geo- graphic information system. Changes in existing data summarization procedures that may be required to support mecha- nistic pavement design include the creation of some additional summary statistics that not all states currently compute and store. These statistics are intended to provide better site-specific traffic loading estimates and thus provide for better pavement designs. Among the statistics that are computed by TrafLoad for use by the pavement design software are • Seasonal (monthly) patterns of truck volumes; • TOD distributions for truck volumes; • Load spectra for different roads and roadway groups; and • Numbers of axles, by type of axle, for each class of trucks. The last two statistics have been discussed in some detail in Chapter 2.0, and other needed sta- tistics (including the first two) have been discussed in Chapter 3.0. Use of TrafLoad If TrafLoad is used to process traffic data and generate traffic data inputs for the Pavement Design Guide software, then the required data must be loaded into the system. There are two primary forms of data to be entered: • Hourly vehicle classification records from specific count locations and • Axle-load data by vehicle for specific sites. Hourly vehicle classification records are assumed to be available in the FHWA C-card (or four- card) record format. Data from both short-duration and continuous sites should be supplied in this format. While all state highway agencies can currently create C-card records easily, con- siderable change may be needed within current data processing systems in order to make hourly classification data available to pavement designers. Many states only provide access to summary statistics such as average daily traffic and overall percent trucks. Making hourly records available to TrafLoad may require modification to current systems or changes in administrative procedures used to store, request, and report traffic data. Axle-load data for individual vehicles are assumed to be available in the FHWA W-card (or seven-card) record format. As in the case of C-card records, making the W-card records avail- able to TrafLoad may require modification to current systems. It is expected that some software development work will be required at most state highway agencies to simplify the extraction of data items from existing traffic databases and to make the appropriate files accessible to TrafLoad. In most cases, these development efforts should be modest. 2-51

2-52
4.3 Administrative and Institutional Changes In preparing to use the new procedures for pavement design, one of the biggest hurdles for most states is likely not to be technical but institutional. In most highway agencies, pavement design and traffic data collection and analysis are in separate areas. This separation limits interaction between these two groups and reduces the ability of the traffic data-collection group to adopt procedures that satisfy the changing input requirements of pavement design and that meet other needs of pavement designers. Most state highway agencies already collect the data needed for estimating traffic loads for mechanistic design. However, few agencies currently summarize this information and effec- tively report it to their pavement designers. As a consequence, few pavement design groups actually use much of the load data being collected. If the mechanistic design practices are to be implemented effectively, these failings must be remedied. The key to improving the collection of load data and its conversion into effective inputs for the mechanistic design procedures is a substantial increase in the interaction between pave- ment designers and the traffic data-collection and analysis staff. This interaction should include the following: • Training for pavement designers on – What traffic data are needed, – Why those data are important, – What effect the data have on the resulting pavement designs, – Where to get the data that are collected, – How to request more data when the available data do not meet design requirements, and – How to review the traffic estimates being provided. • Training for data-collection and analysis staff on – What data are important for pavement design and what data have the largest effect on pavement design, – How the data collected are used in the design process, and – What the flow of traffic load data is in the pavement design process. • Increased communication that – Allows data-collection staff to correctly anticipate (and schedule) the data needs of the designers,

2-53 – Ensures that the data and summary statistics produced by the data-collection staff meet the needs of the pavement designers, – Ensures that the data required are transmitted to the pavement design staff in a timely fashion and in a format that can be easily loaded into the mechanistic design soft- ware, and – Involves both pavement design and data-collection staff in the review and refinement of the data-collection and summarization process used to feed the design process. (For example, are the Truck Weight Road Groups correctly defined? If not, how should they be revised? Where should additional truck weight data be collected? What other holes in the available traffic data should be remedied?) • Reviews of – The resources that are spent collecting traffic data, – The relative value to the pavement designers of the various resources, and – The potential value to the pavement program of addition expenditures on data collection. Although state highway agencies are already doing much of what is needed to meet the traf- fic data requirements of mechanistic pavement design, considerable work is still required to refine the existing procedures and software. Although data are collected, they often are not adequately summarized and reported. Resources will be needed to address these deficiencies. Resources are traditionally in very short supply for traffic data collection and analysis, and it will be necessary for the pavement design group to lend political support to the data-collection group if those resources are to be obtained. This will only happen if the pavement design group understands the importance of the traffic load data and if the traffic data-collection group can be relied on to provide those load estimates in a responsive and efficient manner. Neither of these conditions is currently met in very many highway agencies. The recommended increases in both communication and training should result in many improvements in both data-collection and pavement design processes. Fostering effective communication between the data-collection team and the pavement design group should result in traffic data-collection decisions that consider the needs of the pavement designers more effectively. Similarly, improved communication will enable pavement designers to use traffic data more effectively, thus allowing the development of more reliable designs.

2-54 Glossary AADT Annual average daily traffic. AADTi Annual average daily traffic for vehicle class i. AADTT Annual average daily truck traffic (all classes, combined). AEPV Average ESALs per vehicle. ATR Automatic traffic recorder. AVC Automatic vehicle classifier. DDF Directional distribution factor. DOW Day-of-week. ESALs (18,000-pound) equivalent single-axle loads. HDF Hourly distribution factor. LDF Lane distribution factor. MADW Monthly average days of the week. TOD Time-of-day. TTC Truck traffic classification. 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. 2-55

2-56 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.