MEPDG Traffic Loading Defaults Derived From Traffic Pooled Fund Study

CHAPTER 7-DATA SELECTION CRITERIA FOR DEVELOPMENT OF MEPDG AXLE LOADING DEFAULTS AND RESULTS OF DATA ASSESSMENT

This chapter contains the following information:

• A review of the data selection criteria used for generation of the NCHRP 1-37A MEPDG traffic defaults.⁽³⁾
  
  
• A recommended approach for developing new data selection criteria.
  
  
• A summary of the new proposed data selection criteria.
  
  
• An assessment of the extent of SPS TPF data available in LTPP SDR 24 that satisfy data selection criteria for development of the updated MEPDG traffic loading defaults.

REVIEW OF DATA SELECTION CRITERIA USED FOR ORIGINAL DEFAULTS

The requirements for sufficient traffic data used by the NCHRP 1-37A research team were based on criteria summarized in the earlier FHWA LTPP data analysis report.⁽¹⁵⁾ These criteria took into consideration findings presented in an internal working paper of the LTPP Traffic ETG in July 1997. The criteria for sufficient data availability were defined as follows:

• The availability of at least 210 days of vehicle classification data (to develop vehicle classification and truck volume based defaults).
  
  
• The availability of at least 1 weekday and 1 weekend of WIM data per quarter, preferably 1 week per quarter as a minimum (to develop traffic loading defaults).
  
  
• Availability of these data items for at least 2 years in a 5-year period.

In addition, the NCHRP 1-37A team developed WIM and AVC data collection guidelines to achieve several user-specified levels of data reliability (or maximum expected error) based on selected levels of confidence.⁽³⁾ These guidelines were based on observed error in traffic estimates based on data availability for selected LTPP GPS sites and did not take into account errors associated with data quality.

When these criteria were developed in the late 1990s, very limited research had been done on the quality of LTPP WIM data. The logic behind the 210 days per year and the weekday/weekend availability per quarter was to make sure that researchers included the effects of seasonality since there was minimal knowledge at the time of what those seasonal effects actually were. As a result, the criteria represent a conservative approach that assumes that such variation exists when, in fact, in many cases there are no seasonal effects.

Another limitation of these criteria is the focus on data availability and lack of documented criteria to address data quality (such as WIM equipment precision and bias requirements).

APPROACH FOR DEVELOPMENT OF NEW DATA SELECTION CRITERIA

Focus on WIM-Based Traffic Loading Defaults

The scope of this study was to use SPS TPF WIM data to improve existing MEPDG traffic loading defaults, such as NALS and number of axles per class and axle group type. As such, data selection criteria for developing MEPDG loading defaults focused on identifying WIM data quality and availability requirements to develop unbiased and accurate NALS estimates.

Maximize Use of Available Quality WIM Data

To take full advantage of the available SPS TPF WIM study, the data selection criteria focused on maximizing the use of available data from SPS TPF WIM sites without compromising the quality of NALS estimates. WIM data selection criteria identified for this study address the following three data selection categories:

• Data availability: Maximized use of available data that satisfy data quality criterion.
  
  
• Data quality: Identification and utilization of reliable, high-quality axle loading data.
  
  
• Data reasonableness: Identification of data that are not only accurate but also represent typical/expected conditions for a given site or a group of sites that have similar traffic loading patterns.

The third category addresses cases when data may be valid but are atypical due to a special short-term non-repeatable event (typically lasting less than 6 months) that took place at a site, when the loading pattern observed at a site is significantly different from other sites due to unique truck traffic generators in the vicinity of the site or adjustments made to vehicle classification algorithm, or when truck volumes for a particular vehicle class are so low, given a specific truck class distribution at the site, that data are not sufficient to produce a representative axle load spectrum for that class.

For analysis of traffic patterns observed at the individual SPS TPF sites, atypical is defined as a traffic trend that occurs during a short period of the pavement life, typically less than 6 months (but possibly longer), such as during a construction event occurring near the site. Traffic is correctly measured during that period, but it is not representative of the long-term traffic conditions for the site. These data would be useful for site-specific MEPDG analysis but should not be used to generate global defaults.

The following sections provide detailed requirements that were developed for the three data selection criteria based on analyses presented in chapter 6.

AXLE LOADING DATA AVAILABILITY CRITERIA

MEPDG NALS defaults are built based on data from annualized NALS rather than monthly NALS and represent percentile distribution of axle counts by axle weight categories for a typical day of the year. The reason is that monthly fluctuations in percentages of heavy or light axle loads are likely to be driven by local conditions and are not applicable on the national default level. As such, site-specific annual NALS selected for generation of defaults should be representative of annual axle loading conditions for each site included in computation of defaults. Therefore, minimum data availability criteria should be set to assure that site-specific axle load spectra developed based on the selected data will be representative of loading conditions for a typical day of the year for each site.

To define the criteria, it is important to understand how RANALS are computed for each site. Annual NALS estimates are based on averaging of monthly NALS estimates, and monthly NALS estimates are based on averaging of DOW estimates. DOW spectra are based on averaging of load spectra for the same DOW within a month. Based on the computational procedure and intended use of NALS, is it important to use procedures that would avoid introducing bias in the computation process. More details on computational procedure for RANALS development are provided in chapter 8.

Since NALS are subject to DOW and monthly variation, a number of statistical analyses to evaluate the potential for DOW and monthly errors in annual NALS computations were conducted in this study. Based on the statistical analyses presented in previous chapters, the following conservative minimum data availability criteria were identified to remove any potential DOW and monthly bias from the computation of RANALS for individual WIM sites:

• Availability of at least 1 of each 7 DOW per month to avoid potential DOW bias. Days do not need to be consecutive.
  
  
• Availability of at least 1 of each 12 calendar months to avoid potential seasonal bias. Months do not need to be consecutive or from the same calendar year (use of data from a dataset longer than 4 consecutive years was not tested in this study).

These criteria account for temporal axle load data variability and limit bias due to the computation process, thus facilitating the development of representative annual axle load spectra for individual WIM sites.

When more than 7 DOW per month or more than 12 calendar months of acceptable quality data are available, all available data should be used in the computation of site-specific RANALS.

AXLE LOADING DATA QUALITY CRITERIA

WIM data selected for the development of traffic loading defaults should accurately represent axle loading conditions observed at each site selected for the generation of defaults. To achieve this objective, the data should be of a known acceptable quality. The following acceptable data quality criteria were identified for this study based on a review of documentation from the SPS TPF study, results of the statistical and MEPDG analyses presented in the previous chapter, and recommendations from the LTPP Traffic ETG:

• Data should be from a WIM site conforming to the ASTM E1318-02 specification for type I WIM system accuracy and performance requirements specified in the LTPP Program's internal field operations guide for SPS WIM sites, as evidenced by documented WIM installation reports, and annual or semiannual validation/calibration reports.^(6,12)
  
  
• Selected data must pass a quality review using LTPP QC procedures for the SPS TPF study and be identified as level E data in the LTPP SDR database.

Analysis of NALS that had been adjusted based on representative precision and bias values obtained from SPS TPF data indicated that these values are not likely to produce significant differences in MEPDG outcomes compared to the true axle weight values.

Based on these criteria, maintaining current LTPP requirements for WIM accuracy is an important measure to have reliable data for studies in this area. Current requirements could be further enhanced by limiting the maximum acceptable bias in axle weight measurements to less than 5 percent. Potential calibration drift between field calibration sessions should be checked periodically (by comparing current NALS against NALS computed using 2 weeks of data after calibration), and calibration activities should be scheduled or data flagged if data analysis indicates likely bias over 5 percent.

AXLE LOADING DATA REASONABLENESS CRITERIA

Temporal Consistency

In addition to data quality and availability criteria, another set of data checks was developed in this study called "data reasonableness criteria." The purpose of these checks was to assure that RANALS computed for each WIM site were not biased by short-term events that could take place at the site. These events could include calibration drifts, equipment malfunction, or local business events affecting axle load distribution over some period of time.

Data reasonableness checks focused on evaluating the temporal consistency in monthly NALS distributions, primarily for class 9 vehicles, and included the following:

• Excessive cumulative absolute difference in monthly NALS check.
  
  
• Significant shift in monthly peak loads for class 9 single and tandem axles check.
  
  
• Load peaks outside typical range for the site check.
  
  
• Load peaks outside typical range for class 9 check.
  
  
• Load peaks spread over four load bins check.
  
  
• Cumulative percentage of overloads is over typical upper range for a site check.
  
  
• Cumulative percentage of overloads is over typical upper range for single axle class 9 check.

In addition, monthly NALS for all vehicle classes were checked to identify possible vehicle misclassification cases based on the following:

• High percentages of light loads in the first load bin.
  
  
• Axle group type and APC values checks.

Detailed descriptions of these checks and outcomes of data reasonableness assessment were provided in the previous chapter. Data that were flagged based on data reasonableness checks were excluded from computation of defaults.

Representativeness

Data reasonableness criteria also address the evaluation of whether a spectrum computed for a site can be considered representative of a specific loading condition and, therefore, used for development of the defaults. NALS with unusual loading distributions were flagged and investigated further to determine the reasons for such distributions. NALS based on valid and sufficient data were then identified as special cases and were not included in the computation of typical axle loading conditions. Invalid data were removed from further use.

Where very few axles were observed in a specific load spectra over the course of a year, the load spectra at that site was considered unstable and was not included in the computation of defaults. For example, there may be 3 years of data collected at a WIM site that has low volumes of class 7 vehicles. These data are valid and correctly represent loading condition observed at a site, with valid zero volumes for some days of the month. But the amount of data may be too low to develop NALS with well-defined loading distribution representative of a typical loading condition associated with a given vehicle class and axle group type, and these data should not be used as inputs for developing national default NALS for that vehicle class.

To be included in the computation of the default axle load spectra, a site had to produce at least 100 axles of that axle configuration for that vehicle class to ensure that the shape of the normalized load spectra was not unduly affected by a single random axle.

DATA ASSESSMENT SUMMARY

Data selection criteria were used to assess the extent of SPS TPF WIM data included in LTPP SDR 24 that satisfied these criteria. Table 22 shows axle loading data availability by the number of occurrence of different calendar months for each SPS TPF site. This table includes only the months that pass all data selection criteria (quality, availability, and reasonableness) for a given site.

SHRP = Strategic Highway Research Program.

Data availability was assessed using data for all truck classes (FHWA classes 4 through 13) combined. Each site in table 22 has at least 1 of each 12 calendar months with at least 7 DOW of axle loading data to facilitate development of the site-specific RANALS and APC coefficients and the updated MEPDG traffic loading defaults. For example, for class 9 vehicles, all 26 SPS TPF sites have sufficient data to compute the defaults. However, based on the specific VCDs observed at each of the SPS TPF WIM sites, not all vehicle classes and axles type have the same data availability. For some sites, underrepresented vehicle classes (such as classes 7, 11, 12, and 13) and axle group types (such as, tridems and quads) may have less than 12 calendar months of non-zero data with less than one week of non-zero axle counts per month due to low volumes observed for these vehicle classes.