Estimating AADT on Non-Coverage Roads

This project developed several models to obtain accurate estimates of annual average daily traffic (AADT) at non-coverage locations. The developed models include kriging, regression and point-based. The standard kriging approach is modified in this project to use a default value when its predicted value is over a user-specified threshold (referred to as “hybrid kriging model”). Specifically, when a sampled coverage location is found to have a high absolute error using the kriging method, it is assumed that the surrounding non-coverage locations will also have AADT errors if kriging is used. In such cases, the mean AADT, based on county and functional class, is used as the AADT estimate. Two types of regression models were developed: regular and quantile. The statistically significant variables in the regular regression model are Urban, Single Line, Other Type Median, and Left-Turn Lane. The statistically significant variables in the quantile regression model are Urban, Single Line, Other Type Median, Right-Turn Lane, Left-Turn Lane, and Parking Lot. The point-based model, based on the work of Portland State University and sponsored by Oregon Department of Transportation (Oregon DOT), was calibrated using South Carolina Department of Transportation (SCDOT) data. It predicts that the AADT is 125 vehicles per day when a non-coverage location has zero points (i.e., features), 175 with one point, 350 with two points, 650 with three points, 900 with four points, 1,600 with five points, and 1,800 with six or seven points. Comparison of these models against the use of default values shows a 21.37% improvement for the hybrid kriging model, 22.82% for the point-based model, 17.03% for the regular regression model, and 23.19% for the quantile regression model. To facilitate the implementation of the developed models, an Excel-based tool was created, where the hybrid kriging model serves as the primary model because it provides comparable improvement to other models, but it does not require the SCDOT to collect any additional data. The tool also allows the user to use the predicted AADT from either regression models or point-based model if the road features are available. Other configurable parameters include an absolute error threshold for when a default value should be used instead of the kriging estimate and a reduction factor to account for discrepancy between coverage counts’ mean AADT and non-coverage counts’ mean AADT.

Language

  • English

Media Info

  • Media Type: Digital/other
  • Edition: Final Report
  • Features: Figures; References; Tables;
  • Pagination: 61p

Subject/Index Terms

Filing Info

  • Accession Number: 01835282
  • Record Type: Publication
  • Report/Paper Numbers: FHWA-SC-21-07
  • Contract Numbers: SPR No. 749
  • Files: NTL, TRIS, ATRI, USDOT, STATEDOT
  • Created Date: Feb 7 2022 9:41AM