Application of Crowd-Sourced Speed Data in Developing Free-Flow Speed Models

Abstract

Free-flow speed (FFS) is a critical input to many transportation engineering applications, including capacity estimation, congestion measurement, level of service assessment, and speed limit setting. Traditionally, practitioners have relied on prediction models from the Highway Capacity Manual (HCM) and other existing methods to estimate FFS. These methods were primarily developed and calibrated using fixed location speed data collected during nighttime or other low-volume periods. However, these data are often very limited in amount and spatial coverage due to the resources needed to collect them. This study leverages extensive GPS-based probe speed data sets to develop FFS models for various facilities, including freeways, multilane highways, rural two-lane highways, and interrupted facilities. We employed a random forest tool to identify key variables influencing FFS for each facility type, such as degree of curvature and median width for freeways, area type for multilane highways, and degree of curvature and pavement roughness for rural two-lane highways and interrupted facilities. Simplified linear regression models developed using these variables outperformed existing methods, particularly the HCM approach. The findings of this study can help transportation practitioners enhance the prediction of FFS and contribute to the knowledge base for future model improvements.