Enhancing the Spatial Transferability of Direct Demand Models for Estimating Pedestrian Volumes at Intersections

Abstract

Direct demand (DD) models are an important tool for estimating annual average daily pedestrian traffic (AADPT) for all intersections in a jurisdiction. These models associate socioeconomic and land-use variables with pedestrian exposure and allow the estimation of AADPT for sites where pedestrian counts are not (readily) available. However, some jurisdictions lack pedestrian volume counts from a sufficiently large number of intersections to develop their own DD model or do not have the institutional resources to carry out the model development. Under these circumstances, a cost-effective alternative is to use DD models that were developed in other jurisdictions. Previous research evaluated the spatial transferability of DD models in scenarios where no pedestrian counts are available (i.e., naïve transferability) and showed that this resulted in large estimation errors. This paper examines methods to improve the estimation accuracy of spatially transferred DD models by using AADPT that is readily accessible to jurisdictions (we call this local calibration). Five local calibration models were proposed and evaluated using observed field counts and synthesized counts from three jurisdictions. The best model to use is a function of the number of local jurisdiction sites for which pedestrian counts are available. When pedestrian volume is available for 10% of the sites, Model C presented the best results for the synthetic approach: an average improvement of 8.7% when comparing the locally calibrated and naïve estimates. Using real AADPTs and very limited samples for local calibration, Model C also presented the best performance: an average improvement of 35.0%.