Comparing Direct Demand Models for Estimating Pedestrian Volumes at Intersections and Their Spatial Transferability to Other Jurisdictions

Abstract

Direct demand (DD) models are used to estimate pedestrian volumes at intersections as a function of readily available variables, such as land use and socioeconomic features. The objectives of this paper are: (1) to identify and qualitatively assess existing DD models in the literature; and (2) to evaluate the spatial transferability of DD models for estimating annual average daily pedestrian traffic (AADPT) at signalized intersections. Six DD models developed from jurisdictions with varying characteristics were selected for spatial transferability assessment. The models were applied to three jurisdictions (Milton, Canada; Pima County, U.S.; and Downtown Toronto, Canada) that had notable differences in the level of pedestrian activity, land use, and socioeconomics. Observed pedestrian volumes were obtained for sites in each jurisdiction. The DD models performed considerably differently across jurisdictions. Five of the models performed reasonably well for Milton, a jurisdiction that is comparable to those considered in the calibration of the selected DD models and that shares characteristics with many suburban Canadian and U.S. jurisdictions. Overall, the applications for Pima County and Downtown Toronto, which have extremely low and high pedestrian volumes, respectively, provided poor accuracy. This paper demonstrated the potential for transferring existing DD models to other jurisdictions; but also identified the clear need for further research to improve the spatial transferability of DD models.

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

“…Sobreira and Hellinga investigated the spatial transferability of DD models available in the literature (9). The authors chose six DD models with varying features in relation to the number of significant explanatory variables in the model, calibration dataset sample size, and level of pedestrian activity in the original calibration jurisdiction.…”

“…In addition, when CCs are available from a sufficient number of intersections then the CC data can be used to expand STCs to AADPT. The literature suggests that CCs should be available from between 6 and 25 stations as a minimum, depending on the characteristics of the jurisdiction (9), for the expansion of STCs to provide reliable estimates of AADPT. Because of resource limitations, jurisdictions may have sites for which no recent pedestrian counts are available, which prevents appropriate jurisdictionwide exposure estimates (according to the Federal Highway Administration [10], pedestrian STCs must be renewed every 3 to 7 years).…”

“…There is a relatively large body of literature on DD models. Different models considering varied explanatory variables, calibration sample sizes, modeling techniques, and types of jurisdictions have been developed and are available for the community (a comprehensive literature review on the development of DD models is available in other works [9,11,12]). This leads us to the prime motivator behind this study: the development of adequate DD models requires that jurisdictions have valid pedestrian data (ideally in the form of AADPT) for a sufficiently large number of sites to support the development (i.e., calibration and validation) of a DD model.…”

“…To the authors' knowledge, only one study investigates the spatial transferability of DD models to other jurisdictions. Sobreira and Hellinga evaluated the transferability of six DD models available in the literature to three different jurisdictions and found that the accuracy of the model estimation varied considerably depending on the characteristics of the jurisdiction that the model is being transferred to (referred to in this work as the target jurisdiction) (9). The evaluation methodology assumed that no information on pedestrian volume was known for any sites in the target jurisdiction, resulting in a ''naı¨ve'' transferability.…”

“…The next sections of this work are organized as follows: Section 2 summarizes the results found by Sobreira and Hellinga (9). Sections 3, 4, and 5 present the methods proposed for the local calibration.…”