Examining Spatial Transferability of Direct-Demand Models for Estimating Cyclist Counts at Intersections

Abstract

Jurisdictions need estimates of bicycle activity levels for road safety and infrastructure planning studies. When counts are not available, direct-demand (DD) models can be used to estimate bicycle activity (often as annual average daily bicycle [AADB] counts) as a function of demographic and network data. However, not all jurisdictions have sufficient count data or resources to develop their own DD model and would benefit from applying a DD model developed in another jurisdiction. However, there is little prior evidence of the spatial transferability performance of DD models or of the factors that affect spatial transferability performance. This paper addresses this gap. The spatial transferability of five DD models from the literature was evaluated across four target jurisdictions: the City of Toronto, the City of Milton, and the Region of Waterloo in Canada, and Pima County, AZ, USA. True AADB data from continuous counts were available for all four target jurisdictions. Spatial transferability was quantified using four metrics. Results demonstrated generally poor spatial transferability, with root mean squared errors (RMSE) up to 600 times higher than those reported in the original model development data sets. Moreover, analysis showed only moderate correlation between model accuracy and the similarity of the target and development jurisdictions (both for average levels of cycling activity and site and network characteristics). These findings underscore a pressing need for enhanced methods to improve the spatial transferability of DD models for estimating bicycle counts.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 4 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.