Transferability of Bicycle Volume Models Built with Crowdsourced Data Including Applications to Statewide Prediction and Corridor Analysis

Abstract

Bicycle volume estimates play an essential role in active transportation planning, policy, and analysis. To date, agencies have relied primarily on permanent and short-term counts to capture bicycle traffic at relatively few locations around their state, region, or locality. Count program challenges have led to enthusiasm for newly emerging crowdsourced “big” data on cycling activity. Crowdsourced user data collected via apps such as Strava or by passively recorded location-based service smartphone data present a new option, but concerns persist about representativeness. This study extends previous research that developed data fusion methods using various combinations of contextual location “static” data and crowdsourced bicycle user data. We report on various attempts to transfer the existing pooled models to new locations around Washington state, both before (2019) and after (2022) the COVID-19 pandemic. Consistent with previous research, models combining crowdsourced (in this paper, Strava) and static contextual variables perform best. Combined models are also less reliant on re-estimating transferred models. Somewhat surprisingly, prepandemic 2019 models held up reasonably well when faced with postpandemic 2022 data. To conclude, final models are applied to estimate volumes along potentially high-risk crash corridors in a novel proof of concept study for future safety analysis.