Mapping ridership using crowdsourced cycling data

Cycling volumes are necessary to understand what influences ridership and are essential for safety studies. Traditional methods of data collection are expensive, time consuming, and lack spatial and temporal detail. New sources have emerged as a result of crowdsourced data from fitness apps, allowing cyclists to track routes using GPS enabled cell phones. Our goal is to determine if crowdsourced data from fitness apps data can be used to quantify and map the spatial and temporal variation of ridership. Using data provided by Strava.com, we quantify how well crowdsourced fitness app data represent ridership through comparison with manual cycling counts in Victoria, British Columbia. Comparisons are made for hourly, AM and PM peak, and peak period totals that are separated by season. Using Geographic Information Systems (GIS) and a Generalized Linear Model we modelled the relationships between crowdsourced data from Strava and manual counts and predicted categories of ridership into low, medium, and high for all roadways in Victoria. Our results indicate a linear association (r2 0.40 to 0.58) between crowdsourced data volumes and manual counts, with one crowdsourced data cyclist representing 51 riders. Categorical cycling volumes were predicted and mapped using data on slope, traffic speeds, on street parking, time of year, and crowdsourced ridership with a predictive accuracy of 62%. Crowdsourced fitness data are a biased sample of ridership, however, in urban areas the high temporal and spatial resolution of data can predict categories of ridership and map spatial variation. Crowdsourced fitness apps offer a new source of data for transportation planning and can increase the spatial and temporal resolution of official count programs.