Predicting cycling volumes using crowdsourced activity data

Planning for cycling is often made difficult by the lack of detailed information about when and where cycling takes place. Many have seen the arrival of new forms of data such as crowdsourced data as a potential saviour. One of the key challenges posed by these data forms is understanding how representative they are of the population. To address this challenge, a limited number of studies have compared crowdsourced cycling data to ground truth counts. In general, they have found a high correlation over the long run but with limited geographic coverage, and with counters placed on routes already known to be popular with cyclists. Little is known about the relationship between cyclists present in crowdsourced data and cyclists in manual counts over shorter periods of time and on non-arterial routes. We fill this gap by comparing multi-year crowdsourced data to manual cyclist counts from a cordon count in Scotland’s largest city, Glasgow. Using regression techniques, we estimate models that can be used to adjust the crowdsourced data to predict total cycling volumes. We find that the order of magnitude can be predicted but that the predictions lack the precision that may be required for some applications.