From Counting Stations to City-Wide Estimates: Bicycle Volume Extrapolation With Multi-Source and Sample Data

Abstract

Switching to cycling in urban areas reduces greenhouse gas emissions and improves the health of society as a whole. In order to promote cycling as a mode of transport, accurate information on the volume of passing bicycles is essential for cities to plan infrastructure development strategically. Currently, most cities can only rely on data from sparsely located counting stations. To address this problem, we extrapolate data from these stations to estimate city-wide bicycle volumes for Berlin. Our work involves machine learning models and various public data sources, including app-based crowdsourcing, bike sharing, motorized traffic data, and more. In addition, we simulate performance improvements by conducting sample counts at predicted locations. By providing the model with ten days of count samples for the predicted locations, we can cut the error in half and significantly minimize the variation in performance between predicted locations.