Skip to main content

Non-Motorized Planning & Development

Department of Transportation

Non-Motorized Planning & Development

Leveraging Crowd-sourced Data in Planning, Design, Analysis, and Evaluation of Pedestrian and Bicycle Traffic 


Project Number: SPR-1741

Contract Number: 

Status: Complete

Start Date: 10/01/2021

End Date: 08/31/2024

Summary:

Data on active transportation are difficult to collect for DOTs, MPOs and local agencies because of the short duration of trips, non-traditional time and routes used by the bicyclists and pedestrians as well as lack of extensive resources needed to track, count and map such movements. Crowdsourced data can be a great resource in this case to fill in the gap. However, crowdsourced data comes with its own data quality and coverage challenges which makes it difficult to use it as available. Through this project, the project team developed an analytical data-driven framework, resulting in reliable, reproducible and transferable statistical models that can be used to predict bicycling and pedestrian volumes in road networks using crowdsourced data. The project team first surveyed multiple state DOTs through an online survey to identify relevant crowdsourced data sources and their advantages and disadvantages. Then the project team acquired and processed data from 43 permanent Using volume data from 43 counters in Michigan and crowdsourced data from Strava along with several other land use, network and weather-related variables from multiple open access resources, the project team developed two types of models separately for bicycling volume, pedestrian volume, and bicycling and pedestrian volume for Michigan. The first set of models, inspired by the gravity models from traditional travel demand forecasting models, developed the ‘Active Expansion Factor’ models with prediction accuracy for bicycling volume ranging between -20% to +50%, while that for pedestrian volume ranging between 100 to 200%. The second set of models used mixed effect zero-inflated count models to identify more nuanced causal relationships between volumes and land-use, weather and infrastructure-related variables. Findings indicate that Strava data are significantly correlated with active transportation volume data, especially bicycling volume data, but better model fits and prediction can be achieved when Strava data is augmented by land use and in case of bicycling, weather-related variables.

 

Publications:

  • Final Report

  •