Estimation of Average Annual Daily Bicycle Counts using Crowdsourced Strava Data

Abstract

Traffic volumes are fundamental for evaluating transportation systems, regardless of travel mode. A lack of counts for non-motorized modes poses a challenge for practitioners developing and managing multimodal transportation facilities, whether they want to evaluate transportation safety or the potential need for infrastructure changes, or to answer other questions about how and where people bicycle and walk. In recent years, researchers and practitioners alike have been using crowdsourced data to supplement the non-motorized counts. As such, several methods and tools have been developed. The objective of this paper is to take advantage of new data sources that provide a limited and biased sample of trips and combine them with traditional counts to develop a practical tool for estimating annual average daily bicycle (AADB) counts. This study developed a direct-demand model for estimating AADB in Texas. Data from 100 stations, installed in 12 cities across the state, was used together with the crowdsourced Strava, roadway inventory, and American Community Survey data to develop the count model for estimating AADB. The results indicate that crowdsourced Strava data is an acceptable predictor of bicycle counts, and when used with the roadway functional class and number of high-income households in a block group, can provide quite an accurate AADB estimate (29% prediction error).

Evaluating the Potential of Crowdsourced Data to Estimate Network-Wide Bicycle Volumes

“…A study in Miami found that Strava sampling rates were higher on streets than trails (26), while another found areas with more bike lanes to have lower sampling rates (25). However, these third-party data sources, of which most have focused on Strava thus far, have usually improved bicycle volume and safety performance models significantly, after controlling for built environment and sociodemographic factors (1,(28)(29)(30)(31)(32)(33)(34). Despite consistent findings of association, a few details are worth noting: Correlation does not necessarily track sampling ratio.…”

Quantifying changes in bicycle volumes using crowdsourced data

“…The XGBoost comes in two versions: the Linear Booster and Tree Booster (Chen and Guestrin, 2016). It is noteworthy that these machine learning algorithms have been used in many application fields in the literature including cycling and transportation (Dadashova et al, 2020; Khasawneh et al, 2020; Litzenberger et al, 2018; Mahmoud et al, 2021; Munira and Sener, 2020; Sun and Mobasheri, 2017). To compare their performance, we tuned each model parameter separately to minimize the mean absolute error (MAE) and to maximize R 2 .…”

Biking Where Black: Connecting Transportation Planning and Infrastructure to Disproportionate Policing

“…The data represent up to five percent of total cycling volume and tend to overrepresent white, middle-aged men (Lee and Sener, 2020;Nelson et al, 2020). Research has shown that, with adjustment, Strava Metro data can predict cycling volumes with under 30 percent measurement error (Dadashova et al, 2020). But others have shown that the data by themselves serve as reasonable proxies for cycling exposure (Lee and Sener, 2020), and so I use the Strava data here without adjustment.…”