Spatially varying impacts of the built environment on physical activity from a human-scale view: Using street view data

Abstract

Developing evidence-based planning interventions for promoting physical activity (PA) is considered an effective way to address urban public health issues. However, previous studies exploring how the built environment affects PA over-relied on small-sample survey data, lacked human-centered measurements of the built environment, and overlooked spatially-varying relationships. To fill these gaps, we use cycling and running activity trajectories derived from the Strava crowdsourcing data to comprehensively measure PA in the central city area of Chengdu, China. Meanwhile, we introduce a set of human-scale, eye-level built environment factors such as green, sky, and road view indexes by extracting streetscape characteristics from the Baidu street-view map using the fully Convolutional Neural Network (CNN). Based on these data, we utilize the geographically weighted regression (GWR) model to scrutinize the spatially heterogeneous impact of the built environment on PA. The results are summarized as follows: First, model comparisons show that GWR models outperform global models in terms of the goodness-of-fit, and most built environment factors have spatially varying impacts on cycling and running activities. Second, the green view index restrains cycling activities in general. In contrast, it has a wide-ranging and positive impact on running activities while hampers them in the PA-unfriendly old town. Third, the sky view index stimulates cycling activities in most areas. However, it has a mixed influence on running activities. Fourth, the road view index widely promotes cycling and running activities but hinders them in some areas of the old town dominated by automobiles and under construction. Finally, according to these empirical findings, we propose several recommendations for PA-informed planning initiatives.