Exploring multiscale relationships between environmental characteristics and recreational trail-based activities in urban natural areas: A regional study leveraging user-generated big data and machine learning

Abstract

Human recreational behaviours in nature represent context- and scale-dependent phenomena. Sustainable urban planning and natural resource management call for effective interventions that target local contexts, which require a refined understanding of human-nature interactions. However, existing studies have failed to concurrently examine multiple spatial scales, nonlinearity and spatial variation in these associations. This study leveraged big data (i.e., Strava sports tracking data for running and cycling) and advanced machine learning techniques (partial dependence plots and geographically weighted random forest) to disentangle the complex relationships between environmental characteristics and recreational activities. Key findings from our study include: (1) the influence of environmental variables on running and cycling varies across perceptual scales, as represented by route-based and survey-based navigation; (2) moreover, specific scales are more appropriate for characterising these relationships; and (3) important variables were identified both at the local level and across the region for running and cycling. Our study proposed multiscale approaches for modelling and understanding the effects of environmental characteristics. The results reinforce the need for context-specific strategies in urban planning. Practical considerations are provided for the planning and design of urban natural areas to promote recreational trail-based activities.