BackgroundUnderstanding how urban environments stimulate routine physical activity is a central issue in public health. Running, as a low-threshold and widely accessible exercise, serves as a sensitive indicator of environmental influence. Yet, few studies have used methods capable of capturing the non-linear and context-dependent interactions through which multiple built-environment factors jointly shape running. This study investigates the spatiotemporal patterns of running and identifies how combinations of environmental features across different urban scenarios affect behavioural activation.MethodsWe analysed 83,302 GPS-tracked running trajectories from Beijing. Built-environment indicators were integrated across five dimensions and examined using Light Gradient Boosting Machine with SHapley Additive exPlanations (SHAP). To enhance behavioural interpretability, variables were grouped into scenario-informed contexts representing commuting, restorative, and training environments. Temporal and spatial analyses were also conducted to identify diurnal, weekly, and spatial clustering patterns of running activity.ResultsRunning exhibited clear morning-evening peaks, an inverted-U weekly rhythm, and a concentric spatial structure concentrated in central areas and along continuous spaces such as waterfronts and forest sports parks. SHAP interaction analysis further shows that non-linear built-environment effects are organised through configuration-specific interaction structures rather than marginal feature responses. Three dominant configurations are identified: (i) visual–landscape structures combining vegetation texture, sky openness, and water proximity; (ii) urban density configurations linking residential intensity, building form, accessibility, and nightlight intensity; and (iii) training configurations integrating facility density, route continuity, surface flatness, and shading.ConclusionsRunning behaviour is jointly shaped by temporal rhythms, spatial clustering, and structured interactions among built-environment features. The findings demonstrate that physical activity patterns reflect the conditional contribution of co-occurring spatial attributes, providing evidence for designing more health-supportive and activity-friendly urban environments.