Unraveling nonlinear impacts of seasonal climate and built environments on exercise walking in high-density cities via a modified machine learning approach.

backgroundPhysical inactivity is a major health risk worldwide, while walking is one of the most accessible forms of exercise that improves public health and supports sustainable urban mobility. Yet the combined and nonlinear effects of the built environments and seasonal climate on exercise walking in high-density cities remain insufficiently explored. This study aims to uncover these relationships and provide insights for health-oriented and climate-adaptive urban planning.

methodsCrowdsourced walking trajectory data were analyzed for three representative high-density Chinese cities, Beijing, Wuhan, and Guangzhou, covering both summer and winter. A comprehensive variable system was established, incorporating built environments, seasonal climate, and socioeconomic factors. A geographically weighted extreme gradient boosting model was developed with Bayesian optimization and cross-validation to improve robustness. Interpretability was achieved through Shapley Additive Explanations, partial dependence plots, and clustering analysis to identify global and local drivers of walking activity.

resultsThe geographically weighted extreme gradient boosting model outperformed traditional regression and other machine learning models in prediction accuracy. Walking trajectories showed clear spatial clustering, with central urban cores as hotspots, and seasonal differences most pronounced in Beijing. Walk Score was consistently the most stable and influential factor across cities and seasons. Among climatic variables, air quality and temperature had the strongest impacts, particularly in winter. Variables exhibited three types of nonlinear responses: sustained growth (such as Walk Score and pedestrian street length), threshold-sensitive (such as intersection density and population density), and fluctuating patterns (such as air quality and housing prices). Local cluster analysis revealed three context-specific patterns: environment-driven areas such as parks and campuses, function-driven commercial centers, and structurally imbalanced or transitional zones.

conclusionsExercise walking in high-density cities is shaped by both seasonal climate variability and spatial heterogeneity of the built environments. Improving pedestrian infrastructure, managing density thresholds, and implementing climate sensitive design can mitigate adverse weather impacts and foster year-round walking. Tailored strategies, including enhancing microclimate resilience in ecological zones, optimizing density and functional mix in commercial districts, and restructuring fragmented large blocks, are essential to create pedestrian friendly, health oriented, and climate adaptive cities.