The Nonlinear Influence of Built Environment on Multi Period Running Activities in Streets Based on Random Forest Model: A Case Study of Shenzhen

Investigating the spatiotemporal characteristics of how a built environment influences street-running activities is crucial for advancing public health and constructing healthy cities. However, existing research lacks a detailed analysis of the mechanisms between the built environment and running behavior, particularly from a spatiotemporal perspective, often assuming a singular linear relationship. Taking Shenzhen as a case study, this research employed various urban spatial data sources, including running trajectories, street views, and POI data, to analyze the spatiotemporal differentiation characteristics of street-running activities. Random forest analysis was used to explore the non-linear relationship and threshold effects between street-running flow and the built environment, which consists of six dimensions: natural environment, density, diversity, street design, transportation convenience, and landscape accessibility. The results indicate that (1) there are significant differences in the spatiotemporal distribution of street-running flow in Shenzhen. Temporally, residents' running behavior on the streets was characterized by distinct morning (T 05:00-09:00) and evening (T 17:00-22:00) rush hours, with the most notable activity occurring around 7:00. Spatially, the street-running flow in Shenzhen exhibited a pattern of "scattered points and continuous lines." (2) Among the variables influencing street-running flow, the distance to water, sky openness, and the NDVI had the most significant impacts. The average importance scores for these three factors across the four time-period models are 14.39%, 8.53%, and 8.30%, respectively. Conversely, the impact of transportation facility density and park or square density is minimal, averaging below 5% across the four time-period models. (3) A pronounced nonlinear relationship and a threshold effect were evident between the built environment variables and street-running flow across the four time periods. NDVI and sky openness exhibited a nonlinear positive correlation with the overall street-running flow across various time periods, with thresholds of 0.4 for sky openness. A non-linear negative correlation was noted between slope, interface richness, distance to water and street-running flow in different time periods. A complex nonlinear relationship was observed between functional density, green vision rate, relative walking width, proportion of flow safety facilities, NQPDA and street-running flow, and the impact effects of the five variables vary at different time periods. This research addresses the limitations of prior studies that primarily considered singular temporal features and single relationships when examining the relationship between the built environment and running behavior. By identifying the optimal built environment range that influences street-running flow using random forests, this study offers nuanced guidance for shaping fitness spaces and designing pedestrian-friendly streets in Shenzhen. For example, reasonably optimize the ecological space of street landscapes, reduce the inhibitory effects of street terrain and interfaces , and finely adjust street functional facilities and interface layout.