U‐Shaped Stability in Urban Mobility Flows: Revealing Environmental Drivers via Explainable Machine Learning in Wuhan and New York City

Understanding the inter‐day stability of urban mobility flows is critical for city planning. However, current literature often relies on binary backbone extraction that overlooks the full statistical distribution of recurrence and lacks quantitative rigor in explaining its environmental drivers. Based on taxi trajectory datasets from Wuhan, China, and New York City, United States, spanning approximately half a year, our analysis first reveals that flow stability follows a characteristic U‐shaped distribution, comprised of vast sporadic movements and a distinct, near‐daily persistent core. Second, we demonstrate that this polarized structure is systematically encoded in the urban built environment, as evidenced by the high predictive accuracy of our class‐balanced XGBoost classification model. Third, we use SHAP to quantify the underlying mechanisms, identifying spatial impedance as the primary filter that restricts high‐stability flows to short‐range and predominantly intra‐district connections. Furthermore, persistent flows are functionally anchored by the simultaneous presence of key facilities, such as healthcare and transportation hubs, at both the origin and destination. Ultimately, these results demonstrate that the built environment acts as a decisive filter that drives this polarized U‐shaped structure, strictly enforcing the formation of a rigid, near‐daily core, thereby providing a systematic basis for optimizing urban spatial structure and strategic infrastructure planning.