Examining nonlinear relationships and spatial heterogeneity of influencing factors on potential ridesourcing and public transit substitution through geographic explainable artificial intelligence

Understanding the substitution between ridesourcing and public transit is crucial for developing governance strategies. However, the nonlinear relationships and spatial heterogeneity of the influencing factors remain underexplored. This study applies a geographic explainable artificial intelligence (GeoXAI) framework to investigate these relationships at a fine-grained scale in Shanghai. Potential substitution trips are first identified from spatiotemporal availability and service comparability to calculate the potential substitution rate (PSR) for each spatial unit. The PSR is then modeled by integrating geographic coordinates and other features into automated machine learning. GeoShapley decomposes predictions into global feature, intrinsic location, and spatial interaction effects, while a consistency index assesses explanatory reliability. Trip duration, bus stop density, and intrinsic location are the most important variables. Socioeconomic, built-environment, transit network, and ridesourcing factors exhibit substantial nonlinear relationships and spatial heterogeneity. These findings support location-specific strategies for coordinated multimodal transport governance.