Interpreting the role of the urban-rural gradient in the spatial variation of heat exposure: Path analysis and nonlinear effects across 24 Chinese megacities

Urban heat exposure (UHE) has intensified with rapid urban expansion, yet most studies still treat urban-rural differences as a binary contrast. This study develops a gradient-based framework to examine standardized log-population-weighted UHE (sLPUHE) across 24 major Chinese cities. Using globally comparable GHS-SMOD data, grid-level urban-rural gradients (URGs) were defined from rural clusters to urban centres. Double machine learning (DML)-based pathway analysis was used to examine the overall URG-sLPUHE association and the contributions of environmental pathways, while accumulated local effects (ALE) analysis was applied to identify nonlinear local responses and gradient-specific interactions. The results show that sLPUHE does not increase uniformly along the URG. The strongest positive association occurs during the transition from dense urban clusters to urban centres (g6-g7), where the overall association reaches 0.636 and the pathway-explained component reaches 0.514. By contrast, the g4-g5 transition shows a negative association, indicating that heat exposure does not follow a simple linear increase from rural to urban space. Pathway analysis identifies LST as the dominant contributor to the URG-sLPUHE association, while AH, BD, NDVI, and DEM show transition-specific contributions. ALE results further show that LST has the largest local response range, whereas AH, BD, NDVI, and DEM exhibit nonlinear responses that vary across gradient transitions. The strongest local heterogeneity is again concentrated in g6-g7. The findings suggest that URG is useful for identifying where environmental conditions and heat-exposure burdens change most sharply along urban development. The proposed framework provides evidence for gradient-sensitive heat mitigation strategies across expansion, development, and consolidation stages of urbanization.