Spatial Patterns and Drivers of Surface Urban Heat Islands in East China: A County‐Level Study Based on the RH‐ SHAP Model

ABSTRACT A comprehensive understanding of the spatiotemporal heterogeneity of the urban heat island in a region is essential for formulating evidence‐based urban climate strategies. This study employed a land surface temperature (LST) dataset to calculate surface urban heat island intensity (SUHII) for 2000, 2005, 2010, 2015, and 2020 across 626 county‐level administrative units in East China. Based on methods including spatial autocorrelation, the Random Forest (RF) model, and SHapley Additive exPlanations (SHAP), as well as data from 15 influencing factors, the spatial patterns of daytime (SUHII d ) and nighttime SUHIIs (SUHII n ) at annual and seasonal scales and their influencing factors were systematically examined. The findings revealed that the multi‐year average SUHII of all counties in summer was 2.21°C, which was more pronounced than the winter (0.60°C). The values of SUHII d were generally higher than those of SUHII n , with means of 1.93°C and 0.89°C, respectively. Although the SUHII of different counties changed to some extent during the study period, the regional average value and the overall spatial pattern remained relatively stable. SUHIIs showed a significant spatial association and aggregation effects, with an obvious strong south‐weak north gradient, which was most evident for SUHII d . RF model predicted that urban‐rural difference in surface albedo (ΔAlbedo) has the largest contribution to the spatial differentiation of annual mean SUHII d , followed by urban surface albedo (Albedo), precipitation (PRE), solar radiation (SR), nighttime light intensity (NTL), the urban‐rural difference in vegetation greenness (ΔVG), and aerosol optical depth (AOD). The overall contribution of the seven indicators mentioned above was 69.97%. In contrast, ΔVG was the dominant factor influencing annual mean SUHII n (contributing 33.34%), followed by ∆Albedo and potential evapotranspiration (PET). It is worth noting that the relative predictive importance of each indicator should not be interpreted simply as a direct causal effect. For example, the physical influence of ΔAlbedo may be modulated by the cooling capacity of the rural background. Additionally, these factors exhibited complex nonlinear relationships with SUHIIs, which were primarily manifested as clear threshold effects and variable marginal effects. Differences in the dominant factors between SUHII d and SUHII n suggested differentiated mitigation strategies for decreasing SUHII: applying high‐albedo materials and cool roof technologies to reduce SUHII d , while increasing green coverage and optimizing vegetation layout to mitigate SUHII n .