Unveiling differential impacts of multidimensional urban morphology on heat island effect across local climate zones: Interpretable CatBoost-SHAP machine learning model

Classifying local climate zones (LCZ) improves the understanding of how urban morphology affects the urban heat island (UHI) effect. However, differences in driving mechanisms across various LCZ types are often overlooked, leading to a lack of targeted UHI mitigation measures for different LCZ types. In this study, we took Shenzhen as a case study and employed a novel interpretable CatBoost-SHAP machine learning model to evaluate the contributions of multidimensional urban morphology factors to the UHI effect across various LCZ types. We found that the driving mechanisms in different LCZ types have both commonalities and differences. For all LCZ types, the blue-green space ratio (BGSR), digital elevation model (DEM), albedo (AL), and mean building height (MBH) consistently exhibited significant negative impacts on the UHI effect, whereas space crowding degree (SCD) showed a positive impact. Although some factors such as sky view factor (SVF), building shape index (BSI), and surface unfolding ratio (SUR) also contributed to the UHI effect, the positive and negative correlations are reversed across different LCZ types. Additionally, factors such as green view index (GVI) and patch density (PD) had a greater influence on land cover types compared to built types. These findings reveal the complex interactions between urban morphology and the UHI effect, underscoring the necessity of investigating the driving mechanisms across various LCZ types. Our study can provide valuable insights for urban planners and policy- makers to develop targeted mitigation strategies for specific LCZ types, thereby effectively alleviating the UHI effect.