An interpretable–causal learning framework for identifying the drivers of surface urban Heat Island intensity: evidence from 527 urban clusters in China

Under the combined effects of global warming and rapid urbanization, urban morphology has emerged as an important factor associated with variations in surface urban heat island intensity (SUHII) through its influence on surface physical properties and urban energy balance. Ho，wever, existing studies have largely relied on statistical correlation analyses and often lack standardized definitions of reference zones or sufficient control of confounding variables, making it challenging to disentangle environmental disturbances and identify the underlying drivers of SUHII and their spatial variability. To address these limitations, this study employs the Google Earth Engine platform to construct a multi-constraint framework for quantifying SUHII across 527 urban clusters in China and integrates LightGBM-SHAP with Double Machine Learning (DML) to establish an analytical framework that combines correlation analysis, causal inference, and mechanism interpretation. The results indicate pronounced spatial heterogeneity in summer daytime SUHII across China, with generally higher values observed in eastern and southern regions than in western and northern regions. Correlation analysis identifies impervious surface fraction (ISF) and tree canopy cover (PT) as the dominant statistical predictors. However, causal inference suggests that ISF (ATE = 0.374), largest patch index (LPI, ATE = 0.253), and water body proportion (PW, ATE = −0.240) exhibit relatively robust average estimated effects within the causal inference framework at the national scale. By contrast, PT, building height (BH), and patch density (PD) display substantial contextual dependence, with their thermal regulation effects varying across different climatic conditions, which may explain their inconsistent significance within the causal framework. By integrating correlation-based identification with causal validation, this study helps clarify the scope and limitations of urban morphological influences on SUHII and provides scientific support for the development of differentiated and evidence-informed urban planning strategies aimed at improving climate resilience.