Mitigating urban heat island through urban-rural transition zone landscape configuration: Evaluation based on an interpretable ensemble machine learning framework

Research methods for mitigating urban heat islands (UHIs) have been widely documented. Nevertheless, the importance of mitigating UHIs through landscape allocation in urban-rural transition zones (URTZs) has rarely been emphasized in the context of intra-urban land scarcity and urban expansion in China. This study aimed to quantify the binary relationship between URTZ's landscape configuration and urban heat island intensity (UHII) by using an interpretable ensemble learning framework in Harbin, a megacity in China. After URTZ's identification, this study integrated Boruta algorithm, SHAP, ALE (interpretable machine learning techniques) and 7 tree-based machine learning models to assess the importance of URTZ's landscape configuration with both global and local angles. The results indicated that: construction land contributed most, with construction land ratio (23.20 %), separation degree (15.95 %), and maximum patch index (15.03 %) ranking highest. This was followed by agricultural land landscape shape index (10.31 %) and landscape diversity (9 %). Maintaining construction land ratio at 50-70 % can keep UHII unchanged; UHII at the grid landscape level can be alleviated when separation degree between construction land patches was above 0.7. The largest construction land patch within the grid was maintained at 20-40 or 50-70, which will not bring significant changes to UHII. The agricultural land landscape shape should be as simple as possible to reduce UHII; landscape diversity greater than 0.6 can reduce UHII, and <0.6 can increase UHII. These findings provide valuable insights into UHI mitigation and offer strategic guidance for ecological planning to promote sustainable development of large cities in rapidly changing URTZs.