Histogram-based gradient boosting machine with SHAP-driven interpretability for predicting intensity of urban heat Island effect

This study presents a novel framework for the analysis and prediction of urban heat stress. To conduct the analysis, urban Land Surface Temperature (LST) data were retrieved from Landsat 8 during the dry seasons in 2006, 2014, and 2022. Hoi An's urban center, a Vietnam’s UNESCO World Heritage Site, is selected as the study area. The urban LST prediction in this study was driven by a comprehensive set of 23 input features that capture the multifaceted nature of the urban environment. The input features include topographical, spatial, and urban morphological factors. Moreover, for a better explanation of thermal patterns in the study region, past records of LST are also taken into account. The feature maps were standardized to a spatial resolution of 30 × 30 m. Histogram-Based Gradient Boosting Machine (HBGBM), a state-of-the-art machine learning approach, is used to generalize a functional relationship between LST and the influencing factors. Experimental results showed a great capability of HBGBM with a coefficient of determination (R2) = 0.97. Moreover, Shapley Additive Explanations is used to analyze the impact of each explanatory factor on the prediction outcome. HBGBM is also employed to predict land use/land cover (LULC) changes in Hoi An. The model is trained on LULC transition data from 2014 to 2022. Future projections of the urban morphological factors are used to forecast the spatial variation of LST. Based on the predicted LST, the proposed framework is subsequently utilized to perform projections of the urban heat island effect in 2030. The prediction results obtained from the proposed framework help identify hotspots of urban heat stress in the study area ’s heritage sites. The research findings contribute to a deeper understanding of urban heat stress and support informed urban planning in Hoi An.