Predicting surface urban heat island intensity and its drivers using machine learning and spatial XAI for achieving SDG-11

Surface Urban Heat Island (SUHI) is a major environmental concern especially in rapidly expanding cities with substantial impacts on urban climate, human health, and energy demand. Despite extensive documentation of the phenomenon, our understanding of the driving forces behind SUHI remains limited. Therefore, the present study is intended to analysed the drivers of SUHI and their spatial pattern of influence in Mumbai metropolitan region (MMR) using google earth engine (GEE) and machine learning (ML) models. The SUHI intensity has been calculated using ML models like random forest (RF), gradient boosting, XGBoost, and LightGBM. The divers of SUHI have been analysed using explainable artificial intelligence (XAI) and for the spatial pattern of influence of the drivers, a novel spatial XAI model has been estimated. Multi-stage feature selection involving correlation screening and variance inflation factor (VIF) analysis ensured numerical stability and interpretability. Model performance has been evaluated using multiple metrics, including R 2 , RMSE, MAE, explained variance, bias, and maximum error. Result shows that LightGBM has the best overall performance, achieving a test R 2 of 0.86 and the lowest RMSE (1.49 °C), followed by XGBoost (test R 2 = 0.86; RMSE = 1.52 °C). Spatial cross-validation produced conservative yet robust estimates (CV R 2 = 0.77–0.78), confirming strong spatial generalisation. Bootstrap-based uncertainty analysis further shows narrow confidence intervals for both R 2 and RMSE, indicating high model stability and low sensitivity to sampling variability. Graphical and spatial XAI analyses identified built-up intensity (NDBI), bare land index, night-time light intensity, and population density as dominant positive drivers of SUHII. SUHI intensity in low category has decreased from 13.69 percent in 2002 to 6.31 percent in 2024 while SUHI intensity in very high zone has increased significantly. Spatial prediction shows pronounced SUHII hotspots in densely urbanised and industrial zones, whereas coastal areas, vegetated regions, and water bodies consistently exhibited lower SUHII. The spatial heterogeneity captured by XAI highlights the necessity of location-specific mitigation strategies rather than uniform city-wide interventions. These findings provide critical evidence to enhance urban thermal comfort and resilience for achieving sustainable development goals (SDGs) such as SDG-11.3 (sustainable urban growth), SDG-11.5 (resilience to urban hazards), SDG-11.6 (reducing environmental impacts), and SDG-11.7 (inclusive access to green spaces).