An integrated ML-powered geospatial analysis of surface urban heat island and its mitigation in Riyadh City, Saudi Arabia

Understanding the spatiotemporal dynamics of surface urban heat islands (SUHI) and the influence of land features on their formation is crucial for effective climate-resilient urban policies. Using warm-season ECOSTRESS and Landsat data for Riyadh City, this study aimed to provide information on daytime and nighttime land surface temperatures (LSTs) and diurnal ranges, estimate SUHI intensity, and quantify the local influences of four key land features on LSTs. A geospatial modeling framework that leverages the predictive power of machine learning (ML) was applied. The city had a daytime surface urban cool island (SUCI) and a SUHI at night. While SUCI intensity varied from −0.3 to −1.6 °C, SUHI intensity varied from 2.8 to 3.4 °C, depending on how the non-urban reference area is defined. The city exhibited a smaller LST diurnal range than the surrounding desert. Seven ML models were explored and CatBoost and XGBoost demonstrated the best performance for daytime and nighttime LSTs, respectively. Surface albedo, bare ground, built-up surfaces, and vegetation cover have strong predictive modeling power and are important for mitigating LST. However, location was the most important feature for predicting LSTs, indicating that any mitigation action should be location-targeted within the city rather than a one-size-fits-all approach. All the land features demonstrated nonlinear interactions with LSTs, indicating that effective mitigation strategies must target the ranges in which interventions produce the most cooling effects. The findings can play a crucial role in shaping effective climate-resilient urban policies for the city and other hot desert cities worldwide.