What heats cosmopolitan cities in summer? A Remote Sensing and AI study of land surface temperature drivers: Casablanca vs Barcelona

Urbanization and climate change are increasingly intensifying the urban heat island effect, especially during summer, by changing the relationship between land cover and local thermal conditions. This study investigates how five environmental indicators (NDVI, NDMI, NDBI, NDBSI, and albedo) influence land surface temperature (LST) across Local Climate Zones (LCZs) in two cities with contrasting climates and urban forms, Casablanca and Barcelona, during the summer period (July 1–August 31) from 2014 to 2024. Landsat 8 imagery and the WUDAPT database were combined with seven machine-learning regression models. Among these, the Extra Trees Regressor demonstrated the strongest performance, achieving the highest R² values (0.96–0.98 in training and 0.59–0.85 in testing), the lowest error metrics, and the best bias, and was therefore selected to assess the contribution of each predictor by LCZ class and year. Results show that vegetation, represented by NDVI, plays a leading role in regulating LST in Barcelona, with the highest feature importance values (0.30–0.40), consistent with its Mediterranean setting. In Casablanca, thermal variability is more strongly associated with urbanization-related indicators, particularly NDBI and NDBSI, with feature importance reaching 0.57. In addition, Barcelona exhibits a statistically significant upward trend in summer LST, with mean Sen’s slopes of 0.31°C/year in built LCZs and 0.34°C/year in land-cover LCZs, according to the Mann–Kendall test. Casablanca shows a comparatively more stable thermal pattern. These findings highlight that LST–environment relationships depend on both LCZ structure and regional climate, providing insights for urban heat mitigation and climate-adaptation strategies.