Forecasting Urban Heat Island Dynamics in Morocco Using Multi-Sensor Remote Sensing and Machine Learning: A Multi-Decadal Analysis Across Five Contrasting Climatic Settings

Urban heat islands (UHI) represent one of the most consequential manifestations of anthropogenic land-surface modification, yet their behavior in arid and semi-arid environments remains insufficiently characterized over long temporal horizons. This study presents a comprehensive, multi-decadal (1995–2024) analysis of surface UHI dynamics across five Moroccan cities—Laayoune, Béni Mellal, Taza, Tangier, and Ifrane—selected to span the country’s pronounced climatic gradient from hyper-arid to humid mountainous settings. Using a reproducible Google Earth Engine workflow, monthly land surface temperature (LST) and normalized difference vegetation index (NDVI) were derived from multi-mission Landsat Collection 2 Level-2 products, complemented by ERA5-Land air temperature and humidity reanalysis data and Copernicus C3S annual land cover classifications. A standardized eight-direction radial transect sampling design (0–7 km at 1 km intervals) was employed to compute UHI intensity as the thermal contrast between urban cores and peripheral zones. The analytical framework encompasses six successive phases: exploratory spatial-temporal characterization, seasonal decomposition, environmental driver assessment, robust trend detection using Mann–Kendall and Sen’s slope estimators, predictive modeling through SARIMA, Random Forest, XGBoost, and LSTM architectures, and urbanization-impact evaluation through land-cover stratification. Results reveal strongly city-specific thermal regimes: Tangier exhibits a persistent classical UHI pattern, while Ifrane, Béni Mellal, Laayoune, and Taza display recurrent Urban Heat Sink (UHS) episodes. No statistically significant long-term monotonic trend in UHI intensity was detected in any city. Predictive model comparison demonstrates that SARIMA excels in cities with regular seasonal LST structures (R² up to 0.94 in Ifrane), while machine-learning approaches outperform in contexts with irregular thermal signals (R² = 0.90 for Random Forest in Taza). Urbanization amplifies UHI selectively: a clear positive relationship between urban land-cover fraction and UHI intensity emerges in Tangier and Laayoune, whereas local cooling factors dominate in Ifrane, Béni Mellal, and Taza. These findings underscore the primacy of local climatic context and surrounding land-cover characteristics over urbanization level per se in governing UHI behavior, and they support the deployment of multi-model forecasting frameworks for anticipating urban thermal stress under diverse environmental conditions.