This study investigates the spatiotemporal dynamics of Land Surface Temperature (LST) and Surface Urban Heat Island (SUHI) intensity in Kabul Province, Afghanistan (2000-2024), by integrating multi-source satellite data with climatic, topographic, and surface biophysical variables to identify environmental drivers and predict spatial patterns. Monthly LST time series were standardized and analyzed using the Mann-Kendall test and Sen's slope estimator, while the Random Forest (RF) model was applied to classify and predict LST across Land Use and Land Cover (LULC) classes. LST peaked at 47 degrees C in July 2023 and dropped to-4 degrees C in January 2006. Daytime mean LSTs were highest in bare land (41 degrees C) and urban areas (38 degrees C), followed by water (34 degrees C) and vegetation (32 degrees C). At night, urban surfaces remained the warmest (23 degrees C). The Mann-Kendall test revealed a nonsignificant long-term trend (p = 0.145, Z = -1.459), indicating short-term seasonal fluctuations. Hotspot analysis identified significant summer SUHI clustering in highly urbanized and sparsely vegetated areas (Kabul City, Deh Sabz, Bagrami, Surobi), while winter SUHI was minimal due to snow cover and higher surface albedo. The RF model achieved strong performance (RMSE = 2.33-2.46; r = 0.61-0.88) across MODIS, Landsat, and ERA5 datasets. This integrated remote sensing and machine learning framework provides a scalable approach for monitoring urban thermal environments and supports climate-adaptive urban planning and sustainable land management in semi-arid regions.