Key points are not available for this paper at this time.
Soil moisture ( SM ) is a critical variable in hydrological, agricultural, and climatic systems, yet its accurate estimation remains challenging, particularly in arid and semi-arid environments where ground-based observations are scarce. This study evaluates the performance of five machine learning algorithms Decision Tree (DT), k -Nearest Neighbor (kNN), Random Forest (RF), Light Gradient Boosting Machine (LightGBM), and CatBoost for surface soil moisture prediction across three representative Moroccan regions. Multi-source satellite data, including land surface temperature (LST), normalized difference vegetation index (NDVI), precipitation, evapotranspiration, and terrestrial water storage, were integrated to train and validate the models. Model performance was assessed using Root Mean Square Error ( RMSE ), Mean Absolute Error ( MAE ), and Nash–Sutcliffe Efficiency ( NSE ). The results show that CatBoost achieved the highest predictive accuracy ( NSE = 0.94 ), followed by LightGBM and RF, while simpler models such as DT and kNN demonstrated lower generalization ability. These findings confirm the superiority of ensemble learning algorithms for soil moisture prediction under data-scarce conditions. The proposed approach contributes a transferable, satellite-based framework for accurate and scalable soil moisture monitoring, supporting sustainable water resource management and climate-resilient agriculture in arid and semi-arid regions.
Elmotawakkil et al. (Wed,) studied this question.