Accurate prediction of soil hydraulic properties is crucial for effective water resources management in arid and semi-arid regions. This study presents an ensemble machine learning approach that combines Random Forest (RF) and Cubist models to predict and map soil hydraulic properties with improved accuracy and spatial reliability. The method was applied in Qazvin Province, a semi-arid area in central Iran, using 150 surface soil samples (0-30 cm deep) and covariates derived from Landsat 8, Sentinel-2, and SRTM DEM data. Field capacity (FC) and the permanent wilting point (PWP) were modeled separately and combined to derive available water capacity (AWC), enhancing model interpretability. The RF-Cubist ensemble consistently outperformed individual RF and Cubist models for all properties, achieving the highest accuracy for FC (R-2 = 0.70, RMSE = 3.97%), PWP (R-2 = 0.71, RMSE = 1.81%), and AWC (R-2 = 0.65, RMSE = 3.95%). Uncertainty analysis confirmed its robustness, with the lowest standard deviations (FC = 3.98, PWP = 1.81, AWC = 3.96) and coefficients of variation (11.95%, 13.62%, 19.90%). Moran's I revealed significant residual clustering for RF (I = 0.096, p < 0.05), but not for Cubist (I = 0.008, p > 0.05) or RF-Cubist (I = 0.045, p > 0.05), demonstrating improved spatial performance and reduced spatial uncertainty. These results suggest that the ensemble approach enhances both the predictive accuracy and spatial reliability of soil hydraulic property mapping, supporting better land management and precision agriculture practices in semi-arid environments.