Rainfall in arid and semi-arid regions is characterized by strong temporal variability at monthly and annual scales as well as pronounced spatial heterogeneity in its distribution. Therefore, accurate estimation is crucial for analyzing spatio-temporal patterns and supporting water resources management. In this study, monthly precipitation over Khuzestan Province during 2010–2022 was modeled using satellite-based GPM rainfall data, mean air temperature, land surface temperature (LST), elevation, and geographical coordinates. Initial analyses using linear (Pearson) and nonlinear (Spearman, Kendall-Tau, and mutual information) methods indicated that GPM shows the strongest relationship with precipitation (r = 0.85), while LST exhibits a strong negative Spearman correlation (−0.74), air temperature also shows a negative Spearman correlation (−0.73), and elevation has a weaker positive correlation (r = 0.35). Five methods, including Extra Trees (ET), XGBoost (XGB), LightGBM, Kriging, and Inverse Distance Weighting (IDW), were applied. The results demonstrated that ET achieved the highest performance with R² = 0.91, RMSE = 7 mm, MAE = 4.9 mm, MAPE = 26.8%, and MBE = −1.7 mm. XGB ranked second with R² = 0.77 and RMSE = 11.4 mm, although it tended to underestimate peak precipitation values. Among traditional methods, IDW performed better than Kriging with R² values of 0.71 and 0.69, respectively; however, both exhibited higher errors and positive MBE values, indicating overestimation. Elevation-dependent analysis further showed that underestimation errors increased with altitude across all models. Overall, integrating satellite-based and climatological data with the Extra Trees model provides an effective approach for improving precipitation estimation in arid and semi-arid regions.