Abstract: Soil moisture is a fundamental determinant of crop growth, with root-zone soil moisture providing the primary and direct water supply for plants. Consequently, high-precision, real-time, and large-scale monitoring of surface and root-zone soil moisture is essential for effective drought monitoring, mitigation, and disaster risk reduction.Based on FY-3D MERSI data, this study develops a crop root-zone (0-50 cm) soil moisture inversion model for Liaoning Province using a feature selection framework that integrates feature importance analysis with a stepwise incremental strategy. A random forest algorithm is employed for model training and prediction. Building upon soil parameters, meteorological factors, and remote sensing indices, the root-zone model further incorporates surface soil moisture inversion results as an additional input feature, resulting in improved model accuracy.The results demonstrate that (1) surface soil volumetric water content (SVWC) is the dominant factor governing soil moisture inversion, and its combination with soil properties (bulk density, field capacity, and wilting point) and meteorological variables leads to a substantial improvement in model performance.(2) The model demonstrates excellent performance on the validation dataset, achieving an R2 greater than 0.959, with RMSE and MAE values lower than 0.009 and 0.006 cm3/cm3, respectively, reflecting strong predictive accuracy and model stability.(3) Across different crop growth stages from 2021 to 2023, the inversion results yield mean R2 values above 0.80 and RMSE values below 0.035 cm3/cm3, indicating good spatiotemporal generalization and robustness of the proposed model.The results indicate that the proposed model enables large-scale and timely estimation of root-zone soil moisture, providing robust scientific support for agricultural drought assessment and precision irrigation.