Timely and accurate estimation of winter wheat yield is vital for food security under increasing climate variability. However, conventional statistical models often lack generalizability, while process-based models such as AquaCrop suffer from parameter sensitivity and calibration complexity, limiting predictive accuracy in heterogeneous fields. To overcome these limitations, this study developed a hybrid framework integrating UAV-based multispectral and thermal imagery with the PROSAIL radiative transfer model, machine learning algorithms, and the AquaCrop model through a Least Squares Method (LSM) data assimilation approach. Two-year field experiments (2023-2025) were conducted under multiple irrigation and nitrogen treatments. Leaf area index (LAI) was retrieved from PROSAIL simulations coupled with machine learning and used to calculate canopy cover (CC). Soil water content (SWC) at 0-10 cm and 10-20 cm depths was estimated from UAV-derived multispectral, thermal, and texture features. These variables were assimilated into AquaCrop to constrain growth simulations. The hybrid PROSAIL-XGBoost model improved LAI inversion accuracy by 17-19% over empirical methods (R2 up to 0.82). Random Forest achieved stable SWC inversion across depths and stages. AquaCrop simulations calibrated with field data yielded accurate results in 2023-2024 (R2 = 0.81, RMSE = 0.57 t ha-1) but declined in 2024-2025 (R2 = 0.63, RMSE = 0.85 t ha-1). Incorporating UAV-derived CC and SWC via LSM significantly enhanced yield estimation, with the CC + SWC_10 + SWC_20 scheme performing best (R2 = 0.75, RMSE = 0.65 t ha-1). Compared with open-loop simulations, data assimilation reduced RMSE by 17-25% and increased R2 by 7-19%. Beyond improving predictive accuracy, this framework helps explain how real-time biophysical and soil moisture constraints reduce deviations in crop simulations under meteorological stress. The results suggest that soil water dynamics play a more important role in yield formation than canopy development in this semi-arid, film-mulched system, and contribute to a better understanding of crop-water-radiation interactions under variable climate conditions.