Abstract: The WOFOST crop model can be used to simulate the spring maize growth and yield under varying water conditions. This study aimed to improve the representation of the drought-induced water stress. A dynamic and adaptive framework was developed to predict the yield using data assimilation and parameter optimization. Multi-year, multi-sowing-date field experiments were conducted at the Jinzhou Agrometeorological Experimental Station in Northeast China. Three scenarios were selected to capture the soil moisture: normal water years (2011—2013), typical drought years (2014, 2015, and 2018), and mild drought years (2019, 2023, and 2024). A systematic evaluation was also made on the different water regimes. An Ensemble Kalman Filter (EnKF) data assimilation was implemented using Leaf Area Index (LAI) and Soil Moisture (SM) observations after field experiments. Key photosynthetic parameters, including the maximum leaf CO₂ assimilation rate (AMAXTB), were optimized after the iterative calibration against the biomass and yield data. Additionally, the soil water stress response function (RFWS) was substantially modified to better represent the nonlinear effects of the soil moisture deficits on the photosynthetic processes and biomass accumulation. The oversensitivity of the original function to the slight moisture was reduced below the critical thresholds, previously leading to the excessive yield underestimation under drought conditions. Multiple simulation schemes were combined with the different assimilation and optimization strategies, including: S1 (original model), S2 (LAI assimilation only), S3 (LAI assimilation with parameter optimization), S4 (SM assimilation only), S5 (SM assimilation with parameter optimization), and S9 (SM assimilation with both parameter optimization and RFWS modification). Each scheme was evaluated to compare the above-ground biomass and final yield with the field observations under both normal and drought conditions. The better performance was achieved after the integrated approach. In the normal-precipitation years (2011—2013), the LAI assimilation with the parameter optimization (Scheme S3) achieved the highest accuracy, thus reducing the mean absolute relative errors (MARE) of the yield and above-ground biomass from 38.3% and 42.4% in the original scheme to 11.5% and 14.2%, respectively. The effectiveness of the LAI assimilation was observed under adequate moisture conditions, where the leaf development was primarily driven by the yield formation. In drought years (2014—2015), the SM assimilation with the parameter optimization and modified RFWS function (Scheme S9) performed best, thus lowering the yield and biomass MARE from 54.1% and 32.7% to 34.1% and 18.0%, respectively. There was a severe meteorological drought in 2014, while the actual yield exceeded expectations, due to the carry-over soil moisture from the previous wet year (2013). Scheme S9 captured more effectively than the rest, though some yield underestimation persisted, indicating the challenges in the extreme drought simulation. Rolling forecasts were implemented over the growing season. The optimal prediction was represented approximately 60 days after the jointing stage, corresponding to the critical period of the water demand for the maize and the phase when the yield components were determined largely. A differentiated assimilation was developed using a cumulative precipitation threshold of 149 mm during this period: Scheme S3 (LAI assimilation with photosynthetic parameter optimization) was applied when the accumulated precipitation exceeded 149 mm, while Scheme S9 (SM assimilation with photosynthetic parameter optimization and modified water stress function) was implemented when the accumulated precipitation fell below or equal to this threshold. The 2015, 2018, 2019, and 2024 sowing dates were validated for the differences between the simulated and observed yields of -649 kg/hm² in 2018 and 58 kg/hm² in 2019. The yield trends were captured over varying water conditions. The framework of the data assimilation, parameter optimization, and water stress function modification was substantially enhanced to simulate the spring maize growth and yield over diverse water conditions. This framework can provide a practical tool for yield prediction, drought impact assessment, and decision support under varying water conditions. Future work can be expected to refine the drought response and multi-factor parameterization, including the temperature and radiation. The framework can be extended to the crops and ecological regions in order to improve the model stability and applicability.