• Correcting weather forecast data errors based on deep learning algorithms. • Optimizing irrigation frequency and trigger thresholds based on model predictive control. • Integrate weather forecast data for real-time irrigation decision optimization. As climate variability intensifies, short-term weather forecasts have become increasingly important for irrigation decision. However, forecast errors and traditional rule based irrigation practices can jointly reduce decision accuracy, resulting in either excessive water application or crop water stress. To address this issue, this study proposes an integrated irrigation decision framework that combines deep learning-based weather forecast error correction (WFEC), AquaCrop simulation, and model predictive control (MPC), using Shihezi, Xinjiang, China, as a case study. In the proposed framework, deep learning algorithms are first used to correct real time forecast errors in meteorological variables, and the corrected weather inputs are then incorporated into the AquaCrop–MPC framework to optimize irrigation frequency and stage specific irrigation trigger thresholds during the maize growing period, thereby enabling dynamic irrigation adjustment. Results indicate that the mean absolute error (MAE) reductions for maximum temperatures (T-max) and minimum temperatures (T-min) were 10.8% and 12.3%, respectively. The strategy with an irrigation frequency of 20–25 days achieved high water use efficiency, and the optimal irrigation thresholds for the four growth stages during the growing period were G1: 25%–30%, G2: 35%–40%, G3: 45%–50%, and G4: 75%–80% of soil available water (SAW). Compared with traditional irrigation decisions, the proposed framework provides more precise real time irrigation optimization by integrating forecast error correction with crop simulation and predictive control.
Zhang et al. (Wed,) studied this question.