Quantifying crop yield responses to irrigation deficits using aquacrop and XGBoost-SHAP: Insights from the Ningxia irrigation district

Efficient water management in arid regions requires a deep understanding of the nonlinear and temporally sensitive responses of crop yield to irrigation deficits. This study focuses on the Ningxia Irrigation District in the upper Yellow River Basin, a region where climate-induced water scarcity is intensifying irrigation–hydropower conflicts. Using the calibrated AquaCrop model and meteorological data from 1960 to 2019, we simulated the yield responses of spring wheat, spring maize, and rice under uniform and differentiated monthly irrigation deficit scenarios. The Gompertz function was used to identify irrigation deficit sensitivity thresholds, while a machine learning framework combining XGBoost and SHAP values quantified the relative importance of monthly irrigation throughout the growth periods. Results indicated that (1) spring wheat and rice exhibited marked nonlinear yield responses to irrigation deficits, while spring maize displayed a near-linear response within the explored water deficit range. (2) May and June were identified as critical irrigation months for all crops using the SHAP analysis. (3) Implementing a SHAP-weighted irrigation prioritization strategy enhanced yield outcomes under constrained water supply, exhibiting a bell-shaped benefit curve as a function of water deficit. This integrative approach supports the development of reservoirs multi-objective operation models that reconcile food production and hydropower generation.