Data-Driven Modeling of Closed-End Border Irrigation Performance Evaluation Methods

Closed-end border irrigation is widely used in China, yet its performance is influenced by complex interactions among field and management factors. In this study, data-driven methods combining numerical simulation, dimensional analysis, nonlinear regression (NLR), support vector machine (SVM) and gene expression programming (GEP) were developed to estimate closed-end border irrigation performance indicators (application efficiency, E a ; uniformity distribution, D u ; and storage efficiency, E s ). A large data set generated using the WinSRFR software was used for model development and comparison, followed by validation with field-measured data. The results showed that the SVM achieved the highest predictive accuracy [mean absolute percentage error (MAPE) < 9.54 % , coefficient of determination ( R 2 ) > 0.973 ], while GEP provided competitive performance with explicit mathematical expressions that enhance interpretability and practical applicability ( MAPE < 7.48 % , R 2 > 0.959 ). Overall, the proposed framework offers an efficient and reliable tool for evaluating closed-end border irrigation performance and supports improved water management in surface irrigation systems.