Optimal decision support of irrigation drainage combination schemes in lakefront agriculture-wetland systems

The cultivated farmland and surrounding lake wetlands form a complex lake-adjacent agriculture-wetland system, where irrigation drainage plays a pivotal role as a critical link between agricultural production and wetland ecology. To mitigate eutrophication risks caused by nitrogen (N) and phosphorus (P)-rich agricultural drainage, particularly under the trend of expanding irrigation areas, it is essential to set appropriate threshold limits for agricultural drainage volumes to protect wetland ecosystems. This study introduces a decision-support framework for optimizing irrigation-drainage schemes by integrating a physics-based hydrodynamic-water quality model (Delft3D) with a machine learning surrogate model (FC-ANN), coupled with feasible domain analysis. Applied to the Chagan Lake system in Northeast China, the approach achieved over 97 % prediction accuracy and significantly improved computational efficiency. Through the use of Shapley Additive Explanations (SHAP), the study identified and quantified key factors influencing water quality, enabling transparent connections between drainage strategies and ecological outcomes. Results indicate that the optimal annual runoff volumes for the Qianguo and Da'an irrigation areas are 186 million m³ and 70 million m³, respectively, effectively balancing agricultural demands with wetland health. Furthermore, ecological engineering interventions-such as buffer wetlands and external water diversion projects-could expand the set of feasible drainage scenarios under future conditions. This framework provides a systematic and scalable solution for regional water resources management, offering scientific support for balancing agricultural development, wetland conservation, and sustainable resource governance under increasing environmental pressures.