Reliability of Irrigation Water Allocation under Climate Change: A WEAP–MABIA Assessment for Kinmen Using Stochastic Rainfall and ML-Based Weather Generation

Assessing the reliability of irrigation water-allocation plans under climate change is essential for sustainable agriculture, especially when alternative water sources such as reclaimed water are considered. This study evaluates an irrigation allocation plan for a sorghum–wheat rotation in Kinmen, Taiwan, combining stakeholder engagement, climate analysis, stochastic rainfall simulation, and system modeling. Two local workshops with farmers and experts were conducted to refine feasible irrigation practices and design paired experimental and control fields for subsequent calibration. Historical rainfall from the Kinmen weather station (2005–2025) indicates fewer rainy days and more concentrated events in the recent decade, suggesting increasing rainfall extremity. Future climate scenarios (SSP2-4.5 and SSP5-8.5) were bias-corrected using quantile-based adjustment over 2015–2025, then analyzed for seasonal shifts and extremes. Rainfall temporal structure was simulated using the NEOPRENE Neyman–Scott framework, while XGBoost models were trained on observations to generate daily meteorological variables and reference evapotranspiration. These climate inputs drove a WEAP-based irrigation allocation model coupled with the MABIA method to estimate yields, water use, and economic performance over a 20-year planning horizon. Results show that weekly irrigation increases yields but yields the lowest net profit due to higher labor and energy costs. Under SSP2-4.5 (wetter and more evenly distributed rainfall), a three-week irrigation interval maximizes profit, whereas under SSP5-8.5 (more concentrated rainfall and longer dry spells), a two-week interval provides the best balance between yield stability and cost. The framework provides decision support for robust irrigation planning under uncertain future climate conditions.