Diagnosing the Factors of Domestic Water Shortage Under Climate Change Through an XGBoost-SHAP Framework

Climate-change-induced drought can increase domestic water shortage not only by reducing runoff but also by exposing weaknesses in local sources, reservoirs, and wide-area transfer networks. This study simulates domestic water shortage in the Seomjin River water system, South Korea, using the Korea-Water Evaluation and Planning system (K-WEAP), and proposes an integrated eXtreme Gradient Boosting (XGBoost)–Shapley Additive exPlanations (SHAP) framework to diagnose the explanatory factors of the simulated shortage. Representative drought scenarios were selected from an ensemble of climate projections using a dryness score based on precipitation-related extreme climate indices. K-WEAP was used to simulate shortage for eleven administrative districts, and XGBoost (version 3.2.0) models were trained for districts with sufficient shortage samples. SHAP analysis and shortage-event clustering were then applied to identify dominant explanatory factors and recurrent shortage patterns. The XGBoost models reproduced the simulated shortage rates well, with test R2 values exceeding 0.91. Wide-area water supply was the dominant explanatory factor in three inland districts, whereas reservoir supply was more influential in the southern coastal district. The dependence analysis revealed threshold-type responses, in which shortage risk increased sharply once key supply variables fell below district-specific critical ranges. The results indicate that domestic water shortage is governed by district-specific supply pathways and threshold-type system responses rather than by uniform basin-wide drought effects.