A coordination framework between drinking water treatment plants and water distribution networks for dynamic risk threshold identification

In water distribution networks (WDNs) supplied by multiple drinking water treatment plants (DWTPs), water quality is shaped by finished-water inputs, blending, and transport. However, finished-water and network water quality are often examined separately, with limited attention to their linkage under multi-plant supply. This study developed a framework integrating finished-water inputs with WDN responses to identify dynamic risk thresholds for key indicators. In a Suzhou WDN supplied by three DWTPs, hydraulic source-tracing contributions were used to construct node-specific equivalent finished-water inputs. Recursive feature elimination with cross-validation (RFECV), Light Gradient Boosting Machine (LightGBM), and SHapley Additive exPlanations (SHAP) were combined for feature selection, response prediction, and interpretation. Indicator-specific adverse-response constraints were applied at an acceptable risk probability of α = 0.10, using Q90 for total hardness, permanganate index, and nitrate, and Q10 for the low-concentration risk of residual chlorine. The LightGBM models performed well for all four indicators. Equivalent finished-water total hardness and permanganate index were operationally relevant to their corresponding WDN responses, whereas nitrate was directly associated with equivalent finished-water nitrate; total hardness also served as an auxiliary hydrochemical background indicator for nitrate-response risk. For residual chlorine, water temperature and equivalent finished-water residual chlorine were the principal model-associated predictors. The recommended thresholds for equivalent finished-water total hardness, permanganate index, nitrate, and residual chlorine were 152 mg/L, 1.504 mg/L, 0.699 mg/L, and 0.711 mg/L, respectively. The framework links WDN responses to plant-side operational targets and supports coordinated water-quality risk management in multi-DWTP systems.