Early warning of water quality degradation: A copula-based Bayesian network model for highly efficient water quality...

In the context of global climate change and increasingly severe environmental pollution, drinking water quality risk assessments to provide crucial early warnings have become essential routine work. At present, traditional water quality assessment methods are commonly used without considering the correlation among different indicators and the substantial uncertainty from multiple sources, which limit their applications. To address this issue, a copula-based Bayesian network (CBN) method was proposed in this study to concretely evaluate the water quality risk with multiple environmental risk indicators in a large drinking water reservoir in Tianjin city, China. Taking rainfall and water temperature (WT) as external environmental risk indicators and pH, ammonia nitrogen (NH₃-N), total nitrogen (TN), total phosphorus (TP), and permanganate index (COD_Mn) as internal environmental risk indicators, the CBN model was constructed to investigate the interaction between the indicators and water quality state and assess the contingent risk. Our results showed that TN and NH₃-N should be considered key risk indicators. Additionally, we performed forward and backward risk analyses to assess water quality risk during different seasons and determined the distributions of key indicators under different water quality risk grades. From a time perspective, the reservoir's water quality risk is much higher in winter and spring than in other seasons affected by winter snowfall. From a spatial perspective, the water quality risk is much higher at the reservoir's entrance than at other locations affected by water diversion. Furthermore, we found that the probability of water quality risk events may be relatively high when the TN concentration is 3.6 mg/L to 6.4 mg/L at the reservoir's entrance. The results reveal that the CBN method could be an invaluable decision-support tool for reservoir managers and scientists, which could provide an early warning of water quality degradation by only inputting monitoring data.