Predicting water quality during urbanization based on a causality-based input variable selection method modified back-propagation neural network

Rapid urbanization has been recognized as the primary cause of deteriorating water quality. Thus, it is crucial to take into account urbanization in water quality forecasting. The present study aims at finding the causal relationship between urbanization and water quality, and then predicting water quality based on this causality. For this purpose, nine urbanization indicators and 12 water quality parameters from 2006 to 2018 in Nanjing were collected as urbanization and water quality indices. Correlation and path analyses were firstly used to identify causal relationships between urbanization and water quality indices. Based on these causal relationships, comprehensive water quality indicators and their correlated urbanization parameters were input into a back-propagation neural network (BPNN) to predict water quality. In the improved BPNN, the R 2 of the training sets were all greater than 0.99, and those of the test sets were all greater than 0.76, demonstrating that the optimized model is able to predict the water quality with reasonable accuracy. It also showed that the overall water quality in Nanjing will remain good from 2019 to 2028, which means that, when undergoing future urbanization process, water quality is not necessarily negatively affected. The transfer of industrial structure can have a positive influence on water quality. After 2028, the biological water environment index remained in a good state but the volatile phenol index continued to increase, making it a potential threat to future water quality. Industrial wastewater and fertilizer usage, as the primary sources of volatile phenols, should be prioritized for continued governmental control and monitoring into the future. This study provides new insight into the relationship between urbanization and water quality, and the presented models can assist in future-proofing water management strategies.