Author(s): Zhan-Qiang Jian; Fangnan Xiao; Runqiao Zhang; Huapeng Qin
Linked Author(s):
Keywords: Estuarine hypoxia; Wavelet analysis; Random forest algorithm; Multivariate time series prediction model; Explainable machine learning
Abstract: Estuary hypoxia is nowadays a common ecological problem worldwide due to the accumulation of pollutants in the tailwater of sewage treatment plants and the increase in discharges. Shenzhen River is located in the core area of the Guangdong-Hong Kong-Macao Greater Bay Area, and its estuary hypoxia is gradually receiving attention. Therefore, taking the Shenzhen River estuary as an example, this paper collected monitoring data on water quality, meteorology, tides, and wastewater treatment plant (WWTP) discharge related to dissolved oxygen (DO) in the Shenzhen River estuary, and explores the best prediction effect of the Random Forest model on the DO in the estuary. The Pearson correlation coefficient calculation between the indicators showed that the estuarine water quality was mainly affected by conductivity and discharge from the wastewater treatment plant, with the exception of DO and conductivity, which showed a more pronounced negative correlation. It is worth noteworthy that the interpretable random forest model effectively identifies the driving factors of DO changes. Through cross-wavelet transform, the periodic and causal relationship between the driving factors and DO were mined, and the effects of tides and rainfall on DO were revealed. The results showed that the effect of WWTP tailwater discharge on DO was spatially and temporally heterogeneous. Combining machine learning methods, focusing on multi-source data in specific areas and fully exploring the key information implied by DO driving factors, this research has reference value for the application of environmental monitoring indicators.
Year: 2024