Predicting Coastal Water Quality with Machine Learning, a Case Study of Beibu Gulf, China

• Published in: Water (Basel) Vol. 16; no. 16; p. 2253
• Main Authors: Bai, Yucai, Xu, Zhefeng, Lan, Wenlu, Peng, Xiaoyan, Deng, Yan, Chen, Zhibiao, Xu, Hao, Wang, Zhijian, Xu, Hui, Chen, Xinglong, Cheng, Jinping
• Format: Journal Article
• Language: English
• Published: Basel MDPI AG 01.08.2024
• Subjects: Accuracy; Acidification; Algae; Algorithms; Aquatic ecosystems; coastal water; Economic growth; Environmental economics; Eutrophication; GDP; Gross Domestic Product; Machine learning; Pollution; random forest; Salinity; Seawater; spatiotemporal characterization; Underwater resources; Water quality; water quality forecasting; China
• ISSN: 2073-4441; 2073-4441
• DOI: 10.3390/w16162253

Coastal ecosystems are facing critical water quality deterioration, while the most convenient passage to the South China Sea, Beibu Gulf, has been under considerable pressure to its ecological environment due to rapid development and urbanization. In this study, we characterized the spatiotemporal change in the water quality in Beibu Gulf and proposed a machine learning approach to predict the water pollution level in Beibu Gulf on the basis of 5-year (2018–2022) observation data of ten water quality parameters from ten selected sites. Random forest (rf) and linear algorithms were utilized. Results show that a high frequency of exceedance of water quality parameters was observed particularly in summer and autumn, e.g., the exceeding rate of Dissolved Inorganic Nitrogen (DIN) at GX01, GX03, GX06, and GX07 station were 28.2~78.1% (average is 52.0%), 6.0~21.7% (average is 52.0%), 23.0~44.7% (average is 31.9%), and 5.2~33.4% (average is 21.2%), respectively. With regard to the spatial distribution, the pH, Water Salinity (WS), and Dissolved Oxygen (DO) values of stations inside the bay were overall lower than those of corresponding stations at the mouth of the bay and stations outside the bay. The concentrations of Chlorophyll-a concentration (except QZB) and nutrient salts showed a clearly opposite trend compared with the above concerned three parameters. For instance, the average Chl-a value of station GX09 was 22.5% higher than that of GX08 and GX10 between 2018 and 2022. Correlation analysis among water quality factors shows a significant positive correlation (r > 0.85) between Dissolved Inorganic Nitrogen (DIN) and NO3-N, followed by NO2-N and NH4-N, indicating that the main component of DIN is NO3-N. The forecasting results with machine learning also demonstrate the possibility to estimate the water quality parameters, such as chl-a concentration, DIN, and NH4-N in a cost-effective manner with prediction accuracy of approximately 60%, and thereby could provide near-real-time information to monitor the water quality of the Beibu Gulf. Predicting models initiated in this study could be of great interest for local authorities and the tourism and fishing industries.