Establishment of season-specific nutrient thresholds and analyses of the effects of nutrient management in eutrophic lakes through statistical machine learning

• Published in: Journal of hydrology (Amsterdam) Vol. 578; p. 124079
• Main Authors: Tong, Yindong, Xu, Xiwen, Zhang, Shuliang, Shi, Limei, Zhang, Xiaoyan, Wang, Mengzhu, Qi, Miao, Chen, Cen, Wen, Yingting, Zhao, Yue, Zhang, Wei, Lu, Xuebin
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.11.2019
• Subjects: Lake eutrophication; Nutrient management; Nutrient thresholds; Random forest models; Season-specific; Random forest models; Season-specific; Nutrient thresholds; Nutrient management; Lake eutrophication
• ISSN: 0022-1694; 1879-2707
• DOI: 10.1016/j.jhydrol.2019.124079

•Chlorophyll a predicted by statistical machine learning successfully displayed seasonal variations.•Estimated nutrient thresholds were quite variable and higher in summer than in winter.•Reducing TP concentration in these lakes in summer is important to control algal growth.•Seasonal variation in nutrient concentration should be considered when establishing criteria. Eutrophication and subsequent harmful cyanobacteria blooms are global water quality problems, and identifying the key drivers of water eutrophication and estimating nutrient thresholds for it in waterbodies have long been challenges for water quality managers. Data-intensive machine learning models have been shown to be better able to reveal the nonlinear relationships between variables in the study of complex biotic community dynamics than traditional mechanistic models. In this study, we applied random forest models to long-term datasets from nutrient monitoring and meteorological observations to characterize the relationships between algal growth and different environmental drivers in three eutrophic lakes in China. We further attempted to estimate the season-specific nutrient thresholds in these lakes, and assess the potential decreases in chlorophyll a concentrations that could be achieved through nutrient management. In general, chlorophyll a concentrations predicted by the random forest models were consistent with the values observed in the lakes, and successfully displayed the same seasonal variations. The estimated total nitrogen (TN) and total phosphorus (TP) nutrient thresholds were quite variable among months, and were higher in summer than in winter. To maintain chlorophyll a concentrations below 20 μg/L, the estimated TN thresholds in Lakes Taihu, Dianchi, and Chaohu in August were 2145 ± 683, 2372 ± 918 and 1527 ± 71 μg/L (mean ± standard deviation), respectively, and the corresponding TP thresholds were 82 ± 24, 149 ± 22, and 120 ± 22 μg/L. The modelling results indicated that it was more important to control the TP concentrations in these lakes than the TN concentrations to control algal growth in summer. In summary, the strong seasonal variation in the estimated nutrient thresholds suggests that a ‘one-size-fits-all’ nutrient control target could overprotect these water bodies. Seasonal variation in nutrient concentrations and environmental drivers should thus be considered when establishing nutrient criteria and setting nutrient control targets.