Tailor-made ammonia nitrogen risk management with machine learning models for aquatic environments in the Mainland of China

Efficient management of pollutant risks in water bodies is crucial for public health and aquatic ecosystem sustainability. However, the toxicities of pollutants, such as ammonia nitrogen (NH₃-N), are often affected by multiple water quality factors, including the pH and water temperature. Extensive spatial and temporal variability in these factors hinders tailor-made management of risk. This study used high-frequency monitoring data collected over 1 year to evaluate the long-term NH₃-N risk in China's aquatic ecosystems. High accuracy and interpretability were achieved by decomposing NH₃-N risk into the contributions of key influencing factors using random forest models and Shapley Additive Explanations. Two distinct types of NH₃-N risk hotspots were identified across 18 cities: 15 cities with high NH₃-N concentrations and 3 cities with low environmental carrying capacity due to high pH levels or elevated water temperatures. For the former, rapid NH₃-N abatement measures are necessary to bring NH₃-N concentrations back below the environmental capacity. For the latter, it is recommended that NH₃-N related industries are relocated to regions with high environmental capacities because fragile environments are not suitable for such industries. Importantly, this study investigated methods for attributing pollutant risks in the context of non-linear influencing factors, and the risk of NH₃-N was predicted to increase by 6.1 % by the end of 2100 in the context of increasing temperatures under the SSP 2–4.5 scenario. The methodology is also adaptable and suitable for integration into global ecosystem risk management efforts to balance development and aquatic ecological sustainability.