Simulating heavy metal concentrations in an aquatic environment using artificial intelligence models and physicochemical indexes

Abstract

Although heavy metal monitoring campaigns are established worldwide, it is still difficult to model heavy metals in aquatic environments with limited monitoring data. In this study, surface water physicochemical indexes and heavy metal concentrations were measured in a drinking water source in the Taihu Lake region, China. Afterwards, indexes including water temperature, pH, suspended matter, turbidity, and total nitrogen, nitrate nitrogen, ammonia nitrogen, total phosphorous, orthophosphate and permanganate index were used to simulate dissolved, particulate and total heavy metal concentrations using artificial neural network (ANN) and support vector machine (SVM) models. Sensitivity analysis showed that simulated heavy metal concentrations were most sensitive to pH. Thereafter, quick simulation models based on five sensitive parameters (pH, suspended matter, water temperature, total phosphorus and permanganate index) allowed for quick simulations of heavy metal concentrations were built. Both ANN and SVM quick simulation models simulated particulate heavy metal concentrations well with most Nash-Sutcliffe efficiency coefficients >0.8. Models performed worse when simulating dissolved and total heavy metal concentrations. Results demonstrate that artificial intelligence models like ANN and SVM are alternative ways to simulate heavy metal concentrations with limited monitoring data. Furthermore, sensitivity analysis may help to identify key factors affecting heavy metal behavior, and to improve environmental monitoring campaigns and management strategies.