A Predictive Model of Chlorophyll a in Western Lake Erie Based on Artificial Neural Network

Abstract

The reoccurrence of algal blooms in western Lake Erie (WLE) since the mid-1990s, under increased system stress from climate change and excessive nutrients, has shown the need for developing management tools to predict water quality. In this study, process-based model GLM-AED (General Lake Model-Aquatic Ecosystem Dynamics) and statistical model ANN (artificial neural network) were developed with meteorological forcing derived from surface buoys, airports, and land-based stations and historical monitoring nutrients, to predict water quality in WLE from 2002 to 2015. GLM-AED was calibrated with observed water temperature and chlorophyll a (Chl-a) from 2002 to 2015. For ANN, during the training period (2002–2010), the inputs included meteorological forcing and nutrient concentrations, and the target was Chl-a simulated by calibrated GLM-AED due to the lack of continuously daily measured Chl-a concentrations. During the testing period (2011–2015), the predicted Chl-a concentrations were compared with the observations. The results showed that the ANN model has higher accuracy with lower Chl-a RMSE and MAE values than GLM-AED during 2011 and 2015. Lastly, we applied the established ANN model to predict the future 10-year water quality of WLE, which showed that the probability of adverse health effects would be moderate, so more intense water resources management should be implemented.