Predicting Escherichia coli loads in cascading dams with machine learning: An integration of hydrometeorology, animal density and grazing pattern

Accurate prediction of Escherichia coli contamination in surface waters is challenging due to considerable uncertainty in the physical, chemical and biological variables that control E. coli occurrence and sources in surface waters. This study proposes a novel approach by integrating hydro-climatic variables as well as animal density and grazing pattern in the feature selection modeling phase to increase E. coli prediction accuracy for two cascading dams at the US Meat Animal Research Center (USMARC), Nebraska. Predictive models were developed using regression techniques and an artificial neural network (ANN). Two adaptive neuro-fuzzy inference system (ANFIS) structures including subtractive clustering and fuzzy c-means (FCM) clustering were also used to develop models for predicting E. coli. The performances of the predictive models were evaluated and compared using root mean squared log error (RMSLE). Cross-validation and model performance results indicated that although the majority of models predicted E. coli accurately, ANFIS models resulted in fewer errors compared to the other models. The ANFIS models have the potential to be used to predict E. coli concentration for intervention plans and monitoring programs for cascading dams, and to implement effective best management practices for grazing and irrigation during the growing season