Artificial Neural Networks to Predict the Nitrate Distribution in Cimarron Terrace Aquifer, Oklahoma

Abstract

The artificial neural network (ANN) models were used to expand the groundwater nitrate data taken at well locations to the entire space of Cimarron Terrace Aquifer near the City of Enid, Oklahoma. These neural kriging methods were able to adequately extrapolate the nitrate concentrations from the measured point values over the area of concern. Additional, a series management option models each with a 10 percent decrease nitrogen application rate were developed for the City of Enid's Ames wellfield. These models showed that significant differences existed between nitrate concentrations at threshold levels of 4.0 mg/L and 10.0 mg/L. A reduction of 80 percent of the surface nitrogen application rate was needed to cause a decrease to the nitrate MCL level. Further, neural conditional simulation subsequently was applied to the Ames wellfield data to determine the probability and cumulative density functions and respective best fit curves for the predicted nitrate concentrations.