ESTIMATING AQUIFER SUSCEPTIBILITY TO ENHANCED CHEMICAL TRANSPORT NEAR IRRIGATION WELLS USING ARTIFICIAL NEURAL NETWORKS, HIGH PLAINS AQUIFER

A novel aquifer-susceptibility model is presented that uses artificial neural networks (ANN) to investigate focused recharge and enhanced chemical transport near 128,700 irrigation wells across the regionally extensive High Plains aquifer (450,700 km²). A recent study of the High Plains aquifer conducted by the U.S. Geological Survey's National Water Quality Assessment (NAWQA) Program revealed elevated levels of agrichemicals in ground water resulting from the introduction of irrigated agriculture in the mid 20th century. Unsaturated-zone chemical profiles from the High Plains indicated that percolation rates beneath agricultural fields are enhanced by irrigation, but fluxes are not sufficient to explain the elevated levels of agrichemicals in the underlying ground water given 50 years of transport. During irrigation periods, ponded water has been observed around irrigation wells and is hypothesized to be a mechanism for focused recharge and enhanced chemical migration to the water table. Providing the foundation for ANN development, transient VS2DT model simulations were designed to represent seasonally wetter conditions in a 1-m radius around a generic irrigation well and indicated conservative chemical migration to the water table in <50 years in many cases. Sensitivity analysis revealed soil texture, soil-water potential, and water-table depth to be the most important controlling factors influencing chemical transport to ground water. The results of 336 permutations of VS2DT simulations using various combinations of the controlling factors were used to train and test a probabilistic ANN that estimates chemical transit to the water table in <50 years. Approximately 91% of the 336 permutations were correctly identified during ANN testing, suggesting a strong predictive ability. Preliminary ANN predictions were compared to tritium values at monitoring wells to validate the predictive ability of the model. GIS-based application of the trained ANN at the 128,700 irrigation wells to predict the occurrence of positive results (conservative chemical reaching ground water in <50 years) has important implications for the conceptual understanding of chemical transport mechanisms and management of agricultural practices across the High Plains aquifer.