Neural network approach to stream-aquifer modeling for improved river basin management

Abstract

Artificial neural networks (ANNs) are applied to efficient modeling of stream-aquifer responses in an intensively irrigated river basin under a variety of water management alternatives for improving irrigation efficiency, reducing soil water salinity, increasing crop yields, controlling nonbeneficial consumptive use, and decreasing salt loadings to the river. Two ANNs for the main stem river and the tributary regime are trained and tested using solution datasets from a high resolution, finite difference MODFLOW–MT3DMS groundwater flow and contaminant transport model of a representative subregion within the river basin. Stream-aquifer modeling in the subregion is supported by a dense field data collection network with the ultimate goal of extending knowledge gained from the subregion modeling to the sparsely monitored remainder of the river basin where data insufficiency precludes application of MODFLOW–MT3DMS at the desired spatial resolution. The trained and tested ANNs capture the MODFLOW–MT3DMS modeled subregion stream-aquifer responses to system stresses using geographic information system (GIS) processed explanatory variables correlated with irrigation return flow quantity and quality for basin-wide application. The methodology is applied to the Lower Arkansas River basin in Colorado by training and testing ANNs derived from a MODFLOW–MT3DMS modeled subregion of the Lower Arkansas River basin in Colorado, which includes detailed unsaturated and saturated zone modeling and calibration to the extensive field data monitoring network in the subregion. Testing and validation of the trained ANNs shows good performance in predicting return flow quantities and salinity concentrations. The ANNs are linked with the GeoMODSIM river basin network flow model for basin-wide evaluation of water management alternatives.