Conjunctive Stream-Aquifer Modeling Using Artificial Neural Networks

A tool is presented for modeling complex stream-aquifer interactions by integrating an artificial neural network (ANN) with a geographic information system (GIS) and a numerical groundwater model. The proposed methodology allows development of dynamic, spatially dependent relationships between measurable aquifer stresses and river return flow. A well calibrated finite difference groundwater model is used to represent the system stream-aquifer interaction for training the ANN. ArcGIS (ESRI) is used to process and prepare the extensive spatial data sets for training, validation and testing of the ANN. MATLAB (MathWorks, Inc) performs training and validation of the ANN, with VBA interfaces embedded in the GIS for manipulating the finite difference model and preparing the required ANN spatial data set. A significant feature of the proposed tool is the ability to generate the ANN input/output data set based on physical-spatial system characteristics. The tool is applied to the Lower Arkansas River Basin in Colorado as a case study. Preliminary results show the feasibility of using the proposed methodology for developing more robust stream-aquifer modeling tools for incorporation into river basin management models.