Groundwater Level Forecasting in Sunflower County, Mississippi using Artificial Neural Networks

Author(s): Guzm├ín, S.; Paz, J.; Tagert, M.

The Mississippi Delta Region is one of the most important in the United States given the high productivity levels of crops such as corn, cotton, rice, and soybean. Most of these crops require supplemental irrigation to sustain yield and to reduce the impacts of extended periods of dryness during the growing season. Due to the expansion of croplands , the annual volume of groundwater withdrawals have increased dramatically over the past two decades, exceeding aquifer recharge and generating an important reduction in the aquifer levels. In this study, we present the preliminary groundwater level simulation results for a well in Sunflower County that is within the Mississippi River Valley Shallow Alluvial (MRVA) aquifer. The performance of two different artificial neural networks (ANN) for groundwater level forecasting was evaluated in order to identify an optimal architecture that can simulate decreasing trends of the groundwater level in summer season. Two algorithms, Levenberg-Marquardt and Bayesian Regularization, were evaluated in order to obtain a model that shows better results in the simulation of changes in groundwater level and provide acceptable predictions up to 3 months ahead. The ANN predictive performance was assessed based on the comparison between Root Mean Square Error (RMSE) for each algorithm. Neural networks learn and recognize patterns in the nonlinear temporal data through mathematical analysis and computational architecture inspired by how the human brain works given a set of examples. This methodology is a tool to predict in a short period of time, groundwater levels at specific control points that would be used in an optimized regional plan to manage water withdrawals, and help farmers and water managers decide how to implement plan control procedures and conservation practices.

Go back


Past Conference Archive