Groundwater Table Estimation Using MODFLOW and Artificial Neural Networks

Abstracst

The use of numerical models to simulate groundwater flow has been addressed in many research studies during the past decade. The main drawback with these models is their enormous and generally difficult or costly data requirements. On the other hand, artificial neural networks (ANNs) are offering a simple but precise solution to many simulation problems. In this chapter, the applicability of ANN models in simulating groundwater levels has been investigated. In order to be able to use ANN models for aquifers with limited data, MODFLOW was used to simulate the groundwater flow and the calibrated model was then applied to generate hundreds of data sets for the training of the ANN model. Another purpose of this chapter is to identify ANN models that can capture the complex dynamics of water table fluctuations, even with relatively short lengths of training data. MODFLOW outputs and measured water table elevations were used to compare the performance of the ANN models. The average regression coefficients for multi-layer perceptrons and time lag recurrent neural networks were 0.865 and 0.958, respectively.