Application of Artificial Neural Networks for Predicting Groundwater Levels in Hard Rock Basins

Abstract

The objectives of this study were to determine the factors that influence and control the water table fluctuation in a 
specific geomorphologic situation, to develop a forecasting model and examine its potential in predicting water table 
depth using limited data. Prediction of region specific water table fluctuation would certainly guide the way towards 
conceiving, designing and taking scientific measures to ensure sustainable groundwater management. Analysis of 
change in groundwater table depth, groundwater flow directions within the watershed showed that the influencing factors 
of rainfall, groundwater draft from near by structures and the resulting fluctuation in groundwater table depth were well 
correlated in a specific geological situation. Models for prediction of water table depth were developed based on artificial 
neural networks (ANNs). The study employed multilayer feed forward neural network with backpropagation learning 
method to develop the model. The neural networks with different numbers of hidden layer neurons were developed using 
4 years (2005 - 2008) monthly rainfall, potential evapotranspiration (PET), and water table depth from nearby, influencing 
wells data as input and one month ahead water table depth as output. The best model was selected based on the root 
mean square error (RMSE) of prediction using independent test data set. The results of the study clearly showed that 
ANN can be used to predict water table depth in a hard rock aquifer with reasonably good accuracy even in case of 
limited data situation.