A comparative study of artificial neural network and hybrid model for prediction of Groundwater level

Abstract

The depletion of groundwater table in some areas causing concerns for the long term sustainability of groundwater resources in India. For the effective management of groundwater, it is important to know fluctuations of ground water level (GWL). In hydrology, the artificial neural network (ANN) models have been satisfactorily applied for prediction of stream flow, precipitation, and water quality modelling but their application to the groundwater sector has been found to be limited. In this study, an ANN model and Hybrid model (HM) have been developed to predict the groundwater level fluctuations and their performances were compared. HM is developed by comprising both ANN and regression modeling technique, the output layer of ANN was replaced by a Non-linear regression model (NLRM). The models were developed by employing rainfall and the past GWL as input and the present GWL as output. The data for the modelling task have been used from a watershed located in the Sagar distt., Madhya Pradesh, India. The performances of the developed ANN model and hybrid model has been evaluated using standard statistical measures viz. average arithmetic relative error (AARE), correlation coefficient (R), Nash- Sutcliff Efficiency (E), and Threshold Statistics (TS). The result indicates that the models can successfully be used for prediction of GWL and comparative study shows that the performance of hybrid model is better than ANN model.