A Hybrid-Wavelet Artificial Neural Network Model for Monthly Water Table Depth Prediction

Groundwater is an essential natural resource in the country to fulfil the irrigation, domestic, industrial and other needs. In order to ensure sustainable use of groundwater resources, the groundwater level is used as an important indicator for balancing the groundwater withdrawal rate and replenishment rate through the recharge. Quantitatively, the recharge rate is governed by various complex large-scale hydrological processes and hence achievement of sustainability of groundwater supplies, through sustainable withdrawal rate is a complicated issue. In the present study, a data-driven prediction model by combining discrete wavelet transform (DWT) with artificial neural network (ANN) called as wavelet artificial neural network (WANN) is proposed for the groundwater table prediction. The simulation results obtained by regular ANN model were compared with those obtained by WANN model to prove the superiority of the latter model over the former. WANN model was developed using decomposed signals of rainfall, evapotranspiration and water table depth time series as inputs in the ANN model to arrive at a prediction of monthly fluctuation of the groundwater table. Rainfall time series was decomposed using Haar wavelet at third decomposition level and evapotranspiration and water table depth time series was decomposed using Daubechies wavelet at second decomposition level. The RMSE value of ANN and WANN model during validation were found to be 0.3648 m and 0.1695 m respectively, which showed decrease in RMSE value by 0.195 m when WANN was applied. Model efficiencies of ANN and WANN model during validation were 84.65% and 95.68%, indicating excellent improvement of model accuracy after applying WANN. Hence, the proposed WANN model seems to be a promising tool to predict the monthly water table fluctuation.