A combined Wavelet- Artificial Neural Network model and its application to the prediction of groundwater level fluctuations

Accurate groundwater level modeling and forecasting contribute to civil projects, land use, citys planning and water resources management. Combined Wavelet-Artificial Neural Network (WANN) model has been widely used in recent years to forecast hydrological and hydrogeological phenomena. This study investigates the sensitivity of the pre-processing to the wavelet type and decomposition level in WANN model for groundwater level forecasting. To this end, the monthly groundwater level time series were collected from October 1997 to October 2007 in 26 piezometers of Qorveh aquifer, Iran. Using discrete wavelet transform method and different mother wavelets (Haar, db2, db3 and db4), these time series were decomposed into sub-signals in various resolution levels. Then, these sub-signals entered to the ANN model to reconstruct the original forecasted time series for 6 months ahead. The Root Mean Square Errors (RMSE) and coef?cient of determination (R2) statistics were used for evaluating the accuracy of the model. The results showed merits of db2 and db4 wavelets in comparison with Haar and db3 because of similarity between the signal of groundwater level and the functions of mother wavelets.