Wavelet-based de-noising in groundwater quality and quantity prediction by an artificial neural network

Abstract

Abstract The present study uses a wavelet-based clustering technique to identify spatially homogeneous clusters of groundwater quantity and quality data and to select the most effective input data for the feed-forward neural network (FFNN) model to predict the groundwater level (GL), pH and HCO3− in groundwater. In the second stage of this methodology, first, GL, pH and HCO3− time series of different piezometers were de-noised using a threshold-based wavelet method and the impact of de-noised and noisy data were compared in temporal GL, pH and HCO3− modeling by the artificial neural network (ANN). The results suggest that the proposed model decreases the dimensionality of the input layer and consequently the complexity of the FFNN model with acceptable efficiency in the spatiotemporal simulation of GL and groundwater quality parameters. Also, the application of wavelet-based de-noising for modeling GL, pH and HCO3− parameters with ANN increases the accuracy of predictions, respectively, up to 11.53, 11.94 and 38.85% on average.