Groundwater level forecasting using Wavelet-Artificial Neural networks (Case study: Maragheh Plain-East Azarbaijan)

Understanding the behavior of the groundwater system and forecasting it’s fluctuations in the future are essential to achieve comprehensive and sustainable management of groundwater resources. The purpose of this study was clustering of Maragheh Aquifer’s observation wells and groundwater level prediction using Wavelet-Artificial Neural networks. Initially, 20 observation wells of Maragheh Aquifer with 15 years and more groundwater level records were clustered using hierarchical-WARD clustering method. Cluster with 6 homogenous subcluster and representative well of each subcluster were selected. Using wavelet, input time series noise were removed. Then groundwater level of representative wells were forecasted by Artificial Neural Networks. Results showed, considering of temperature time series data as input was confused Artificial Neural Networks and Wavelet-Artificial Neural Networks. On results, taking 3 to 12 months consecutive time delay in input data decreased different between recorded and forecasted data. Minimum value of RMSE (0.03 m) and maximum value of (0.999) were in WNN. Mentioned values in ANN were 0.32 m and 0.885 (respectively). Based on the results of this research, de-noising of input data decreased difference between recorded and forecasted data as 11 cm averagely.