Using SOM and Wavelet Transform pre-processing methods in groundwater level prediction (Case Study: AzarShahr plain)

Prediction of groundwater level play an important role in the groundwater source management. Groundwater plays most important role in providing required water for agricultural, urban and industrial uses, especially in semi-arid regions. The Present study focused on predicting groundwater level in the AzarShahr plain using pre-processing tools in two scenarios. Clustering tool was used by means of Self-Organized Maps (SOM) for conducting spatial pre-processing and wavelet transform (WT) for time pre-processing and also artificial neural system for modeling. SOM based clustering technique was used to identify spatially homogeneous clusters of groundwater data to use in artificial neural network to model groundwater resources. The WT was also used to extract dynamic and multi-scale features of the non-stationary GWL, runoff and rainfall time series. Results showed that using the WT and combining it with artificial neural system in groundwater level modeling of AzarShahr plain led to 11.6 percent improvement in the modeling accuracy, in verification stage the in the first scenario and 23.5 percent improvement in the second scenario. It can be concluded that using new modeling methods such as applying time and spatial pre-processing tools leads to significant increase in the modeling accuracy.