Groundwater level fluctuations simulation and prediction by ANFIS- and hybrid Wavelet-ANFIS/Fuzzy C-Means (FCM) clustering models: Application to the Miandarband plain

Abstract

The Miandarband plain is one of the most fertile plains of the Kermanshah province, Iran. The major water supply for agriculture is groundwater. In this regard, simulation and prediction of groundwater level (GL) fluctuations plays an important role for effective water resources management. GL-changes are complex to model, as they depend on many nonlinear and uncertain factors, thus selecting suitable numerical or stochastic models that could simulate the nonlinearity and complex patterns is of great importance. Wavelet transform (WT) integrating with artificial intelligence methods like ANFIS are one family of models that have proven to be very useful to that regard. In this study, after data completion using a novel multiple linear regression approach, the ANFIS model using the Fuzzy Clustering Model (FCM) has been applied. From a cluster sensitivity analysis two clusters have been determined. As the GL-time series are non-stationary for which artificial intelligence methods sometimes have problems to adapt the weights, the additional use of WT in ANFIS is able to enhance the model results. This new hybrid Wavelet-ANFIS model has been used with several combinations of inputs and mother wavelets to simulate and predict GL-fluctuations in the Miandarband plain, Iran. The results show that all model approaches can be used with acceptable accuracy, wherefore the hybrid Wavelet-ANFIS model with the Symlet mother wavelet performs better than other model variants with values for R2 and RMSE of 0.996 and 0.1, respectively, in the training phase and 0.984 and 0.17 in the testing phase.