A hybrid ensemble learning wavelet-GARCH-based artificial intelligence approach for streamflow and groundwater level forecasting at Silakhor plain, Iran

Today, due to the surface and ground water crisis, these components become vital and valuable resources throughout the world. Being aware of surface and ground water quantities for each period is necessary to correct planning. This study extends the literature of streamflow and groundwater level prediction by applying a hybrid model which combines wavelet transform (WT) and generalized autoregressive conditional heteroskedasticity (GARCH) algorithms with M5p model tree (M5pMT) to predict streamflow and groundwater level at Silakhor plain for a period of 17 years, between 1997 and 2014. To assess the ability of the proposed hybrid model for the prediction of streamflow and groundwater level at monthly scale, several benchmark models are constructed and compared using several evaluation measures. Results indicate that artificial intelligence (AI) models when coupled with GARCH model characterizing the error for the remaining subseries which are decomposed by WT. The value of RMSE and MAE of W-GARCH-M5pMT is reduced by about 14 and 2% for streamflow prediction when compared with standalone M5pMT model. For groundwater prediction, accuracy of W-GARCH-based models improved from 0.953 to 0.965 for ANN, from 0.955 to 0.976 for ELM, and from 0.95 to 0.994 for M5pMT model.