GROUNDWATER LEVEL PREDICTIONS USING HYBRID WAVELET MACHINE LEARNING METHODS
Abstract - Groundwater level (GWL) prediction is fundamental for agriculture and water distribution. Machine learning (ML) models,such as artificial neural networks (ANN), have been gradually moreacceptedby scholars to predict GWL becauseof theircapability to simulate the nonlinearities between GWL and its drivers (e.g.,precipitation). The present researchsuggested a traditionalsupervised MLmethod to describe the GWL variationsin the Tehran-Karaj plain in Iran using 14 years monthly interval dataset.The wavelet transform (WT) method was also employedto advance the GWL simulationcapability of ML approaches for three monthsaheadusing various input combinations of temperature, precipitation, evapotranspiration and GWL. The methods were compared based on different error criteria, andthe comparison of the results proved thatthe hybrid-wavelet MLsubstantiallyadvanced the standalone model outcomes. The best GWL results were acquired from the hybrid WT-ANNmodel results forscenario one,and this model produced MAE, RMSE, NSE, and R as 0.20, 0.25, 0.99 and 0.99 for one month ahead of GWL prediction.The results of the two models were similar in predicting groundwater level fluctuations one month ahead of the GWL simulation.
Keywords - Artificial Intelligence, Groundwater Level Forecasting, Wavelet Transform, Standalone Model.