Investigating the Effect of Noise on Temporal Prediction of Grouneater Flow and Contaminant Transport in Porous Media using Artificial Intelligence Models

Uncertainties of the field parameters such as hydraulic conductivity and dispersion coefficient, unknown boundary conditions and the noise of the measured data are among the main limiting factors in the groundwater flow and contaminant transport (GFCT) modeling.

Miandoab plain was investigated as a case study for simulating groundwater level (GL) and chloride concentration (CC). This paper presents an artificial intelligence-meshless model for temporal GFCT modeling. In this study, time series of groundwater level (GL) and chloride concentration (CC) observed at different piezometers of Miyandoab plain (in Iran) were firstly de-noised by the wavelet-based data de-noising approach. Then, the effect of noisy and de-noised data on the performance of artificial intelligence model was compared. For this end, time series of GL and CC observed in 14 different piezometers were trained and verified via artificial neural network (ANN) and adaptive neuro-fuzzy inference system (ANFIS) models to predict the GL and CC at one month ahead.

The results showed that the threshold-based wavelet de-noising approach can enhance the performance of the modeling up to 25%. Reliability of ANFIS model is more than ANN model in both calibration and verification stages duo to the efficiency of fuzzy concept to overcome the uncertainties of the phenomenon. Discussion and

Waveletde-noising approach as a data preprocessing method enhances the performance of the artificial intelligence model in temporal modeling of GFCT.