Groundwater Level Short-Term Forecasting Under Tailings Recharge Using Wavelet-Bootstrap-Neural Network Models

Abstract:

In this study, five data-driven models were evaluated for groundwater level short-term forecasting under tailings recharge from an abandoned mine in Quebec, Canada. Multiple linear regression (MLR) models were used as a linear model, while artificial neural network (ANN) models were used as a non-linear model. In addition, two hybrid models that utilize wavelet transforms for data preprocessing with MLR or ANN models (W-MLR, W-ANN) were considered for the evaluation of the usefulness of wavelet analysis with linear and nonlinear models. The fifth model was a wavelet bootstrap ANN (W-B-ANN) model. Three predictors were considered as inputs: the tailing recharge, total precipitation, and mean air temperature. Results showed that models using wavelets for data preprocessing (W-MLR and W-ANN) performed better than their corresponding basic models (MLR and ANN), which highlights the ability of wavelet transforms to decompose non-stationary data into discrete wavelet components, highlighting cyclic patterns and trends in the time series at varying temporal scales, rendering the data usable in forecasting. In general, with or without wavelets, ANN models performed better than MLR models; this indicates the nonlinear relationship between the three predictors and the groundwater level. Overall, the W-B-ANN model outperformed all models for each of the three lead-times, which highlights the usefulness of bootstrap modeling, and ensuring model robustness along with improved reliability by reducing variance.