A maximal overlap discrete wavelet packet transform coupled with an LSTM deep learning model for improving multilevel groundwater level forecasts

Authors: ROY, DILIP - Arkansas State University; HASHEM, AHMED - Arkansas State University; Reba, Michele; LESLIE, DEBORAH - University Of Memphis; NOWLIN, JOHN - Arkansas State University

Submitted to: Water Resources Research
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 4/3/2024
Publication Date: 4/8/2024
Citation: Roy, D., Hashem, A., Reba, M.L., Leslie, D., Nowlin, J. 2024. A maximal overlap discrete wavelet packet transform coupled with LSTM deep learning model for improving multilevel groundwater level forecasts. Water Resources Research. 4(16):1-20. https://doi.org/10.1007/s43832-024-00073-1.
DOI: https://doi.org/10.1007/s43832-024-00073-1

Interpretive Summary: Sustainable planning and management of limited groundwater resources is a growing concern in Bangladesh and worldwide. We proposed an improved method of predicting groundwater levels using advanced modeling tools. We found that our proposed approach could produce more accurate forecasts of groundwater level compared to the traditional approaches. Our approach was found promising for groundwater level forecasts at the specified observation wells of a water-scarce region in Bangladesh that could be extended to other geographical areas. Our results will interest policy makers and managers of water resources and may improve decisions on the utilization of groundwater resources for irrigation scheduling, water supply, and land development.

Technical Abstract: Developing precise groundwater level (GWL) forecast models is essential for the optimal usage of scarce groundwater resources and sustainable planning and management of water resources. In the present research, an improved forecasting accuracy (three weeks-ahead) of GWLs in Bangladesh was achieved by a coupled Long Short Term Memory (LSTM) network-based deep learning algorithm and a Maximal Overlap Discrete Wavelet Packet Transform (MODWPT) data preprocessing. The Coupled LSTM-MODWPT model performance was compared with the LSTM model. For both standalone LSTM and LSTM-MODWPT models, the Random Forest-based feature selection approach was employed to select the ideal inputs from the candidate GWL lags. In the LSTM-MODWPT model, input GWL timeseries were decomposed using MODWPT. The ‘Fejér-Korovkin’ mother wavelet with a filter length of 18 was used to obtain a collection of scaling coefficients and wavelets for every single input timeseries. Model performance was assessed using five performance indices: Root Mean Squared Error, RMSE; Scatter Index, SI; Maximum Absolute Error, MAE; Median Absolute Deviation, MAD; and a-20 index. The LSTM-MODWPT model outperformed standalone LSTM models for all time horizons in GWL forecasting (the percentage improvements in the forecasting accuracies were 36.28%, 32.97%, and 30.77%, respectively, for one-, two-, and three-weeks-ahead forecasts at the observation well GT3330001). Accordingly, the coupled LSTM-MODWPT model could potentially be used to enhance multiscale GWL forecasts. This research demonstrates that the coupled LSTM-MODWPT model could generate more precise GWL forecasts at the study site, with potential applications in other geographic locations globally.