Related Experiment Videos
Forecasting of groundwater level fluctuations using ensemble hybrid multi-wavelet neural network-based models
Rahim Barzegar1, Elham Fijani2, Asghar Asghari Moghaddam1
1Department of Earth Sciences, Faculty of Natural Sciences, University of Tabriz, Tabriz, Iran.
The Science of the Total Environment
|May 3, 2017
Summary
Accurate groundwater level (GWL) forecasting is crucial for water management. Hybrid wavelet-extreme learning machine (WA-ELM) models, especially those using boosting, show superior performance for multi-step-ahead predictions compared to wavelet-group method of data handling (WA-GMDH) models.
Area of Science:
- Environmental Science
- Hydrology
- Data Science
Background:
- Groundwater level (GWL) fluctuations significantly impact water resources management.
- Accurate forecasting of GWL is essential for sustainable water use and planning.
- Existing models require enhancement for reliable multi-step-ahead GWL prediction.
Purpose of the Study:
- To evaluate and compare hybrid wavelet-group method of data handling (WA-GMDH) and wavelet-extreme learning machine (WA-ELM) models for GWL forecasting.
- To develop and assess combined wavelet-based models for predicting GWL one, two, and three months ahead.
- To identify the most effective wavelet decomposition and machine learning combination for GWL prediction.
Main Methods:
- Utilized 367 monthly GWL datasets from the Maragheh-Bonab plain, NW Iran (Sep 1985-Mar 2016).
- Employed stepwise selection for input lag times and split data into 85% training and 15% testing sets.
- Applied maximal overlap discrete wavelet transform (MODWT) with various functions (Daubechies, Symlet, Haar, Dmeyer) at level two, feeding decomposed series into GMDH and ELM models.
- Integrated a least squares boosting (LSBoost) algorithm to combine multiple wavelet-based models.
Main Results:
- Extreme Learning Machine (ELM) models demonstrated superior performance over Group Method of Data Handling (GMDH) models.
- Wavelet-based models significantly improved the forecasting accuracy of both GMDH and ELM for multi-step-ahead GWL prediction.
- The boosting multi-wavelet-extreme learning machine (multi-WA-ELM) models achieved the best forecasting performance compared to single WA-ELM models.
Conclusions:
- Hybrid wavelet-based models, particularly WA-ELM, enhance the accuracy of multi-step-ahead groundwater level forecasting.
- Combining multiple wavelet transforms using boosting techniques (LSBoost) leads to optimal GWL prediction performance.
- The developed models offer a robust approach for effective water resources management through accurate GWL forecasting.