Groundwater level modeling framework by combining the wavelet transform with a long short-term memory data-driven model.

Autor: Wu C; College of Hydrology and Water Resources, Hohai University, Nanjing 210098, Jiangsu, China., Zhang X; College of Hydrology and Water Resources, Hohai University, Nanjing 210098, Jiangsu, China., Wang W; College of Hydrology and Water Resources, Hohai University, Nanjing 210098, Jiangsu, China., Lu C; College of Hydrology and Water Resources, Hohai University, Nanjing 210098, Jiangsu, China. Electronic address: luchengpeng@hhu.edu.cn., Zhang Y; Department of Geological Sciences, University of Alabama, Tuscaloosa, AL 35487, USA. Electronic address: yzhang264@ua.edu., Qin W; College of Hydrology and Water Resources, Hohai University, Nanjing 210098, Jiangsu, China., Tick GR; Department of Geological Sciences, University of Alabama, Tuscaloosa, AL 35487, USA., Liu B; College of Hydrology and Water Resources, Hohai University, Nanjing 210098, Jiangsu, China., Shu L; College of Hydrology and Water Resources, Hohai University, Nanjing 210098, Jiangsu, China.
Jazyk: angličtina
Zdroj: The Science of the total environment [Sci Total Environ] 2021 Aug 20; Vol. 783, pp. 146948. Date of Electronic Publication: 2021 Apr 08.
DOI: 10.1016/j.scitotenv.2021.146948
Abstrakt: Developing models that can accurately simulate groundwater level is important for water resource management and aquifer protection. In particular, machine learning tools provide a new and promising approach to efficiently forecast long-term groundwater table fluctuations without the computational burden of building a detailed flow model. This study proposes a multistep modeling framework for simulating groundwater levels by combining the wavelet transform (WT) with the long short-term memory (LSTM) network; the framework is named the combined WT-multivariate LSTM (WT-MLSTM) method. First, the WT decomposes the groundwater level time series (i.e., the training stage) into a self-control term and a set of external-control terms. Second, Pearson correlation analysis reveals the correlations between the influencing factors (i.e., river stage) and the groundwater table, and the multivariate LSTM model incorporating external factors is built to simulate the external-control terms. Third, the spatiotemporal evolution of the groundwater level is modeled by reconstructing the sequence of each term of the groundwater level time series. Methodological applications in the Liangshui River Basin, Beijing, China and the Cibola National Wildlife Refuge along the lower Colorado River, United States, show that the combined WT-MLSTM model has a higher simulation accuracy than the standard LSTM, MLSTM, and WT-LSTM models. A comparison between the combined WT-MLSTM model and support vector machine (SVM) also demonstrates the advantage of the proposed model. Additional comparison between model forecasts and observed groundwater levels shows the model predictability for short-term time series. Further analysis reveals that the applicability of the combined WT-MLSTM model decreases with increasing distance between the groundwater well and adjacent river channel, or with the increasing complexity of the changing groundwater level patterns, which may be driven by additional controlling factors. This study therefore provides a new methodology/approach for the rapid and accurate simulation and prediction of groundwater level.
Competing Interests: Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
(Copyright © 2021 Elsevier B.V. All rights reserved.)
Databáze: MEDLINE
Externí odkaz: