Design data decomposition-based reference evapotranspiration forecasting model: A soft feature filter based deep learning driven approach
Article
Article Title | Design data decomposition-based reference evapotranspiration forecasting model: A soft feature filter based deep learning driven approach |
---|---|
ERA Journal ID | 32032 |
Article Category | Article |
Authors | Zheng, Zihao, Ali, Mumtaz, Jamei, Mehdi, Xiang, Yong, Karbasi, Masoud, Yaseen, Zaher Mundher and Farooque, Aitazaz Ahsan |
Journal Title | Engineering Applications of Artificial Intelligence |
Journal Citation | 121 |
Article Number | 105984 |
Number of Pages | 21 |
Year | 2023 |
Publisher | Elsevier |
Place of Publication | United Kingdom |
ISSN | 0952-1976 |
1873-6769 | |
Digital Object Identifier (DOI) | https://doi.org/10.1016/j.engappai.2023.105984 |
Web Address (URL) | https://www.sciencedirect.com/science/article/pii/S0952197623001689?via%3Dihub |
Abstract | Reference evapotranspiration can cause huge discrepancies in soil moisture and runoff which is responsible for uncertainties in drought warning systems. Reference evapotranspiration (ETo) is one of the major drought elements that leads to soil dryness, vegetation surfaces and transpiration. An innovative strategy is proposed based on Multivariate Variational Mode Decomposition hybridized with Soft Feature Filter and Gated Recurrent Unit to design MVMD-SoFeFilterGRU to forecast one-day daily ETo at (t+1). Initially, the importance of each predictor was determined using correlation matrix to identify significant lags at (t+1). Next, the intrinsic mode functions (IMFs) in terms of signals were obtained via MVMD to decompose the lags. The SoFeFilter approach was employed to select the most relevant IMFs which were then incorporated into the GRU to construct the MVMD-SoFeFilter-GRU model to forecast one-day ahead daily ETo. For comparison, the LSTM, BiLSTM, RNN, BiRNN, and BiGRU models were combined with MVMD and SoFeFilter to create MVMD-SoFeFilterLSTM, MVMD-SoFeFilterBiLSTM, MVMD-SoFeFilterRNN, MVMD-SoFeFilterBiRNN, and MVMD-SoFeFilterBiGRU models. Further, the results were also compared against the standalone GRU, BiGRU, LSTM, BiLSTM, RNN, and BiRNN models based on goodness-of-fit metrics for two stations in Queensland, Australia. For example, in Gympie station, the MVMD-SoFeFilterGRU model produced highest values of WIE=0.9795, NSE=0.9234, LME=0.7645, and for Redcliffe station, these metrics are WIE=0.9800, NSE=0.9257, LME=0.7580. The findings confirm that the MVMD-SoFeFilterGRU is the most precise to forecast one-day ahead ET0. |
Keywords | Evapotranspiration; MVMD; LSTM; BiLSTM; GRU; BiGRU; RNN; BiRNN |
Contains Sensitive Content | Does not contain sensitive content |
ANZSRC Field of Research 2020 | 410402. Environmental assessment and monitoring |
Public Notes | Files associated with this item cannot be displayed due to copyright restrictions. |
Byline Affiliations | Deakin University |
UniSQ College | |
University of Prince Edward Island, Canada | |
Shahid Chamran University of Ahvaz, Iran | |
University of Zanjan, Iran | |
School of Mathematics, Physics and Computing |
https://research.usq.edu.au/item/w8w19/design-data-decomposition-based-reference-evapotranspiration-forecasting-model-a-soft-feature-filter-based-deep-learning-driven-approach
55
total views0
total downloads3
views this month0
downloads this month