View in EDS

A novel adaptive time-scale decomposition fused LSTM-transformer framework (ATSD-LT) for TCWV prediction.

Saved in:
Bibliographic Details
Title: A novel adaptive time-scale decomposition fused LSTM-transformer framework (ATSD-LT) for TCWV prediction.
Authors: Xu, Longhao1,2 (AUTHOR), Mao, Kebiao1,2 (AUTHOR) maokebiao@caas.cn, Guo, Zhonghua1 (AUTHOR) guozhh@nxu.edu.cn, Shi, Jiancheng3 (AUTHOR) shijiancheng@nssc.ac.cn, Bateni, Sayed M.4,5 (AUTHOR) smbateni@hawaii.edu, Yuan, Zijin2 (AUTHOR) yuanzijin@caas.cn
Source: Environmental Modelling & Software. Jun2026, Vol. 201, pN.PAG-N.PAG. 1p.
Subject Terms: *Water management, *Environmental monitoring, Transformer models, Long short-term memory, Spatiotemporal processes, Precipitable water
Abstract: Total column water vapor (TCWV) is vital for the water cycle, precipitation, and extreme weather events. This study proposes a novel adaptive time-scale decomposition fused LSTM-Transformer framework (ATSD-LT). ATSD employs gradient-optimized filters to resolve sequences into trends and multi-scale fluctuations, suppressing aliasing and frequency leakage. Through heterogeneous ensemble modeling, LSTM extracts trend evolution while Transformer captures global long-range spatiotemporal correlations. Results yield superior performance, particularly in spring and summer (R2 of 0.959 and 0.967, respectively). Compared to Autoformer, Informer, Transformer, and LSTM, ATSD-LT demonstrates more significant gains in winter (improving R2 by ∼18%–55% and reducing MAE by 64%–76%). Ablation tests confirm the proposed ATSD-LT framework achieves a total 79.4% reduction in summer MAE, while feature weighting provides an additional 12.3% goodness-of-fit gain. Spatial mapping reproduces tropical highs (e.g., Amazon Basin) and high-latitude lows (e.g., northern North America); cross- and site-validation further confirm framework robustness. ATSD-LT effectively supports global climate monitoring and water management. [Display omitted] • Propose ATSD-LT framework for multi-scale TCWV spatiotemporal prediction. • Fuse ATSD with LSTM-Transformer, boosting R2 by 28% and cutting MAE by 70%. • Accurately map tropical high-value and high-latitude low-value TCWV zones. • Validate stability via cross-validation for agricultural water management. [ABSTRACT FROM AUTHOR]
Copyright of Environmental Modelling & Software is the property of Elsevier B.V. and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)
Database: GreenFILE
Description
Abstract:Total column water vapor (TCWV) is vital for the water cycle, precipitation, and extreme weather events. This study proposes a novel adaptive time-scale decomposition fused LSTM-Transformer framework (ATSD-LT). ATSD employs gradient-optimized filters to resolve sequences into trends and multi-scale fluctuations, suppressing aliasing and frequency leakage. Through heterogeneous ensemble modeling, LSTM extracts trend evolution while Transformer captures global long-range spatiotemporal correlations. Results yield superior performance, particularly in spring and summer (R2 of 0.959 and 0.967, respectively). Compared to Autoformer, Informer, Transformer, and LSTM, ATSD-LT demonstrates more significant gains in winter (improving R2 by ∼18%–55% and reducing MAE by 64%–76%). Ablation tests confirm the proposed ATSD-LT framework achieves a total 79.4% reduction in summer MAE, while feature weighting provides an additional 12.3% goodness-of-fit gain. Spatial mapping reproduces tropical highs (e.g., Amazon Basin) and high-latitude lows (e.g., northern North America); cross- and site-validation further confirm framework robustness. ATSD-LT effectively supports global climate monitoring and water management. [Display omitted] • Propose ATSD-LT framework for multi-scale TCWV spatiotemporal prediction. • Fuse ATSD with LSTM-Transformer, boosting R2 by 28% and cutting MAE by 70%. • Accurately map tropical high-value and high-latitude low-value TCWV zones. • Validate stability via cross-validation for agricultural water management. [ABSTRACT FROM AUTHOR]
ISSN:13648152
DOI:10.1016/j.envsoft.2026.106967