Sun et al. (2026) A novel approach for predicting soil water storage in agricultural areas: An optimized stacked ensemble model based on variational mode decomposition and empirical wavelet transform

Identification

• Journal: Agricultural Water Management
• Year: 2026
• Date: 2026-08-01
• Authors: Tianyu Sun, Liangliang Zhang, Dong Liu, Nan Sun, Xiaochen Qi, Mo Li, Muhammad Abrar Faiz, Tianxiao Li, Song Cui, Muhammad Imran Khan
• DOI: 10.1016/j.agwat.2026.110671

Research Groups

• College of Hydraulic Science and Engineering, Northeast Agricultural University, Harbin, Heilongjiang 150030, China
• Joint International Research Laboratory of Habitat Health of Black Soil in Cold Regions, Ministry of Education, Northeast Agricultural University, Harbin, Heilongjiang 150030, China
• Key Laboratory of Effective Utilization of Agricultural Water Resources of Ministry of Agriculture and Rural Affairs, Northeast Agricultural University, Harbin, Heilongjiang 150030, China
• Heilongjiang Provincial Key Laboratory of Water Resources and Water Conservancy Engineering in Cold Region, Northeast Agricultural University, Harbin, Heilongjiang 150030, China
• Research Center for Eco-Environment Protection of Songhua River Basin, Northeast Agricultural University, Harbin, Heilongjiang 150030, China
• Department of Irrigation and Drainage, University of Agriculture Faisalabad, Pakistan

Short Summary

This study developed a novel approach to predict soil water storage in agricultural areas using an optimized stacked ensemble model based on variational mode decomposition and empirical wavelet transform. The model was applied to the Songnen Plain in Northeast China from 1982-2024 and showed high accuracy in predicting spatiotemporal evolution of soil reservoirs.

Objective

• To develop a hybrid model that combines variational mode decomposition (VMD) and empirical wavelet transform (EWT) for signal decomposition and reconstruction.
• To construct an optimized stacked ensemble model using NSGA-III to integrate the strengths of LSTM, CNN, and TFT in predicting soil water storage capacity.

Study Configuration

• Spatial Scale: Regional scale, focusing on the Songnen Plain in Northeast China.
• Temporal Scale: Monthly data from 1982-2024.

Methodology and Data

• Models used: Variational Mode Decomposition (VMD), Empirical Wavelet Transform (EWT), Long Short-Term Memory (LSTM), Convolutional Neural Network (CNN), Temporal Fusion Transformer (TFT), Non-dominated Sorting Genetic Algorithm III (NSGA-III).
• Data sources: FEWS NET Land Data Assimilation System (FLDAS) on the NASA Earth Science Data portal.

Main Results

• The VE-NLCT model showed high accuracy in predicting spatiotemporal evolution of soil reservoirs, with a mean C value of 0.085 and 96.93% of sites rated "excellent".
• The model outperformed benchmarks across all metrics, including R² = 0.9883, error reductions of 30.72-71.09% in mean absolute error, root mean square error, mean absolute scaled error, dynamic threshold error, and shape-based error.

Contributions

• This study provides a novel approach to predict soil water storage capacity using an optimized stacked ensemble model.
• The VE-NLCT model can be applied to other agricultural regions worldwide with similar characteristics.

Funding

• This research was funded by the National Natural Science Foundation of China (Grant No. 52109243), the Heilongjiang Provincial Key Laboratory of Water Resources and Water Conservancy Engineering in Cold Region (Grant No. 2022-KF01), and the Research Center for Eco-Environment Protection of Songhua River Basin, Northeast Agricultural University.

Citation

@article{Sun2026novel,
  author = {Sun, Tianyu and Zhang, Liangliang and Liu, Dong and Sun, Nan and Qi, Xiaochen and Li, Mo and Faiz, Muhammad Abrar and Li, Tianxiao and Cui, Song and Khan, Muhammad Imran},
  title = {A novel approach for predicting soil water storage in agricultural areas: An optimized stacked ensemble model based on variational mode decomposition and empirical wavelet transform},
  journal = {Agricultural Water Management},
  year = {2026},
  doi = {10.1016/j.agwat.2026.110671},
  url = {https://doi.org/10.1016/j.agwat.2026.110671}
}