Feng et al. (2026) Reconstructing terrestrial water storage and quantifying drought evolution in Australia using LSTM networks

Identification

• Journal: Journal of Hydrology
• Year: 2026
• Date: 2026-09-01
• Authors: Bo Feng, Yan Xu, Nan Jiang, Biaobiao Guo, Peng Yin, Ao Guo, Tianhe Xu, Harald Schuh
• DOI: 10.1016/j.jhydrol.2026.136335

Research Groups

• School of Space Science and Technology, Shandong University, China
• Beijing Huitu Technology (Group) CO., LTD., China
• School of Geographic Sciences, Xinyang Normal University, China
• Institute of Geodesy and Geoinformation Science, Technical University of Berlin, Germany

Short Summary

The study implements a Long Short-Term Memory (LSTM) framework to reconstruct high-resolution Terrestrial Water Storage (TWS) in Australia by fusing GNSS and GRACE data with meteorological inputs, effectively filling GRACE mission gaps and improving drought monitoring.

Objective

• To overcome the coarse spatial resolution and data gaps of GRACE and the high-frequency noise and uneven distribution of GNSS to achieve a continuous, high-quality reconstruction of TWS for quantifying drought evolution in Australia.

Study Configuration

• Spatial Scale: Australia (continental scale)
• Temporal Scale: 2011–2023

Methodology and Data

• Models used: Long Short-Term Memory (LSTM) networks
• Data sources:
  • GNSS (Equivalent Water Height - EWH time series)
  • GRACE (Gravity Recovery and Climate Experiment)
  • Meteorological data
  • Global Land Data Assimilation System (GLDAS)
  • NscPDSI (Palmer Drought Severity Index)

Main Results

• TWS Reconstruction: The LSTM framework significantly improved the agreement between GNSS-derived EWH and GRACE benchmarks, with correlations reaching 0.854 in the Northern Territory and increasing from 0.290 to 0.656 in Western Australia.
• Continuity: The model successfully bridged the GRACE mission gap, ensuring a continuous TWS time series.
• Drought Quantification: The derived Drought Severity Index (LSTM-DSI) showed a correlation of 0.663 with the NscPDSI, demonstrating its ability to capture extreme hydrological signals.

Contributions

• Developed a multi-source fusion approach that integrates the high temporal resolution of GNSS with the physical stability of GRACE.
• Provided a method to mitigate the inherent limitations of individual geodetic sensors (spatial resolution of GRACE and noise/distribution of GNSS).
• Established a robust foundation for investigating hydroclimatic interactions and drought evolution in Australia.

Funding

• Not provided in the text.

Citation

@article{Feng2026Reconstructing,
  author = {Feng, Bo and Xu, Yan and Jiang, Nan and Guo, Biaobiao and Yin, Peng and Guo, Ao and Xu, Tianhe and Schuh, Harald},
  title = {Reconstructing terrestrial water storage and quantifying drought evolution in Australia using LSTM networks},
  journal = {Journal of Hydrology},
  year = {2026},
  doi = {10.1016/j.jhydrol.2026.136335},
  url = {https://doi.org/10.1016/j.jhydrol.2026.136335}
}