Li et al. (2026) Observation‐Driven Forecast of Global Terrestrial Water Storage and Evaluation for 2010–2024

⚠️ Warning: This summary was generated from the abstract only, as the full text was not available.

Identification

• Journal: Water Resources Research
• Year: 2026
• Date: 2026-02-01
• Authors: F. Li, Oliver Baur
• DOI: 10.1029/2025wr041710

Research Groups

• European Centre for Medium‐Range Weather Forecasts (ECMWF)
• University of Bonn, Institute of Geodesy and Geoinformation (implied by data repository)

Short Summary

This study develops a machine learning approach to forecast GRACE-like terrestrial water storage changes (TWSC) up to 12 months ahead, addressing the latency of GRACE/GRACE-FO products. The method demonstrates improved accuracy and robustness compared to ECMWF's seasonal forecasts, providing a viable data-driven solution for operational TWSC forecasting.

Objective

• To address the latency of GRACE/GRACE-FO terrestrial water storage change (TWSC) products, which limits their utility for real-time and operational forecasting.
• To develop and evaluate a machine learning-based method for forecasting GRACE-like TWSC up to 12 months ahead, using only observational and reanalysis-based inputs.
• To benchmark the developed method against state-of-the-art seasonal forecasts from the European Centre for Medium‐Range Weather Forecasts (ECMWF)’s new long‐range forecasting system (SEAS5).

Study Configuration

• Spatial Scale: Global, at 1° resolution.
• Temporal Scale: Forecasts up to 12 months ahead; evaluation period from 2010 to 2024; semi-operational forecast data set from 2024 onward.

Methodology and Data

• Models used: Machine learning model (specific architecture not detailed in abstract).
• Data sources:
  • Observational and reanalysis-based inputs (for machine learning model).
  • GRACE and GRACE Follow‐On (GRACE/‐FO) satellite mission data (for terrestrial water storage change measurements, which are being forecasted).
  • European Centre for Medium‐Range Weather Forecasts (ECMWF)’s new long‐range forecasting system (SEAS5) (for benchmarking).

Main Results

• The developed machine learning method offers improved accuracy and robustness in forecasting GRACE-like TWSC compared to the ECMWF SEAS5 seasonal forecasts.
• The method provides a viable data-driven alternative for operational TWSC forecasting, effectively addressing the latency issue of GRACE/GRACE-FO products.
• Global forecast data sets at 1° resolution are generated and made publicly available, extending GRACE-like insights into the near future.

Contributions

• Provides a novel, robust, and publicly available machine learning-based solution for real-time and operational forecasting of GRACE-like terrestrial water storage changes, overcoming the inherent latency of satellite products.
• Demonstrates superior performance of the data-driven approach compared to a state-of-the-art seasonal forecasting system (ECMWF SEAS5).
• Generates and disseminates global, high-resolution (1°), long-term (2010-2024 hindcast, 2024 onward forecast) TWSC forecast datasets, serving as a valuable resource for various applications.
• Enables critical applications such as drought early warning, sea level prediction, hydrological model validation, and geodetic applications (e.g., forecasting Earth orientation parameters, estimating loading corrections in GNSS and altimetry data analysis).

Funding

Not specified in the abstract.

Citation

@article{Li2026ObservationDriven,
  author = {Li, F. and Baur, Oliver},
  title = {Observation‐Driven Forecast of Global Terrestrial Water Storage and Evaluation for 2010–2024},
  journal = {Water Resources Research},
  year = {2026},
  doi = {10.1029/2025wr041710},
  url = {https://doi.org/10.1029/2025wr041710}
}