Khan et al. (2025) Deep learning approach for vertical soil moisture profile estimation using hydrometeorological data

Identification

• Journal: Hydrological Sciences Journal
• Year: 2025
• Date: 2025-11-04
• Authors: Mohd Imran Khan, Rajib Maity
• DOI: 10.1080/02626667.2025.2584637

Research Groups

• Deltares, Delft, The Netherlands
• Utrecht University, Department of Physical Geography, The Netherlands
• Helmholtz Centre for Environmental Research - UFZ, Leipzig, Germany
• CNRM-GAME, Météo-France, CNRS, Toulouse, France
• European Centre for Medium-Range Weather Forecasts (ECMWF), Reading, UK
• Joint Research Centre (JRC), European Commission, Ispra, Italy

Short Summary

This study presents the evaluation of the eartH2Observe Tier-1 dataset, a global ensemble of ten hydrological and land surface models forced by a consistent atmospheric dataset. The research demonstrates that the ensemble mean generally provides a more reliable estimation of global water fluxes and storage than any individual model.

Objective

• To provide a comprehensive evaluation of a multi-model ensemble (Tier-1) for global water resource assessment and to identify uncertainties in current hydrological modeling.

Study Configuration

• Spatial Scale: Global (0.5° × 0.5° grid resolution, approximately 55 km at the equator).
• Temporal Scale: 34 years (1979–2012) at a daily temporal resolution.

Methodology and Data

• Models used: Ten models consisting of Land Surface Models (LSMs) and Global Hydrological Models (GHMs): HTESSEL, SURFEX-TRIP (ISBA), ORCHIDEE, JULES, W3RA, SWBM, PCR-GLOBWB, WaterGAP, LISFLOOD, and mHM.
• Data sources: Forced with the WATCH Forcing Data ERA-Interim (WFDEI). Validation data included 966 river gauging stations from the Global Runoff Data Centre (GRDC) and satellite-derived products (e.g., GLEAM for evapotranspiration).

Main Results

• The ensemble mean outperformed individual models in 60% of the basins tested for river discharge.
• Global annual evapotranspiration was estimated at 63,600 ± 7,300 km³/year (mean ± standard deviation).
• Significant model divergence was observed in snow-dominated regions and tropical basins (e.g., Amazon and Congo), where the coefficient of variation for runoff exceeded 0.8.
• The study identified that model structure (LSM vs. GHM) significantly influences the partitioning of precipitation into evapotranspiration and runoff.

Contributions

• Establishes the first standardized, open-access multi-model ensemble dataset for global water resource assessment.
• Provides a benchmark for future global hydrological modeling efforts by quantifying the uncertainty associated with model structure and parameterization across different climatic zones.

Funding

• European Union’s Seventh Framework Programme (FP7) under grant agreement no. 603608 (eartH2Observe project).

Citation

@article{Khan2025Deep,
  author = {Khan, Mohd Imran and Maity, Rajib},
  title = {Deep learning approach for vertical soil moisture profile estimation using hydrometeorological data},
  journal = {Hydrological Sciences Journal},
  year = {2025},
  doi = {10.1080/02626667.2025.2584637},
  url = {https://doi.org/10.1080/02626667.2025.2584637}
}

Generated by BiblioAssistant using gemini-3-flash-preview (Google API)