Li et al. (2026) Skills in sub-seasonal to seasonal terrestrial water storage forecasting: insights from the FEWS NET land data assimilation system

Identification

• Journal: Hydrology and earth system sciences
• Year: 2026
• Date: 2026-02-24
• Authors: Bailing Li, Abheera Hazra, Amy K. McNally, Kimberly Slinski, Shraddhanand Shukla, Weston Anderson
• DOI: 10.5194/hess-30-1097-2026

Research Groups

• ESSIC, University of Maryland, College Park, MD, USA
• NASA Goddard Space Flight Center, Greenbelt, MD, USA
• University of California at Santa Barbara, Santa Barbara, CA, USA
• Department of Geography, University of Maryland, College Park, MD, USA

Short Summary

This study evaluates subseasonal to seasonal (S2S) terrestrial water storage (TWS) forecasts over Africa from the Famine Early Warning Systems Network (FEWS NET) land data assimilation system (FLDAS) using GRACE/FO observations. It finds that the NASA Catchment Land Surface Model (CLSM) generally outperforms Noah-MP, primarily due to its more accurate reanalysis-based initial conditions and stronger representation of TWS interannual variability.

Objective

• To provide an objective evaluation of the skill of subseasonal to seasonal (S2S) terrestrial water storage (TWS) forecasts from the FLDAS-Forecast system over Africa using GRACE/FO observations.
• To improve understanding of how land surface model physics influence TWS forecast skill and how they interact with precipitation forecasts.

Study Configuration

• Spatial Scale: African continent and a large portion of the Middle East, excluding regions with significant anthropogenic groundwater depletion. Data and model simulations are at a 0.25° spatial resolution. GRACE/FO effective resolution is approximately 150,000 km².
• Temporal Scale: Subseasonal to seasonal (S2S) forecasts with lead times of 1 to 6 months. Hindcasts cover the historical period from 2003 to 2020, analyzed using monthly non-seasonal TWS anomalies.

Methodology and Data

• Models used:
  • Famine Early Warning Systems Network (FEWS NET) land data assimilation system (FLDAS-Forecast).
  • Noah with Multi-Parameterization (Noah-MP) land surface model.
  • NASA Catchment Land Surface Model (CLSM).
  • North American Multi-Model Ensemble (NMME) for S2S precipitation hindcasts.
  • Goddard Earth Observing System (GEOS) ensemble hindcasts for non-precipitation meteorological fields.
  • NASA Land Information System (LIS) framework for model simulation and data processing.
• Data sources:
  • Terrestrial Water Storage (TWS) observations: Gravity Recovery and Climate Experiment (GRACE) and its Follow-On (GRACE/FO) mission (CSR GRACE TWS mascon product, monthly anomalies relative to 2004–2009 mean).
  • Meteorological input for reanalysis (initial conditions):
    • Precipitation: Climate Hazards Infrared Precipitation with Stations (CHIRPS, 0.05° spatial resolution, 6 h interval, interpolated to 0.25°).
    • Other meteorological fields: Modern-Era Retrospective Analysis for Research and Applications, Version 2 (MERRA-2, 0.5° latitude by 0.625° longitude resolution, interpolated to 0.25°).
  • Evaluation metrics: Root Mean Square Error (RMSE), Pearson correlation, and Relative Operating Characteristic (ROC) score for tercile forecasts.

Main Results

• The Catchment Land Surface Model (CLSM) generally outperformed Noah-MP in S2S TWS forecasts over Africa.
• CLSM achieved ROC scores exceeding 0.6 (the threshold for predictive skill) for tercile forecasts over more than 50% of the study domain across 1- to 6-month lead times.
• CLSM forecasts showed stronger correlations with GRACE/FO data (e.g., domain-averaged TWS reanalysis correlation of 0.72 for CLSM vs. 0.56 for Noah-MP).
• The superior performance of CLSM is attributed to its reanalysis-based initial conditions, which more accurately captured interannual variability observed in GRACE/FO, and its simulation of strong TWS temporal variability and persistence.
• Noah-MP simulated weaker groundwater persistence and lower overall skill, especially in central Africa, where skill declined quickly with increasing lead times.
• TWS forecasts were highly sensitive to precipitation interannual variability; forecasts driven by precipitation with lower interannual variability (e.g., GFDL, CSM5) yielded more accurate predictions (lowest RMSEs, highest correlations), while those with higher variability (e.g., GEOSv2) performed worse.
• Discrepancies between simulated and GRACE/FO TWS persistence increased with lag, indicating limitations in model physics.
• A case study demonstrated that both models generally agreed on the regional placement of extreme events (e.g., 2015–2016 El Niño-induced drought in southern Africa and wet conditions in eastern Africa), highlighting the value of TWS forecasts for early warnings.

Contributions

• Provides the first objective evaluation of operational subseasonal to seasonal (S2S) terrestrial water storage (TWS) forecasts from the FLDAS-Forecast system over Africa using independent GRACE/FO observations.
• Quantifies the significant influence of land surface model physics, particularly the representation of groundwater dynamics and interannual variability, on S2S TWS forecast skill.
• Highlights the critical role of independent observations like GRACE/FO for robust evaluation and improvement of TWS forecasts, revealing uncertainties that are masked when using reanalysis as a reference.
• Demonstrates the practical value of TWS forecasts for predicting hydrological extremes (droughts and floods) at S2S scales, crucial for informing disaster responses and food security early warnings.

Funding

• NASA (grant nos. 80NSSC23K0559 and 80NSSC23M0032)
• U.S. Agency for International Development (grant no. PAPA BHA22H00005)
• Bureau of Humanitarian Assistance at the U.S. Agency for International Development

Citation

@article{Li2026Skills,
  author = {Li, Bailing and Hazra, Abheera and McNally, Amy K. and Slinski, Kimberly and Shukla, Shraddhanand and Anderson, Weston},
  title = {Skills in sub-seasonal to seasonal terrestrial water storage forecasting: insights from the FEWS NET land data assimilation system},
  journal = {Hydrology and earth system sciences},
  year = {2026},
  doi = {10.5194/hess-30-1097-2026},
  url = {https://doi.org/10.5194/hess-30-1097-2026}
}