Mwangi et al. (2026) Uncertainties in long-term ensemble estimates of contextual evapotranspiration over southern France

Identification

• Journal: Hydrology and earth system sciences
• Year: 2026
• Date: 2026-02-26
• Authors: Samuel Mwangi, Albert Olioso, Jordi Etchanchu, Kanishka Mallick, Aolin Jia, Jérôme Demarty, Nesrine Farhani, Emmanuelle Sarrazin, P. Gamet, Jean-Louis Roujean, Gilles Boulet
• DOI: 10.5194/hess-30-1117-2026

Research Groups

• Unité de Recherche Ecologie des Forêts Méditerranéennes, INRAE, Avignon, France
• Université de Toulouse, CESBIO, CNES/CNRS/IRD/UPS/INRAE, Toulouse, France
• UMR HSM, IRD-CNRS-Université de Montpellier, Montpellier, France
• Institute of Agricultural Sciences (IAS), Dept. of Environmental Systems Science (D-USYS), ETH Zürich, Zurich, Switzerland
• Indo-French Cell for Water Sciences (IFCWS), IISc, Bangalore, India

Short Summary

This study quantifies uncertainties in long-term contextual evapotranspiration (ET) estimates over southern France using the EVASPA ensemble model, revealing that Land Surface Temperature (LST) inputs and evaporative fraction (EF) formulations are the dominant sources of variability, with uncertainties peaking in summer. The ensemble approach provides reliable flux estimates and a meaningful uncertainty spread, enhancing ET retrieval robustness.

Objective

• To preliminarily assess the temporally continuous ensemble estimates of contextual ET in preparation for the TRISHNA SWEAT products.
• To quantify and analyze the inherent uncertainties introduced by various factors/variables in long-term contextual ET estimates.

Study Configuration

• Spatial Scale: Southern France (two regions: South East and South West, combined), covering diverse ecosystems (croplands, grasslands, forests). Data resampled to a uniform 1 km resolution. Evaluation performed across eight specific flux sites.
• Temporal Scale: 2004–2024 (21 years), yielding 972 instantaneous ET estimates, aggregated to daily values.

Methodology and Data

• Models used:
  • EVapotranspiration Assessment from SPAce (EVASPA) contextual tool.
  • Surface Energy Balance (SEB) methods: Temperature–Vegetation Index (T-VI) and Simplified Surface Energy Balance Index (S-SEBI).
  • Ensemble modeling framework integrating:
    • 4 Land Surface Temperature (LST) datasets.
    • 3 radiation datasets.
    • 9 evaporative fraction (EF) formulations.
    • 9 ground heat flux (G) parameterizations.
    • Total of 972 ensemble members.
  • Temporal interpolation for missing daily ET using continuous daily solar radiation data.
• Data sources:
  • Satellite:
    • MODIS (TERRA: MOD11A1, MOD21A1D LST/E; AQUA: MYD11A1, MYD21A1D LST/E; MOD13A2 NDVI; MCD43A3 Albedo; MCD15A3 LAI; MOD44W Water mask; MODDEM DEM).
    • VIIRS (VNP21A1D LST/E, 2012–2024).
    • MSG (Meteosat Second Generation) radiation products (0.05°, 15 min).
  • Reanalysis:
    • ERA5 (clear-sky) and ERA5-Land (all-sky) surface/single-level radiation (hourly, ~25 km / ~9 km).
    • MERRA2 radiation data (hourly, 0.5° × 0.625°).
  • Observation (for validation): Fluxnet2015, ICOS, and other independent flux sites (daily ET observations).

Main Results

• LST inputs and evaporative fraction (EF) formulations are the dominant sources of uncertainty in EVASPA ET estimates, with ground heat flux (G) methods having the least influence.
• Absolute uncertainties in ET estimates peak during summer, following the annual cycle of radiation.
• Satellite overpass time introduces more uncertainty to gap-filled daily ET estimates than the specific LST/emissivity (LSE) separation methods (Generalized Split-Window vs. Temperature-Emissivity Separation).
• A simple ensemble average provides reasonable agreement with in-situ observations (RMSD ranging from 0.7 to 1.6 mmd−1), outperforming most individual ensemble members.
• Performance is best over forested sites and lowest over heterogeneous grassland sites within urban environments.
• Seasonal performance shows best agreement (lowest RMSD and MAE) in winter, degrading through spring and summer, and improving in autumn.
• Interpolated daily ET estimates, gap-filled using daily average solar radiation, demonstrate better performance than non-interpolated estimates.
• MSG-based daily ET estimates consistently outperform MERRA- and ERA5-based estimates, likely due to higher spatial resolution.

Contributions

• Provides a preliminary assessment and uncertainty quantification for future operational evapotranspiration products of the TRISHNA satellite mission.
• Highlights the critical importance of both spatial and temporal resolution in remote sensing of ET, particularly the sensitivity of ET estimates to satellite overpass time.
• Demonstrates the value of ensemble-based modeling frameworks (like EVASPA) for characterizing uncertainty, improving ET estimation accuracy, and enhancing the resilience of Earth observation products.
• Suggests that ensemble modeling, by allowing optimal member selection based on surface and climatic conditions, can provide more robust and informative ET estimates for agro-hydrology and land surface modeling.

Funding

• CNES through the APR program (no. 240894/00), evaluated by the TOSCA committee, in preparation for the TRISHNA satellite mission.

Citation

@article{Mwangi2026Uncertainties,
  author = {Mwangi, Samuel and Olioso, Albert and Etchanchu, Jordi and Mallick, Kanishka and Jia, Aolin and Demarty, Jérôme and Farhani, Nesrine and Sarrazin, Emmanuelle and Gamet, P. and Roujean, Jean-Louis and Boulet, Gilles},
  title = {Uncertainties in long-term ensemble estimates of contextual evapotranspiration over southern France},
  journal = {Hydrology and earth system sciences},
  year = {2026},
  doi = {10.5194/hess-30-1117-2026},
  url = {https://doi.org/10.5194/hess-30-1117-2026}
}