Rouzies et al. (2026) Comparison of ensemble assimilation methods in a hydrological model dedicated to agricultural best management practices

Identification

• Journal: Hydrology and earth system sciences
• Year: 2026
• Date: 2026-01-05
• Authors: Émilie Rouzies, Claire Lauvernet, Arthur Vidard
• DOI: 10.5194/hess-30-1-2026

Research Groups

• INRAE, RiverLy, Lyon-Villeurbanne, 69625 Villeurbanne Cedex, France
• Univ. Grenoble-Alpes, Inria, CNRS, Grenoble-INP, LJK, 38000 Grenoble, France

Short Summary

This study compares three ensemble data assimilation methods (EnKF, ES-MDA, iEnKS) for jointly estimating vertical moisture profiles and soil water retention properties within the PESHMELBA hydrological model. Using synthetic surface moisture images from a virtual agricultural catchment, the research aims to reduce model uncertainties and improve water quality management.

Objective

• To propose and evaluate a data assimilation (DA) framework for the process-oriented, modular PESticide and Hydrology Modelling at the catchment scale (PESHMELBA) model in risk assessment applications.
• To compare the performance of three ensemble DA algorithms (Ensemble Kalman Filter - EnKF, Ensemble Smoother with Multiple Data Assimilation - ES-MDA, and Iterative Ensemble Kalman Smoother - iEnKS) for jointly estimating vertical moisture profiles and saturated water content (θs) parameters using synthetic surface soil moisture images.

Study Configuration

• Spatial Scale: A virtual catchment (Morcille-like) composed of 10 vineyard plots, 4 vegetative filter strips (VFSs), and a river portion. Soil columns are 4 meters deep, vertically discretized into 25 numerical cells with thicknesses ranging from 0.05 meters to 1 meter.
• Temporal Scale: A 78-day simulation period, representing a typical winter period, with a nominal observation frequency of 6 days (144 hours).

Methodology and Data

• Models used:
  • Hydrological Model: PESticide and Hydrology Modelling at the catchment scale (PESHMELBA)
  • Data Assimilation Methods: Ensemble Kalman Filter (EnKF), Ensemble Smoother with Multiple Data Assimilation (ES-MDA), Iterative Ensemble Kalman Smoother (iEnKS)
• Data sources: Synthetic images mimicking satellite surface moisture data in the top 5 centimeters, generated from a "true" PESHMELBA reference run and perturbed with Gaussian, non-biased noise. These synthetic data mimic observations from the synergistic use of Sentinel-1 and Sentinel-2 satellites.

Main Results

• The ES-MDA method consistently outperformed EnKF and iEnKS in estimating surface moisture and saturated water contents (θs) for surface horizons, achieving average Continuous Ranked Probability Skill Scores (CRPSS) of 0.38.
• All methods showed limited performance in correcting subsurface moisture variables (at 0.2 meters and 4 meters depth) and parameters, with the iEnKS often degrading the estimation in deeper layers.
• For surface parameters (θs), all methods significantly improved estimation (CRPSS > 0.58), with EnKF and ES-MDA performing best. For subsurface parameters, the iEnKS showed slightly better, though still limited, performance.
• Computational costs varied significantly: EnKF (277 CPU hours) was the fastest, followed by ES-MDA (558 CPU hours), and iEnKS (2143 CPU hours) was the slowest.
• Sensitivity analysis showed that for significant improvements in surface moisture estimation, observation errors should be less than 0.05 cm³ cm⁻³ for EnKF and 0.1 cm³ cm⁻³ for ES-MDA.
• An observation frequency of 144 hours (6 days) was found to be nearly optimal for ES-MDA, while EnKF performed best at 72 hours.
• An ensemble size of 100 members is recommended for stable performance for both ES-MDA and EnKF.
• A lack of correlation between surface and subsurface moisture was identified as the primary reason for the limited ability of surface observations to correct deeper soil layers.

Contributions

• Establishes a rigorous data assimilation framework for the modular, process-oriented PESHMELBA hydrological model, specifically designed for water quality management and pesticide transfer assessment in agricultural catchments.
• Provides a comprehensive comparison of three representative ensemble data assimilation methods (EnKF, ES-MDA, iEnKS) for joint state and parameter estimation in a complex hydrological model.
• Identifies ES-MDA as the most suitable method for joint estimation of surface moisture and saturated water content parameters in this context, offering a good balance between performance and computational cost.
• Offers practical guidelines for optimizing data assimilation setup parameters (observation error, frequency, ensemble size) for future operational applications of PESHMELBA.
• Highlights the inherent limitation of using only surface moisture observations for correcting subsurface hydrological variables and parameters due to poor vertical correlation, suggesting the need for multi-source data assimilation in future work.

Funding

No specific funding projects or reference codes were explicitly mentioned in the provided text.

Citation

@article{Rouzies2026Comparison,
  author = {Rouzies, Émilie and Lauvernet, Claire and Vidard, Arthur},
  title = {Comparison of ensemble assimilation methods in a hydrological model dedicated to agricultural best management practices},
  journal = {Hydrology and earth system sciences},
  year = {2026},
  doi = {10.5194/hess-30-1-2026},
  url = {https://doi.org/10.5194/hess-30-1-2026}
}