Alonso‐González et al. (2026) Ensemble-based data assimilation improves hyperresolution snowpack simulations in forests

Identification

• Journal: ˜The œcryosphere
• Year: 2026
• Date: 2026-01-14
• Authors: Esteban Alonso‐González, A. A. Harpold, Jessica Lundquist, Cara R. Piske, Laura Sourp, Kristoffer Aalstad, Simon Gascoin
• DOI: 10.5194/tc-20-209-2026

Research Groups

• Instituto Pirenaico de Ecología, Consejo Superior de Investigaciones Científicas (IPE-CSIC), Jaca, Spain
• Department of Natural Resources and Environmental Science, University of Nevada, Reno, Nevada, USA
• Civil and Environmental Engineering, University of Washington, Seattle, Washington, USA
• Airborne Snow Observatories Inc., Mammoth Lakes, California, USA
• Université de Toulouse, CNRS, CNES, IRD, INRAE, Centre d’études spatiales de la biosphère, Toulouse, France
• MAGELLIUM, Ramonville Saint-Agne, France
• Department of Geosciences, University of Oslo, Oslo, Norway

Short Summary

This study proposes and tests ensemble-based data assimilation (DA) experiments to improve hyperresolution snowpack simulations in forested environments by propagating information from forest clearings to sub-canopy areas. The successful experiments significantly improved snowpack simulations in terms of validation metrics (e.g., correlation coefficient from 0.1 to 0.8) and spatial patterns compared to a deterministic reference run.

Objective

• To explore the potential of lidar-derived real observations to update distributed snowpack simulations at hyperresolution (10 m) scales in forest environments.
• To test different spatiotemporal data assimilation configurations for estimating snow under the canopy when only observations in forest gaps are available.
• To propagate snowpack information obtained in forest clearings to sub-canopy areas where remote sensing observations are occluded.

Study Configuration

• Spatial Scale: A 2 km × 2 km domain within the Sagehen Creek watershed (California, USA), with simulations and observations at 10 m spatial resolution.
• Temporal Scale: Data assimilation was performed using a single airborne lidar snow depth map collected on 21 March 2022, with model simulations covering a water year.

Methodology and Data

• Models used:
  • Flexible Snow Model (FSM2)
  • Multiple Snow data Assimilation system (MuSA)
  • MicroMet (for meteorological forcing downscaling)
  • Deterministic Ensemble Smoother with Multiple Data Assimilation (DES-MDA)
• Data sources:
  • Airborne lidar-derived snow depth maps (National Center for Airborne Laser Mapping, 21 March 2022, and a snow-off flight)
  • Lidar-derived vegetation parameters (vegetation height, Vegetation Area Index, in-forest sky view factor)
  • ERA5 atmospheric reanalysis (for meteorological forcing)
  • LiDAR-based digital elevation model
  • SNOTEL data (for reference snow water equivalent, SWE)

Main Results

• Ensemble-based data assimilation (DA) experiments using Euclidean (Eu) and Mahalanobis (Ma) distance metrics significantly improved snowpack simulations compared to the deterministic reference run.
• The correlation coefficient (R) between simulated and observed snow depth improved from 0.1 (reference) to an average of 0.8 for successful DA experiments.
• Root Mean Square Error (RMSE) improved by approximately 30% (from 0.32 m to 0.2 m).
• Spatial patterns, quantified by the Frechet distance (FrDist), showed significant improvement (from 0.029 for reference to an average of 0.005 for successful DA experiments).
• Mahalanobis distance configurations (e.g., Ma0.5, Ma1) consistently yielded lower FrDist values (e.g., 0.003) than Euclidean configurations (e.g., 0.006), indicating a better representation of small-scale spatial patterns.
• Principal Component Analysis (PCA)-based experiments showed only slight improvements in correlation (R = 0.20 to 0.46) but degraded other metrics or were similar to the reference, failing to adequately simulate spatial patterns.
• The posterior mean multiplicative precipitation parameter was 1.06 (±0.30) and the additive temperature parameter was -0.04 °C (±0.73 °C) for the Ma0.5 experiment, suggesting that DA primarily corrected small-scale meteorological forcing redistribution rather than large biases.

Contributions

• Demonstrates the successful application of ensemble-based data assimilation to improve hyperresolution (10 m) snowpack simulations in complex forested terrain by effectively propagating information from clearings to sub-canopy areas.
• Evaluates and compares different spatio-temporal data assimilation configurations (Euclidean, Mahalanobis, and PCA-based distances) for information propagation in environments with spatially incomplete observations.
• Highlights the superior performance of Mahalanobis distance in a topographical feature space for preserving realistic small-scale spatial snowpack patterns compared to Euclidean distance.
• Introduces and validates significant improvements to the open-source Multiple Snow data Assimilation (MuSA) system, including efficient sparse matrix operations, k-d tree implementation for distance mapping, and regularization (jitter technique) for enhanced numerical stability in prior covariance matrix sampling.
• Provides insights into the flexibility of choosing correlation length scale hyperparameters in DA for forested environments and the benefits of updating meteorological correction parameters rather than snowpack states directly.

Funding

• European Space Agency Climate Change Initiative (ESA-CCI) Research Fellowship (SnowHotspots project)
• “Ramon y Cajal” Fellowship RYC2023-044416-I
• AD-For project no. 20253AT017 from the CSIC talent attraction programme
• ERC-2022-ADG under grant agreement No 101096057 GLAC-MASS
• ESA-CCI Research Fellowship (PATCHES project)
• NSF EAR #2012310
• NSF EAR #1723990
• CSIC Open Access Publication Support Initiative (for article processing charges)
• Centre National d’Études Spatiales (CNES) (supercomputing infrastructure through TREX cluster)

Citation

@article{AlonsoGonzález2026Ensemblebased,
  author = {Alonso‐González, Esteban and Harpold, A. A. and Lundquist, Jessica and Piske, Cara R. and Sourp, Laura and Aalstad, Kristoffer and Gascoin, Simon},
  title = {Ensemble-based data assimilation improves hyperresolution snowpack simulations in forests},
  journal = {˜The œcryosphere},
  year = {2026},
  doi = {10.5194/tc-20-209-2026},
  url = {https://doi.org/10.5194/tc-20-209-2026}
}