Mazzolini et al. (2025) Spatio-temporal snow data assimilation with the ICESat-2 laser altimeter

Identification

• Journal: ˜The œcryosphere
• Year: 2025
• Date: 2025-09-16
• Authors: Marco Mazzolini, Kristoffer Aalstad, Esteban Alonso‐González, Sebastian Westermann, Désirée Treichler
• DOI: 10.5194/tc-19-3831-2025

Research Groups

• Department of Geosciences, University of Oslo, Oslo, Norway
• Pyrenean Institute of Ecology, CSIC, Jaca, Spain

Short Summary

This study presents a novel spatio-temporal data assimilation framework to integrate sparse ICESat-2 laser altimeter snow depth profiles with Sentinel-2 fractional snow-covered area (fSCA) observations into a snow model. It demonstrates that jointly assimilating both data types significantly improves the accuracy and spatial distribution of simulated snow depth, particularly during the accumulation season, compared to using fSCA alone.

Objective

• To determine if information from sparse ICESat-2 snow depth profiles can be used to infer catchment-scale average snow depth and its complete spatial distribution.
• To compare the performance of assimilating sparse ICESat-2 snow depth retrievals against more commonly used fSCA observations from optical satellites.
• To investigate if ensemble-based data assimilation can effectively leverage information from both fSCA and sparse snow depth observations in a joint assimilation experiment.

Study Configuration

• Spatial Scale: Izas experimental catchment (55 hectares) in the central Spanish Pyrenees, with elevations ranging from 2075 to 2325 meters above sea level. Simulations were performed at a 20 meter spatial resolution within an extended domain of approximately 5 kilometers by 3 kilometers.
• Temporal Scale: Water year 2020 (November to July). ICESat-2 data from 5 February 2020. Sentinel-2 fSCA data from 5 February to 17 July 2020. Snow melt-out date climatology derived from 2017–2021.

Methodology and Data

• Models used:
  • Flexible Snow Model (FSM2) for snowpack simulation.
  • Multiple Snow Data Assimilation System (MuSA) for ensemble-based data assimilation.
  • Deterministic Ensemble Smoother with Multiple Data Assimilation (DES-MDA) algorithm.
  • TopoSCALE for downscaling meteorological forcing.
• Data sources:
  • ICESat-2 (ATL03 product) laser altimeter for sparse snow depth profiles.
  • Sentinel-2A and Sentinel-2B (MSI Level-2A product) for fractional snow-covered area (fSCA).
  • ERA5 reanalysis for meteorological forcing.
  • Plan Nacional de Ortofotografía Aérea (PNOA) airborne lidar initiative (2 meter resolution) for snow-off Digital Elevation Model (DEM).
  • Drone-based snow depth maps (1 meter original resolution, resampled to 20 meters) for independent validation.
  • Theia land data centre for Climatology of Snow Melt-out Date (CSMD).

Main Results

• Assimilating ICESat-2 snow depth profiles successfully updated the snow model in the unobserved experimental catchment, improving simulated average snow depth compared to the prior run.
• Joint assimilation of ICESat-2 snow depth and Sentinel-2 fSCA led to an accurate reconstruction of the snow depth spatial distribution.
• For the accumulation season, the joint assimilation experiment (J) improved the Continuous Ranked Probability Score (CRPS) by 19% (CRPS of 35 ± 7 centimeters) compared to assimilating fSCA alone (C) (CRPS of 44 ± 10 centimeters).
• Experiment (C) (fSCA only) overestimated catchment-average snow depth at peak snow water equivalent (SWE) by 51%.
• Experiment (D) (ICESat-2 only) overestimated catchment-average snow depth by 23%, showing improved precision and accuracy during the winter accumulation season but degraded performance in the melt season.
• Experiment (J) accurately reconstructed catchment-average snow depth, with only a 7 centimeter difference from observed on 11 March, and reproduced the general shape of the snow depth distribution.

Contributions

• First demonstration of hyper-resolution multivariate snow data assimilation.
• First study to assimilate ICESat-2 snow depth observations and show how sparse profiles located outside the area of interest can be leveraged through spatio-temporal data assimilation using a feature space defined by topographical index and melt-out date climatology.
• Modified the MuSA system to effectively exploit sparse ICESat-2 observations and jointly assimilate them with spatially complete fSCA data.
• Demonstrated the potential of ICESat-2 data to enhance current state-of-the-art snow reanalyses, particularly for estimating maximum seasonal snow accumulation.
• Proposed workflow utilizes globally available datasets, making it applicable to inaccessible regions.

Funding

• Research Council of Norway (SNOWDEPTH project, grant no. 325519)
• ESA (Glaciers_cci+, grant no. 4000127593/19/I-NB)
• Research Council of Norway (Spot-On project, grant no. 301552)
• ERC (grant no. 01096057 GLAC-MASS)
• ESA (CCI Research Fellowship project PATCHES)
• ESA (CCI Research Fellowship project SnowHotspots)

Citation

@article{Mazzolini2025Spatiotemporal,
  author = {Mazzolini, Marco and Aalstad, Kristoffer and Alonso‐González, Esteban and Westermann, Sebastian and Treichler, Désirée},
  title = {Spatio-temporal snow data assimilation with the ICESat-2 laser altimeter},
  journal = {˜The œcryosphere},
  year = {2025},
  doi = {10.5194/tc-19-3831-2025},
  url = {https://doi.org/10.5194/tc-19-3831-2025}
}