Montpetit et al. (2025) Snow Water Equivalent from airborne Ku-band data: the Trail Valley Creek 2018/19 snow experiment

Identification

• Journal: ˜The œcryosphere
• Year: 2025
• Date: 2025-11-07
• Authors: Benoît Montpetit, Julien Meloche, Vincent Vionnet, Chris Derksen, Georgina Woolley, Nicolas R. Leroux, Paul Siqueira, J. Adam, Mike Brady
• DOI: 10.5194/tc-19-5465-2025

Research Groups

• Climate Research Division, Environment and Climate Change Canada, Ontario, Canada
• Meteorological Research Division, Environment and Climate Change Canada, Quebec, Canada
• Northumbria University, Newcastle upon Tyne, UK
• College of Engineering, University of Massachusetts Amherst, MA, USA

Short Summary

This study develops and validates a snow water equivalent (SWE) retrieval algorithm for a proposed Ku-band synthetic aperture radar (SAR) satellite mission by combining a priori snow conditions from a land surface model (SVS-2) with a Markov Chain Monte Carlo (MCMC) Bayesian model coupled with the Snow Microwave Radiative Transfer (SMRT) model, achieving a root-mean-square error (RMSE) of 15.8 mm (16.4 %) for SWE retrieval.

Objective

• To develop and validate a robust snow water equivalent (SWE) retrieval algorithm for the proposed Terrestrial Snow Mass Mission (TSMM) Ku-band SAR satellite, utilizing a Bayesian MCMC approach with land surface model priors and airborne Ku-band data.

Study Configuration

• Spatial Scale: Trail Valley Creek (TVC) watershed, near Inuvik, Northwest Territories, Canada. Airborne SAR measurements covered a 2 km swath with a 2 m ground-range resolution. Ground surveys were conducted over 100 m × 100 m areas.
• Temporal Scale: The 2018/2019 snow experiment, with specific focus on airborne SAR measurements and SVS-2 model outputs from 12–15 January 2019, representing dry snow conditions.

Methodology and Data

• Models used:
  • Canadian Soil Vegetation Snow version 2 (SVS-2) model: Used to generate prior distributions of snow properties.
  • Ensemble System Crocus (ESCROC) model: Integrated within SVS-2 for multi-layered snow information.
  • Snow Microwave Radiative Transfer (SMRT) model (v1.0): Used to simulate radar backscatter (σ0) from snowpack variables, employing the Improved Born Approximation (IBA) for scattering.
  • Markov Chain Monte Carlo (MCMC) Bayesian model (PyMC v5.16.2): Used for iterative optimization of snow properties, employing Adaptive Differential Evolution Metropolis (DEMCZ) sampling.
• Data sources:
  • Airborne Ku-band SAR data: Acquired by a University of Massachusetts (UMASS) instrument on a Cessna-208, operating at 13.285 GHz in VV polarization, with incidence angles from approximately 20° to 70°.
  • Ground-based snow and soil measurements: Collected at 20 surveyed sites, including snowpit profiles (temperature, density, specific surface area (SSA)), Snow Micro Penetrometer (SMP) profiles, and MagnaProbe snow depth measurements. Soil temperature, moisture, and permittivity were continuously measured at static sites.
  • ArcticDEM: Digital Elevation Model used for context.
  • Vegetation classification map (Grünberg and Boike, 2019).

Main Results

• The MCMC Bayesian model, when initialized with a priori snow conditions from the Arctic version of the SVS-2 land surface model, coupled with the SMRT model, achieved a SWE retrieval RMSE of 15.8 mm (16.4 %) and a retrieved SWE uncertainty (quartile deviation) of 23.4 mm (25.2 %). This was achieved by increasing the uncertainty on the a priori grain size estimation and incorporating four radar observations from different incidence angles.
• Using priors from the Arctic SVS-2 version generally yielded better SWE retrieval accuracy (RMSE of 20.9 mm) compared to the default SVS-2 version (RMSE of 27.6 mm) when using all 120 ensemble members.
• Increasing the prior uncertainty for Specific Surface Area (SSA), a parameter known to be underestimated by SVS-2, improved SWE retrieval accuracy (RMSE of 18.7 mm).
• The most significant improvement in SWE accuracy (RMSE of 15.8 mm) was observed when including four σ0 observations from different incidence angles, which are sensitive to various scattering mechanisms within the snowpack.
• While the SVS-2 model accurately reproduces bulk SWE over TVC, it struggles with representing snow stratigraphy, particularly underestimating SSA and the density of the rounded grain (R) layer.
• Applying inter-layer constraints on snow properties (e.g., density of R layer > depth hoar (DH) layer, thickness of R layer < DH layer, SSA of R layer > DH layer) was crucial for obtaining realistic vertical snow profiles, even if similar bulk SWE accuracy could be achieved without them.

Contributions

• This study presents a validated workflow for retrieving SWE from Ku-band SAR measurements using a Bayesian MCMC approach, which is crucial for the development of the Canadian Terrestrial Snow Mass Mission (TSMM).
• It demonstrates the feasibility of using dynamic, spatially and temporally evolving prior distributions from a land surface model (SVS-2) to constrain the MCMC retrieval, enhancing operational efficiency and accuracy for future satellite missions.
• The research highlights the importance of incorporating multiple radar observations (e.g., different incidence angles, dual-frequency, dual-polarization) to improve SWE retrieval accuracy and constrain snow microstructure parameters, thereby validating the TSMM mission concept.
• It provides insights into the limitations of current snow physical models (like Crocus within SVS-2) for Arctic snowpacks and suggests pathways for mutual improvement between land surface models and radar measurements through data assimilation.
• The study emphasizes the necessity of applying physical constraints on snow properties between layers within the MCMC framework to ensure retrieved snow profiles are physically realistic, which is vital for hydrological applications and numerical prediction systems.

Funding

• Environment and Climate Change Canada
• Canadian Space Agency
• NASA’s THP and ESTO-IIP programs (grant nos. 80NSSC20K1592, 80NSSC22K0279)

Citation

@article{Montpetit2025Snow,
  author = {Montpetit, Benoît and Meloche, Julien and Vionnet, Vincent and Derksen, Chris and Woolley, Georgina and Leroux, Nicolas R. and Siqueira, Paul and Adam, J. and Brady, Mike},
  title = {Snow Water Equivalent from airborne Ku-band data: the Trail Valley Creek 2018/19 snow experiment},
  journal = {˜The œcryosphere},
  year = {2025},
  doi = {10.5194/tc-19-5465-2025},
  url = {https://doi.org/10.5194/tc-19-5465-2025}
}