Machine learning tools for predicting snow depth from remote sensing data.
deep-snow provides models and workflows to estimate snow depth from Sentinel-1, Sentinel-2, SNODAS, terrain, and forest-cover inputs.
The focus is on a simple, reproducible pipeline that can be run locally or via GitHub Actions. For model details and evaluation, see Brencher et al., 2026.
There are two main ways to use deep-snow. Use GitHub Actions for quick runs or large batch jobs with minimal setup. Use a local install if you want to develop, debug, or build custom workflows with the CLI or Python API.
Fork the repo and run workflows from the Actions tab:
batch_predict_sd: generate one snow-depth map for a target date over an area of interestbatch_sd_timeseries: generate a time series of snow-depth maps over a date range
See docs/github-actions.md for details.
git clone https://github.com/geo-smart/deep-snow.git
cd deep-snow
conda install mamba -n base -c conda-forge
mamba env create -f environment.yml
conda activate deep-snow
pip install -e .Make a prediction using the CLI:
deep-snow predict-sd 20240320 20230910 "-108.20 37.55 -108.00 37.75" 25Add --predict-swe True if you also want Hill-model SWE and density outputs alongside depth.
Or make a prediction using the Python API:
from deep_snow import predict_sd
aoi = {
"minlon": -108.20,
"minlat": 37.55,
"maxlon": -108.00,
"maxlat": 37.75,
}
ds = predict_sd(
aoi=aoi,
target_date="20240320",
snowoff_date="20230910",
out_dir="data/application",
predict_swe=True,
)See docs/local-prediction.md for details.
- docs/scientific-context.md: data sources, model formulation, resolution, validation domain, and limitations
- docs/github-actions.md: running batch workflows in GitHub
- docs/local-prediction.md: local CLI and Python usage
The model uses the following as inputs:
- Sentinel-1 RTC backscatter data (snow-on and snow-off)
- Sentinel-2 imagery (snow-on)
- SNODAS snow depth
- Fractional forest cover
- COP30 digital elevation model
Sentinel-1 and Sentinel-2 inputs are selected close in time to the target date. Inputs are co-registered to a common grid and assembled into model-ready datasets. Airborne Snow Observatory lidar snow depth maps are used for training and evaluation, but not for inference.
Contributions are welcome!
- Quinn Brencher, gbrench@uw.edu
- Eric Gagliano, egagli@uw.edu
2023 GeoSMART Hackweek team:
- Bareera Mirza
- Ibrahim Alabi
- Dawn URycki
- Taylor Ganz
- Mansa Krishna
- Taryn Black
- Will Rosenbluth
- Yen-Yi Wu
- Fadji Maina
- Hui Gao
- Jacky Chen Xu
- Nicki Shobert
- Kathrine Udell-Lopez
- Abner Bogan (Helper)
2024 NASA Earth Sciences and UW Hackweek team:
- Ekaterina (Katya) Bashkova
- Manda Chasteen
- Sarah Kilpatrick
- Isabella Chittumuri
- Kavita Mitkari
- Shashank Bhushan (Helper)
- Adrian Marziliano (Helper)
George Brencher, Eric Gagliano, Taylor Ganz, Dawn URycki, Taryn Black, Mansa Krishna, Manda Chasteen, Isabella Chittumuri, Zachary Hoppinen, wrosenbluth, Yen-Yi Wu, nshobert, kudelllopez, Ibrahim O Alabi, fadjimaina, Shashank Bhushan, Hui, Handsome Jacky Chen, Bareera Mirza, … Abner Bogan. (2026). geo-smart/deep-snow: v0.1.0 (v0.1.0). Zenodo. https://doi.org/10.5281/zenodo.18968780
- Background notebook
- Spicy-snow tutorial background
- Spicy-snow paper
- Lievens et al. (2022)
- What is SAR? (ASF)
- What is SAR? (NASA Earthdata)
- Sentinel-1 SAR user guide
- ASF HyP3 RTC product guide
This project draws on code, ideas, and inspiration from:
- https://github.com/SnowEx/spicy-snow
- https://github.com/relativeorbit/fufiters
- https://github.com/spestana/goes-ortho-builder
- https://github.com/uwescience/GitHubActionsTutorial-USRSE24
