Satellite and data assimilation¶
Satellite snow cover¶
MODIS (on NASA's Terra satellite) observes the whole globe daily at 500 m. SnowMapper uses the MOD10A1F product: the NDSI snow cover, cloud-gap-filled with the most recent clear observation. From it the board derives the fraction of each terrain unit covered by snow (fSCA). The Model fSCA layer is the model's equivalent, so the two can be compared directly.
| Strength | Limitation |
|---|---|
| independent of the model, daily, complete coverage | sees snow cover, not depth or SWE |
| shows the snow line and its movement well | cloud gap-filling repeats old observations during long cloudy spells |
| forests hide snow on the ground; steep shaded slopes and low sun angles in winter are harder |
Sentinel-2 gives 10–20 m true-colour images every few days, where cloud allows. They are shown for selected areas as Satellite image layers, for visual checks.
Stations¶
Where available, station measurements (temperature, precipitation, snow depth) from national services and global archives (NOAA ISD and GHCN-Daily) are compared with the model in the Val-Lab. They are checked for quality first. A station is compared with the terrain unit it sits in, or with a dedicated point run (see Validation).
Data assimilation (selected regions)¶
Data assimilation (DA) corrects the model with observations. SnowMapper uses two methods:
- Particle Batch Smoother (PBS; Fiddes et al., 2019). An ensemble of snow model runs is made with perturbed precipitation and temperature. Each run is weighted by how well its snow cover matches the MODIS fSCA over the season. The weighted ensemble gives a corrected snowpack. Snow cover constrains when snow appears and melts, and through that the amount.
- Ensemble Kalman filter updates with station snow depth, where real-time station data exist.
Where DA runs, the Map tab offers a DA Corrected switch and the Model-DA fSCA layer.
Snow cover alone cannot pin down the amount
A pack of 1 m and one of 2 m both show 100 % snow cover in mid-winter. MODIS only constrains the amount through the date of melt-out. DA therefore improves timing more reliably than peak SWE. Evaluation against held-out stations shows modest gains in depth.