Sentinel-2 snow cover extent and albedo mapping

Sentinel-2's shortwave infrared bands separate snow from cloud where visible light cannot, enabling 10–20 m snow-cover maps and surface albedo estimates critical for hydrology, climate and resort operations.

Sensors

Sentinel-2 MSI (MultiSpectral Instrument): 13 spectral bands from 443 nm to 2190 nm. Snow-relevant bands: B3 (green, 10 m), B4 (red, 10 m), B8 (NIR, 10 m), B11 and B12 (SWIR at 1610 nm and 2190 nm, 20 m). Twin-satellite constellation (2A and 2B) gives 5-day repeat at the equator, roughly 2–3 days at mid-latitudes above 50°.
MODIS (Terra and Aqua): 500 m resolution snow product (MOD10A1 / MYD10A1) at daily revisit. Coarser than Sentinel-2 but provides the long heritage baseline (2000 to present) against which Sentinel-2 anomalies are contextualised. NDSI computed from bands 4 (green) and 6 (SWIR).
VIIRS I-band (Suomi NPP and NOAA-20): 375 m resolution, daily global coverage. VIIRS Snow Cover (VNP10A1) fills temporal gaps between Sentinel-2 acquisitions and extends coverage to polar night edges using the day-night band as a secondary input. Useful for basin-scale runoff forecasting when Sentinel-2 is cloud-obscured.
Sentinel-2 red-edge bands (B5, B6, B7): 20 m resolution at 705–783 nm. Not used for NDSI directly, but contribute to atmospheric correction quality and help distinguish wet snow from saturated vegetation at the snowline, where spectral confusion is highest.

Why shortwave infrared is the deciding vote

Snow is highly reflective in visible wavelengths, peaking near 90 % albedo for fresh dry snow. Cloud is also highly reflective in visible wavelengths. In the shortwave infrared, however, snow absorbs strongly: reflectance at 1610 nm drops to below 10 % for most snow types, while water-droplet cloud remains bright. That contrast is the physical basis of the Normalised Difference Snow Index (NDSI), computed as (Green − SWIR) / (Green + SWIR). Values above 0.4 are conventionally classified as snow-covered; cloud pixels cluster near zero or negative.

Sentinel-2 delivers NDSI at an effective spatial resolution governed by its 20 m SWIR bands (B11 at 1610 nm, B12 at 2190 nm), even though the green band (B3) is natively 10 m. The distinction matters operationally: a 20 m pixel can resolve individual ski runs, small glacial lobes and narrow valley-floor snowpack that a 500 m MODIS pixel smears into a fractional estimate. The trade-off is coverage rate. Sentinel-2 images a 290 km swath; MODIS covers the globe daily. For a river basin under persistent cloud, MODIS or VIIRS may be the only source for days at a time.

Wet snow, dry snow and the ambiguity no index fully resolves

Dry snow and wet snow have different albedos and very different hydrological implications. Dry snow stores water; wet snow is actively melting. B12 at 2190 nm is more sensitive to liquid water content than B11 at 1610 nm because the ice absorption feature deepens at longer wavelengths. A pixel with high B11 reflectance but depressed B12 reflectance is a reasonable indicator of wet or metamorphosed snow, though the signal is subtle and requires careful atmospheric correction to be trusted.

Thin snow over bright soils is the other persistent confusion. Arid soils, salt flats and light-coloured limestone all produce moderate NDSI values. A threshold of 0.4 on NDSI alone will misclassify some bright-soil pixels as snow, particularly in semi-arid regions at the start and end of the snow season. Combining NDSI with a NIR reflectance threshold (B8 > 0.11 is a common filter in the literature) reduces false positives, but does not eliminate them. Honest snow-cover products carry a confusion probability layer, not just a binary mask.

From reflectance to albedo: the correction chain that matters

Surface albedo is not the same as top-of-atmosphere reflectance. Getting from one to the other requires atmospheric correction (removing aerosol scattering and gaseous absorption) and, in mountain terrain, topographic correction (adjusting for slope, aspect and shadowing). ESA's Sen2Cor processor converts Sentinel-2 Level-1C top-of-atmosphere data to Level-2A bottom-of-atmosphere reflectance. That is a necessary step, not a sufficient one for albedo.

Narrowband-to-broadband conversion then translates individual MSI band reflectances into a single broadband shortwave albedo figure, using empirical coefficients derived from field campaigns and radiative transfer modelling. Published coefficients exist for snow surfaces specifically, accounting for the fact that snow grain size and impurity content shift the spectral shape. The resulting broadband albedo feeds directly into energy-balance models: a 1 % albedo change over a large snowfield alters the absorbed solar flux by roughly 3–4 W/m² under typical mid-latitude insolation, which is climatologically significant. Topographic correction in steep alpine terrain can shift apparent reflectance by 20–30 % on shadowed slopes, so omitting it produces systematically biased albedo on north-facing aspects.

The 5-day constraint and how heritage sensors fill the gap

At the equator, the Sentinel-2A/2B pair achieves a 5-day repeat. At 45°N (central Alps, Rockies, Hindu Kush), orbit geometry compresses this to roughly 2–3 days under clear sky. Under cloud, which is the dominant condition over mountain ranges in winter, the effective revisit for usable optical data can stretch to weeks. This is not a flaw unique to Sentinel-2; it is a physics constraint on passive optical sensors.

Operational snow services address this by fusing Sentinel-2 with MODIS or VIIRS on days when no Sentinel-2 clear acquisition exists. The Copernicus Global Land Service publishes a daily 500 m fractional snow cover product built on this multi-sensor logic. For hydrological forecasting, a daily fractional estimate at 500 m is often more useful than a fortnightly binary map at 20 m. The right choice depends on basin size and forecast lead time, not on which sensor is nominally superior.

Applications where the resolution and physics actually matter

Hydrological forecasting is the highest-stakes application. Snow water equivalent (SWE) cannot be measured directly by optical sensors, but snow cover extent combined with a snowpack model constrains SWE estimates substantially. Basins where meltwater feeds irrigation or hydropower generation need accurate extent maps at the start of the melt season. Errors of 5–10 % in snow-covered area translate directly into forecast uncertainty for reservoir inflow.

Ski-resort management is a more immediate commercial use. At 20 m resolution, Sentinel-2 can distinguish open pistes from forested areas and track natural snow cover day by day during marginal conditions. Resorts use this to calibrate snowmaking decisions and communicate conditions to visitors. The data is free and open; the value is in the processing pipeline and interpretation.

Climate monitoring requires long, consistent time series. Sentinel-2 archive depth runs from 2015, which is short for climatology. MODIS extends to 2000, Landsat to 1972 for selected scenes. A credible climate snow-cover product stitches these together with careful inter-calibration. Satellize applies this multi-source approach in its analytics work, including the spectral consistency checks developed for the Tonga crop-estimation programme, where distinguishing bright soil from vegetated surface under variable illumination required the same atmospheric correction discipline that snow mapping demands.

What the data cannot tell you, and what to do about it

Optical snow mapping stops at the cloud base. It cannot see through forest canopy, which can mask 20–40 % of actual snow-covered area in boreal and sub-alpine zones. It measures surface conditions only: a thin crust of snow over bare ground looks identical to deep snowpack in spectral terms. And it produces no direct SWE figure without an ancillary model.

SAR sensors (covered separately in the SAR sea ice and soil moisture pages in this library) penetrate cloud and provide some sensitivity to snowpack structure, but C-band SAR has limited depth penetration in dry snow and its own wet-snow ambiguities. The practical answer for high-value applications is multi-sensor fusion: Sentinel-2 for spatial detail when skies are clear, VIIRS for daily continuity, and in-situ snow depth gauges to anchor the model. No single source is sufficient.

Typical figures

Primary spatial resolution (snow mask)	20 m (governed by SWIR bands B11/B12)
Revisit period (mid-latitudes, 2-satellite)	2–3 days under clear sky; effective optical revisit under cloud cover can exceed 2 weeks
NDSI spectral bands	B3 green (560 nm) and B11 SWIR (1610 nm); NDSI threshold ≥ 0.4 for snow classification
Albedo retrieval accuracy	±5–10 % broadband shortwave albedo after atmospheric and topographic correction, per published validation studies on alpine terrain
Heritage baseline (MODIS)	500 m daily, 2000 to present; MOD10A1 / MYD10A1 products
VIIRS gap-fill resolution	375 m I-band, daily global; VNP10A1 snow cover product
Minimum detectable snow patch	Approximately one 20 m pixel (~400 m²) under clear sky; sub-pixel fractional cover estimated at 500 m via MODIS
Sentinel-2 archive depth	2015 to present (systematic global coverage from 2017 with launch of 2B)
Latency (ESA Level-2A product)	Typically 3–5 hours after acquisition for near-real-time processing via Copernicus Data Space
Delivery formats	GeoTIFF (snow mask, NDSI raster, albedo raster), NetCDF for time-series stacks, vector polygons for snow-line elevation

Analytics Satellize can run

Binary snow cover mask	NDSI thresholding (≥ 0.4) with NIR reflectance filter (B8 > 0.11) on Sentinel-2 Level-2A; cloud masking via SCL layer	GeoTIFF raster at 20 m, updated per clear acquisition; cumulative seasonal extent polygon layer
Fractional snow cover (daily)	Multi-sensor fusion of Sentinel-2 and VIIRS/MODIS using linear spectral unmixing; gap-filled by temporal interpolation on cloud days	Daily 250–500 m fractional snow cover grid in NetCDF; basin-average time series as CSV
Broadband surface albedo	Narrowband-to-broadband conversion from atmospherically corrected Sentinel-2 reflectance using published snow-specific coefficients; topographic correction via DEM-derived illumination model	GeoTIFF albedo raster per acquisition; monthly mean albedo anomaly map relative to MODIS baseline
Wet versus dry snow classification	Dual-SWIR index using B11/B12 ratio to infer liquid water content; combined with NDSI for snow/no-snow discrimination	Classified raster (dry snow / wet snow / mixed / no snow) at 20 m; alert trigger when wet-snow fraction in a basin exceeds defined threshold
Snow-line elevation	Intersection of snow mask with a 30 m Copernicus DEM; median elevation of the snow boundary per aspect quadrant	Daily snow-line elevation value per basin or sub-basin, delivered as structured JSON feed or tabular report
Seasonal snow cover anomaly report	Comparison of current-season cumulative extent against MODIS 20-year climatology; z-score anomaly per month	PDF seasonal report with maps and basin-level statistics; suitable for water authority or reinsurance briefing
Ski-resort natural snow monitoring	High-frequency Sentinel-2 tasking over resort polygon; NDSI and albedo tracked per piste segment	Weekly GIS layer of snow presence by piste; dashboard-ready JSON for integration with resort operations systems

Sources

Who does the work

We can get this done for you. Satellize runs its own analyst desk and a strong science team. You do not buy a data feed and work out what it means; our people source the imagery, run the analysis described on this page, and hand you the answer with its confidence limits stated. Discuss this requirement.