Alpine snowpack retreat and high-altitude habitat phenology
The date snow leaves an alpine patch determines when plants green up, insects emerge, and specialist montane species can breed. Satellites track that date across entire mountain ranges, though cloud and terrain shadow make it harder than it sounds.
Sensors
- MODIS MOD10A1 (Terra): Daily 500 m snow-cover product using the Normalised Difference Snow Index (NDSI). Sufficient temporal resolution to detect clearance dates to within a few days, but 500 m pixels blend snow-free patches with residual snow in complex terrain. Archive runs from February 2000, giving more than two decades of phenological trend data.
- VIIRS VNP10A1 (Suomi-NPP / NOAA-20): Daily 375 m NDSI-based snow cover, finer than MODIS and with a slightly later archive start (2012). The two NOAA-20 and Suomi-NPP overpasses per day can be combined to reduce cloud-gap frequency, though persistent orographic cloud still blocks both.
- Sentinel-2 MSI: 10 m to 20 m multispectral imagery with a 5-day revisit at mid-latitudes (combined Sentinel-2A and -2B). Resolves individual snow-free patches too small for MODIS or VIIRS to detect, and supports NDVI-based greenup detection at the same spatial scale. Cloud cover in alpine zones frequently reduces effective revisit to 10 to 20 days or worse.
- Sentinel-1 SAR (C-band): Cloud-penetrating synthetic aperture radar at 5 m to 20 m resolution. Wet snow strongly attenuates C-band backscatter, making it a useful proxy for snow presence when optical sensors are blocked. Dry snow is largely transparent to C-band, so Sentinel-1 is most reliable during the melt season rather than mid-winter accumulation.
Why a date matters more than a depth
Most remote-sensing work on alpine snow focuses on snow-water equivalent or total extent. For biodiversity, the critical variable is simpler and harder to pin down: the calendar date on which a given patch of ground becomes snow-free long enough for plants to begin growing. That date, sometimes called the snow-off date or snow-clearance date, sets the length of the growing season available to alpine plants, invertebrates, and the vertebrates that depend on them. A shift of two weeks earlier or later can determine whether a ptarmigan chick hatches into a food-rich or food-poor environment.
At high altitude, the growing season is already compressed to perhaps eight to twelve weeks at the elevation of permanent snowfields. Even modest interannual variation in clearance timing has outsized ecological consequences. Satellite time series long enough to detect trends, currently around 24 years for MODIS and 12 years for VIIRS, are therefore among the more ecologically informative data products available to conservation planners.
Reading clearance dates from daily snow products
MODIS MOD10A1 and VIIRS VNP10A1 both use NDSI, the ratio of green-band to shortwave-infrared reflectance, to classify each pixel as snow-covered or snow-free each day. The standard threshold of NDSI greater than 0.4 performs well on open terrain but degrades under forest canopy and on north-facing slopes where shadows mimic snow spectral signatures. In practice, analysts apply terrain-corrected versions and often raise the threshold slightly in densely shadowed cirques.
From a daily binary time series, the snow-off date for each pixel is conventionally defined as the first date after peak accumulation on which the pixel remains snow-free for a specified run of consecutive days, commonly five to seven. Requiring a run rather than a single clear day reduces false clearances caused by mid-season melt events followed by re-accumulation. The resulting annual snow-off map can then be differenced across years to quantify trends, or correlated with NDVI greenup dates derived from the same sensors.
Cloud is not a minor nuisance in mountain terrain
Orographic cloud formation means that alpine zones are frequently obscured precisely when snow is melting. In the European Alps, cloud-free MODIS observations during April and May can fall below 30 percent of available days in some grid cells. A naive clearance-date algorithm that simply looks for the first snow-free observation will return systematically late estimates wherever cloud gaps are sparse.
The standard mitigations are compositing, interpolation, and SAR fusion. Temporal compositing, taking the minimum snow-cover fraction over a rolling window, reduces cloud contamination but smooths the very transitions analysts want to detect. Linear or spline interpolation across cloud gaps is widely used and works reasonably well when gaps are short, typically under ten days. For longer gaps, Sentinel-1 C-band backscatter provides a cloud-independent signal: a sustained drop in backscatter relative to a dry-snow or bare-ground baseline indicates wet-snow presence, and a return to baseline suggests melt completion. The limitation is that dry snow and bare frozen ground produce similar C-band returns, so SAR alone cannot distinguish early-season from late-season states without optical anchoring.
Sentinel-2 at 10 m adds spatial detail that MODIS and VIIRS cannot provide, resolving snow-free corridors along ridge crests and south-facing micro-slopes that may be ecologically significant even when surrounding terrain remains snow-covered. But Sentinel-2's five-day revisit, already modest, degrades severely under persistent cloud. It is most useful for characterising spatial heterogeneity in years or windows when cloud is cooperative, rather than as a reliable phenological clock.
From snow-off to greenup: the phenological chain
Snow clearance is the trigger, not the outcome. What conservation managers usually need is the date and spatial extent of vegetation greenup, which follows snow-off by days to a few weeks depending on soil temperature and species composition. NDVI derived from Sentinel-2 or MODIS can track this transition, but NDVI in alpine zones is noisy at low biomass levels. The Enhanced Vegetation Index (EVI), which reduces soil-background contamination, is often preferred for sparse alpine swards.
Linking snow-off dates to greenup dates pixel by pixel produces a lag map. Where the lag is short and consistent, snowmelt directly drives soil warming and plant emergence. Where the lag is variable, other factors, aspect, soil drainage, late frost, are mediating the response. This spatial structure is ecologically meaningful: patches with short, reliable lags are likely to be the most productive foraging areas for early-season specialists and therefore the highest-priority conservation targets.
What the record shows, and what it cannot tell you
The MODIS archive from 2000 onwards documents a general trend towards earlier snow-off dates across many mid-latitude mountain ranges, consistent with published analyses of the Hindu Kush, the Alps, and the Rockies. The magnitude varies considerably by elevation band and aspect. These are real signals in the public record, not modelling artefacts, though attribution to specific climate drivers requires ancillary meteorological data.
What satellite snow products cannot resolve directly: sub-pixel heterogeneity below 375 m (VIIRS) or 500 m (MODIS), the difference between a thin late-season snow layer and a bare bright rock surface in some spectral conditions, and the phenological responses of individual plant species rather than the integrated community signal. Ground-truth from phenological cameras or field surveys remains necessary to validate clearance-date estimates and to interpret what a given greenup NDVI value means for a specific focal species. Satellize incorporates that ground-truth layer when clients supply it, as with the crop-calibration field data used in the Tonga programme.
Designing an operational monitoring system
A practical alpine phenology system combines three layers. A daily MODIS or VIIRS backbone provides the temporal resolution needed to detect clearance events as they happen, with cloud gaps flagged and interpolated. Sentinel-1 SAR acquisitions, typically available every six to twelve days in most mountain regions depending on orbit geometry, fill the longest optical gaps during the melt season. Sentinel-2 acquisitions, processed when cloud-free, anchor the fine-scale spatial characterisation needed to map habitat heterogeneity.
Outputs are most useful when expressed as anomalies relative to a baseline period rather than absolute dates. A map showing that snow-off in a given year arrived fifteen days earlier than the 2000 to 2020 median, at a spatial resolution of 500 m with Sentinel-2 detail overlaid at 10 m in cloud-free areas, gives a ranger or conservation biologist an immediately actionable picture. Alert thresholds can be set for anomalies exceeding one or two standard deviations, triggering field surveys timed to the actual ecological window rather than a fixed calendar date.
Typical figures
| Spatial resolution (snow cover) | 500 m (MODIS MOD10A1); 375 m (VIIRS VNP10A1); 10–20 m (Sentinel-2); 5–20 m (Sentinel-1 SAR) |
| Temporal revisit (optical) | Daily (MODIS, VIIRS); 5 days at mid-latitudes combined Sentinel-2A+2B; effective cloud-free revisit in alpine zones often 10–20 days |
| Temporal revisit (SAR) | 6–12 days typical for Sentinel-1 in mountain regions, depending on orbit and acquisition mode |
| Key spectral bands / frequency | Green (~0.55 µm) and SWIR (~1.6 µm) for NDSI snow detection; Red and NIR for NDVI/EVI greenup; C-band (5.4 GHz) for SAR wet-snow proxy |
| Archive depth | MODIS: February 2000 to present; VIIRS: 2012 to present; Sentinel-2: 2015 to present; Sentinel-1: 2014 to present |
| Snow-off date detection precision | Typically ±3–7 days under good cloud conditions; degrades to ±10–14 days or worse with frequent cloud gaps without gap-filling |
| Minimum detectable snow-free patch | ~1 ha reliably with MODIS/VIIRS; ~0.01 ha with Sentinel-2 in cloud-free conditions |
| Latency (operational products) | MODIS and VIIRS standard products available within 1–2 days of acquisition; Sentinel-2 L2A typically 1–3 days via Copernicus Data Space |
| Delivery formats | GeoTIFF snow-off date rasters, NetCDF time-series stacks, GIS-ready anomaly layers, tabular phenological statistics per habitat zone |
Analytics Satellize can run
| Annual snow-off date map | NDSI threshold classification on MODIS MOD10A1 or VIIRS VNP10A1 daily time series, with consecutive-day run-length criterion and temporal interpolation across cloud gaps | GeoTIFF raster (day-of-year per pixel) per season, stacked multi-year archive |
| Snow-off anomaly map | Pixel-wise deviation from 2000–2020 baseline median, expressed in days; standard deviation envelope computed from full archive | Annual GeoTIFF anomaly layer with per-zone summary statistics in CSV |
| Cloud-gap-filled snow-cover time series | Fusion of Sentinel-1 C-band backscatter change detection with optical NDSI, using SAR to bridge gaps exceeding 7 days during melt season | Daily gap-filled binary snow-cover raster stack in NetCDF, flagged by data source per pixel |
| Vegetation greenup onset map | EVI or NDVI time-series inflection-point detection (logistic curve fitting or piecewise linear regression) applied to Sentinel-2 or MODIS surface-reflectance composites | GeoTIFF greenup day-of-year raster per season, with confidence interval layer |
| Snow-off to greenup lag surface | Pixel-wise differencing of snow-off date and greenup onset rasters; spatial clustering to identify lag-homogeneous habitat patches | GIS polygon layer of lag-classified habitat patches, ranked by ecological priority score |
| Fine-scale snow heterogeneity characterisation | Sentinel-2 NDSI at 20 m, mosaicked from cloud-free acquisitions within a ±14-day window of the MODIS-estimated clearance date | High-resolution snow-free patch map in GeoTIFF, with patch area and connectivity statistics |
| Phenological early-warning alert | Threshold trigger when rolling snow-off anomaly exceeds 1.5 standard deviations from baseline in a defined protected-area boundary | Automated alert report (PDF + GeoJSON boundary) delivered within 48 hours of threshold crossing |
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.