Sensors
- AMSR2 (JAXA/GCOM-W): Passive microwave radiometer at 6.9 to 89 GHz. Retrieves snow water equivalent (SWE) at roughly 5–25 km grid resolution depending on frequency channel. Daily global coverage. Retrievals saturate above approximately 150 mm SWE and lose accuracy when the snowpack is wet or refrozen in layers.
- Sentinel-2 (ESA): 13-band multispectral imager at 10–20 m resolution, 5-day revisit at mid-latitudes with both satellites. The normalised difference snow index (NDSI) using green (Band 3) and SWIR (Band 11) distinguishes snow cover from cloud and bare roof with high spatial fidelity, though cloud cover is a persistent obstacle at high latitudes in winter.
- MODIS Terra/Aqua (NASA): Daily global snow-cover product (MOD10A1/MYD10A1) at 500 m resolution using NDSI. Useful for tracking snow-cover duration and onset dates across a network footprint, though too coarse for individual rooftop assessment. Combined Terra and Aqua passes reduce cloud-gap probability.
- Landsat 8/9 (USGS/NASA): OLI sensor at 30 m resolution, 16-day revisit per satellite (8-day combined). SWIR bands support NDSI computation and can detect partial snow clearance on larger rooftop surfaces. Archive depth back to 1972 (Landsat 1) supports multi-decade seasonal climatology for design-load benchmarking.
What a passive microwave radiometer actually measures
AMSR2 does not photograph snow. It listens to microwave emission from the Earth's surface across frequencies from 6.9 to 89 GHz. Dry snow scatters upwelling microwave radiation from the soil beneath it, depressing the brightness temperature measured at the satellite. The magnitude of that depression, particularly the difference between 19 GHz and 37 GHz channels, correlates with snow water equivalent. JAXA publishes daily global SWE grids derived from this principle.
The physics imposes hard limits. Wet snow, which occurs during melt events or rain-on-snow, absorbs rather than scatters microwaves, making the snowpack appear shallow or absent. Deep, dense snowpacks scatter so strongly that the signal saturates: retrievals above roughly 150 mm SWE become unreliable. For telecoms engineers, this matters most in alpine or subarctic deployments where the heaviest structural loads occur precisely where the retrieval is least accurate. That is not a reason to discard the method; it is a reason to pair it with optical data and local climatological priors.
Optical change detection fills the gaps passive microwave cannot
Sentinel-2's NDSI is computed as (Green minus SWIR) / (Green plus SWIR). Values above approximately 0.4 reliably indicate snow cover. At 10–20 m resolution, the sensor can distinguish a snow-covered rooftop from an adjacent cleared one, and track the spatial extent of accumulation across a city block. That is far more actionable than a 25 km passive microwave grid cell that averages over hundreds of buildings.
The limitation is cloud cover, which is not trivial at high latitudes in winter. A Sentinel-2 pass over Helsinki or Anchorage in January may be cloud-obscured for days at a time. MODIS's twice-daily cadence helps maintain temporal continuity at 500 m, and multi-day compositing using the MODIS gap-filled snow product (MOD10A1) can recover coverage extent even through persistent cloud. The practical workflow fuses AMSR2 SWE estimates for load magnitude with Sentinel-2 NDSI for spatial distribution, using MODIS to bridge cloud gaps.
From snow depth to structural moment: the engineering translation
Snow water equivalent in millimetres converts directly to load in kilograms per square metre: 1 mm SWE equals 1 kg/m². A 200 mm SWE event, which is not unusual in maritime subarctic climates, imposes 200 kg/m² on a flat roof. Eurocode 1 Part 1-3 and equivalent national standards define characteristic ground snow loads by location; the roof load is then modified by shape coefficients that account for pitch, exposure and thermal properties of the roof.
Antenna masts and small-cell cabinets introduce a different failure mode. Snow accumulating on a horizontal antenna element or dish acts as an eccentric load, generating a bending moment at the mast base or wall bracket. The moment scales with both the mass of accumulated snow and its distance from the attachment point. A 0.6 m dish with 20 kg of wet snow at its centre creates a substantial overturning moment on a bracket designed for wind load, not vertical eccentric loading. Satellite data cannot resolve individual dishes, but it can flag the site as being in a high-accumulation zone during a period when the structural design threshold is likely exceeded.
Building a risk flag: combining SWE, roof geometry and asset inventory
The analytical pipeline has three inputs: a SWE estimate from AMSR2 (or a downscaled product blended with Sentinel-2 snow extent), a building footprint and roof-pitch layer from a national cadastre or photogrammetric model, and an asset inventory of rooftop telecoms installations. Where the asset inventory does not exist, rooftop obstruction mapping from high-resolution optical imagery can approximate it, though that is covered separately in the sibling page on rooftop obstruction mapping.
Roof pitch matters because shallow-pitched roofs retain snow; steep roofs shed it. The shape coefficient in Eurocode 1-3 drops from 0.8 on a flat roof to near zero above a 60-degree pitch. A flat-roofed urban building in a high-SWE grid cell is a structurally different risk from a steeply pitched rural building in the same cell. The risk flag therefore combines SWE magnitude, pitch-adjusted load, and whether the estimated load exceeds a threshold fraction of the design load for the structure type. Thresholds are set by the operator; the satellite layer provides the spatial and temporal SWE field.
Revisit cadence determines how much warning is practical. AMSR2 updates daily. Sentinel-2 revisits every five days at best. For a rapidly developing storm event, the AMSR2 daily product is the primary operational input, with Sentinel-2 providing post-event spatial verification once cloud clears.
Honest limits and what they mean for operational use
This method is a screening tool, not a structural survey. AMSR2 SWE at 25 km resolution cannot distinguish one building from its neighbour. Sentinel-2 at 10 m can map snow extent on rooftops but cannot measure depth directly. Neither sensor sees through cloud in real time. In wet-snow conditions, which are common during the most damaging loading events, passive microwave retrievals may underestimate SWE by 30–50% or more according to published validation studies.
The method is most defensible as a portfolio-level risk prioritisation tool: identifying which sites in a network of hundreds or thousands are in elevated-risk zones during a given accumulation event, so that ground inspection or remote structural monitoring can be directed efficiently. It does not replace a structural engineer's assessment of any individual installation. Satellize applies this kind of satellite-derived screening to network asset portfolios; the Tonga crop-estimation programme is an example of how open-constellation analytics can be operationalised for a specific national context, and the same pipeline logic applies here.
Archive depth and climatological design value extraction
One underused application is retrospective. Landsat's archive extends to 1972 and MODIS to 2000. AMSR-E, AMSR2's predecessor on Aqua, ran from 2002 to 2011. Together these archives support extraction of return-period snow-cover statistics for any location: the 1-in-10 or 1-in-50 year SWE event that should inform the design load for a new antenna installation. Published methods for fitting extreme-value distributions to satellite-derived SWE time series exist in the peer-reviewed literature, and the Copernicus Climate Change Service publishes gridded snow climatologies derived from these datasets.
For a network operator planning a high-latitude small-cell rollout, this is arguably more valuable than real-time monitoring. Knowing that a candidate rooftop in Tromsø has experienced three events exceeding 180 mm SWE in the past 20 years changes the structural specification conversation with the landlord before the installation contract is signed.
Typical figures
| SWE spatial resolution (AMSR2) | 5–25 km depending on frequency channel; 10 km for the standard JAXA Level 3 product |
| Snow-cover spatial resolution (Sentinel-2 NDSI) | 10–20 m (Band 3 at 10 m, Band 11 at 20 m) |
| Revisit cadence | AMSR2: daily global; Sentinel-2: 5 days at mid-latitudes (both satellites); MODIS: 1–2 passes per day |
| SWE detection range (passive microwave) | Approximately 10–150 mm SWE; retrievals unreliable above 150 mm and in wet-snow conditions |
| Key frequency channels (AMSR2) | 6.9, 10.65, 18.7, 23.8, 36.5, 89.0 GHz (dual polarisation) |
| Key spectral bands (Sentinel-2 NDSI) | Band 3 (green, 560 nm) and Band 11 (SWIR, 1610 nm) |
| Archive depth | AMSR2 from 2012; AMSR-E (predecessor) 2002–2011; MODIS from 2000; Landsat from 1972 |
| Latency (operational products) | AMSR2 Level 3 SWE: typically 1–2 days; Sentinel-2 L2A: 3–5 hours after acquisition via Copernicus Data Space |
| Minimum rooftop area detectable (Sentinel-2) | Approximately 100 m² for reliable NDSI classification; sub-pixel mixing affects smaller surfaces |
| Delivery format | GeoTIFF risk-flag rasters, GeoJSON site-level alert feed, CSV portfolio risk table |
Analytics Satellize can run
| Daily SWE load estimate per site | AMSR2 passive microwave SWE retrieval (JAXA algorithm) spatially disaggregated using Sentinel-2 snow-extent fraction within each grid cell | Daily GeoJSON alert feed with estimated kg/m² load per registered rooftop asset, flagged against operator-defined thresholds |
| Pitch-adjusted roof load layer | SWE load multiplied by Eurocode 1-3 shape coefficient derived from roof pitch extracted from building footprint data | GeoTIFF raster and site-level CSV with adjusted design load fraction for each asset |
| Snow-cover onset and duration map | MODIS MOD10A1 daily NDSI compositing with gap-filling; Sentinel-2 NDSI for spatial verification | Seasonal summary GeoTIFF showing first-snow date, peak-cover date and melt-out date per pixel |
| Extreme SWE return-period climatology | Generalised extreme value (GEV) distribution fitted to annual maximum SWE from AMSR2/AMSR-E archive (2002 to present) | Gridded GeoTIFF of 1-in-10 and 1-in-50 year SWE design values for candidate site locations |
| Post-storm accumulation change detection | Sentinel-2 NDSI differencing between pre- and post-storm acquisitions to map new accumulation extent at 10–20 m | Change-detection GeoTIFF and ranked site list by accumulation increase |
| Portfolio risk tier classification | Multi-factor scoring combining SWE percentile rank, roof pitch class, asset type (dish, panel, cabinet) and structural age proxy | Tabular risk-tier report (high/medium/low) with recommended inspection priority order |
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.