Sensors
- Sentinel-1 SAR (C-band, ESA): Interferometric Wide swath mode delivers 20 m resolution across a 250 km swath with 6-day repeat at mid-latitudes, reducing to near-daily coverage above 70°N where orbital tracks converge. Backscatter contrast between first-year ice, multi-year ice and open water allows ice-type discrimination and floe delineation that passive microwave cannot resolve.
- AMSR2 (JAXA, aboard GCOM-W): Passive microwave radiometer providing daily, all-weather sea-ice concentration at 3.125 km (89 GHz channel) to 25 km (lower-frequency channels). The standard Bootstrap and NASA Team algorithms retrieve concentration basin-wide but carry a known low bias of 5–15 % in summer melt conditions and struggle to separate thin new ice from open water during autumn freeze-up.
- RADARSAT-2 (C-band SAR, MDA): Commercial SAR with selectable modes from 3 m Spotlight to 100 m ScanSAR. Polarimetric quad-pol modes improve ice-type classification beyond what single-pol Sentinel-1 achieves, at the cost of narrower swath and tasking fees. Useful for high-confidence ice-type surveys at specific candidate turbine locations.
- SSMI/S (DMSP, NSIDC archive): The SSMI/S passive microwave record extends back to 1987, giving a 35-plus-year climatology of ice extent and concentration. Essential for establishing the statistical distribution of ice-free window length and inter-annual variability at any proposed site, well before Sentinel-1 or AMSR2 data existed.
What the passive microwave record actually tells you
AMSR2 and the long SSMI/S archive answer the first question any offshore wind developer must ask: how many ice-free days per year can this site expect, and how variable is that number across decades? The Bootstrap algorithm applied to SSMI/S data resolves ice concentration to roughly 5–10 % accuracy under winter conditions, with daily global coverage regardless of cloud or darkness. That combination is irreplaceable for climatology.
The limit is spatial resolution. At 3.125 km even on AMSR2's 89 GHz channel, a pixel the size of a small town is reported as a single concentration value. Individual floes, leads and polynyas are invisible. Worse, during autumn freeze-up the thin grease ice and nilas that form first have emissivity signatures close enough to open water that standard retrieval algorithms systematically underestimate concentration by up to 15 %. For a developer trying to determine when the site transitions from operable to ice-affected, that ambiguity is precisely the period that matters most.
Where SAR resolves the ambiguity passive microwave cannot
C-band SAR backscatter responds to surface roughness and dielectric properties in ways that separate ice types the microwave radiometer conflates. New ice (nilas and grey ice, typically less than 30 cm thick) returns lower backscatter than consolidated first-year ice in calm conditions, and its texture is visibly different from the specular return of open water. Sentinel-1's dual-pol HH/HV mode sharpens that contrast: HV cross-polarisation is particularly sensitive to volume scattering in older, more deformed ice.
At 20 m resolution, Sentinel-1 resolves individual floe boundaries, ridges and leads down to a few hundred metres wide. That matters for two reasons. First, it allows engineers to characterise the floe-size distribution, which feeds directly into ice-load calculations for monopile and jacket foundations. Second, it catches the early stages of freeze-up that the passive microwave misses, giving a more accurate estimate of when the operational window actually closes. Above 70°N, ascending and descending Sentinel-1 passes can combine to near-daily revisit, which is sufficient to track rapid freeze events.
Translating ice maps into an operability window
The operational window for an offshore wind site is not simply the ice-free period. Most installation vessels and crew-transfer vessels have ice-class ratings that allow limited operation in low-concentration or thin-ice conditions. The International Association of Classification Societies defines ice classes from PC1 (unrestricted polar) down to PC7 and the Baltic ice classes; each implies a maximum ice concentration and thickness the vessel can safely work in. Satellite-derived concentration maps, calibrated against SAR ice-type data, can be used to estimate the number of days per year each vessel class could operate at a given site.
The design envelope question is separate. Ice-load standards such as ISO 19906 (Arctic offshore structures) require statistical characterisation of extreme ice conditions: maximum floe thickness, ridge keel depth and drift speed. SAR can constrain floe geometry and surface roughness as a proxy for deformation, but keel depth requires sonar or mooring data. Satellite imagery defines the boundary conditions; it does not replace in-situ measurement for the extreme-load case.
The fusion workflow: neither sensor alone is enough
A practical analytical workflow runs in two stages. The SSMI/S and AMSR2 archive establishes the climatological baseline: mean ice-free window, inter-annual standard deviation, trend over the satellite record, and the probability distribution of early freeze or late melt events. This is the layer that answers whether a site is worth investigating at all.
Sentinel-1 then provides the high-resolution seasonal layer for the years it covers (2014 onwards for Sentinel-1A, 2016 for Sentinel-1B, with Sentinel-1C launched in late 2024 maintaining continuity). Change detection between sequential SAR acquisitions tracks freeze-up progression at 20 m. Ice-type classification using established algorithms such as the Nansen Environmental and Remote Sensing Centre's RTIC product maps first-year versus multi-year ice fraction. The two data streams are reconciled by spatially aggregating the SAR classification to the passive microwave grid and correcting the concentration bias during freeze-up. The result is a merged daily product that is more accurate during the ambiguous transition periods than either input alone.
Cloud cover is not a constraint for either sensor. Both SAR and passive microwave operate at wavelengths that penetrate cloud and Arctic darkness. That is one of the few genuinely straightforward aspects of this problem.
Honest limits and what they mean for project decisions
Several limitations are worth stating plainly. SAR ice-type classification accuracy degrades in summer melt conditions when surface flooding and melt ponds alter backscatter in ways that mimic open water. Published studies using Sentinel-1 report overall classification accuracies of roughly 80–90 % for two-class (ice/water) discrimination in winter, dropping in the melt season. Multi-class ice-type maps carry higher uncertainty still.
RADARSAT-2 quad-pol data improves classification but is not freely available; it requires commercial tasking and adds cost. For a feasibility study, Sentinel-1 is usually sufficient. For a final investment decision at a marginal site, the additional discrimination from quad-pol SAR or from airborne ice radar surveys is likely justified.
Archive depth matters too. The Sentinel-1 record is just over a decade old. For sites where inter-annual variability is high, ten years may not capture the tail of the distribution. The SSMI/S record from 1987 remains the only way to characterise rare but damaging late-season ice incursions. Satellize runs the merged passive microwave and SAR workflow as a standard analytical product, and has applied similar time-series methods to the Tonga crop-estimation programme, where combining sensors of different resolution is equally central to getting the answer right.
From ice map to bankable document
What a lender or insurer wants from this analysis is a probability distribution, not a single number. The deliverable should state, for a named site and a defined vessel ice class, the exceedance probability for the operational window falling below a given threshold: for example, the probability that the ice-free window is shorter than 90 days in any given year. That requires the full SSMI/S climatology, trend-adjusted for recent Arctic change, cross-validated against the Sentinel-1 record where it overlaps.
The ice-load design input is a separate document: the statistical distribution of floe size, concentration and thickness during the shoulder seasons, drawn from SAR classification and, where available, ice-thickness products from CryoSat-2. These two outputs together, the operability window distribution and the ice-load envelope, constitute the satellite contribution to an Arctic offshore wind bankability package. Neither replaces a metocean campaign, but both substantially reduce the uncertainty a campaign must resolve.
Typical figures
| Spatial resolution (SAR, Sentinel-1 IW mode) | 20 m range × 22 m azimuth |
| Spatial resolution (passive microwave, AMSR2 89 GHz) | 3.125 km |
| Spatial resolution (SSMI/S climatology) | 25 km (standard grid) |
| Revisit (Sentinel-1, above 70°N) | Near-daily with combined ascending/descending passes |
| Revisit (AMSR2) | Daily, global |
| Sensor frequency / band | Sentinel-1: C-band 5.405 GHz; AMSR2: 6.9–89 GHz multi-channel; SSMI/S: 19–91 GHz |
| Ice concentration retrieval accuracy (winter) | ±5–10 % for consolidated ice; bias up to 15 % during freeze-up thin-ice conditions |
| SAR ice/water classification accuracy (winter) | ~80–90 % two-class; lower in melt season |
| Archive depth | SSMI/S from 1987; AMSR2 from 2012; Sentinel-1 from 2014 |
| Delivery formats | GeoTIFF concentration grids, GeoJSON floe polygons, NetCDF time-series, PDF operability-window report |
Analytics Satellize can run
| Ice-free window climatology | Bootstrap or NASA Team algorithm applied to SSMI/S archive; trend-adjusted linear regression over the satellite record | Annual exceedance probability table and GIS layer per candidate site |
| Daily merged ice-concentration product | SAR-calibrated passive microwave fusion; bias correction during freeze-up using Sentinel-1 ice/water mask | Daily GeoTIFF concentration grid with uncertainty band, delivered within 24 hours of SAR acquisition |
| Ice-type classification map | Sentinel-1 dual-pol backscatter classification (HH/HV); established NRCS threshold or Random Forest classifier trained on ice-type reference data | Seasonal GeoJSON polygon layer: open water, new ice, first-year ice, multi-year ice |
| Floe-size distribution analysis | Automated floe boundary delineation from Sentinel-1 SAR segmentation; area-weighted histogram | Statistical summary table and floe polygon shapefile for ice-load input to ISO 19906 calculations |
| Freeze-up and break-up date series | Threshold-crossing detection (e.g. 15 % concentration) applied to daily merged product; bootstrapped confidence intervals | Annual date series with inter-quartile range, exportable as CSV or integrated into a project metocean database |
| Vessel operability day count by ice class | Concentration threshold lookup per IACS ice-class rating applied to merged daily product climatology | Annual operability-day distribution per vessel class, formatted for lender technical due-diligence reports |
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.