Multi-year versus first-year sea-ice classification

Multi-year ice and first-year ice look similar to the eye but behave very differently under a microwave sensor. Distinguishing the two, at basin scale and weekly cadence, is the sharpest available measure of whether the Arctic is losing its structural resilience.

Sensors

AMSR2 on GCOM-W1 (JAXA): Passive microwave radiometer measuring brightness temperatures at 6.9 to 89 GHz across multiple polarisations. Spatial resolution ranges from roughly 3 km (89 GHz) to 62 km (6.9 GHz) at the surface. The polarisation ratio and gradient ratio at 18.7 and 36.5 GHz are the primary inputs to NASA Team and Bootstrap ice-type algorithms. Daily global coverage, archive from 2012.
Sentinel-1 A/B C-band SAR (ESA): Synthetic aperture radar at 5.4 GHz, imaging in IW (Interferometric Wide Swath, 250 km swath, 5 × 20 m resolution) and EW (Extra Wide, 400 km swath, 25 × 100 m) modes. Multi-year ice returns higher HH-polarised backscatter than first-year ice at C-band due to volume scattering from air bubbles and brine-free crystal structure. Texture metrics (GLCM entropy, homogeneity) over 100–500 m windows sharpen classification boundaries. Revisit 6–12 days at the poles with both satellites combined.
SSMIS on DMSP (NOAA/USAF): Passive microwave successor to SSM/I, operating at 19 to 91 GHz. Provides continuity with multi-decade brightness-temperature records back to 1978, essential for trend analysis. Spatial resolution 13–70 km depending on channel. Still operational on DMSP F-17 and F-18, though instrument degradation on some channels requires careful intercalibration.
RADARSAT-2 (MDA/CSA): C-band SAR with selectable polarisation modes including full quad-pol. Quad-pol decompositions (Pauli, Freeman-Durden) provide additional discrimination between volume and surface scatterers, helping separate multi-year ice from deformed first-year ice, which can mimic multi-year backscatter signatures. Spatial resolution 3–100 m depending on mode; tasking required, so coverage is not systematic.

Why brine drainage is the whole story

Sea ice begins life as frazil crystals that trap brine in small pockets as they consolidate. First-year ice (FYI) is ice that has not yet survived a full melt season; it retains a relatively high brine volume, which gives it a high dielectric constant and makes it nearly opaque to microwave energy. Multi-year ice (MYI) has survived at least one summer. During that melt season, brine drains downward under gravity, leaving behind a lattice of air-filled voids. The dielectric constant drops sharply, and the ice becomes a volume scatterer rather than a surface one.

That physical difference is what remote sensing exploits. At passive microwave frequencies, FYI emits more efficiently than MYI, so MYI appears colder in brightness temperature despite being the same physical temperature. At C-band radar frequencies, MYI backscatter is typically 2–5 dB higher than FYI in winter conditions, because the air-bubble lattice scatters energy back toward the sensor. These are not subtle signals; they are the basis of operational ice-type products that have run continuously since the late 1970s.

What the passive microwave record actually shows

The SSMIS and AMSR2 time series, stitched together with SSM/I data going back to 1978, constitute the longest continuous satellite-derived climate record in the cryosphere. The NASA Team algorithm uses polarisation ratios at 19 and 37 GHz alongside a spectral gradient ratio to separate open water, FYI and MYI. The Bootstrap algorithm takes a different route, fitting brightness temperatures to tie-points in a two-dimensional polarisation space. Both approaches agree well in mid-winter but diverge near the ice edge and during the melt onset period, when surface wetness confounds the emissivity signal.

AMSR2's 89 GHz channel, with its roughly 3 km effective resolution, allows finer spatial detail than the lower-frequency channels used for ice-type retrieval, but 89 GHz is strongly affected by atmospheric water vapour and cannot be used for ice-type classification directly. Operational products from JAXA and the National Snow and Ice Data Center (NSIDC) blend channels to balance resolution against atmospheric contamination. Buyers of these products should understand that the quoted 12.5 km grid spacing on AMSR2 ice-type maps does not reflect independent spatial information at that scale; the underlying retrieval is smoother.

Where SAR earns its place

Passive microwave gives daily, basin-wide coverage but cannot resolve individual floes smaller than several kilometres. Sentinel-1 in Extra Wide mode closes part of that gap: 25 × 100 m pixels across a 400 km swath, all-weather, day and night. The standard classification approach applies grey-level co-occurrence matrix (GLCM) texture features, computed over sliding windows of 100–500 m, to separate the rough, heterogeneous surface of MYI from the smoother, more homogeneous surface of FYI. Random forest and convolutional neural network classifiers trained on ice-chart labels from national ice services (Norwegian, Canadian, Danish) now achieve reported overall accuracies of 80–90 % in winter, with confusion rising sharply in spring.

The persistent ambiguity is deformed FYI. Pressure ridging and rafting can roughen first-year ice until its SAR texture mimics MYI. Polarimetric data from RADARSAT-2 in fine quad-pol mode helps here: volume-scattering decomposition components isolate the brine-free bubble structure of true MYI from the geometrically rough but brine-rich surface of deformed FYI. The limitation is coverage. Quad-pol RADARSAT-2 scenes are narrow (25 km swath in fine mode) and require tasking, so they are most useful for targeted validation rather than basin-wide mapping.

Fusion of passive microwave and SAR is now the operational standard. The microwave product supplies the basin-wide prior; SAR refines boundaries and resolves features below the microwave resolution floor. NSIDC and several national ice services publish blended products at 1–5 km grid spacing.

Honest limits of the method

Three failure modes matter most. First, the melt season. From roughly May to September in the Arctic, surface meltwater pools on ice and saturates the snowpack, driving emissivity toward that of open water regardless of ice age. Both passive microwave and C-band SAR ice-type retrievals become unreliable during this period. L-band SAR (ALOS-2 PALSAR-2) penetrates wet snow better and retains some discriminating power, but coverage is not systematic. Second, new ice in leads and polynyas can be misclassified as MYI by some algorithms because nilas and young ice have distinctive emissivity signatures that sit outside the simple two-type model. Third, the tie-points used in passive microwave algorithms are seasonally and regionally variable; algorithms calibrated on Arctic conditions perform poorly applied to Antarctic sea ice without retraining.

Archive depth is a genuine asset but also a source of error. Intercalibrating brightness temperatures across SSM/I, SSMIS and AMSR2 introduces uncertainties of 1–3 K that propagate into ice-type fraction estimates. Long-term trend studies should use intercalibrated datasets such as the ESA CCI Sea Ice products rather than raw sensor records.

From classification to resilience indicator

The ratio of MYI to total ice area is used as a proxy for Arctic sea-ice resilience because MYI is thicker, stronger and more likely to survive a subsequent melt season than FYI. In the late 1980s, MYI accounted for roughly 60–70 % of Arctic winter ice area by some estimates from passive microwave records. By the 2010s that fraction had dropped to below 30 % in some years, a shift with direct consequences for shipping-route accessibility, polar bear habitat and the albedo feedback that amplifies Arctic warming. The classification product is therefore not merely a scientific curiosity; it is an operational input to ice-load models used in offshore engineering and to seasonal forecasts used in maritime insurance.

Satellize runs ice-type classification pipelines on open Sentinel-1 and AMSR2 archives, producing gridded MYI fraction layers and anomaly alerts calibrated against the ESA CCI baseline. For clients with specific area-of-interest requirements, RADARSAT-2 tasking can be added on licence. The methodology follows published operational schemes from national ice services rather than proprietary black-box models, which matters when a classification product needs to be defensible in a regulatory or insurance context.

What a buyer should ask for

A credible ice-type product should state its algorithm lineage, its training data source, the season and region for which accuracy figures apply, and how it handles the melt-onset transition. Any product claiming better than 85 % accuracy in spring or in the marginal ice zone should be treated with scepticism until the validation methodology is clear.

Useful deliverables include: weekly gridded MYI fraction at 1–5 km resolution for a defined region of interest; anomaly maps showing deviation from the 2003–2023 AMSR2 climatology; and event alerts when MYI fraction in a routing corridor drops below a defined threshold. The last of these is the form most directly useful to an operator planning ice-class vessel transits or subsea infrastructure surveys.

Typical figures

Passive microwave spatial resolution (AMSR2)	3 km at 89 GHz; 12–62 km at ice-type retrieval frequencies (18.7–36.5 GHz)
SAR spatial resolution (Sentinel-1 EW mode)	25 × 100 m ground range detected; texture features computed at 100–500 m windows
SAR spatial resolution (RADARSAT-2 fine quad-pol)	~8 m, 25 km swath; tasked acquisition
Revisit cadence	AMSR2: daily global; Sentinel-1 A+B combined: 6–12 days over Arctic; RADARSAT-2: on demand
Frequency bands used	Passive microwave 6.9–89 GHz (AMSR2/SSMIS); C-band SAR 5.4 GHz (Sentinel-1, RADARSAT-2)
Classification accuracy (mid-winter, open Arctic)	80–90 % reported for SAR-based schemes; lower in spring and marginal ice zone
Archive depth	Passive microwave: 1978–present (SSM/I, SSMIS, AMSR2); Sentinel-1: 2014–present; AMSR2: 2012–present
Seasonal reliability	Most reliable October–April; melt-season retrievals (May–September) unreliable at C-band and passive microwave
Typical gridded product resolution	1–12.5 km for blended passive microwave/SAR operational products
Delivery formats	NetCDF (gridded fraction), GeoTIFF (classified raster), GeoJSON (anomaly polygons), scheduled API feed

Analytics Satellize can run

Weekly MYI fraction map	Passive microwave polarisation ratio (NASA Team / Bootstrap) fused with Sentinel-1 GLCM texture classification	Gridded NetCDF or GeoTIFF at 1–5 km resolution, covering user-defined Arctic sub-region
MYI anomaly alert	Deviation from AMSR2 2003–2023 climatological baseline, thresholded by user-defined percentage points	Automated alert (email or API webhook) with anomaly polygon in GeoJSON
Ice-type boundary delineation	Sentinel-1 EW SAR segmentation with random forest classifier trained on national ice-service chart labels	Vector polygon layer (GeoPackage or Shapefile) of MYI, FYI and open-water zones
Deformed FYI vs MYI disambiguation report	RADARSAT-2 quad-pol Freeman-Durden or Pauli decomposition, volume-scattering component analysis	Scene-level PDF report with classified image and confusion assessment, for targeted validation areas
Multi-decadal MYI fraction trend series	Intercalibrated SSM/I, SSMIS and AMSR2 brightness temperatures processed through Bootstrap algorithm with ESA CCI tie-points	Time-series CSV and chart, 1979–present, for defined region of interest
Routing-corridor ice-type risk score	MYI fraction and concentration thresholds applied along defined vessel route waypoints	Weekly tabular risk score per corridor segment, suitable for integration into voyage planning 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.