Thin ice thickness from thermal infrared heat-flux inversion
For ice below roughly 50 cm, surface temperature betrays thickness through conductive heat flux. MODIS, VIIRS and Sentinel-3 SLSTR make the measurement possible; cloud and wind make it hard.
Sensors
- MODIS Terra / Aqua: Bands 31 and 32 (10.78–12.27 µm) at 1 km nadir resolution; combined Terra and Aqua orbits give up to four passes per day at mid-latitudes, more at high latitudes. Brightness-temperature precision is approximately 0.05 K under calibrated conditions, sufficient to resolve thickness differences of a few centimetres in the 0–20 cm range.
- VIIRS (Suomi-NPP / NOAA-20): Thermal bands I5 (11.45–12.26 µm) at 375 m and M15/M16 at 750 m nadir; reduced bow-tie effect relative to MODIS improves swath-edge geometry at high latitudes. The finer I-band resolution allows smaller leads and thin-ice patches to be resolved.
- Sentinel-3 SLSTR: Dual-view (nadir and 55° oblique) thermal channels at 1 km, with split-window bands at 10.85 and 12.0 µm. The dual-view geometry provides an independent atmospheric correction path, reducing one of the dominant error sources in the inversion.
- AVHRR (historical archive): Channels 4 and 5 (10.3–12.5 µm) at approximately 1.1 km; operational since 1978 on NOAA and MetOp platforms. Radiometric calibration varies across the archive but the record is indispensable for multi-decade trend analysis of thin-ice production in polynyas and seasonal ice zones.
Why cold ice tells you its own thickness
The physics is straightforward. Heat conducts through ice at roughly 2.1 W m⁻¹ K⁻¹. When ice is thin, that flux is large, the surface stays relatively warm, and a thermal sensor looking down from orbit sees a surface temperature noticeably higher than the ambient air. As ice thickens beyond about 50 cm, the surface cools to near air temperature and the thermal signal becomes indistinguishable from noise. The method therefore occupies a niche that radar and laser altimetry struggle with: very thin, newly formed ice where freeboard is a few centimetres at most.
The inversion is a one-dimensional thermodynamic balance. Given the observed surface brightness temperature, an estimate of air temperature and wind speed from atmospheric reanalysis (ERA5 is the standard choice), and assumed values for snow cover and ice emissivity, you solve for the ice thickness that closes the energy budget. Published implementations, including the MODIS-based thin-ice thickness product developed by Maykut and others and later refined by Yu and Lindsay, report uncertainties of roughly ±5 cm for ice thinner than 20 cm, growing to ±15 cm near the 50 cm ceiling.
The cloud problem is not a detail, it is the dominant constraint
Thermal infrared cannot see through cloud. In the Arctic and Southern Ocean, cloud cover exceeds 70 % on average in many seasons, and the thin-ice formation events that matter most, new ice forming in leads and coastal polynyas, often occur under synoptic systems that bring cloud. A single clear overpass is useful; a time-series of clear overpasses over the same location is rare.
The practical consequence is that the method works best in the central Arctic winter, when persistent high-pressure systems bring clear skies and the coldest air temperatures that drive rapid ice growth. The Antarctic sea-ice zone is cloudier and the method is harder to apply systematically. Compositing multiple clear overpasses over a few days can build spatial coverage, but introduces temporal smearing in a medium that changes thickness on timescales of hours. Users should treat cloud-gap-filled products with caution.
Wind adds a second constraint. The heat-flux inversion assumes conductive transport dominates. In winds above roughly 5–7 m s⁻¹, turbulent sensible-heat exchange at the ice surface becomes significant and the simple conductive model underestimates thickness. ERA5 wind fields are used to flag or correct affected pixels, but the correction is imperfect.
What the reanalysis brings and what it costs you
The inversion requires air temperature at a reference height (typically 2 m) and wind speed. ERA5 provides these at 0.25° spatial resolution and hourly temporal resolution, which is adequate for the large-scale thermodynamic forcing but cannot resolve the sharp temperature gradients that occur at lead edges or in coastal katabatic-wind zones. The mismatch between a 1 km thermal pixel and a 25 km reanalysis cell is a genuine source of error, particularly near coastlines and in the marginal ice zone where mesoscale atmospheric variability is high.
Some operational implementations substitute in-situ meteorological observations from drifting buoys or coastal stations where available. This improves accuracy locally but creates spatial inhomogeneity in the product. The honest position is that the reanalysis-driven inversion gives a useful regional picture of thin-ice distribution, not a precise point measurement.
Complementarity with altimetry and what falls through the gap
ICESat-2 and CryoSat-2 measure ice freeboard with centimetre-level precision, but their skill degrades for ice thinner than roughly 30–50 cm because freeboard becomes comparable to surface roughness and wave-induced tilt. The thermal inversion is at its most useful precisely where altimetry is weakest. The two methods are genuinely complementary rather than redundant, and combined products that use thermal-derived thickness for thin ice and altimetry-derived thickness for thicker ice have been published in the peer-reviewed literature.
The gap that neither fills well is ice in the 0–5 cm range, including frazil, grease ice and nilas. At these thicknesses, emissivity assumptions break down, the ice surface is not flat, and the thermal signal is contaminated by open-water pixels within the sensor footprint. Sub-pixel mixing is a persistent problem at 1 km resolution. VIIRS at 375 m helps but does not eliminate it.
Operational uses: where this product earns its keep
Ice thickness in the 0–50 cm range matters for three distinct communities. Polar shipping operators need to know whether new ice in a lead is thin enough to break or thick enough to trap a vessel; even a rough thickness map at 1 km resolution is decision-relevant. Oceanographers use thin-ice area and thickness to estimate brine rejection and freshwater flux into the polar ocean, which drives thermohaline circulation models. Climate scientists track thin-ice production in polynyas as a proxy for ocean-atmosphere heat exchange in winter, a variable that is poorly constrained by any other observation type.
Satellize runs thermal inversion analytics on open MODIS, VIIRS and Sentinel-3 SLSTR archives, combining ERA5 reanalysis fields and a quality mask that flags cloud-contaminated and high-wind pixels before the inversion is attempted. The output is a gridded thickness estimate with per-pixel uncertainty, delivered as a GeoTIFF or NetCDF layer. The approach is the same physics-based method used in published polar research; what changes is the operational packaging and the integration with client ice-routing or oceanographic modelling workflows. For context on how Satellize structures analytics programmes around open data, the Tonga crop-estimation programme is the one publicly described example.
Typical figures
| Spatial resolution (MODIS) | 1 km at nadir; degrades to ~2 km at swath edge |
| Spatial resolution (VIIRS I-band) | 375 m at nadir; moderate bow-tie correction applied |
| Revisit frequency | Up to 4 passes per day at 70°N from combined Terra/Aqua; higher at greater latitudes |
| Thickness range | 0–50 cm; method loses skill above this threshold |
| Thickness uncertainty (published) | ±5 cm for ice <20 cm; ±10–15 cm approaching 50 cm ceiling |
| Thermal spectral bands | Split-window pair ~10.8 µm and ~12.0 µm on all listed sensors |
| Atmospheric input | ERA5 reanalysis at 0.25° / 1 h; in-situ substitution where available |
| Cloud limitation | Method fails under cloud; Arctic winter clear-sky fraction typically 20–40 % |
| Archive depth | AVHRR from 1978; MODIS from 2000; VIIRS from 2012; Sentinel-3 SLSTR from 2016 |
| Delivery format | Gridded GeoTIFF or NetCDF with per-pixel uncertainty and quality flag |
Analytics Satellize can run
| Thin-ice thickness map | 1-D conductive heat-flux inversion (Yu & Lindsay formulation) applied to cloud-screened brightness temperatures with ERA5 forcing | Per-orbit or daily composite GeoTIFF with uncertainty layer, polar stereographic projection |
| Clear-sky composite thickness field | Multi-day aggregation of cloud-free pixels; temporal weighting by recency | NetCDF time-series stack covering user-defined region and date range |
| Thin-ice area fraction by region | Pixel classification into thickness bins (0–10, 10–20, 20–50 cm) with area summation | CSV time-series report, weekly or monthly cadence, by named basin or polynya |
| Ocean-to-atmosphere heat-flux estimate | Bulk flux calculation from inverted thickness and reanalysis atmospheric state; published Maykut parameterisation | Gridded heat-flux layer (W m⁻²) for integration into ocean or climate model boundary conditions |
| High-wind contamination flag | ERA5 10 m wind speed threshold mask (configurable, default 6 m s⁻¹) applied before inversion | Quality-flag band included in all raster outputs; summary statistics of flagged fraction per scene |
| Polynya thin-ice production anomaly | Climatological baseline from MODIS/AVHRR archive; z-score anomaly detection against seasonal mean | Anomaly map and alert when production exceeds configurable threshold, delivered as GIS layer and email digest |
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.