Sensors
- ICESat-2 ATLAS: Photon-counting lidar at 532 nm. Six beams arranged in three pairs, each pair separated by 90 m across-track. Along-track photon aggregation yields surface height estimates at roughly 11 m intervals. 91-day exact repeat orbit at 92-degree inclination, reaching 88°N/S. Vertical precision on flat sea ice approximately 2–3 cm per 150-photon aggregate.
- CryoSat-2 SIRAL: Ku-band radar altimeter operating in SAR Interferometric mode over sea ice. Footprint approximately 300 m along-track in SARIn mode. Penetrates dry snow to some degree, which introduces a systematic freeboard bias relative to lidar; the two instruments are therefore complementary rather than interchangeable.
- Sentinel-3 SRAL: Ku-band SAR altimeter with roughly 300 m along-track resolution in closed-loop mode. Provides basin-wide sea ice freeboard at lower spatial detail than CryoSat-2 but with a 27-day repeat and open-access data, useful for change monitoring at seasonal timescales.
- Operation IceBridge (airborne): NASA airborne campaign using ATM scanning lidar and snow radar. Spatial resolution of order 1 m. Served as the primary validation bridge between ICESat (2003–2009) and ICESat-2 (2018–present) and continues to provide snow depth profiles used to calibrate freeboard-to-thickness conversion.
What freeboard actually measures, and what it does not
Freeboard is the height of the ice or snow surface above the local sea surface. It sounds simple. The complication is that roughly 90 per cent of sea ice sits below the waterline, so a 20 cm freeboard corresponds to something in the range of 1.5 to 2.5 metres of total thickness, depending on how much of that freeboard is snow versus ice. ATLAS does not see below the surface. It measures the elevation of whatever photons bounce back from first, which in practice is the top of the snow layer on multi-year ice or the ice surface on bare first-year ice.
The conversion from freeboard to thickness uses Archimedes' principle and requires three inputs: freeboard height, snow depth on top of the ice, and assumed densities for snow, ice and seawater. Published values for Arctic sea ice density range from roughly 860 to 940 kg/m³ depending on ice age and brine content. Snow depth is the weakest link. It is typically taken from climatological models such as Warren et al. (1999) or the more recent NESOSIM model, neither of which resolves the spatial heterogeneity that airborne snow radar reveals. A 5 cm error in assumed snow depth propagates to roughly 30–50 cm error in derived thickness, which is not a small number when mean first-year ice is around 1.5 metres.
How ATLAS counts photons across six beams
ATLAS fires 532 nm green laser pulses at 10 kHz. The detector is a single-photon avalanche diode array, sensitive enough to record individual returning photons. At typical orbital altitude of 496 km, each pulse illuminates a spot roughly 11 m in diameter on the surface. The satellite moves fast enough that consecutive pulses are spaced about 0.7 m apart along track, but the standard geophysical data product aggregates returns over roughly 150 photons to produce a reliable surface height estimate, which corresponds to segments of approximately 11–17 m depending on surface reflectance.
The six beams are arranged as three pairs. Within each pair, one beam is stronger (the reference beam) and one weaker, separated by 90 m across-track. The three pairs are spread across approximately 6 km in total cross-track width. This geometry is a deliberate trade-off: wider swath would improve coverage, but the photon budget from a single laser limits how many beams can be run simultaneously at useful signal-to-noise. The 91-day exact repeat cycle means the same ground tracks are revisited four times per year, which is adequate for seasonal cycle studies but coarse for tracking the rapid changes in Arctic sea ice extent that can occur over days during melt season.
Leads give you the sea surface. They also try to fool the algorithm.
To convert ice surface elevation to freeboard, ATLAS needs a reference sea surface height along the same track. Open water leads between ice floes provide that reference: calm water produces a specular reflection that returns a tight cluster of photons at a well-defined elevation. The ATL07 sea ice product identifies these lead segments and fits a local sea surface height, which is then interpolated under the ice to give a freeboard for each ice segment.
The problem is that specular leads and smooth ice can look similar in photon return statistics, particularly when a thin skim of new ice covers a lead and the surface is still nearly flat. The ATL07 algorithm uses a combination of return photon density, surface roughness statistics and background noise rates to classify segments, but misclassification remains a documented source of error, especially in the marginal ice zone where ice and open water are intermixed at fine scales. Melt ponds on top of multi-year ice in summer present a related ambiguity: the pond surface can return photons at a lower elevation than the surrounding ice, mimicking a lead. Summer freeboard retrievals are therefore less reliable than winter ones, and published uncertainty estimates reflect this asymmetry.
Validation numbers from published campaigns
NASA and ESA have published validation comparisons between ICESat-2 freeboard and independent measurements from Operation IceBridge, CryoSat-2 and moored upward-looking sonar. For Arctic sea ice freeboard, ICESat-2 ATL10 product comparisons with IceBridge ATM show mean biases typically below 2 cm and root-mean-square differences of 5–8 cm at the segment level, which is genuinely impressive for a space-based measurement. Thickness derived from those freeboards carries considerably larger uncertainty: published comparisons with submarine sonar and IceBridge snow radar suggest thickness RMSE of 0.4 to 0.8 metres for first-year ice and somewhat larger for multi-year ice, where snow depth variability is higher.
Antarctic sea ice is harder. The Warren snow climatology was built from Arctic data and performs poorly in the Southern Ocean, where snow accumulation rates and ice growth mechanisms differ substantially. Several published studies have used alternative snow depth inputs, including ERA5 reanalysis precipitation, and find that the choice of snow product changes derived Antarctic thickness by 20–30 per cent in some regions. This is an active research problem, not a solved one.
What analysts can do with the data, and where it runs out
ICESat-2 data are freely available through NASA Earthdata. The ATL07 (sea ice height) and ATL10 (freeboard) standard products are distributed in HDF5 format with documented uncertainty flags per segment. Analysts working on shipping route assessment, climate model validation or polar logistics planning typically start with ATL10, apply quality filters on the provided confidence flags, and then grid the along-track data to regular spatial bins, commonly 25 km EASE-Grid cells to match passive microwave sea ice concentration products.
The 91-day repeat means that between repeats, coverage of any specific area depends on the density of adjacent tracks, which varies with latitude. Near the pole the tracks converge and coverage is dense. At lower latitudes, gaps between tracks can exceed 50 km. For operational sea ice monitoring at daily or weekly timescales, ICESat-2 freeboard is best treated as a periodic calibration layer rather than a continuous feed. SAR-based ice type classification and drift tracking, covered separately in this library, fills the temporal gaps. Satellize incorporates ICESat-2 freeboard retrievals as a calibration input in multi-source polar analytics, alongside open radar altimetry from Sentinel-3.
One genuinely underused application is fresh-water ice on large lakes and rivers. ATLAS resolves lake ice freeboard at the same precision as sea ice, and several published studies have used it to infer lake ice thickness in regions where in-situ measurement is sparse. The physics of the freeboard-to-thickness conversion is identical; the density assumptions are different and arguably better constrained for fresh water.
Typical figures
| Instrument | ATLAS photon-counting lidar, 532 nm, 10 kHz pulse rate |
| Along-track segment spacing | ~11–17 m (150-photon aggregates on sea ice) |
| Across-track beam spread | ~6 km total (three beam pairs, 90 m within each pair) |
| Vertical precision (flat ice surface) | ~2–3 cm per aggregated segment |
| Freeboard RMSE vs IceBridge ATM | 5–8 cm (published validation, Arctic first-year ice) |
| Derived thickness uncertainty | 0.4–0.8 m RMSE (first-year Arctic ice; larger for multi-year and Antarctic) |
| Repeat cycle | 91-day exact repeat; sub-cycle tracks fill some gaps at ~29-day intervals |
| Orbital inclination / poleward limit | 92° inclination; data to 88°N and 88°S |
| Data products | ATL07 (sea ice height), ATL10 (freeboard); HDF5 via NASA Earthdata |
| Archive depth | October 2018 to present |
Analytics Satellize can run
| Gridded seasonal freeboard maps | ATL10 quality filtering, along-track-to-grid interpolation onto 25 km EASE-Grid cells, gap-filling with kriging or optimal interpolation | GeoTIFF or NetCDF seasonal freeboard grids with per-cell uncertainty estimates |
| Derived ice thickness with uncertainty bounds | Isostatic freeboard-to-thickness conversion using published density climatologies and NESOSIM or ERA5 snow depth; Monte Carlo uncertainty propagation over density and snow depth ranges | Thickness raster with 5th–95th percentile uncertainty layer; seasonal report |
| Multi-year vs first-year ice freeboard discrimination | Freeboard distribution analysis combined with ATL07 surface roughness statistics; threshold classification validated against passive microwave ice-type products | Ice-type fraction map per grid cell; time-series chart of multi-year ice volume proxy |
| Shipping route ice thickness profiles | Along-track ATL10 extraction along candidate Northern Sea Route or Northwest Passage corridors; comparison against vessel draft requirements | Route-specific thickness profile PDF and tabular summary for voyage planning |
| ICESat-2 / CryoSat-2 freeboard fusion | Cross-calibration of lidar and radar freeboard using co-located track pairs; radar penetration correction derived from residual analysis | Fused freeboard product at improved spatial sampling; methodology note |
| Interannual volume anomaly monitoring | Repeat-cycle freeboard differencing on exact-repeat tracks; volume anomaly relative to 2019–present baseline | Annual polar ice volume anomaly report with trend significance assessment |
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.