Iceberg freeboard and keel-depth estimation from altimetry
Satellite altimetry measures how high an iceberg sits above sea level. Combined with ice-density assumptions, that freeboard converts to total draft, often exceeding 200 metres, with direct consequences for subsea pipelines, cables and drilling infrastructure.
Sensors
- ICESat-2 ATLAS (NASA): Photon-counting lidar at 532 nm. Along-track shot spacing of approximately 0.7 m, across-track beam separation of 3.3 km between beam pairs. Surface elevation precision better than 3 cm over flat ice. Six beams in three pairs allow slope estimation. Repeat ground track every 91 days; exact-repeat sub-cycle roughly 29 days. No penetration of cloud; cloud cover is the primary data-loss cause in polar regions.
- CryoSat-2 SIRAL (ESA): Ku-band radar altimeter with three operating modes. SARIn (synthetic-aperture interferometric) mode over ice-sheet margins and ocean provides a footprint of roughly 300 m along-track. Freeboard precision over sea ice approximately 2–5 cm after waveform retracking, though iceberg surfaces introduce retracking ambiguity. 369-day repeat; sub-cycle of 30 days gives denser sampling in practice. Ku-band penetrates thin snow layers, which can bias freeboard low by a few centimetres.
- WorldView stereo pairs (Maxar): Panchromatic resolution 0.3–0.5 m. Stereo acquisition on the same pass or across passes yields digital surface models with vertical accuracy of 0.5–2 m depending on base-to-height ratio and surface texture. Useful for large bergs where the surface is textured enough for image matching. Tasked on demand; latency depends on cloud and tasking queue. No penetration of cloud or snow surface.
- Sentinel-6 Michael Freilich (ESA/EUMETSAT): Ku- and C-band radar altimeter primarily designed for sea-surface height. Covers latitudes to 66° N/S in its standard orbit, limiting polar coverage. Useful for cross-calibrating freeboard retrievals from CryoSat-2 over lower-latitude icebergs drifting into shipping lanes. 10-day exact repeat. Freeboard retrieval from Sentinel-6 over icebergs is experimental rather than operational.
What freeboard actually tells you, and what it does not
Freeboard is the height of an iceberg's waterline above mean sea level. It sounds simple. The complication is that the number you want, keel depth, is not directly observable from any satellite. It is inferred by assuming isostatic equilibrium: the berg floats because the weight of ice displaced equals the weight of seawater displaced. For a tabular iceberg with ice density near 900 kg/m³ and seawater density near 1025 kg/m³, the ratio of submerged draft to freeboard works out to approximately 7:1. A berg showing 30 metres of freeboard carries a keel near 210 metres.
That ratio is not a constant. Ice density varies with firn compaction, melt-pond infiltration and the proportion of bubbly versus dense ice. Published studies using in-situ measurements alongside altimetry find density uncertainty of roughly 10–20 kg/m³ in the firn layer alone, which propagates to keel-depth uncertainty of 10–20% on a large berg. For subsea infrastructure risk assessment, that is the honest error budget you are working with before any sensor uncertainty is added.
ICESat-2 and CryoSat-2: complementary, not interchangeable
ICESat-2's ATLAS instrument fires 10,000 laser pulses per second across six beams. Over a flat iceberg surface it can resolve freeboard to better than 3 cm in clear conditions, and the beam geometry allows surface slope to be estimated, which matters for irregular bergs. The instrument's weakness is cloud. Polar maritime air masses produce persistent low cloud, and ATLAS returns nothing useful through them. Revisit at a given point is 91 days for the exact repeat, though the 29-day sub-cycle improves sampling at high latitudes where ground tracks converge.
CryoSat-2 SIRAL in SARIn mode operates at Ku-band and penetrates most cloud. Its freeboard precision over icebergs is lower than ICESat-2, partly because radar waveforms from a tilted or irregular berg surface are harder to retrack than the smooth returns the algorithms were designed for. A further subtlety: Ku-band penetrates dry snow by several centimetres, so the radar range references a surface slightly below the true top. The two instruments therefore bracket the truth rather than duplicate it, and combining them, where their ground tracks intersect near a berg, reduces systematic bias.
WorldView stereo adds a third independent geometry. It is not a repeat measurement but it provides spatial context that altimetry cannot: the planimetric shape of the berg, its surface roughness, and visible crevassing that signals density heterogeneity. The vertical accuracy of a stereo DSM over a textured iceberg surface is typically 0.5–2 m, coarser than ICESat-2 but sufficient to validate the broad freeboard estimate and flag tilted bergs where the altimeter footprint may be sampling a slope rather than the mean surface.
Where the physics gets awkward
The 7:1 ratio assumes the berg is in hydrostatic equilibrium and that ice density is uniform. Neither is guaranteed. Newly calved tabular bergs from ice shelves carry firn layers with density well below 900 kg/m³, making them ride higher and giving a falsely optimistic keel estimate if a bulk density is assumed. Bergs that have rolled or partially melted develop irregular keels that bear no simple relationship to the visible freeboard at any single point.
Melt ponds on the surface add mass without changing freeboard in the way solid ice does, introducing a further bias. Sea-surface height itself must be known accurately; in the Southern Ocean, dynamic topography can vary by tens of centimetres over short distances, and errors in the reference sea-surface model propagate directly into freeboard. These are not hypothetical concerns. They set a practical floor on keel-depth accuracy that no sensor improvement alone can remove.
Operational relevance: pipelines, cables and freshwater volume
The Grand Banks off Newfoundland see several hundred icebergs per year in active seasons, and subsea pipelines and wellheads in that region sit at depths where large berg keels are a credible hazard. The International Ice Patrol, operating since 1913, tracks surface positions but keel depth is not directly observed operationally. Altimetry-derived keel estimates, combined with drift tracking, give operators a probabilistic hazard envelope rather than a point position.
Freshwater volume is the other application. A tabular berg 10 km across and 250 m thick carries on the order of tens of cubic kilometres of fresh water. Freeboard-derived volume estimates, integrated over a population of bergs tracked through a season, contribute to freshwater flux budgets in the Southern Ocean and North Atlantic. These budgets feed ocean circulation models. The uncertainty in individual berg volume is large, but systematic biases tend to partially cancel across a population.
Satellize can run freeboard retrieval pipelines against ICESat-2 ATL07 and CryoSat-2 baseline products, combining them with optical geometry from commercial tasking where cloud permits, and deliver keel-depth probability distributions for named bergs of interest. The same density-assumption framework we use for the Tonga crop-estimation programme, where careful propagation of input uncertainty into output confidence intervals is the point, applies directly here.
Honest limits and what to do about them
Cloud blocks ICESat-2 entirely and degrades optical stereo. CryoSat-2 covers a given berg only when its ground track passes nearby, which may be infrequent for a fast-drifting berg at lower latitudes. Keel-depth uncertainty from density assumptions alone is 10–20%, and this cannot be reduced by better sensors without in-situ density profiles from the berg itself, which are rarely available.
The practical response is to treat keel depth as a probability distribution, not a number. A berg with 35 m freeboard has a most-likely keel near 245 m but a 90th-percentile keel that may reach 290 m if density is at the low end of the plausible range. Infrastructure risk decisions should be made against the tail of that distribution, not the median. Altimetry gives you the distribution; assuming a single ratio gives you false precision.
Typical figures
| ICESat-2 ATLAS along-track shot spacing | ~0.7 m |
| ICESat-2 freeboard precision (flat ice, clear sky) | <3 cm |
| CryoSat-2 SARIn along-track footprint | ~300 m |
| CryoSat-2 freeboard precision (after retracking) | 2–5 cm; higher uncertainty over irregular berg surfaces |
| WorldView stereo DSM vertical accuracy | 0.5–2 m depending on surface texture and B/H ratio |
| Typical keel-to-freeboard ratio (tabular berg) | ~7:1 (ice density ~900 kg/m³, seawater ~1025 kg/m³) |
| Keel-depth uncertainty from density assumptions | 10–20% of derived draft |
| ICESat-2 exact-repeat cycle | 91 days; sub-cycle ~29 days |
| CryoSat-2 repeat cycle | 369 days; effective sub-cycle ~30 days |
| Archive depth (ICESat-2) | September 2018 to present |
Analytics Satellize can run
| Per-berg freeboard map | ICESat-2 ATL07 sea-ice and ice-surface photon product, filtered to berg extent defined by SAR or optical mask; elevation referenced to EGM2008 geoid | GeoTIFF surface elevation grid with per-pixel uncertainty, delivered per tasking event |
| Keel-depth probability distribution | Monte Carlo propagation of freeboard estimate through ice-density prior (mean and variance from published firn-density literature) using hydrostatic equilibrium | PDF report with median, 10th and 90th percentile keel depth; GIS point feature with attribute table |
| CryoSat-2 / ICESat-2 cross-validated freeboard | Temporal and spatial co-location of CryoSat-2 SARIn waveform retracking with ICESat-2 photon returns; bias correction for Ku-band snow penetration using published seasonal offsets | Merged freeboard dataset with provenance flags, in NetCDF or GeoPackage |
| Stereo-derived surface model for large bergs | WorldView stereo pair processed to DSM via semi-global matching; co-registered to ICESat-2 ground control points on the berg surface | 0.5 m DSM GeoTIFF, hillshade visualisation, freeboard summary statistics |
| Freshwater volume estimate | Freeboard-to-total-thickness conversion integrated over planimetric area from SAR or optical outline; uncertainty bounds propagated from density and freeboard errors | Tabular report with volume estimate in km³, uncertainty range, and time-stamped berg position |
| Subsea hazard clearance depth alert | 90th-percentile keel depth compared against user-supplied infrastructure depth threshold; alert triggered when berg drift track (from separate tracking product) brings hazard zone within specified radius | Automated alert via API or email, with supporting freeboard data and drift trajectory |
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.