Iceberg size-distribution and volume estimation from SAR

SAR amplitude imagery from Sentinel-1 and RADARSAT Constellation Mission can detect icebergs down to roughly 100 m in calm seas, enabling the population statistics that govern freshwater flux and iron fertilisation across polar oceans.

Sensors

Sentinel-1 A/B (C-band SAR, 5.4 GHz): Extra-Wide Swath mode delivers 400 km swath at 20 × 40 m ground range resolution with 6-day repeat at the equator, shrinking to 1–3 days near the poles. VV+VH dual-polarisation allows cross-polarisation ratio analysis that separates ice from wind-roughened sea surface. Free and open archive from 2014.
RADARSAT Constellation Mission (C-band SAR): Three identical satellites provide daily revisit at latitudes above 50°N and near-daily coverage of the Southern Ocean. Medium-resolution ScanSAR modes at 50 m and 100 m resolution support wide-area population surveys; spotlight modes reach 3–5 m for individual berg geometry.
TerraSAR-X / TanDEM-X (X-band SAR, 9.65 GHz): Spotlight mode achieves 1–2 m resolution, resolving surface texture and enabling freeboard estimation on individual large bergs. Repeat-pass InSAR pairs can constrain volume change. Commercial tasking adds latency and cost, so use is selective rather than systematic across a full population.
Copernicus Emergency Management Service (CEMS): Activates rapid SAR-based mapping products, including iceberg delineation layers, for declared events. Useful as a validation or cross-check dataset rather than a primary population-statistics source, given its event-driven rather than routine cadence.

Why population statistics matter more than individual tracks

Most operational iceberg services focus on a single question: where is this berg, and will it hit something? That is a legitimate safety problem. But for oceanographers and climate modellers, the more consequential question is how much fresh water a calving event injects into the ocean over months, and across what spatial extent. That requires knowing the full size-frequency distribution of a berg population, not just the position of the largest hazard.

Iceberg size distributions follow approximate power-law relationships, meaning small bergs collectively carry a significant fraction of the total ice volume even though individually they are invisible to most sensors. A survey that detects only bergs above 1 km misses the tail of the distribution and can underestimate total freshwater flux by a substantial margin. SAR is the only spaceborne modality that can survey these populations systematically in polar darkness and through cloud, which is why it anchors this analysis.

What the radar return actually tells you, and what it does not

A smooth iceberg surface produces a strong specular return in C-band SAR when the incidence angle is low, and a weaker return at steeper angles. In practice, berg surfaces are rarely smooth: crevasses, meltwater ponds and rough fracture faces create a complex backscatter signature. The key discriminator is contrast against the surrounding sea surface. In calm conditions, the ocean return is low (dark in the image) and bergs appear as bright targets. In high sea states, ocean clutter rises and small bergs become difficult to separate from wave features. Detection of bergs below roughly 100 m diameter becomes unreliable in sea states above Beaufort 4–5.

Cross-polarisation ratios help. Ice depolarises the radar signal more than the sea surface, so the VH/VV ratio is elevated over ice. This is not a perfect separator: ships and some wave features also show elevated cross-pol returns. Automated detection pipelines therefore combine amplitude thresholding, cross-pol ratio filters, and shape metrics (compactness, elongation) to reduce false alarms. Residual false-alarm rates in published literature vary widely depending on sea state and incidence angle, which is an honest reason to report detection confidence alongside any population count.

From detected objects to a size-frequency distribution

Once bergs are detected and delineated in a SAR scene, their plan-view area is measured directly from the image. Converting area to volume requires an assumption about berg shape. The standard approach uses empirically derived relationships between waterline length and total volume, with the Barker et al. (2004) and subsequent studies providing shape coefficients for different berg morphologies (tabular, domed, pinnacled, drydocked). These coefficients carry real uncertainty: a tabular berg and a pinnacled berg with the same plan-view area have very different volumes. Where TerraSAR-X or ICESat-2 freeboard data are available for a subset of bergs, they can calibrate the shape assumption for that calving event.

The resulting size-frequency histogram, plotted as log(count) against log(area), typically shows a power-law slope in the range of roughly minus 1.5 to minus 2.5, consistent with published studies from Greenland and Antarctic source regions. The slope itself carries information: a steeper slope means relatively more small bergs, which affects how quickly the population melts and where freshwater is released. Tracking how the slope evolves over successive SAR passes reveals whether a calving event is producing a self-similar fragmentation cascade or a bimodal distribution from a single large tabular break-up.

Revisit rate and the temporal aliasing problem

A population survey is only meaningful if the revisit interval is short relative to the drift speed and melt rate of the bergs being counted. Icebergs in the Labrador Sea can drift at 0.1 to 0.5 m/s in surface currents, meaning a berg can move 40 km or more between a 3-day Sentinel-1 pass and the next. Small bergs melt on timescales of days to weeks. If the revisit is too long, the population counted in pass two is not the same population as pass one, and apparent changes in the distribution conflate drift, melt and new calving.

RADARSAT Constellation Mission's daily polar revisit largely solves this for Arctic regions. For the Southern Ocean, Sentinel-1 coverage is denser near the Antarctic coast but thins at lower latitudes where most drift occurs. Combining ascending and descending passes from both constellations is standard practice to reduce the effective revisit gap. Even so, any population estimate should state the observation window explicitly and treat the distribution as a snapshot rather than a census.

Freshwater flux and biological consequences

The reason governments and research agencies commission this analysis is ultimately biogeochemical. Melting icebergs release cold, fresh water that stratifies the upper ocean and suppresses vertical mixing. They also release terrigenous iron scavenged from subglacial sediment. Published work, including studies using Southern Ocean drifter and satellite chlorophyll data, has shown elevated phytoplankton concentrations in the wakes of large icebergs, attributed to this iron fertilisation. Quantifying that effect requires knowing not just where the large bergs are but how much ice is melting across the full population, including the small bergs that contribute disproportionately to melt flux per unit volume.

Volume estimates from SAR-derived size distributions feed directly into ocean biogeochemical models as spatially distributed freshwater and micronutrient source terms. The uncertainty in those estimates, driven by shape-coefficient ambiguity and detection incompleteness below 100 m, is large enough that honest model inputs should carry error bounds rather than point estimates. Satellize runs this pipeline on open Sentinel-1 and RCM data, with optional TerraSAR-X tasking for shape calibration, and delivers georeferenced population layers and flux estimates to clients working on polar ocean monitoring. The same analytical infrastructure underpins the firm's crop-estimation work for the Kingdom of Tonga, adapted here for a very different detection problem.

Honest limits of the method

No SAR-based survey is a complete census. Bergs below roughly 100 m in diameter in moderate sea states will be missed. Very large tabular bergs can be misclassified as sea-ice floes if they lack strong edge contrast. Subsurface keel volume, which can be four to five times the freeboard volume for typical berg density ratios, is inferred rather than measured; altimetry from ICESat-2 or CryoSat-2 is the appropriate complement for keel estimation, covered in a separate page in this library. Finally, SAR amplitude imagery cannot distinguish between a freshly calved berg and one that has been drifting and melting for weeks, which matters for flux timing. Combining SAR population surveys with drift-tracking from sequential passes, and with optical imagery when cloud permits, gives the most defensible estimate of both total volume and its temporal evolution.

Typical figures

Typical spatial resolution (survey mode)	20–100 m (Sentinel-1 EW/IW and RCM ScanSAR); 1–5 m with TerraSAR-X spotlight for individual berg characterisation
Revisit interval (polar regions)	1–3 days with combined Sentinel-1 A/B ascending and descending passes; daily with RADARSAT Constellation Mission above 50°N
Minimum detectable berg diameter (calm seas)	Approximately 100 m in Sentinel-1 EW; degrades to 300–500 m in Beaufort 5+ sea states
Radar frequency and polarisation	C-band (5.4 GHz) for Sentinel-1 and RCM; X-band (9.65 GHz) for TerraSAR-X; dual-pol VV+VH standard for ice discrimination
Swath width (survey mode)	400 km (Sentinel-1 EW); 350–500 km (RCM ScanSAR wide); 10–100 km (TerraSAR-X depending on mode)
Archive depth	Sentinel-1 from 2014; RADARSAT-2 heritage data from 2007; TerraSAR-X from 2007
Volume estimation uncertainty	Typically ±30–50% driven by shape-coefficient ambiguity; reducible with freeboard calibration from altimetry
Delivery formats	GeoTIFF detection masks, GeoJSON berg polygons with area/volume attributes, CSV size-frequency tables, NetCDF flux grids

Analytics Satellize can run

Berg population map	Amplitude thresholding combined with cross-polarisation ratio filter and morphological shape metrics on Sentinel-1 or RCM imagery	GeoJSON polygon layer of detected bergs with plan-view area, centroid coordinates, and detection-confidence score per scene
Size-frequency distribution	Power-law fitting to log-log histogram of detected berg areas; slope and intercept reported with bootstrap confidence intervals	CSV table and PDF chart showing size-frequency distribution per observation epoch, with slope evolution time series
Volume estimate per population	Empirical area-to-volume conversion using published morphological shape coefficients (tabular, domed, pinnacled classes); uncertainty bounds propagated from coefficient ranges	Tabular volume estimate with ±1-sigma uncertainty range; optional NetCDF grid of spatially distributed volume density
Freshwater flux estimate	Volume time-series differencing between successive passes combined with published melt-rate parameterisations as a function of sea-surface temperature	Time-stamped flux report in km³ per month with spatial distribution map; suitable for direct input to ocean model boundary conditions
False-alarm flagged detection log	Ship-registry cross-check via AIS position matching; wave-clutter rejection using local variance texture filter	Cleaned detection GeoJSON with false-alarm candidates flagged and annotated; audit trail for quality control
Multi-epoch population change report	Sequential scene comparison tracking distribution-slope evolution and total ice area following a named calving event	Periodic PDF briefing or API feed showing population evolution from calving through fragmentation to dispersal

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.