Iceberg detection and drift tracking
From Antarctic tabular giants to sub-10-metre growlers, satellite SAR and optical imagery can locate, classify and track icebergs across open ocean and sea-ice fields. The hazard to shipping lanes and subsea pipelines makes positional accuracy and revisit rate the critical variables.
Sensors
- Sentinel-1 SAR (C-band, ESA): 5 x 20 m resolution in IW mode, 250 km swath; operates day and night through cloud and polar darkness. Primary workhorse for large-berg detection and the source imagery for US National Ice Center tabular-berg analysis. Distinguishing icebergs from sea-ice floes in C-band backscatter is genuinely difficult and often requires multi-pass or cross-polarisation (VV+VH) analysis.
- RADARSAT-2 (C-band, MDA/CSA): Spotlight mode reaches 1 m resolution; ScanSAR Wide covers 500 km swaths at 100 m. Used alongside Sentinel-1 by the National Ice Center for Antarctic berg cataloguing. Fine-resolution modes can resolve medium bergs (roughly 100–500 m length) with enough contrast to estimate shape and area.
- Sentinel-2 MSI (ESA): 10 m resolution in visible and near-infrared bands; 5-day revisit at mid-latitudes, less predictable at high latitudes due to cloud and polar night. Excellent for confirming berg presence and estimating surface area in cloud-free conditions, but entirely blind in darkness and severely limited by the persistent cloud cover typical of the Southern Ocean.
- Planet SuperDove: 3 m resolution, up to daily revisit at mid-latitudes. Useful for precise positional fixes and surface-area estimates of medium and large bergs when skies are clear. Cloud penetration is zero; the constellation's strength is temporal density in the sub-daily window, not all-weather coverage.
Why a floating mountain is hard to find
The scale of the largest Antarctic tabular icebergs is genuinely absurd. A-76, which calved from the Ronne Ice Shelf in May 2021, measured roughly 170 km by 25 km. Finding it is not the problem. Tracking the thousands of smaller bergs, and reliably detecting the sub-500-metre 'medium' bergs that pose the real shipping hazard, is where satellite methods earn their keep.
The physics of SAR backscatter is what makes detection possible in polar darkness and through cloud. An iceberg's freeboard presents a rough, highly reflective surface relative to calm open water, which returns very little energy to the radar. The contrast is strong in open ocean. The difficulty arises when bergs sit within sea-ice fields: both the ice floe and the berg surface can produce similarly high backscatter values in C-band imagery, and the spatial texture difference is often subtle. Multi-temporal analysis, comparing two passes separated by hours or days to see which features have moved independently, is one practical way to separate drifting bergs from the surrounding pack.
The National Ice Center's catalogue and what it actually covers
The US National Ice Center (NIC) maintains the authoritative catalogue of Antarctic icebergs, assigning alphanumeric designations (A-, B-, C-, D- by quadrant) to bergs that exceed roughly 10 nautical miles (18.5 km) in at least one dimension. Analysis is primarily manual, drawing on Sentinel-1 and RADARSAT-2 imagery. Positional updates are typically weekly, which is adequate for strategic maritime routing but not for real-time collision avoidance.
The NIC catalogue is a public resource and a useful baseline, but it covers only the largest bergs. The hazard population for most commercial vessels and subsea infrastructure operators is the medium-to-small range: bergs from a few hundred metres down to growlers (under 1 m above waterline) and bergy bits (1–5 m above waterline). These are largely absent from the NIC catalogue and require dedicated tasking at higher spatial resolution to detect reliably.
Drift trajectories: what repeat passes can and cannot tell you
A drift trajectory is built by matching the centroid or distinctive edge feature of a berg across two or more SAR or optical acquisitions and computing displacement over the known time interval. Sentinel-1's 6-day repeat (or 3-day with both satellites in constellation) gives a coarse positional update; commercial SAR tasking can reduce this to hours. Ocean current models such as HYCOM or the Copernicus Marine Service outputs can then be used to interpolate positions between satellite passes, though berg drift is also influenced by wind drag on the freeboard and keel depth, introducing model error that grows with forecast horizon.
Positional uncertainty compounds quickly. A berg moving at 0.3 knots accumulates roughly 13 km of displacement per day. If the last confirmed fix is 48 hours old and the current forecast has a 20% speed error, the uncertainty ellipse around the true position can reach tens of kilometres. For subsea pipeline operators, that is an unacceptable margin. High-revisit commercial SAR tasking, not open-constellation repeat cycles, is the correct tool when the asset at risk is expensive.
Growlers: the detection floor and honest limits
A growler sits less than 1 metre above the waterline and typically extends 5–10 metres horizontally. At Sentinel-1's 5 x 20 m pixel, a growler may occupy a single pixel or less. Detection is unreliable at that scale, particularly in the presence of wave clutter or wind roughening. RADARSAT-2 in Spotlight mode (1 m) improves the odds considerably, but even then, wave state and incidence angle strongly affect the signal-to-clutter ratio.
Airborne radar and ship-borne X-band radar remain the only reliable methods for growler detection at close range. Satellite imagery can narrow the search area by confirming that a parent berg is present and fragmenting, but it cannot substitute for proximity sensors in the final approach. Any analytics product that claims satellite-only growler detection at high confidence in rough sea states should be read sceptically.
Hazard products for shipping and subsea operators
For shipping, the practical output is a drift forecast with a confidence envelope: current position, projected track over 24–72 hours, and a polygon representing positional uncertainty at each forecast step. Vessels transiting the Labrador Sea, Drake Passage or Grand Banks benefit most from these products, as iceberg density in those corridors is high and seasonal. The International Ice Patrol, which covers the North Atlantic shipping lanes under the SOLAS convention, publishes its own iceberg limit broadcasts, but these are zone-level advisories rather than individual-berg tracking.
Subsea infrastructure presents a different problem. A pipeline or mooring system cannot move. The relevant question is whether a berg's predicted track intersects the infrastructure footprint within the operational response window, which is typically days to weeks for a managed tow or shutdown decision. Keel depth is a critical variable here: a berg with 200 m of draft in 150 m of water will scour the seabed. Keel depth cannot be measured directly from satellite imagery; it must be estimated from freeboard using a density ratio (roughly 1:7 freeboard-to-draft for typical iceberg ice), with significant uncertainty.
Satellize's analytics pipeline ingests open Sentinel-1 acquisitions and can add commercial SAR tasking under client licence for higher-priority assets, following the same approach used in our Tonga crop-estimation programme: open data as the baseline, commercial data where the decision stakes justify the cost.
Sensor combinations and what they add up to
No single sensor solves the problem. Sentinel-1 provides all-weather, all-darkness coverage at moderate resolution and is the only realistic option for systematic monitoring of large ocean areas. Sentinel-2 and Planet SuperDove add geometric precision and surface-area confidence when cloud cooperates. RADARSAT-2 in fine-resolution modes bridges the gap for medium bergs where Sentinel-1 pixels are too coarse to resolve shape reliably.
The sensible architecture for a serious iceberg-hazard programme is a tiered system: automated Sentinel-1 change detection for wide-area alerting, followed by commercial SAR or optical tasking triggered on features of interest, followed by drift modelling with explicit uncertainty bounds. The honest answer is that this architecture still has a detection floor around 50–100 m berg length in open water under moderate sea states, and the floor rises sharply inside dense sea-ice fields or in high-wave conditions.
Typical figures
| Spatial resolution (SAR, standard mode) | Sentinel-1 IW: 5 x 20 m; RADARSAT-2 ScanSAR Wide: ~100 m; RADARSAT-2 Spotlight: ~1 m |
| Spatial resolution (optical) | Sentinel-2 MSI: 10 m (VIS/NIR); Planet SuperDove: 3 m |
| Revisit (open constellation) | Sentinel-1: 6-day single satellite, ~3-day dual; Sentinel-2: 5-day at mid-latitude, variable at high latitude |
| Revisit (commercial tasking) | RADARSAT-2 and commercial SAR constellations: hours to sub-daily depending on latitude and tasking priority |
| All-weather / all-darkness capability | SAR only; optical sensors (Sentinel-2, Planet) require daylight and cloud-free conditions |
| Minimum reliably detectable berg (SAR, open water) | Approximately 50–100 m length in Sentinel-1 IW; ~10–20 m in RADARSAT-2 Spotlight under favourable sea state |
| Keel depth measurement from satellite | Not directly measurable; estimated from freeboard using ~1:7 density ratio, uncertainty typically ±20–30% |
| Archive depth | Sentinel-1: from 2014; RADARSAT-2: from 2007; Sentinel-2: from 2015 |
| Typical positional accuracy (centroid) | 10–50 m (SAR geocoding); drift uncertainty grows with time since last fix and current-model error |
| Delivery formats | GeoJSON point/polygon features, GeoTIFF, KML, drift-track GIS layer, PDF hazard report |
Analytics Satellize can run
| Berg detection and classification | SAR backscatter thresholding and texture analysis (GLCM or equivalent) on Sentinel-1 VV/VH; cross-polarisation ratio to separate berg from sea-ice floe | GeoJSON polygon layer of detected bergs with estimated surface area and classification confidence score |
| Positional fix and size estimate | Centroid extraction from SAR or optical imagery; surface-area calculation from polygon perimeter; freeboard-to-draft ratio applied for volume estimate | Point feature with centroid coordinates, acquisition timestamp, estimated length/width, freeboard estimate where optically visible |
| Drift trajectory reconstruction | Multi-temporal feature matching across repeat SAR passes; displacement vector calculation; interpolation against HYCOM or Copernicus Marine Service current fields | Drift-track GIS layer with timestamped positions, speed and heading at each step, and uncertainty ellipse per position |
| 72-hour drift forecast with uncertainty envelope | Kalman-filter or ensemble-based forward projection using observed displacement and ocean current model; wind-drag parameterisation applied to freeboard estimate | Forecast track polygon (confidence envelope at 24/48/72 h) delivered as GeoJSON or KML; updated on each new satellite pass |
| Infrastructure intersection alert | Forecast track intersected against client-supplied pipeline, mooring or platform footprint buffer; threshold trigger on probability of intersection within response window | Automated alert (email or API push) with time-to-intersection estimate and recommended response window |
| Seasonal berg density climatology | Historical detection archive aggregated over Sentinel-1 record (2014–present) to produce monthly probability grids of berg presence by 0.1-degree cell | Raster climatology layer (GeoTIFF) and summary PDF for route-risk assessment and insurance underwriting |
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.