Sea-ice drift and deformation velocity fields

Sequential SAR and passive-microwave imagery reveal how sea ice drifts, shears and piles up. Velocity fields derived from feature-tracking and cross-correlation expose lead openings, pressure ridges and divergence zones that static concentration maps cannot show.

Sensors

Sentinel-1 A/B SAR (C-band, 5.4 GHz): Provides 20 m resolution in Extra Wide Swath mode (400 km swath) with 6-day repeat at mid-latitudes, improving to near-daily coverage poleward of 70°N when both satellites are active. Cloud-independent, day/night capable. The primary source for high-resolution drift and deformation vectors over short time baselines (1–3 days).
RADARSAT-2 (C-band SAR): Offers ScanSAR Wide modes at 50–100 m resolution with 500 km swath, and Ultrafine modes at 3 m for localised deformation detail. Commercially tasked; useful when Sentinel-1 revisit is insufficient or when a specific deformation event requires sub-daily repeat.
AMSR2 on GCOM-W1 (passive microwave, 6.9–89 GHz): Daily global coverage at 3.5–62 km effective resolution depending on frequency channel. Brightness-temperature sequences at 89 GHz (approximately 3.5 km footprint) support large-scale drift estimation via cross-correlation over 1–3 day windows. Penetrates cloud and darkness; resolution is too coarse for lead-scale deformation but captures basin-scale kinematics reliably.
OSI SAF passive-microwave motion products: EUMETSAT's Ocean and Sea Ice Satellite Application Facility produces operational daily low-resolution (60–80 km) ice-motion vectors from merged SSM/I, SSMIS and AMSR2 brightness temperatures. Archive extends back to 1991 for the Arctic and 1992 for the Antarctic. Provides the climatological baseline against which anomalous drift events are assessed.

What feature-tracking actually measures

Deriving a velocity field from two SAR images separated by 24–72 hours is conceptually simple: find the same patch of ice in both images, measure how far it moved, divide by time. In practice, the difficulty lies in the matching. Maximum cross-correlation (MCC) and normalised cross-correlation (NCC) slide a template window across the second image to find the displacement that maximises similarity. Pattern-matching feature trackers go further, identifying distinctive surface textures such as ridges and floe edges and following them explicitly. Both approaches produce a grid of displacement vectors, typically at spacings of 5–20 km for Sentinel-1 EW pairs.

The vector field is then differentiated spatially to yield the deformation invariants: divergence (positive where ice pulls apart, opening leads; negative where it converges), shear (the rotational component), and total deformation. These quantities matter operationally because divergence exposes open water that refreezes rapidly as thin nilas, while convergence piles ice into ridges that can exceed 30 m keel depth. Neither process is visible in a single image. It is the difference between frames that carries the physical information.

Why the time baseline is a design choice, not a given

Short baselines (12–24 hours) resolve fast-moving marginal-ice-zone floes and capture rapid lead-opening events, but they demand near-daily SAR coverage. Over the central Arctic in winter, Sentinel-1 revisit at high latitudes can approach 1–2 days when both satellites are operational, making this feasible. Longer baselines (3–6 days) are more practical at lower latitudes or when only one satellite is available, but fast deformation events alias: a lead that opened and refroze within 48 hours leaves no trace.

Passive-microwave products from AMSR2 and OSI SAF typically use 2-day composites to suppress noise, which smooths out sub-daily kinematics entirely. They are well-suited to basin-scale transport questions, such as how much multi-year ice is exported through Fram Strait in a given season, but they cannot resolve the kilometre-scale deformation zones that govern local ice strength and navigability. The two sensor families answer different questions; conflating them is a common source of misinterpretation.

Deformation fields reveal what concentration maps hide

A passive-microwave concentration product might report 90% ice cover across a 25 km grid cell. That cell could be mechanically quiescent or it could be actively shearing, with linear kinematic features (leads and ridges) forming and closing on timescales of hours. High-resolution SAR deformation fields expose this internal structure. Studies using RADARSAT and Sentinel-1 have documented total deformation rates in the Arctic that span more than two orders of magnitude across the basin, with the highest values concentrated in narrow deformation zones that occupy a small fraction of the total area but account for a disproportionate share of ice mass redistribution.

Lead opening is particularly important for atmosphere-ocean heat exchange. A 1 km wide lead in winter can transfer heat fluxes of several hundred watts per square metre to the atmosphere, compared with single-digit fluxes through consolidated pack ice. Identifying where and when leads form from divergence fields is therefore relevant to weather forecasting, sea-ice thickness modelling and polar shipping risk assessment, though the last of those is covered separately in the route-condition page.

Honest limits of the method

Matching fails when ice surface texture is homogeneous, which happens most often over newly formed thin ice, during summer melt when the surface becomes spectrally flat, or in heavily deformed zones where the reference texture has been destroyed between acquisitions. Correlation scores drop, and the standard practice is to apply a threshold on peak correlation value and a median-filter consistency check to reject spurious vectors. Typical rejection rates in difficult conditions can reach 20–40% of the grid, leaving spatial gaps in the deformation field.

SAR geometry introduces its own ambiguity. Ice displacement has three components but SAR measures only the range-direction projection of motion. Azimuth offsets can be estimated from along-track displacement, but the vertical component is negligible for horizontal drift and is generally ignored. More practically, the 6-day Sentinel-1 repeat cycle at mid-Arctic latitudes means that fast-moving events near the ice edge can produce displacements that exceed the search window of the cross-correlator, causing the tracker to fail entirely. Analysts must tune the search radius to the expected drift speed, which requires some prior knowledge of the regime.

From vectors to operational products

The OSI SAF archive offers a ready-made climatology of Arctic and Antarctic ice motion back to the early 1990s, which is invaluable for anomaly detection. Comparing a current season's drift pattern against the 30-year record can flag unusually rapid Transpolar Drift Stream acceleration or anomalous Beaufort Gyre spin-up months before they become visible in extent products. Sentinel-1 pairs then provide the spatial resolution to examine the deformation consequences of those anomalies at the floe scale.

Satellize runs this workflow on open Sentinel-1 acquisitions and OSI SAF archives, producing gridded divergence and shear layers for client-defined regions of interest. The Tonga crop-estimation programme is the company's best-known analytics engagement, but the underlying pipeline for sequential-image feature-tracking transfers directly to polar motion problems. Outputs are delivered as georeferenced GeoTIFF or NetCDF layers with associated uncertainty masks, ready for ingestion into ice-routing or climate-modelling workflows.

Archive depth and what it enables

The OSI SAF low-resolution motion product archive begins in 1991 for the Arctic, giving three decades of basin-scale kinematics. AMSR2 brightness-temperature data from GCOM-W1 are available from 2012. Sentinel-1 EW acquisitions over the Arctic are systematic from late 2014, providing nearly a decade of high-resolution SAR pairs from which deformation climatologies can be constructed. RADARSAT-2 extends the high-resolution SAR record back to 2008 for areas of commercial interest.

This depth matters because sea-ice deformation is not stationary. The Arctic has lost roughly half its September sea-ice extent since the late 1970s, and the remaining ice is younger, thinner and mechanically weaker than the multi-year pack that dominated mid-century. Deformation rates have increased as a result. A drift-vector analysis that uses only recent years will not capture the full range of variability; one that uses only the passive-microwave climatology will miss the structural shift in the deformation regime that high-resolution SAR reveals.

Typical figures

Spatial resolution (SAR-derived vectors)	5–20 km vector spacing from Sentinel-1 EW (20 m pixel); 1–5 km from RADARSAT-2 fine modes
Spatial resolution (passive-microwave vectors)	60–80 km (OSI SAF operational product); ~25 km from AMSR2 89 GHz channel
Temporal baseline	1–6 days for SAR pairs; 1–2 day composites for passive-microwave products
Revisit (Sentinel-1 at >70°N)	1–2 days with both satellites active; 3–6 days with one
Sensor frequency / band	C-band SAR at 5.4 GHz; passive microwave 6.9–89 GHz (AMSR2)
Minimum detectable displacement	Approximately 200–500 m over 24 hours for Sentinel-1 MCC; basin-scale drift of >5 km/day reliably detected by passive microwave
Archive depth	OSI SAF from 1991; AMSR2 from 2012; Sentinel-1 systematic Arctic coverage from 2014; RADARSAT-2 from 2008
Coverage	Arctic and Antarctic; Sentinel-1 EW covers polar regions systematically; OSI SAF products are hemispheric daily
Delivery formats	GeoTIFF, NetCDF-CF, GeoJSON vector layers with uncertainty masks
Latency (near-real-time)	OSI SAF operational product: ~24 hours; Sentinel-1 NRT processing: 1–3 hours post-acquisition via Copernicus services

Analytics Satellize can run

Gridded ice-drift velocity field	Maximum cross-correlation (MCC) or normalised cross-correlation (NCC) on sequential Sentinel-1 EW image pairs	NetCDF or GeoTIFF layer with eastward/northward displacement components and per-vector correlation score
Divergence and convergence map	Spatial differentiation of the velocity field; positive divergence flags lead-opening zones, negative flags convergence and ridging	Gridded deformation-rate layer (units: day⁻¹) with lead-probability overlay
Shear and total deformation field	Rotational component extracted from velocity-gradient tensor; total deformation as quadratic sum of divergence and shear	GeoTIFF raster; high-deformation zones flagged as polygon GIS layer
Seasonal drift anomaly report	Comparison of current-season OSI SAF vectors against 30-year climatological mean and standard deviation	PDF or interactive report showing anomaly maps for Transpolar Drift Stream and Beaufort Gyre, with time-series plots
Lead-opening event detection	Threshold on positive divergence combined with SAR backscatter change detection to confirm new open-water or thin-ice signature	Event alert with polygon extent, opening rate (km²/day) and timestamp; delivered via API feed or email
Ridge-formation intensity index	Convergence magnitude integrated over time window; calibrated against published relationships between convergence rate and ridge keel depth	Raster index layer and summary statistics table for user-defined area of interest
Multi-year drift trajectory archive	Lagrangian integration of gridded velocity fields (OSI SAF + Sentinel-1 fusion) to reconstruct ice-parcel pathways	GeoJSON trajectory file and visualisation; exportable to climate or routing model input format

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.