Ice-sheet and glacier surface velocity from InSAR and offset tracking
SAR interferometry and pixel-offset tracking convert repeat-pass radar imagery into ice-flow velocity fields, exposing acceleration events, surge cycles and dynamic instability months before visible change reaches the terminus.
Sensors
- Sentinel-1 A/B (C-band SAR, ESA): Primary workhorse for ice velocity. Six-day exact repeat at high latitudes (three days when both satellites operate) enables short-baseline InSAR pairs that preserve coherence on slow-to-moderate glaciers. Ground range resolution approximately 5 x 20 m in IW mode, 25 m in EW mode. Free and open archive from 2014.
- ALOS-2 PALSAR-2 (L-band SAR, JAXA): L-band penetrates surface scattering layers and maintains coherence over longer baselines and in areas of heavy snowfall where C-band decorrelates. Spotlight mode reaches 3 m resolution; ScanSAR covers 350 km swaths at 100 m. Fourteen-day repeat; tasking required for non-default coverage.
- RADARSAT-2 (C-band SAR, MDA): Commercial C-band with flexible incidence angles and multiple beam modes from 3 m (Ultra-Fine) to 500 km swath (ScanSAR Wide). Particularly useful for offset tracking on fast outlet glaciers where InSAR phase is unwrappable only at very short baselines. Twenty-four-day repeat unless tasked.
- TerraSAR-X / TanDEM-X (X-band SAR, DLR): Eleven-day repeat; Staring Spotlight mode achieves 0.25 m azimuth resolution, enabling sub-pixel offset tracking on rapidly calving fronts. X-band loses coherence quickly over snow-covered surfaces, so best suited to bare-ice zones or short-interval pairs of one to three days via bistatic or pursuit-monostatic configurations.
- Landsat 8/9 OLI (optical, USGS/NASA): Not a SAR system, but optical feature tracking on cloud-free acquisitions provides independent velocity estimates at 15 m resolution. Sixteen-day repeat limits temporal resolution; cloud and polar night restrict coverage. Used primarily to cross-validate SAR-derived products in summer.
Two methods, one physical quantity
InSAR measures the phase difference between two SAR acquisitions taken from nearly identical orbital positions. Each full phase cycle represents half the radar wavelength of displacement along the line of sight, so Sentinel-1's C-band (5.6 cm wavelength) can theoretically resolve displacements of a few millimetres. In practice, ice-sheet applications use the along-track and range components together, or combine ascending and descending passes, to decompose motion into horizontal velocity. The method works well on slow-to-moderate glaciers, roughly below a few metres per day, where the ice surface remains coherent between passes.
Faster outlet glaciers, particularly tidewater systems that calve at rates of tens of metres per day, shear and fracture so severely that phase coherence is lost within the six-day Sentinel-1 window. There, pixel-offset tracking (also called intensity or speckle tracking) cross-correlates small image patches between acquisitions and measures the sub-pixel shift of surface texture. Precision is coarser, typically one-tenth to one-twentieth of a pixel, but the method tolerates decorrelation that would make InSAR unusable. The two approaches are therefore complementary rather than competing: InSAR for the slow interior, offset tracking for the dynamic margins.
What the velocity field actually reveals
A single velocity map is a snapshot. A time series is a diagnostic. Seasonal acceleration tied to surface meltwater reaching the bed through moulins is distinguishable from multi-year acceleration driven by terminus retreat and loss of buttressing. Jakobshavn Isbræ in Greenland, for example, has been documented accelerating from roughly 20 m per day in the 1990s to over 40 m per day in some seasons, a change captured in successive ERS, RADARSAT and Sentinel-1 analyses. The NASA MEaSUREs programme synthesises these records into annual and quarterly mosaics covering the Greenland and Antarctic ice sheets, and the ITS_LIVE project extends this to glaciers globally using Landsat and Sentinel-2 feature tracking.
Acceleration events are often the first observable signal of dynamic instability. A surge-type glacier may show a velocity increase of one to two orders of magnitude over weeks, with the fast-flow front propagating down-glacier in a pattern that InSAR time series can map at the scale of individual flow units. Subglacial hydrology changes, detectable as sudden velocity pulses that decay over days, are a separate phenomenon from surges but equally important for ice-dynamics modelling. Neither is visible in optical imagery until surface morphology changes substantially.
The reference products and their honest limits
The NASA MEaSUREs velocity mosaics are the community standard. Annual Greenland mosaics at 200 m resolution are publicly available; quarterly products exist for recent years. ITS_LIVE provides 120 m annual mosaics globally from 1985 onward, derived from Landsat feature tracking, with Sentinel-2 added for recent epochs. Both products carry per-pixel uncertainty estimates, which is important: errors are not uniform. Slow-moving interior regions where phase gradients are small have uncertainties of a few metres per year; fast outlet glaciers tracked by offset correlation can carry uncertainties of tens of metres per year.
Cloud and polar night do not affect SAR, which is its principal advantage over optical tracking. But SAR has its own failure modes. Temporal decorrelation from surface melt or wind-driven snow redistribution degrades coherence even at C-band within six days during summer. Steep valley walls cause geometric distortions (layover and shadow) that blank out data in narrow fjord glaciers. Atmospheric path-delay variations, particularly in coastal Greenland where water vapour gradients can be steep, introduce apparent displacement signals that must be modelled or filtered. No single acquisition pair should be interpreted without checking these artefacts.
Processing chain from raw SAR to velocity
For InSAR, the standard chain runs through co-registration of the image pair to sub-pixel accuracy, interferogram formation, filtering, phase unwrapping and conversion from line-of-sight displacement to surface-parallel flow assuming the ice moves parallel to the surface slope (the surface-parallel flow assumption, which breaks down near grounding lines and ice falls). Software such as ISCE, SNAP or ROI_PAC handles most steps; the unwrapping stage remains the most error-prone, particularly across fast-flowing shear margins where phase gradients exceed half a cycle per pixel.
Offset tracking pipelines cross-correlate image chips, typically 64 x 64 to 256 x 256 pixels, using normalised cross-correlation or phase correlation in the frequency domain. The output is a dense displacement field at the chip spacing chosen, commonly one-quarter to one-half the chip size. Post-processing filters outliers using median filters or consistency checks between overlapping chip windows. Final products are georeferenced to a polar stereographic grid and delivered as GeoTIFF or NetCDF with separate uncertainty layers.
Where the data feeds downstream
Velocity fields are boundary conditions and validation targets for ice-dynamics models such as ISSM, Elmer/Ice and PISM. A model constrained by observed velocities can invert for basal friction coefficients, which are otherwise unobservable, and then project future mass loss under different climate scenarios. This chain connects satellite measurements directly to sea-level rise projections used in IPCC assessments.
For operational users, the more immediate application is change detection. A government or infrastructure operator monitoring a glacier that threatens a downstream community, or a hydropower operator tracking a glacier that feeds a reservoir, needs to know whether the ice is accelerating. A quarterly velocity update with anomaly flagging is a more actionable product than a single mosaic. Satellize structures its analytics delivery around this kind of time-series monitoring rather than one-off snapshots, drawing on the same open Sentinel-1 archive that feeds the MEaSUREs programme but with client-specific area-of-interest processing and reporting cadence. The Tonga crop-estimation programme is a different domain, but the underlying workflow, regular ingestion of open-constellation data, anomaly detection against a baseline, structured reporting, is the same.
Minimum detectable velocity and practical thresholds
InSAR can detect displacements of a few centimetres over a six-day Sentinel-1 pair, corresponding to annual velocities of roughly 2 to 3 metres per year in favourable conditions. That is sufficient to characterise most of the slow-moving interior of the Greenland and Antarctic ice sheets, where velocities range from near-zero to a few hundred metres per year. Offset tracking with Sentinel-1 IW mode at 5 x 20 m resolution and a six-day baseline has a practical detection floor around 0.5 to 1 m per day, adequate for outlet glaciers but not for the slow interior.
TerraSAR-X Staring Spotlight pairs separated by one to three days push offset-tracking precision to sub-metre-per-day levels on fast outlet glaciers, at the cost of narrow swath and commercial tasking expense. The choice of sensor and method is therefore driven by the velocity regime of the target glacier, not by a single preferred platform. Any credible velocity product for a dynamic system like Sermeq Kujalleq (Jakobshavn) or Thwaites will combine InSAR for the slow flanks and offset tracking for the fast trunk, validated against independent optical feature tracking where cloud permits.
Typical figures
| Spatial resolution (InSAR, Sentinel-1 IW) | 5 x 20 m single-look; typically multi-looked to 40–100 m for velocity products |
| Spatial resolution (offset tracking, TerraSAR-X Staring Spotlight) | ~0.25 m azimuth; chip spacing typically 10–50 m in output velocity grid |
| Revisit interval | 6 days (Sentinel-1, single satellite); 3 days at high latitudes with two satellites; 14 days ALOS-2; 11 days TerraSAR-X |
| Radar frequency / wavelength | C-band 5.405 GHz / 5.6 cm (Sentinel-1); L-band 1.2 GHz / 23.6 cm (ALOS-2); X-band 9.65 GHz / 3.1 cm (TerraSAR-X) |
| Minimum detectable velocity (InSAR, 6-day pair) | ~2–3 m yr⁻¹ in coherent slow-ice zones; limited by atmospheric noise floor |
| Minimum detectable velocity (offset tracking, Sentinel-1 IW) | ~0.5–1 m day⁻¹ practical floor |
| Swath coverage | 250 km (Sentinel-1 IW); 350 km (ALOS-2 ScanSAR); 30 km (TerraSAR-X Stripmap) |
| Archive depth | Sentinel-1 from 2014; ALOS-2 from 2014; ERS/Envisat legacy SAR from 1992 (via ESA) |
| Reference velocity mosaic | NASA MEaSUREs annual/quarterly (200 m); ITS_LIVE annual global (120 m) from 1985 |
| Delivery formats | GeoTIFF (velocity magnitude and direction), NetCDF with uncertainty layers, polar stereographic projection (EPSG:3413 Greenland / EPSG:3031 Antarctic) |
Analytics Satellize can run
| Baseline velocity mosaic | InSAR time-series stack or offset tracking over user-defined area of interest, averaged across multiple pairs to reduce atmospheric noise | GeoTIFF velocity magnitude and azimuth layers at 50–200 m resolution, with per-pixel uncertainty estimate |
| Quarterly velocity change map | Differencing of consecutive seasonal mosaics; anomalies flagged where change exceeds two standard deviations of the historical baseline | GIS layer with anomaly polygons and tabulated statistics per glacier or flow unit; PDF summary report |
| Surge onset detection alert | Automated monitoring of velocity time series for step-change signatures consistent with surge initiation; threshold set from glacier-specific historical variance | Email or API alert with map attachment within 48 hours of qualifying acquisition pair |
| Outlet glacier acceleration profile | Longitudinal velocity profiles extracted along glacier centrelines; rate-of-change analysis referenced to terminus position | Time-series chart and tabulated centreline velocity data; updated quarterly or on request |
| Ice-flux cross-section estimate | Velocity field multiplied by ice thickness from published BedMachine or BEDMAP datasets across user-specified gate; propagated uncertainty reported | Annual flux estimate in Gt yr⁻¹ or km³ yr⁻¹ with uncertainty bounds; formatted for input to ice-dynamics models |
| Coherence-loss mapping | Interferometric coherence computed for each pair; low-coherence zones indicate fast flow, melt or surface disruption | Coherence raster overlaid on velocity product; used to flag where offset tracking replaces InSAR in the processing chain |
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.