Glacier surge and rapid flow velocity mapping by SAR offset tracking

Surging glaciers can accelerate from metres to kilometres per year in weeks, breaking conventional InSAR. SAR amplitude offset tracking and multiple-aperture InSAR recover full 2-D velocity fields where phase-based methods fail entirely.

Sensors

Sentinel-1 A/C (C-band SAR, ESA): 6- or 12-day repeat in IW mode; 5 × 20 m resolution. Amplitude offset tracking resolves displacements from roughly 1/20th of the pixel spacing, giving a practical velocity floor around 0.1–0.3 m/day at 12-day repeat. Archive from 2014; free and open.
ALOS-2 PALSAR-2 (L-band SAR, JAXA): L-band (23.6 cm wavelength) penetrates surface snow better than C-band, preserving coherence longer into a surge. Multiple-aperture InSAR (MAI) splits the aperture to yield along-track displacements with precision roughly 1/100th of the range resolution, around 0.03–0.1 m per acquisition pair. Repeat is 14 days in standard mode.
Landsat 8/9 OLI (optical, USGS/NASA): 15 m panchromatic band enables sub-pixel feature tracking (normalised cross-correlation on crevasse patterns or surface texture). Velocity precision typically 3–10 m over a 16-day pair. Global free archive from 2013 (Landsat 8) and 2021 (Landsat 9); cloud cover is the binding constraint at high latitudes.
Planet SuperDove (optical, commercial): 3 m resolution, near-daily revisit globally. High cadence reduces the temporal baseline, which matters when a surge accelerates rapidly. Feature-tracking precision scales with resolution; at 3 m pixels, sub-pixel offsets of 1–2 m per day are feasible over short baselines, though cloud and solar illumination still limit coverage in polar winters.

Why conventional InSAR breaks down at surge velocities

Standard differential InSAR measures the phase difference between two SAR acquisitions. The method works elegantly when surface displacement between passes is a small fraction of the radar wavelength: Sentinel-1's C-band wavelength is 5.6 cm, so displacements larger than roughly 2–3 cm per day at a 6-day repeat start to cause phase aliasing, and anything beyond about 10–15 cm per day produces complete decorrelation in the interferogram. A surging glacier routinely moves 1–10 m per day. The phase is simply gone.

Amplitude offset tracking sidesteps the phase problem entirely. Instead of comparing phase, it cross-correlates the intensity pattern of speckle or surface texture between two SAR images, finding the peak of the correlation surface to estimate the shift. The measurable displacement range is bounded below by roughly 1/20th of the pixel spacing (noise floor) and above only by the requirement that the surface texture remains recognisable between acquisitions. For Sentinel-1 IW data at 5 × 20 m, that gives a practical ceiling of tens of metres per day, which comfortably covers all known surge velocities.

Multiple-aperture InSAR: recovering the along-track component

Standard offset tracking measures range (roughly cross-track) and azimuth (along-track) displacements separately, but azimuth precision is weaker because azimuth pixel spacing is coarser and the cross-correlation peak is broader in that direction. Multiple-aperture InSAR (MAI) addresses this by splitting the synthetic aperture into forward-looking and backward-looking sub-apertures and forming an interferogram between them. The result is an along-track phase measurement with sensitivity roughly equivalent to one-quarter of the azimuth resolution, substantially better than amplitude offset tracking alone in the along-track direction.

For ALOS-2 PALSAR-2 in Fine Beam mode (approximately 10 m azimuth resolution), published studies report MAI along-track precision of around 0.03–0.08 m per acquisition pair under good coherence conditions. Combining range offset tracking with MAI along-track measurements gives a 2-D surface velocity vector without needing a third acquisition geometry. The L-band wavelength also helps: longer wavelengths maintain coherence over rough, crevassed ice surfaces for weeks rather than days.

What Karakoram and Svalbard surges have revealed

The Karakoram is unusual in hosting a disproportionate number of the world's actively surging glaciers. Studies using Sentinel-1 amplitude offset tracking on glaciers including Hispar, Skamri and Kyagar have documented surge velocities reaching 15–25 m per day at peak flow, with the surge front propagating down-glacier at detectable rates over months. The kinked medial moraines visible in optical imagery confirm what the SAR velocity maps quantify: mass is being redistributed faster than the glacier's long-term balance can absorb.

Svalbard provides a different regime. Surges there tend to be slower (typically 1–5 m per day at peak) and longer in duration, sometimes persisting for several years. Sentinel-1's 6-day repeat and free archive have made it possible to track the full lifecycle of surges on glaciers such as Nathorstbreen and Negribreen at temporal resolution that was simply unavailable before 2014. Published work using these data has shown that the surge front can be tracked week by week as a sharp velocity gradient migrating down-glacier, which is directly relevant to proglacial lake hazard assessment.

Optical feature tracking on Landsat pairs complements the SAR work in summer when snow-free surfaces offer strong texture contrast. The 15 m panchromatic band on Landsat 8 and 9 yields velocity maps comparable in spatial coverage to Sentinel-1 offset tracking, though at coarser precision. The two datasets are often run together: SAR provides all-weather, all-season coverage; optical provides validation and higher spatial detail in clear conditions.

Precision limits and where the methods genuinely struggle

Amplitude offset tracking is not a precision geodetic tool in the way that InSAR is for slow motion. The noise floor depends on image co-registration quality, surface coherence and the chosen correlation window size. Larger windows improve signal-to-noise but blur spatial detail. A 64 × 64 pixel window on Sentinel-1 IW data (roughly 320 × 1280 m on the ground) is common in published work; it will miss the sharp velocity gradients at a surge front. Smaller windows reduce blur but increase noise. There is no free lunch.

Seasonal snow accumulation resets surface texture, which can break the cross-correlation entirely in winter. Fresh snowfall between acquisitions adds a displacement signal unrelated to ice flow. Steep valley walls cause layover and shadow in SAR geometry, masking the glacier margins precisely where shear zones are most dynamic. Planet SuperDove's high cadence helps with the temporal aliasing problem, but optical sensors go dark in polar night and are frequently obscured by cloud in maritime climates like Svalbard. Any honest velocity product should carry an uncertainty layer, not just a velocity magnitude.

From velocity maps to hazard numbers

A surge velocity map on its own is scientifically interesting but operationally incomplete. The downstream questions are: is the glacier terminus advancing toward a proglacial lake? Is ice flux through a cross-section increasing in a way that suggests a calving event or ice-dammed lake formation? Answering these requires combining the velocity field with a bed topography model (typically from airborne radar sounding or published datasets such as BedMachine) and a surface DEM to compute volumetric flux. The velocity map is the input; the hazard number is the output.

Satellize processes Sentinel-1 and Landsat time-series for surge monitoring as part of its open-constellation analytics work. The Tonga crop-estimation programme is a different domain, but the underlying pipeline for time-series offset tracking on open SAR data is the same infrastructure. For glacier clients, the deliverable is typically a quarterly velocity mosaic with uncertainty bounds, flagged against a surge-onset threshold, alongside a GIS layer of the surge front position through time.

Choosing the right sensor pair for a given glacier

The decision tree is straightforward once you know the expected velocity range. For velocities above roughly 2 m per day, Sentinel-1 amplitude offset tracking is the workhorse: free, frequent, and well-validated in the published literature. For velocities in the 0.1–2 m per day range (slow surges, or the quiescent phase of a recovering glacier), MAI on ALOS-2 or careful InSAR on Sentinel-1 with short baselines becomes necessary. Below about 0.05 m per day, you are in the territory of conventional InSAR and the sibling page on ice sheet grounding line migration.

High-cadence Planet optical data is most valuable during the surge onset, when the acceleration is rapid and the precise timing of the velocity peak matters for hazard warning. A 1-day revisit can resolve the acceleration curve in ways that a 6- or 12-day SAR repeat cannot. The cost is cloud and darkness. A monitoring system that combines all three, Sentinel-1 for all-weather continuity, Landsat for free optical validation, and Planet for burst-phase resolution, is more defensible than any single-sensor approach.

Typical figures

SAR spatial resolution (Sentinel-1 IW)	5 × 20 m (range × azimuth); offset-tracking output typically resampled to 40–100 m grid
SAR spatial resolution (ALOS-2 Fine Beam)	~3–10 m depending on mode; MAI output typically 20–50 m grid
Optical resolution (Landsat 8/9 pan)	15 m; feature-tracking output typically 300–500 m grid
Optical resolution (Planet SuperDove)	3 m; feature-tracking output typically 30–60 m grid
Revisit period	Sentinel-1: 6–12 days (polar regions often 6-day); ALOS-2: 14 days; Landsat: 16 days; Planet: 1 day (cloud-permitting)
Velocity detection floor (Sentinel-1 offset tracking)	~0.1–0.3 m/day at 12-day repeat; dependent on window size and coherence
Velocity detection ceiling	No hard ceiling for offset tracking; practically limited by surface texture survival, typically >50 m/day measurable
MAI along-track precision (ALOS-2)	~0.03–0.08 m per pair under good coherence; degrades over heavily crevassed ice
Archive depth	Sentinel-1: from 2014 (free); Landsat: from 1972 (free); ALOS-2: from 2014 (licensed); Planet: from ~2016 (commercial)
Key limiting factors	Cloud (optical); polar night (optical); fresh snowfall resetting texture; SAR layover in steep terrain; temporal decorrelation on wet ice in summer

Analytics Satellize can run

Surge velocity mosaic	SAR amplitude offset tracking (normalised cross-correlation on Sentinel-1 IW SLC pairs)	Quarterly or monthly GeoTIFF velocity magnitude and direction rasters with per-pixel uncertainty layer
Surge front position time-series	Velocity gradient detection on successive offset-tracking maps; front defined as leading edge of anomalous velocity zone	GIS polyline shapefile of surge front position at each epoch, with advance rate in m/day
Along-track velocity component (MAI)	Multiple-aperture InSAR on ALOS-2 PALSAR-2 SLC pairs; combined with range offset for 2-D vector	GeoTIFF 2-D velocity vector field; supplied as separate range and azimuth components for downstream flux modelling
Optical feature-tracking velocity map	Normalised cross-correlation on Landsat 8/9 panchromatic or Planet SuperDove image pairs	Seasonal velocity raster at 30–300 m grid; used as independent validation of SAR-derived products
Surge onset alert	Threshold detection on rolling 30-day velocity anomaly relative to multi-year quiescent baseline	Email or API alert with velocity map attachment when velocity exceeds configurable threshold
Ice flux cross-section time-series	Velocity field integrated over glacier cross-section using published surface DEM and ice thickness estimate	Tabular flux estimate (m³/day) at user-defined gate, with uncertainty range; suitable for proglacial lake hazard assessment

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.