Glacial isostatic adjustment and ice-unloading uplift measurement
As ice sheets shed mass, the mantle rebounds upward at millimetres to centimetres per year. Sentinel-1 InSAR stacking and continuous GNSS now resolve that signal, but disentangling ancient GIA from present-day elastic rebound demands careful modelling.
Sensors
- Sentinel-1 A/B (C-band SAR): 5 m x 20 m ground range resolution in Interferometric Wide Swath mode, 6-day repeat at mid-latitudes when both satellites are active. InSAR stacking over multi-year archives resolves line-of-sight displacement rates of 1-2 mm/yr in coherent terrain. Decorrelation over snow and ice limits usable pixels to bedrock outcrops and moraine fields at glacier margins.
- TanDEM-X (X-band SAR): Bistatic X-band pair with a baseline adjustable to centimetres. Shorter wavelength gives higher sensitivity to small displacements but also higher decorrelation over wet surfaces. Published studies have used TanDEM-X repeat-pass stacks at Patagonian ice-field margins to extract uplift rates where Sentinel-1 coherence is marginal.
- Continuous GNSS (SONEL / UNAVCO networks): Ground-truth for InSAR: long-running bedrock GNSS stations in Greenland and Patagonia achieve vertical precision of 0.5-1 mm/yr after multi-year averaging. SONEL archives tide-gauge-collocated stations; UNAVCO/GAGE archives polar and subpolar networks. Both are open access. Station density is sparse, which is precisely why InSAR spatial coverage matters.
- ICESat-2 (ATL06/ATL11 products): Photon-counting lidar with a 91-day exact repeat and stated surface-elevation accuracy of roughly 3 cm over flat ice. Differencing repeat passes quantifies ice-surface lowering independently of InSAR, providing the mass-change input needed to separate elastic rebound from long-period GIA in combined inversions.
Why the ground moves after the ice has gone
Glacial isostatic adjustment (GIA) is the Earth's slow response to being relieved of ice that, in some regions, was kilometres thick during the Last Glacial Maximum roughly 20,000 years ago. The mantle is not a rigid solid. It behaves as a viscoelastic fluid on timescales of thousands of years, and the lithosphere above it is still rising in Scandinavia and northern Canada at rates of up to 10 mm/yr, long after the ice itself disappeared.
Present-day ice loss adds a second, faster signal on top. When a glacier thins by tens of metres over a decade, the elastic response of the crust is nearly instantaneous on geological timescales, producing uplift that can reach 20-30 mm/yr at the margins of rapidly retreating Patagonian ice fields. That elastic signal and the ancient viscous GIA signal are superimposed in every geodetic observation. Separating them requires either a well-constrained GIA forward model (ICE-6G, W12 and similar) or a joint inversion that uses both the surface deformation field and independent ice-mass-change data.
What InSAR stacking can and cannot resolve
A single interferogram spanning 6 or 12 days captures displacement but is dominated by tropospheric noise, which can reach 10-20 mm of apparent range change in a single pass over mountainous terrain. Stacking dozens or hundreds of interferograms over several years suppresses that noise by roughly the square root of the number of pairs, pushing the detectable rate signal toward 1-2 mm/yr in favourable conditions. That is the regime where GIA and elastic rebound become visible.
The hard constraint is coherence. C-band Sentinel-1 loses coherence over snow, ice and dense vegetation within days. Published results from Greenland margins, such as work using Sentinel-1 ascending and descending stacks over the Kangerlussuaq and Jakobshavn forelands, achieve coherent pixels only on exposed bedrock. Coverage is therefore patchy: you get uplift rates at the exact locations where rock outcrops, not a continuous field. X-band TanDEM-X has shorter wavelength, which improves sensitivity to small motions but worsens decorrelation over wet surfaces. Neither sensor gives you the fjord floor or the glacier surface itself.
Atmospheric correction is the other limiting factor. Generic empirical corrections based on elevation (phase-elevation correlation removal) help, but over coastal Greenland and Patagonia, where weather systems are rapid and spatially variable, residual tropospheric noise remains the dominant error source in any single interferogram. ERA5 reanalysis-based corrections reduce this but do not eliminate it.
Published results from Greenland and Patagonia
Studies using Sentinel-1 stacks over the Greenland ice-sheet margin have reported bedrock uplift rates of 5-20 mm/yr at sites within 50 km of the calving front, consistent with elastic rebound driven by accelerating mass loss since the early 2000s. The spatial pattern follows the geometry of recent thinning rather than the broad bowl expected from ancient GIA alone, which is one of the cleaner observational arguments for separating the two components.
In Patagonia, the Southern and Northern Patagonian Ice Fields sit on thin, warm lithosphere with a low-viscosity mantle. Published GNSS and InSAR results report elastic uplift rates of 30-40 mm/yr at some bedrock stations near the ice-field margins, among the highest observed anywhere on Earth outside Antarctica. TanDEM-X bistatic acquisitions have been used here precisely because the rapid motion and high coherence loss over wet rock make standard repeat-pass C-band stacking difficult. The Patagonian case is instructive about limits: even at 30 mm/yr, you need at least two to three years of data to confidently separate signal from atmospheric noise at the millimetre level.
The separation problem: elastic rebound versus ancient GIA
GIA models predict the long-wavelength, slowly varying component of uplift driven by mantle flow responding to ice loads that disappeared millennia ago. The elastic component responds to mass changes over years to decades and has a shorter spatial wavelength tied to the geometry of current ice loss. In practice, both signals are present simultaneously, and neither GNSS nor InSAR can distinguish them without additional information.
The standard approach combines a GIA forward model with an ice-loading history derived from satellite altimetry (ICESat-2, CryoSat-2) and gravimetry (GRACE-FO), then attributes the residual geodetic signal to present-day elastic rebound. Uncertainty in mantle viscosity structure propagates directly into the GIA prediction, and that uncertainty is substantial in regions like West Antarctica and Patagonia where the lithosphere is thin and the mantle is hot. Published estimates of GIA in West Antarctica, for instance, span a range of tens of millimetres per year depending on the viscosity model assumed. That range is not a modelling failure; it reflects genuine physical uncertainty that no amount of InSAR data alone can resolve.
What a monitoring programme looks like in practice
A credible GIA and elastic-rebound monitoring workflow begins with a multi-year Sentinel-1 archive stack, processed through a time-series algorithm such as SBAS or PS-InSAR, with atmospheric correction applied using ERA5 or a similar reanalysis product. The output is a map of line-of-sight velocity at coherent pixels, typically with uncertainties of 1-3 mm/yr after stacking. Ascending and descending geometries are combined to decompose vertical and east-west horizontal components; north-south sensitivity from C-band InSAR alone is poor.
GNSS station data from SONEL or UNAVCO are then used to validate and anchor the InSAR field. Where GNSS and InSAR agree within their stated uncertainties, confidence in the spatial interpolation between sparse GNSS points is higher. Where they diverge, the cause is usually either atmospheric noise in the InSAR stack or a local effect at the GNSS monument (frost heave, monument instability) that is not representative of bedrock motion.
ICESat-2 elevation differencing provides the ice-mass-change input for the elastic-rebound model. The combination of all three data streams, InSAR spatial coverage, GNSS long-term stability, and ICESat-2 mass-change forcing, is what makes a physically interpretable result rather than a geodetic curiosity. Satellize structures this kind of multi-sensor fusion for clients who need defensible numbers for sea-level-rise attribution or infrastructure planning in deglaciated terrain, drawing on the same open-constellation processing approach used in the Tonga crop-estimation programme.
Honest limits and what they mean for buyers
If your area of interest is covered by active ice or dense temperate forest, InSAR stacking will not give you useful uplift rates. You are dependent on GNSS, and if the network is sparse, spatial interpolation introduces large uncertainties. Greenland's west coast has reasonable GNSS coverage; the east coast and much of Antarctica do not.
Rates below about 2 mm/yr are difficult to confirm from InSAR alone without five or more years of archive data, and even then atmospheric noise can masquerade as a coherent signal if the correction is imperfect. For infrastructure decisions, such as port design in a rebounding Arctic fjord, combining InSAR with at least one collocated continuous GNSS station is not optional; it is the minimum defensible standard. The physics is well understood. The data are increasingly available. The remaining challenge is the modelling, and anyone who sells you a GIA map without an explicit uncertainty estimate for mantle viscosity assumptions is omitting the most important number.
Typical figures
| Sentinel-1 spatial resolution (IW mode) | 5 m x 20 m (range x azimuth); multi-looked to ~20-80 m for InSAR stacking |
| Sentinel-1 revisit (dual satellite) | 6 days at mid-latitudes; 12 days with single satellite |
| TanDEM-X bistatic resolution | ~3 m (SpotLight) to ~12 m (StripMap); repeat-pass intervals vary by tasking |
| Minimum detectable uplift rate (InSAR stack) | ~1-2 mm/yr after multi-year stacking in coherent terrain; ~0.5 mm/yr with GNSS anchor |
| ICESat-2 elevation accuracy | ~3 cm over flat ice surfaces (ATL06 product); 91-day exact repeat |
| GNSS vertical precision (long-term) | 0.5-1 mm/yr after 3+ years of continuous observation |
| Sentinel-1 archive depth | From 2014 (Sentinel-1A launch); full open archive via Copernicus Data Space |
| Coherence constraint | Coherent pixels on exposed bedrock only; ice, snow and dense vegetation decorrelate within days at C-band |
| Atmospheric noise (single interferogram) | 10-20 mm apparent range change; reduced to ~2-4 mm residual after ERA5-based correction and stacking |
| Delivery formats | GeoTIFF velocity maps, CSV time-series at GNSS-collocated points, NetCDF stacks, PDF technical report |
Analytics Satellize can run
| Bedrock uplift velocity map | SBAS or PS-InSAR time-series inversion of multi-year Sentinel-1 archive, ascending and descending geometries combined for vertical decomposition | GeoTIFF of vertical and east-west velocity fields with per-pixel uncertainty, clipped to coherent pixels |
| Elastic rebound rate estimate | Residual after subtracting GIA forward-model prediction (ICE-6G or equivalent) from InSAR+GNSS combined velocity field | Gridded NetCDF of elastic component with uncertainty bounds derived from GIA model ensemble spread |
| GNSS-InSAR consistency assessment | Point-by-point comparison of InSAR line-of-sight rates projected to GNSS geometry against SONEL/UNAVCO station velocities | Tabular report of agreement statistics and flagged outliers, with commentary on likely causes |
| Ice-mass-change forcing time-series | ICESat-2 ATL11 repeat-track elevation differencing over glacier and ice-field surface, converted to mass change using firn-compaction correction | Annual mass-change time-series per defined ice basin, for use as elastic-rebound model input |
| Atmospheric noise characterisation | ERA5 reanalysis tropospheric delay correction applied per interferogram, with before/after phase-elevation correlation statistics | Correction quality report and corrected interferogram stack in ISCE or SNAP-compatible format |
| Infrastructure site-specific uplift report | Extraction of InSAR time-series and GNSS velocity at user-specified coordinates, with seasonal signal separation and trend confidence interval | PDF technical report suitable for engineering or planning input, with stated assumptions and uncertainty ranges |
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.