Permafrost thaw settlement and active-layer dynamics
InSAR time series over permafrost terrain separates reversible seasonal frost heave from irreversible thermokarst subsidence, revealing where ground ice is being lost for good. Rates, extents, and the sensor trade-offs that determine what you can actually measure.
Sensors
- ALOS-2 PALSAR-2 (L-band, 1.27 GHz): 23 cm wavelength penetrates dense tundra vegetation canopy, maintaining coherence over intervals of 14 to 46 days where C-band loses it entirely. Stripmap mode delivers 3 m single-look range resolution; ScanSAR covers 350 km swaths at 100 m. The long wavelength makes it the workhorse for vegetated Siberian and Alaskan tundra.
- Sentinel-1 A/B (C-band, 5.405 GHz): 6 or 12 day repeat (single satellite after Sentinel-1B failure in 2021, now 12 days over most Arctic areas). IW mode gives 5 x 20 m resolution across a 250 km swath. Coherence drops sharply over tall shrub tundra between passes, but bare ground, gravel roads, and sparse low-Arctic vegetation remain workable. Free and open archive from 2014.
- TerraSAR-X / TanDEM-X (X-band, 9.65 GHz): 1 m Spotlight and 3 m Stripmap modes resolve individual infrastructure pads and building footprints. Coherence is excellent over bare gravel pads and engineering structures but degrades rapidly over any vegetated surface. Useful for monitoring specific assets such as pipeline corridors and airstrips rather than landscape-scale mapping.
- NISAR (L-band + S-band, launch expected 2025): NASA-ISRO joint mission designed partly for permafrost science. 12-day repeat globally, dual-frequency. L-band will extend PALSAR-era time series; S-band adds a shorter wavelength for comparison. Published mission requirements include a minimum detectable displacement of approximately 1 cm per year over vegetated surfaces.
Two signals buried in the same measurement
Every InSAR time series over permafrost contains two physically distinct processes that must be separated before either means anything. The first is elastic: the active layer freezes each autumn, expands by roughly 9% as water becomes ice, and the surface heaves upward by several centimetres. It thaws the following spring and the ground returns, approximately, to where it started. This cycle repeats annually and is not, by itself, a problem.
The second process is not elastic. Where ground ice in the permafrost body melts, the volume loss is permanent. Ice-rich silts and clays can contain 50 to 80% ice by volume; when that ice goes, the surface drops and does not come back. Published ALOS PALSAR studies over the Tibetan Plateau have documented long-term subsidence rates of 1 to 3 cm per year in ice-rich areas, with localised thermokarst features showing episodic collapse of tens of centimetres over a single thaw season. Distinguishing the reversible seasonal signal from this permanent trend requires time series spanning at least two to three annual cycles, and ideally a decade or more.
What the wavelength actually decides
The central practical problem in permafrost InSAR is temporal decorrelation. Between two SAR acquisitions, vegetation moves, soil moisture changes, and snow redistributes. Each of these randomises the phase of returning radar echoes, destroying the interferometric signal. C-band (Sentinel-1, ~5.6 cm wavelength) is highly susceptible. Over tall shrub tundra or boreal forest edges, coherence at 12-day repeat can fall below usable thresholds, particularly in summer when the active layer is thawing and vegetation is growing. Practical C-band InSAR for permafrost is largely limited to winter acquisitions over low-Arctic or high-Arctic terrain, bare ground, or engineered surfaces.
L-band (ALOS-2, ~23 cm) penetrates the canopy and scatters from the soil surface beneath, maintaining coherence over much longer intervals and across a wider range of vegetation types. This is not a marginal advantage; it is the difference between a usable signal and noise. Published studies using ALOS PALSAR over Alaskan tundra and Siberian lowlands have achieved coherence values above 0.3 at 46-day repeat over areas where C-band produces nothing usable at 12 days. The trade-off is coarser resolution in standard modes and a less dense time series than Sentinel-1 can provide where C-band does work.
X-band sits at the other extreme. Its very short wavelength means it decorrelates almost immediately over any vegetation, but over bare gravel pads, concrete, or metal structures it maintains extraordinary coherence and can detect millimetre-scale displacements. For a pipeline operator monitoring a specific compressor station pad, TerraSAR-X Spotlight is the right tool. For mapping a 10,000 km² watershed, it is not.
Processing a permafrost time series: the honest difficulty
Small Baseline Subset (SBAS) and Persistent Scatterer InSAR (PS-InSAR) are the two main time-series methods applied to permafrost. SBAS connects interferogram pairs with short temporal and spatial baselines to maximise coherence; it works well over distributed targets such as tundra soil. PS-InSAR identifies point-like stable reflectors, which are abundant over infrastructure but sparse over open tundra. In practice, permafrost studies often use a hybrid approach, applying SBAS over natural terrain and PS analysis over engineering structures within the same scene.
The seasonal signal must be modelled and removed before the long-term trend is interpretable. A sinusoidal model fitted to the annual freeze-thaw cycle is the standard approach, but permafrost terrain complicates this because the timing and amplitude of heave vary spatially with soil texture, ice content, drainage, and snow cover. Residuals after seasonal modelling carry real geophysical information about spatial heterogeneity in active-layer thickness and ground-ice distribution. They are not simply noise.
Atmospheric delay is a persistent source of error. The Arctic atmosphere is drier than mid-latitudes, which reduces tropospheric noise, but stratified delays correlated with topography can mimic real displacement signals. Correction using ERA5 reanalysis or GACOS (Generic Atmospheric Correction Online Service) is standard practice but imperfect, particularly over complex terrain such as the Tibetan Plateau where topographic gradients are steep.
What the archive already shows
The scientific record is now substantial enough to give honest expectations. ALOS PALSAR studies over the Lena River delta and Yamal Peninsula in Siberia documented subsidence rates of 1 to 5 cm per year in thermokarst-active areas over the 2006 to 2011 observation window. Sentinel-1 studies over Svalbard and the Qinghai-Tibet Railway corridor have mapped seasonal heave amplitudes of 1 to 4 cm and identified long-term subsidence of 0.5 to 2 cm per year over ice-rich terrain, consistent with independent ground-based measurements.
One honest limit deserves emphasis: InSAR measures displacement in the satellite line-of-sight direction, not vertical displacement directly. Converting line-of-sight to vertical requires either ascending and descending pass combinations or an assumption that horizontal motion is negligible. Over flat tundra, the vertical assumption is usually acceptable. Over slopes with active solifluction or near thermokarst lake margins where lateral collapse is occurring, it is not, and the error can be significant.
Infrastructure risk is the commercial driver
The scientific interest in permafrost dynamics is matched by a very practical engineering problem. Roads, pipelines, airstrips, and buildings across Arctic Russia, Alaska, Canada, and the Tibetan Plateau are founded on permafrost that is warming. The Trans-Alaska Pipeline System traverses approximately 675 km of permafrost terrain. The Qinghai-Tibet Railway crosses more than 550 km of it. Differential settlement of even a few centimetres per year can compromise structural integrity, and the spatial pattern of settlement matters as much as the rate: a uniform 2 cm per year is manageable; a 5 cm differential across a 10 m span is not.
InSAR time series can identify which sections of a linear asset are subsiding fastest and flag anomalous acceleration before it becomes a maintenance emergency. This is where Satellize's analytics pipeline, running on open Sentinel-1 archives supplemented by commercial ALOS-2 tasking on client licence, can deliver structured GIS outputs to engineering teams who have no interest in processing SAR data themselves.
NISAR and the next decade of monitoring
The upcoming NISAR mission changes the resource equation substantially. Free, open, global, 12-day L-band coverage will do for permafrost InSAR what Sentinel-1 did for European ground motion: make systematic monitoring affordable at continental scale. Published mission science requirements specify surface displacement sensitivity of around 1 cm per year over vegetated surfaces, which is precisely the threshold needed to detect early-stage thermokarst development before it becomes visible in optical imagery.
The archive question is the remaining gap. NISAR will build a new time series from its launch date, but understanding whether observed subsidence is accelerating relative to historical rates requires comparison with ALOS PALSAR (2006 to 2011) and ALOS-2 PALSAR-2 (2014 to present) data. Maintaining methodological consistency across sensors with different wavelengths, incidence angles, and noise floors is non-trivial. It is, however, a solved problem in the published literature, and the methods are well documented.
Typical figures
| Typical spatial resolution (InSAR product) | 20 to 100 m for landscape-scale SBAS products; 3 to 10 m for infrastructure-focused PS analysis; 1 m Spotlight for individual structure monitoring (TerraSAR-X) |
| Minimum detectable displacement rate | ~1 mm per year over stable coherent surfaces with a dense time series; ~5 to 10 mm per year is a realistic expectation over tundra with seasonal noise |
| Revisit interval | 12 days (Sentinel-1 single satellite); 14 days (ALOS-2 PALSAR-2 standard); 11 days (TerraSAR-X); 12 days (NISAR, planned) |
| Radar frequency / wavelength | L-band: 1.2 to 1.3 GHz / ~23 cm; C-band: 5.4 GHz / ~5.6 cm; X-band: 9.65 GHz / ~3.1 cm |
| Swath width | 250 km (Sentinel-1 IW); 70 km (ALOS-2 Stripmap); 350 km (ALOS-2 ScanSAR); 30 km (TerraSAR-X Stripmap) |
| Archive depth | Sentinel-1 from 2014; ALOS PALSAR from 2006 to 2011; ALOS-2 PALSAR-2 from 2014; ERS/Envisat from 1992 to 2012 (usable for long-baseline studies) |
| Seasonal signal amplitude (published range) | 1 to 4 cm peak-to-peak heave over ice-rich tundra; up to 10 cm in high-ice-content silts |
| Long-term subsidence rates (published range) | 0.5 to 5 cm per year in actively thawing thermokarst terrain; near-zero over stable permafrost |
| Delivery formats | GeoTIFF displacement rasters, shapefiles of subsidence anomaly polygons, CSV time-series per asset point, PDF engineering summary report |
| Latency from acquisition to product | Typically 3 to 7 days for routine monitoring products; 24 to 48 hours for priority tasking on commercial sensors |
Analytics Satellize can run
| Seasonal heave / subsidence decomposition map | SBAS InSAR time-series analysis with sinusoidal seasonal model fitting; ascending and descending pass combination for vertical/horizontal decomposition | Annual GeoTIFF rasters separating elastic heave amplitude from residual long-term trend, at 20 to 100 m posting |
| Thermokarst hotspot detection and ranking | Anomaly detection on long-term subsidence trend layer; spatial clustering of pixels exceeding a defined rate threshold (e.g. >2 cm per year) | Ranked shapefile of thermokarst-active polygons with subsidence rate, area, and confidence score; updated annually |
| Linear infrastructure differential settlement profile | PS-InSAR along pipeline, road, or railway corridor; displacement extracted at regular intervals and differenced to compute settlement gradient | CSV and PDF engineering report showing displacement rate per kilometre of asset, flagging sections exceeding client-defined differential thresholds |
| Active-layer thickness proxy from seasonal amplitude | Correlation of InSAR heave amplitude with Stefan equation model of freeze-thaw depth; calibrated against available borehole or CALM (Circumpolar Active Layer Monitoring) network data | Gridded active-layer thickness estimate with uncertainty bounds; GeoTIFF format |
| Multi-epoch change detection (ALOS to Sentinel-1 to NISAR) | Cross-sensor time-series stitching with common reference frame; inter-sensor bias correction using stable reference points | Long-baseline (2006 to present) subsidence trend stack showing acceleration or deceleration over the full observation period |
| Atmospheric correction quality assessment | ERA5 or GACOS tropospheric delay correction applied to interferogram stack; residual variance assessed against DEM correlation | Correction quality flag per scene in the time series; included as metadata layer in all displacement products |
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.