Permafrost active-layer dynamics from InSAR surface deformation

Repeat-pass SAR interferometry detects the seasonal heave and subsidence of permafrost active layers at centimetre scale, revealing where frozen ground is losing structural integrity and releasing stored carbon.

Sensors

Sentinel-1 A/B (C-band, 5.6 cm wavelength): Interferometric Wide Swath mode delivers 5 × 20 m resolution across a 250 km swath with 6-day repeat at mid-latitudes; polar orbits give denser revisit above 60°N, which matters for active-layer monitoring. Free and open archive from 2014. C-band coherence degrades quickly over dense vegetation and wet surfaces, limiting usable interferograms in boreal tundra during peak summer.
ALOS-2 PALSAR-2 (L-band, 23.6 cm wavelength): L-band penetrates vegetation canopy and maintains coherence over vegetated tundra far better than C-band, making it the preferred frequency for SBAS or PS-InSAR stacks in shrub-covered terrain. Fine-beam single-polarisation mode achieves 3 m resolution; standard repeat is 14 days. Data access requires a JAXA research agreement or commercial licence.
RADARSAT Constellation Mission (C-band): Three-satellite Canadian constellation achieves daily revisit at high latitudes, which is operationally significant for monitoring rapid thaw events or post-disturbance subsidence. Medium-resolution mode gives 16 m; compact polarimetry modes add surface-roughness information useful for distinguishing water-saturated from dry active-layer conditions.
TerraSAR-X / TanDEM-X (X-band, 3.1 cm wavelength): Stripmap mode at 3 m resolution supports very high spatial detail over infrastructure corridors such as pipeline rights-of-way or road embankments. The 11-day repeat and commercial tasking allow targeted acquisition campaigns. X-band is the most sensitive to small displacements in theory, but also the most vulnerable to temporal decorrelation over bare or wet ground.

What the active layer actually does to the ground surface

Permafrost is perennially frozen ground, but the layer above it, the active layer, thaws each summer and refreezes each autumn. As ice melts, the soil consolidates and the surface drops. As it refreezes, ice formation expands the volume and the surface rises. In undisturbed tundra this seasonal cycle is roughly reversible, with vertical displacements typically in the range of 1 to 5 cm, though values above 10 cm have been measured in ice-rich silts.

When permafrost warms over years or decades, the active layer deepens and the thaw front reaches ground-ice bodies that were previously stable. The melt of that excess ice is not reversible on any human timescale. The result is thermokarst: irregular, often rapid subsidence that can exceed 30 cm in a single season at ice-wedge polygon sites. It is this irreversible signal, superimposed on the seasonal cycle, that infrastructure engineers and carbon scientists most need to detect early.

Phase shifts and what they reveal

InSAR works by comparing the phase of radar echoes from two passes over the same area. A displacement of half the radar wavelength toward or away from the satellite shifts the phase by a full cycle. At Sentinel-1's C-band wavelength of 5.6 cm, the theoretical line-of-sight sensitivity is around 2 to 3 mm under good coherence conditions. In practice, atmospheric water-vapour delays introduce errors of the same order, so individual interferograms over tundra are rarely reliable on their own.

Time-series methods address this by stacking many interferograms. Small Baseline Subset (SBAS) selects pairs with short temporal and spatial baselines to maximise coherence, then inverts the network to recover a deformation time series at each pixel. Persistent Scatterer InSAR (PS-InSAR) instead identifies pixels, typically man-made structures, exposed rock or gravel, that remain coherent across all acquisitions regardless of baseline. Both methods can suppress atmospheric noise to the level of a few millimetres per year when sufficient acquisitions are available, typically 20 or more epochs.

The decorrelation problem over vegetated tundra

Coherence is the measure of how similar the radar scattering pattern is between two passes. Vegetation moves, grows and changes moisture content between acquisitions, randomising the phase and destroying the interferometric signal. Over dense shrub tundra, C-band coherence can collapse to near zero within 6 to 12 days in summer, which is precisely the season when active-layer thaw is most active. This is not a processing artefact; it is a physical consequence of the short wavelength interacting with a changing canopy.

L-band partially solves this. At 23.6 cm, ALOS-2 PALSAR-2 penetrates low shrub canopies and scatters from the soil surface beneath, preserving coherence over timescales of 14 to 46 days in many tundra environments. Published studies using PALSAR-2 stacks over Alaskan and Siberian sites have recovered seasonal displacement signals in areas where Sentinel-1 interferograms were unusable. The trade-off is access: PALSAR-2 data is not freely distributed and the 14-day repeat limits the density of the time series compared with Sentinel-1.

A practical workflow often combines both. Sentinel-1 stacks identify coherent scatterers, typically exposed mineral ground, gravel pads or infrastructure, and provide high temporal density. PALSAR-2 acquisitions extend spatial coverage into vegetated areas. Neither frequency alone covers the full range of tundra surface types.

What the signal cannot tell you without additional data

InSAR measures line-of-sight displacement, not vertical displacement directly. Converting to vertical requires knowing the satellite look angle, typically 30 to 45 degrees from vertical in standard modes, and assuming the dominant motion is vertical. For permafrost subsidence this assumption is usually reasonable, but on slopes or near thermokarst lake margins where horizontal movement is significant, the decomposition introduces error.

The method also cannot distinguish the cause of subsidence. Ground-ice melt, consolidation of thawed organic material, drainage of water from the active layer and loading from snow all produce downward surface motion. Separating these mechanisms requires ground-truth measurements: active-layer thickness probing, soil-moisture sensors or shallow borehole temperature records. InSAR provides the spatial pattern and the magnitude; it cannot replace in-situ process understanding.

Cloud cover is irrelevant to radar, which is one of the genuine advantages over optical monitoring in polar regions. Polar night is similarly irrelevant. However, heavy wet snow on the ground surface changes the dielectric properties and can introduce phase artefacts that mimic subsidence. Winter acquisitions over snow-covered ground need careful interpretation.

Infrastructure stability and carbon flux: two very different uses of the same data

For infrastructure operators, the priority is detecting differential settlement early. A pipeline, road embankment or building foundation that subsides uniformly across a wide area is less dangerous than one where adjacent sections move at different rates. InSAR time series can map strain gradients at metre-to-tens-of-metres scale, flagging sections where differential displacement exceeds engineering thresholds. TerraSAR-X at 3 m resolution is well suited to this, allowing monitoring of individual foundation pads or culvert crossings.

For carbon accounting, the question is different. Thawing permafrost releases CO2 and methane as organic carbon decomposes, but the spatial pattern of thaw is highly heterogeneous. InSAR-derived subsidence maps serve as a proxy for where thaw is deepest and most active, which can constrain the spatial distribution of flux estimates derived from atmospheric or eddy-covariance measurements. This is an active research area rather than a settled operational method, and InSAR subsidence alone cannot quantify carbon release without soil carbon density maps and decomposition models.

Satellize can run SBAS time-series processing on open Sentinel-1 archives and combine results with PALSAR-2 or RCM acquisitions sourced under client licence, producing displacement maps and annual trend layers suited to either application. The approach is the same one we apply in other analytics contexts, including the Tonga crop-estimation programme, where combining open and commercial imagery is routine.

Archive depth and what it means for trend detection

Sentinel-1A began systematic acquisitions in 2014. A decade of C-band data now exists over most permafrost regions, though coverage density varies by acquisition mode and geographic priority. This archive is long enough to detect multi-year subsidence trends at sites with sufficient coherent scatterers, separating interannual variability from a directional signal. ALOS PALSAR (the predecessor to PALSAR-2) extends L-band coverage back to 2006 over selected areas, giving a potential 18-year baseline for sites where both datasets are usable.

Trend detection at the millimetre-per-year level requires careful handling of seasonal signals. The annual freeze-thaw cycle must be modelled and removed before a long-term trend is interpretable. Standard SBAS implementations include seasonal terms in the inversion, but the model assumptions, typically a sinusoidal annual cycle, may not capture the asymmetric timing of thaw and freeze in a warming climate. This is a known limitation acknowledged in the published literature, not a processing oversight.

Typical figures

Spatial resolution (typical operational)	5–20 m (Sentinel-1 IW); 3–10 m (PALSAR-2 Fine Beam); 3 m (TerraSAR-X Stripmap)
Revisit interval	6 days (Sentinel-1, mid-to-high latitudes); 14 days (PALSAR-2); 1–4 days (RCM at high latitudes); 11 days (TerraSAR-X)
Radar frequency / wavelength	C-band 5.6 cm (Sentinel-1, RCM); L-band 23.6 cm (PALSAR-2); X-band 3.1 cm (TerraSAR-X)
Minimum detectable displacement (time series)	2–5 mm line-of-sight per epoch under good coherence; annual trend detection to ~1–3 mm/yr with 20+ acquisitions
Coherence constraint	C-band loses coherence over vegetated tundra within 6–12 days in summer; L-band maintains coherence over 14–46 days in low-shrub terrain
Swath width	250 km (Sentinel-1 IW); 70 km (PALSAR-2 Fine Beam); 30 km (TerraSAR-X Stripmap)
Archive depth	Sentinel-1: 2014–present (open); ALOS PALSAR: 2006–2011; PALSAR-2: 2014–present (licensed)
Atmospheric correction	ERA5 or GACOS tropospheric delay models applied; residual errors ~2–5 mm per interferogram
Typical time-series stack size for reliable trend	Minimum ~20 epochs; 50+ preferred for millimetre-level annual trend
Deliverable formats	GeoTIFF displacement maps, NetCDF time-series stacks, GIS-ready vector anomaly layers, PDF summary reports

Analytics Satellize can run

Seasonal displacement time series	SBAS InSAR inversion with sinusoidal seasonal model; Sentinel-1 or PALSAR-2 stack	GeoTIFF stack of cumulative displacement per epoch, plus amplitude and phase of annual cycle per pixel
Multi-year subsidence trend map	Linear trend extraction from SBAS time series after seasonal decomposition; minimum 3-year archive	GeoTIFF of mm/year vertical trend with uncertainty layer; flagged anomaly zones as vector polygons
Infrastructure corridor differential settlement report	PS-InSAR on coherent scatterers (gravel pads, structures, exposed rock) along pipeline or road alignment; TerraSAR-X or Sentinel-1	PDF engineering report with displacement profiles along corridor centreline, threshold-exceedance alerts
Active thermokarst hotspot detection	Change detection on annual SBAS trend maps; pixels exceeding defined subsidence rate threshold flagged and clustered	GIS vector layer of hotspot polygons with area, mean rate and first-detection date; updated annually
Freeze-thaw cycle characterisation	Seasonal model fitting to multi-year Sentinel-1 time series; extraction of thaw-onset and freeze-onset dates per pixel	Annual raster layers of thaw-onset day-of-year and active-layer displacement amplitude
L-band coherence and vegetation masking	PALSAR-2 coherence estimation across seasonal pairs; coherence thresholding to separate vegetated and bare-ground pixels	Binary and continuous coherence masks as GeoTIFF, used to qualify which pixels support reliable deformation retrieval

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.