Permafrost thaw and thermokarst damage at Arctic heritage sites

Permafrost thaw is physically dismantling Norse, Indigenous and Cold War heritage sites across the Arctic. InSAR displacement time-series and land surface temperature records let site managers see the damage accumulating before a wall collapses or a midden slides into the sea.

Sensors

Sentinel-1 SAR (C-band, 5.6 cm wavelength): Interferometric Wide Swath mode at 5 x 20 m ground resolution, 6-day repeat at high latitudes with two satellites. Phase difference between acquisitions detects surface displacement to sub-centimetre precision; coherence loss flags active slumping or soil disturbance. Arctic coverage is near-daily at latitudes above 70°N due to orbital convergence.
MODIS Land Surface Temperature (Terra and Aqua): 1 km spatial resolution, twice-daily overpass. MOD11A1 and MYD11A1 products provide LST at roughly 1 K accuracy under clear skies. Used to build seasonal thaw-degree-day accumulation records and identify anomalously warm summers that precede accelerated subsidence.
VIIRS Land Surface Temperature (Suomi-NPP and NOAA-20): 375 m resolution in the I-band thermal channel, daily revisit. Finer spatial detail than MODIS LST, useful for resolving temperature contrasts across individual site footprints. VIIRS also extends the consistent thermal record as MODIS instruments age.
ArcticDEM (public mosaic and strip products): Produced by the Polar Geospatial Center from commercial stereo optical imagery. Strip products at 2 m resolution, mosaic at 2 m and 10 m. Differencing strips from different dates reveals volume loss, pond expansion and retrogressive thaw slump headwall retreat at metre-scale precision where strip pairs overlap.
Sentinel-2 Multispectral (10 m visible and NIR): 5-day revisit (two satellites). Used to map thermokarst pond extent, bare sediment exposure and vegetation change around disturbed sites. Cloud cover is a genuine constraint in maritime Arctic climates; summer acquisition windows may be narrow.

What thawing permafrost actually does to a site

Permafrost is not simply frozen ground. It is the structural matrix that holds stratigraphic layers in place, suppresses microbial decomposition of organic material, and keeps waterlogged wooden artefacts, food remains and human tissue intact for centuries. When the active layer deepens, that matrix fails. Buildings tilt. Middens slump. Coastal bluffs calve into the sea, taking with them the entire depositional sequence that archaeologists would otherwise excavate over decades.

Thermokarst, the irregular subsidence terrain produced when ice-rich permafrost melts, is not a slow or uniform process. Retrogressive thaw slumps can retreat at rates of 10 to 30 metres per year, documented at multiple Alaskan and Canadian sites. A Norse farmstead or an Iñupiaq semi-subterranean house that was stable through the twentieth century may now be losing structural integrity within a single decade. The satellite record is long enough to see this acceleration directly.

What InSAR measures, and where it struggles

Sentinel-1 interferometric SAR measures the phase difference between two radar acquisitions of the same ground. Where the surface has moved toward or away from the satellite between passes, the phase shifts proportionally. Over permafrost terrain, seasonal freeze-thaw produces a predictable annual signal of a few centimetres of heave and subsidence. Anomalies on top of that signal, particularly irreversible downward displacement that does not recover each winter, indicate permanent thaw settlement.

The method has real limits that any honest assessment must name. Coherence, the statistical similarity between two SAR images, collapses when the surface changes dramatically between passes: fresh snow, standing water, or rapid slumping all decorrelate the signal and produce data gaps precisely where the damage is most acute. Vegetation also decorrelates C-band phase. Sites surrounded by tall shrub tundra are harder to monitor than bare coastal bluffs. Time-series approaches such as Persistent Scatterer or Small Baseline Subset (SBAS) processing mitigate some of this by stacking many interferograms, but they cannot recover information that was never coherent. Where coherence loss itself is spatially concentrated and temporally sudden, it becomes a detection signal in its own right: the SAR is telling you that something moved violently enough to scramble the phase entirely.

Published studies of Svalbard cultural heritage sites, including work associated with the Norwegian Institute for Cultural Heritage Research, have demonstrated that SBAS time-series can resolve seasonal displacement amplitudes and identify sites where the long-term trend is downward settlement rather than elastic freeze-thaw cycling. Alaskan coastal studies have used similar methods to document bluff retreat rates at Indigenous village sites.

Thermal forcing: reading the summer that caused last year's damage

Ground displacement lags thermal forcing. A particularly warm summer drives active-layer deepening; the visible structural consequence may not appear until the following spring or later. MODIS and VIIRS LST products allow analysts to reconstruct thaw-degree-day accumulation for any summer in the archive going back to 2000 for MODIS, and to correlate anomalous thermal years with subsequent InSAR displacement anomalies. This causal chain matters for site managers: it means damage can be anticipated from thermal data before it is confirmed in the displacement record.

LST products carry their own caveats. Cloud cover in the maritime Arctic routinely interrupts the thermal record for days at a time. The 1 km MODIS pixel integrates across heterogeneous terrain, smoothing out the micro-topographic temperature contrasts that control local thaw depth. VIIRS at 375 m improves this somewhat, but neither product resolves the temperature field at the scale of an individual building or midden. Ground-truth from on-site loggers remains necessary for calibration.

ArcticDEM differencing: the volume question

InSAR gives displacement in the radar line-of-sight direction, which requires geometric decomposition to separate vertical from horizontal motion. ArcticDEM strip differencing gives volumetric change directly in three dimensions, which is more intuitive for site managers and more useful for estimating sediment loss. Where two strips separated by one or more years cover the same site, the difference DEM shows pond expansion, headwall retreat distances and net volume of material lost.

Strip coverage is not systematic. The Polar Geospatial Center produces strips from commercial tasking that was not planned around heritage site locations, so temporal coverage varies considerably by site. Some locations have dense strip pairs; others have gaps of several years. The 2 m mosaic is a useful baseline but is not dated to a single moment. Analysts should check strip availability for a specific site before committing to a differencing approach.

Putting it together for a site manager

A practical monitoring programme for a threatened Arctic heritage site combines three layers. First, an InSAR displacement time-series run over the full Sentinel-1 archive from 2014 to present, processed with SBAS or a similar multi-temporal method, gives a displacement history and flags sites where the trend is consistently downward. Second, MODIS and VIIRS LST records identify the summers of peak thermal forcing. Third, ArcticDEM strip differencing, where coverage allows, quantifies volume loss and maps the spatial extent of active slumping.

The combination does not replace ground survey. It prioritises where ground survey is most urgent. For heritage agencies managing dozens of sites across a territory the size of Alaska or Svalbard, that triage function alone justifies the analysis. Satellize runs this multi-source workflow on open constellation data, with commercial tasking added where higher-resolution optical or SAR coverage is needed. The Tonga crop-estimation programme demonstrated that the same open-data infrastructure can be adapted to very different analytical questions; the Arctic heritage case is methodologically more demanding but rests on the same principle of combining freely available sensors with careful processing.

One practical note on latency: Sentinel-1 data is available within 24 hours of acquisition through the Copernicus Data Space Ecosystem. MODIS and VIIRS LST products are available within hours through NASA FIRMS and EARTHDATA. ArcticDEM strips are released on a rolling basis by the Polar Geospatial Center. A near-real-time alert on coherence loss is technically feasible; a full displacement time-series update is typically a quarterly or annual product.

What the satellite cannot settle

Satellite data identifies where the ground is moving and how fast. It cannot determine what is in the ground at risk, whether a slumping area contains significant buried deposits or is archaeologically sterile, or whether the material being lost is irreplaceable. Those questions require ground investigation, ideally guided by the satellite triage. The method is also blind to subsurface ice content, which controls how rapidly a site will degrade once thaw begins. A site showing modest current displacement may be sitting on massive ground ice that will produce catastrophic collapse once the thermal front reaches it. Integrating borehole data or airborne electromagnetic surveys with the satellite displacement record is the honest way to assess that risk.

Typical figures

InSAR spatial resolution (Sentinel-1 IW mode)	5 x 20 m (range x azimuth); multi-looked to ~20 m for interferogram processing
InSAR displacement sensitivity	Sub-centimetre in line-of-sight; typically 3–5 mm per interferogram under good coherence conditions
Sentinel-1 revisit at high latitudes (>70°N)	Near-daily with two satellites; 6-day exact repeat baseline for interferometry
MODIS LST resolution and revisit	1 km, twice daily (Terra + Aqua combined); ~1 K absolute accuracy under clear sky
VIIRS LST resolution and revisit	375 m (I-band), daily; extends thermal record beyond MODIS instrument lifetime
ArcticDEM strip resolution	2 m horizontal; vertical accuracy typically 0.1–1 m depending on terrain and source imagery
Sentinel-1 archive depth	From April 2014 (Sentinel-1A launch); growing continuously
MODIS LST archive depth	From 2000 (Terra) and 2002 (Aqua)
Cloud impact on optical and thermal products	Significant in maritime Arctic; summer clear-sky windows may be limited to weeks per year
Minimum detectable displacement (time-series SBAS)	Approximately 5–10 mm cumulative vertical displacement over a multi-year stack under coherent conditions

Analytics Satellize can run

InSAR displacement time-series	SBAS or Persistent Scatterer multi-temporal interferometry on Sentinel-1 IW SLC stack	GeoTIFF displacement maps and CSV time-series per site, annual or quarterly update
Coherence-loss alert	Interferometric coherence magnitude thresholding between consecutive 6-day pairs; spatial clustering of decorrelated pixels flags active slump initiation	Near-real-time alert (within 48 hours of Sentinel-1 acquisition) with polygon of affected area
Thaw-degree-day accumulation record	Integration of MODIS MOD11A1 and VIIRS LST daily products above 0°C threshold over the summer season, per site bounding box	Annual thermal forcing report with time-series chart and anomaly ranking against the 2000–present baseline
Thermokarst pond expansion mapping	Multitemporal water-body extraction from Sentinel-2 NDWI (normalised difference water index) at 10 m resolution	Annual pond-extent GIS layer with area change statistics relative to a chosen baseline year
ArcticDEM volume-change assessment	Co-registration and differencing of available strip DEM pairs; volumetric change calculated over site polygon	Difference DEM GeoTIFF and summary table of volume lost, headwall retreat distance and pond infill
Site-triage priority ranking	Composite scoring of displacement trend magnitude, thermal anomaly frequency and coastal proximity; ranked across a portfolio of sites	Tabular priority report with per-site risk scores and recommended ground-survey sequencing

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.