Permafrost coastline erosion rates and cliff-face thaw monitoring
Arctic permafrost coasts can retreat more than 10 metres in a single storm season. Multi-temporal optical shoreline extraction and SAR coherence analysis now make those losses measurable from orbit, at the pace the problem demands.
Sensors
- Planet SuperDove: 3 m resolution, up to daily revisit at Arctic latitudes during the ice-free season (roughly June to October). Captures individual storm events and the episodic retreat pulses that dominate annual budgets.
- Sentinel-2 MSI: 10 m resolution in visible and near-infrared bands, 5-day revisit at high latitudes with two satellites. Provides the decadal archive baseline from 2015 onward for long-term rate calculation and calibrates Planet shoreline picks.
- Sentinel-1 SAR (C-band): Approximately 10 m ground-range resolution in Interferometric Wide swath mode, 6-day repeat. Coherence differencing between image pairs detects cliff-face disturbance and sediment redistribution regardless of cloud cover or polar darkness.
- ArcticDEM (optical stereo archive): Sub-metre posting digital elevation model derived from commercial stereo imagery by the Polar Geospatial Center. Enables volumetric loss calculations and cliff-height measurement; strips updated irregularly but coverage across the Beaufort and Siberian coasts is extensive.
Why permafrost coasts erode faster than almost anywhere else on Earth
Most coastlines erode because waves remove sediment grain by grain. Permafrost coasts do something more dramatic: the cliff face thaws, loses its structural ice cement, and collapses in blocks. The process is called thermoerosion, and it means that a single warm storm can accomplish what would take decades on a temperate shore. Documented mean annual retreat rates on the Beaufort Sea coast of Alaska and Canada exceed 1 to 2 metres per year on average, with local hotspots exceeding 10 metres per year. On the East Siberian coast, multi-decadal studies using aerial photography and satellite imagery have recorded comparable figures.
The consequences extend well beyond the land boundary. Ice-rich permafrost contains large stocks of particulate organic carbon accumulated over millennia. When the cliff collapses, that carbon enters the nearshore ocean directly, fuelling microbial activity and altering local nutrient budgets. Quantifying the flux requires knowing not just the horizontal retreat distance but the cliff height and ice content, which is where elevation data from ArcticDEM becomes indispensable.
What a floating roof gives away: reading shoreline change from orbit
The standard method for satellite-derived shoreline extraction is multi-temporal segmentation of the water-land boundary in optical imagery. In practice, that means applying a water index such as the Modified Normalised Difference Water Index (MNDWI, using Sentinel-2 green and shortwave-infrared bands) to image stacks, then tracking the position of the boundary across dates. At 10 m resolution, Sentinel-2 can resolve retreat pulses of that magnitude reliably. Planet SuperDove's 3 m pixels push the detectable minimum considerably lower, though sub-pixel accuracy claims should be treated with caution: co-registration errors of 1 to 2 pixels are realistic, so reported retreats below about 5 to 6 m in a single interval carry real uncertainty.
The ice-free season at high Arctic latitudes runs roughly June through October, which is also when thermoerosion is most active. Cloud cover is the persistent nuisance. In some years, cloud-free optical coverage over a given cliff segment may amount to only a handful of usable acquisitions. This is precisely why SAR coherence analysis matters: Sentinel-1 images through cloud and polar twilight, and a drop in interferometric coherence between two C-band acquisitions is a reliable indicator of surface disturbance, whether from a slump, a collapse, or sediment reworking in the surf zone.
SAR coherence loss as a collapse detector
Interferometric SAR coherence measures how similar the radar backscatter phase is between two acquisitions. Stable ground maintains high coherence. A cliff face that has partially collapsed, or a beach that has been reworked by a storm, loses coherence sharply because the scattering geometry has changed. With Sentinel-1's 6-day repeat, it is possible to bracket storm events and identify which cliff segments were disturbed.
The method has genuine limits. Wet snow, standing water, and vegetation change also suppress coherence, so Arctic summer coherence maps contain noise from sources unrelated to erosion. Careful masking using ancillary land-cover data reduces false positives, but some ambiguity remains. SAR coherence is best treated as a screening tool that flags segments for closer optical inspection, not as a standalone erosion measurement.
Turning retreat distances into carbon-flux estimates
A shoreline retreat polygon tells you the area lost. Multiply by cliff height (from ArcticDEM) and you have a volume. Multiply by bulk density and organic carbon content (from published soil surveys or field campaigns) and you have a carbon-flux estimate. The uncertainty at each step compounds, so flux figures derived entirely from remote sensing carry wide error bars. Published studies using this approach, including work on the Beaufort Sea coast by the US Geological Survey and Canadian collaborators, report organic carbon fluxes on the order of tens to hundreds of grams of carbon per metre of coastline per year at eroding sites, but the range across sites is large.
Satellize's analytics pipeline can produce the geometric component of this calculation, specifically retreat polygons, volumetric loss estimates, and time-series of cliff-face coherence, as GIS layers ready for fusion with field-measured carbon parameters. The organic-carbon conversion itself requires ground-truth inputs that satellite data alone cannot supply.
Archive depth, revisit, and the episodic-event problem
Long-term rate calculations benefit from the Sentinel-2 archive, which begins in 2015 and grows continuously. For pre-2015 baselines, Landsat 8 (30 m, operational since 2013) and earlier Landsat missions extend the record to the 1970s, though at coarser resolution that limits detection to larger retreat events. ArcticDEM strips span roughly 2010 onward.
The episodic nature of permafrost coastal erosion is the central challenge for any monitoring system. A single storm in late September, when sea-ice extent is minimal and fetch is maximal, can account for the majority of a year's retreat. Miss that event in the optical record due to cloud cover and the annual rate is underestimated. Planet SuperDove's near-daily cadence during the open-water season is the most practical way to reduce that sampling gap, though tasking must be planned in advance of storm season rather than reactively. Satellize applies this approach in its analytics work, scheduling systematic tasking over priority cliff segments from June onward, in the same way it structures seasonal acquisition planning for the Tonga crop-estimation programme.
What this monitoring cannot do
Satellite observation of permafrost coastal erosion is genuinely useful, but several limits are worth stating plainly. Optical methods cannot measure cliff-face temperature or ice content directly. They cannot resolve the internal stratigraphy that determines whether a given bluff will fail catastrophically or retreat gradually. Sub-metre retreat events are below the reliable detection threshold of any freely available sensor. And in the Siberian Arctic, systematic commercial tasking coverage has historically been sparse, meaning that some of the fastest-eroding coastlines on Earth have the thinnest satellite records.
SAR coherence analysis helps with cloud and darkness, but C-band penetration into frozen sediment is limited to the surface layer, so subsurface ice-wedge degradation is not directly visible. L-band SAR (as on ALOS-2 or the forthcoming NISAR mission) penetrates further and may improve detection of near-surface thaw, though that capability is still being validated for this specific application. Any programme that needs sub-annual volumetric accuracy at specific sites will require airborne lidar or UAV surveys to anchor the satellite-derived geometry.
Typical figures
| Optical spatial resolution | 3 m (Planet SuperDove), 10 m (Sentinel-2 MSI), 30 m (Landsat 8/9) |
| SAR spatial resolution | ~10 m ground-range (Sentinel-1 IW mode) |
| Optical revisit at Arctic latitudes | Up to daily (Planet SuperDove, ice-free season); 5-day (Sentinel-2 two-satellite) |
| SAR revisit | 6 days (Sentinel-1); all-weather, including polar twilight |
| Minimum detectable retreat (optical) | ~5 to 6 m per interval at 3 m resolution, accounting for co-registration error; ~15 to 20 m at 10 m resolution |
| Elevation model resolution | ArcticDEM: 2 m posting (strip products), 10 m mosaic; vertical accuracy ~0.2 to 1 m over stable terrain |
| Archive depth | Sentinel-2 from 2015; Landsat from 1972 (30 m); ArcticDEM strips from ~2010 |
| Spectral bands used | Green, NIR, SWIR (MNDWI shoreline extraction); C-band 5.4 GHz (SAR coherence) |
| Operational season (optical) | June to October (ice-free, sufficient solar illumination) |
| Delivery formats | GeoTIFF shoreline polygons, GeoPackage time-series, CSV retreat-rate tables, coherence-change GeoTIFFs |
Analytics Satellize can run
| Annual and seasonal shoreline retreat polygons | Multi-temporal MNDWI water-index segmentation on Sentinel-2 and Planet SuperDove stacks | GeoPackage with dated shoreline vectors and per-segment retreat distances |
| Long-term retreat-rate raster (metres per year) | Linear regression on shoreline position time series across Sentinel-2 and Landsat archive | GeoTIFF rate map with per-pixel uncertainty band, PDF summary report |
| Storm-event collapse detection | SAR coherence differencing (Sentinel-1 pre- and post-storm pairs) | Coherence-change GeoTIFF with flagged disturbance polygons, delivered within 48 hours of SAR acquisition |
| Volumetric loss time series | Retreat polygon area multiplied by cliff height extracted from ArcticDEM strip differencing | CSV table of volume loss per segment per period, with propagated geometric uncertainty |
| Episodic-event capture schedule | Systematic Planet SuperDove tasking plan aligned to open-water season and storm-forecast windows | Acquisition calendar and cloud-cover QA log for each monitored cliff segment |
| Carbon-flux geometry layer | Retreat area and volume estimates formatted for fusion with published bulk-density and organic-carbon soil parameters | GIS layer with geometric inputs ready for client-supplied carbon conversion factors |
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.