Thermokarst lake expansion and drainage in thawing permafrost
Thawing permafrost creates thermokarst lakes that expand laterally for years, then drain catastrophically through ice-wedge breaches. Multitemporal Landsat and Sentinel-1 track area change; InSAR resolves the ground settlement that precedes and follows each event.
Sensors
- Sentinel-1 A/B (C-band SAR, 5.6 cm wavelength): Interferometric Wide swath mode at 20 m resolution, 6-day repeat at mid-latitudes (12-day with one satellite). Backscatter change distinguishes open water from saturated ground after drainage; coherence loss over newly thawed soil flags active settlement zones. Cloud-independent, works through polar night.
- ALOS-2 PALSAR-2 (L-band SAR, 23.6 cm wavelength): L-band penetrates shallow standing water and saturated peat to return signal from the lake bed or underlying frozen layer, which C-band cannot reach. Fine-beam single-polarisation mode achieves 3–10 m resolution. Published studies use PALSAR coherence to map sub-lake talik (unfrozen zone) extent.
- Landsat 8 and 9 (OLI/TIRS): 30 m multispectral, 16-day repeat per satellite (8-day combined). The NDWI (green/NIR ratio) and MNDWI (green/SWIR) reliably delineate lake boundaries down to roughly one to two 30 m pixels in area. The Landsat archive extends to 1972, enabling decadal change detection across Siberia and Alaska. Thermal band (100 m resampled to 30 m) distinguishes ice-covered from open water in shoulder seasons.
- Planet SuperDove (PlanetScope): 3 m resolution, daily revisit globally. Useful for tracking rapid drainage events and resolving small ponds below Landsat's detection floor. Eight spectral bands include red-edge and NIR. No SAR capability; cloud cover in Arctic summers remains a genuine constraint.
- Sentinel-2 MSI: 10 m in visible and NIR bands, 5-day revisit at high latitudes with both satellites. Bridges the resolution gap between Landsat and Planet for lake-area time series. Freely available from 2015; useful for monitoring individual drainage corridors and refill dynamics.
What ice-wedge polygons do when they fail
Permafrost in Arctic lowlands is laced with ice wedges: vertical sheets of ground ice that formed over thousands of years as repeated freeze-thaw cycles widened thermal contraction cracks. From above, these wedges produce the distinctive polygonal patterning visible across Siberia's Lena Delta and Alaska's North Slope. When summer air temperatures rise or snow insulation increases, the wedge tops melt first. The polygon centres, previously elevated above the wedge rims, subside as ice volume is lost. Water pools in the resulting depressions.
That pooling accelerates the process. Water absorbs more solar radiation than tundra vegetation, warming the ground further and deepening the thaw. A single polygon pond can coalesce with neighbours within a decade, producing a thermokarst lake tens of metres across. Once a lake is established, it can persist for centuries or drain within hours if its bank is breached by a stream, a coastal bluff, or simply by the lake itself undercutting its own rim. The drainage leaves a drained thermokarst lake basin, or 'alas', which may revegetate or refill depending on local hydrology. Neither outcome is easily predicted from a single image.
What a floating roof gives away: optical lake-area change
Landsat's archive is the workhorse for decadal thermokarst studies. Published research using Landsat imagery across continuous permafrost zones in Siberia and Alaska has documented both net lake expansion and, in some regions, net lake area decline driven by drainage exceeding new lake formation. The distinction matters: a shrinking lake count in a warming landscape is not good news; it often signals that drainage is outpacing formation, with large pulses of stored organic carbon entering river systems.
NDWI thresholding on Landsat OLI imagery can detect lakes larger than roughly 0.1 hectares with reasonable confidence. Below that, mixed pixels dominate and commission errors rise sharply. Planet SuperDove reduces this floor to around 0.001 hectares (100 square metres) in clear conditions, though systematic Arctic coverage at daily cadence is expensive and cloud contamination in summer is frequent. A practical workflow combines Landsat for the long baseline and Planet for event-scale monitoring of specific drainage episodes.
Sentinel-2 adds value in the 2015-to-present window, particularly for tracking the refill of drained basins. Its 10 m NIR band resolves the shallow, pale-coloured water that indicates partial refill, which is easily confused with wet sedge in 30 m imagery.
SAR sees through the water surface and through the dark
C-band SAR (Sentinel-1) treats open water as a specular reflector: the radar signal bounces away from the sensor and returns low backscatter, making lakes appear dark. That contrast against rougher tundra vegetation is reliable and cloud-independent, which matters enormously in Arctic winters when optical sensors are useless for months. Sentinel-1's 6-day repeat (at mid-latitudes; somewhat longer at high Arctic latitudes depending on orbital geometry) is fast enough to catch rapid drainage events if the acquisition schedule aligns.
L-band SAR tells a different story. At 23.6 cm wavelength, ALOS-2 PALSAR-2 can penetrate shallow lake water and return coherent signal from the lake bed. This makes it possible to map the talik, the unfrozen zone that develops beneath a persistent lake, even while the lake surface is intact. Talik extent is a proxy for how deeply permafrost has been disturbed and how much ground ice remains vulnerable. Published studies have used PALSAR coherence time series to distinguish lakes sitting above continuous frozen ground from those that have already thawed through to the substrate. The practical limit is PALSAR-2's infrequent revisit over any given Arctic site, typically 14 to 42 days depending on observation mode and tasking priority.
InSAR measures the ground sinking around the lake
Lake expansion is visible in optical imagery, but the ground deformation that accompanies and precedes it is not. Differential InSAR using Sentinel-1 pairs can measure line-of-sight displacement to centimetre or sub-centimetre precision over stable coherent surfaces. In permafrost terrain, coherence is maintained over dry tundra and exposed mineral soil but is often lost over wet vegetation. This limits InSAR reliability in the most actively thawing zones, which is an honest constraint worth stating plainly.
Where coherence holds, InSAR time series (using persistent scatterer or small-baseline subset methods) have documented seasonal active-layer heave and subsidence cycles of several centimetres, and multi-year settlement trends of one to several centimetres per year in degrading ice-wedge terrain. Settlement rates accelerate near lake margins, consistent with lateral heat transfer from the water body. After a drainage event, the exposed lake bed can show rapid initial subsidence as saturated sediments consolidate, followed by slower creep as the newly exposed ground refreezes or compacts further. Sentinel-1's C-band is less effective over freshly drained, water-saturated sediment because coherence drops; this is where ALOS-2's L-band again has an advantage.
The honest limits of the method stack
No single sensor resolves the full problem. Optical imagery misses events that occur under cloud or during polar night. SAR backscatter change detects drainage but cannot determine whether a dark area is open water, wet sediment, or smooth ice without ancillary data. InSAR coherence fails precisely where thaw is most active. L-band PALSAR-2 has the physics but not the revisit frequency for operational monitoring; its data is also not freely available in the way Sentinel-1 is.
Lake drainage can occur in under 24 hours, as documented in Alaska's Seward Peninsula. A 6-day Sentinel-1 repeat may entirely miss the event itself and only observe the before and after states. Planet's daily optical cadence helps, but cloud cover during Arctic summer storms is common. The practical answer is a sensor-fusion approach: Landsat and Sentinel-2 for area baselines, Sentinel-1 for SAR change detection, Planet for high-cadence optical confirmation, and PALSAR-2 for targeted sub-lake structural assessment.
Satellize runs this kind of multi-source fusion on open constellations, adding commercial tasking where event cadence demands it. The analytical pipeline is similar in structure to the crop-estimation work done for the Kingdom of Tonga, adapted for binary water-body classification rather than vegetation indices.
What the output actually looks like for an infrastructure client
A government or engineering client monitoring a pipeline corridor or road alignment across discontinuous permafrost typically needs three things: a baseline lake inventory with area and depth estimates; an alert when any lake within a defined buffer zone changes area by more than a threshold (say, 10 percent in a single season); and a ground-settlement time series at specific infrastructure locations.
The lake inventory is produced from a Landsat or Sentinel-2 time series using NDWI thresholding and object-based classification, delivered as a GIS polygon layer with per-lake area statistics across the archive period. The alert layer runs on each new Sentinel-1 acquisition, flagging backscatter anomalies that indicate new open water or drainage. The settlement time series is a small-baseline InSAR product, typically delivered as a displacement map with time-series plots at nominated point locations. Uncertainty bounds should accompany every figure; a claimed precision of 3 mm per year means little without the coherence mask that shows where that precision actually holds.
Typical figures
| Optical lake-area resolution (Landsat 8/9) | 30 m pixel; reliable lake detection above ~0.1 ha |
| Optical lake-area resolution (Sentinel-2) | 10 m in NIR; detection floor ~0.01 ha in clear conditions |
| Optical lake-area resolution (Planet SuperDove) | 3 m; detection floor ~0.001 ha; cloud cover limits Arctic utility |
| SAR backscatter change (Sentinel-1 IW) | 20 m resolution; 6-day repeat at mid-latitudes; cloud- and dark-independent |
| L-band SAR sub-lake sensing (ALOS-2 PALSAR-2) | 3–10 m (mode-dependent); 14–42 day revisit; not freely available |
| InSAR displacement precision (Sentinel-1 SBAS) | ~5–10 mm per epoch line-of-sight; ~1–3 mm/yr in time series over coherent surfaces |
| Landsat archive depth | 1972 to present (Landsat 1–9); enables 50-year change baselines |
| Sentinel-1 archive depth | 2014 to present; free via Copernicus Data Space |
| Minimum detectable drainage event (SAR) | Area change detectable if event occurs between acquisitions; sub-24 hr events may be missed |
| Delivery formats | GeoTIFF lake-area polygons, NetCDF displacement time series, GeoJSON alert feeds, PDF seasonal reports |
Analytics Satellize can run
| Decadal lake inventory and area-change time series | NDWI/MNDWI thresholding and object-based classification on Landsat 8/9 and Sentinel-2 archive | GIS polygon layer with per-lake area statistics per season, 1984 to present |
| Rapid drainage event alert | SAR backscatter anomaly detection on sequential Sentinel-1 IW acquisitions; threshold set per scene | GeoJSON alert feed with lake ID, area-change estimate, and acquisition timestamp; latency under 24 hours of ESA product availability |
| Ground settlement time series at infrastructure locations | Small-baseline subset (SBAS) InSAR on Sentinel-1 SLC pairs; coherence masking applied | Displacement map (GeoTIFF) plus CSV time-series plots at nominated point locations, with coherence-derived uncertainty bounds |
| Sub-lake talik mapping | L-band coherence analysis on ALOS-2 PALSAR-2 time series; distinguishes frozen substrate from thawed talik by coherence retention depth | Classified raster layer indicating talik presence/absence and estimated lateral extent, delivered as GeoTIFF with metadata |
| Ice-wedge polygon degradation classification | Multi-temporal object-based analysis combining Sentinel-2 spectral indices and Sentinel-1 texture; polygon stage classified (low-centred, high-centred, coalescent) | Annual classified polygon map with degradation-stage statistics per watershed or corridor buffer zone |
| Seasonal active-layer heave/subsidence cycle | Differential InSAR between spring freeze-up and late-summer thaw acquisitions; Sentinel-1 ascending and descending orbits combined for 2D decomposition where geometry allows | Seasonal displacement map (GeoTIFF) with amplitude and phase of annual cycle at 20 m posting |
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.