Permafrost thermokarst lake expansion and drainage monitoring
Thermokarst lakes expand as ground ice melts and drain catastrophically when ice-wedge networks fail. Multi-decadal Landsat records, Sentinel-1 InSAR subsidence maps and ICESat-2 elevation data together quantify permafrost degradation rates and infrastructure risk across Arctic regions.
Sensors
- Landsat TM / ETM+ / OLI: 30 m multispectral resolution, 16-day revisit per satellite (8-day with Landsat 8 and 9 combined). The archive runs from 1972 for MSS and from 1984 for TM, making it the only freely available record long enough to track multi-decadal lake-area change. Near-infrared and shortwave-infrared bands separate open water from wet sedge with high confidence in cloud-free acquisitions.
- Sentinel-1 SAR (C-band, 5.6 cm): Interferometric Wide Swath mode at 5 x 20 m resolution, 6-day repeat at high latitudes. C-band coherence is maintained over frozen ground and lake ice, enabling differential InSAR to detect centimetre-scale surface subsidence around actively thawing lake margins. Backscatter intensity also distinguishes open water from ice-covered lake surfaces.
- ALOS-2 PALSAR-2 (L-band, 23.6 cm): L-band penetrates shallow snow and vegetation canopy better than C-band, preserving interferometric coherence over longer temporal baselines (14 to 42 days). Fine-beam single-polarisation mode reaches 3 m resolution. Particularly useful for detecting subsidence in vegetated thermokarst margins where C-band loses coherence, and for mapping talik geometry beneath shallow lakes.
- ICESat-2 ATL03 / ATL13: Photon-counting lidar at 532 nm. Along-track resolution of roughly 0.7 m, with lake surface elevation retrievable to better than 10 cm vertical uncertainty under clear skies. Provides absolute elevation reference to calibrate InSAR relative deformation maps and detects lake-level drawdown associated with drainage initiation. Repeat ground-track interval is 91 days, limiting temporal sampling.
What a thermokarst lake is actually measuring
A thermokarst lake is not simply a hole filled with water. It is a surface expression of subsurface heat transfer. When ice-rich permafrost thaws, the ground volume collapses by the fraction that was ice, typically 20 to 80 percent of the soil column in yedoma deposits. The resulting depression fills with water, which then accelerates thaw at its margins and base because liquid water conducts heat far more efficiently than frozen ground. The lake grows laterally and deepens until it either reaches a drainage threshold or the underlying permafrost table stabilises.
This feedback loop means lake area is a proxy for cumulative thaw energy delivered to the permafrost column beneath it. Measuring area change over years to decades is therefore not just a mapping exercise; it is an indirect measurement of permafrost degradation rate. The catch is that lakes also drain, sometimes within hours when a thaw-lake margin intersects a drainage network or an ice-wedge trough. A drained lake leaves a characteristic oval scar of wet sedge and exposed mineral soil that is detectable in near-infrared imagery but easily confused with a stable wetland if only a single epoch is examined.
Thirty years of Landsat and what it resolves
The Landsat archive from TM onward (1984 to present) is the primary tool for multi-decadal lake-area mapping. The Modified Normalised Difference Water Index (MNDWI), using green and shortwave-infrared bands, separates open water from surrounding tundra reliably at 30 m. Studies across the Siberian yedoma belt and the North Slope of Alaska have used this approach to map lakes down to roughly 0.001 km² (about one hectare), though the 30 m pixel size means smaller ponds are systematically underestimated in count and area.
The honest limit here is cloud cover. Arctic summers are short and frequently overcast; a single Landsat scene may be cloud-free over a study area only a few times per season. Seasonal compositing from multiple acquisitions reduces this problem but introduces uncertainty when comparing epochs: a composite spanning June to September conflates real lake-area change with seasonal water-level fluctuation. Careful selection of late-summer acquisitions, when lakes are at their seasonal maximum, is standard practice and should be specified explicitly in any monitoring protocol.
Drainage events are the most operationally urgent signal. A lake that drains completely can release decades of accumulated carbon in a single season. Landsat's 16-day revisit means a drainage event that completes in hours will appear as a step-change between two acquisitions, not a process. Sentinel-1, with its 6-day repeat and all-weather capability, is better suited to catching the transition.
InSAR subsidence: the signal beneath the water
Surface subsidence around a thermokarst lake margin is the leading indicator of active thaw before the lake visibly expands. Differential InSAR from Sentinel-1 can detect vertical displacement of one to two centimetres per year in favourable conditions: coherent ground surface, short temporal baseline, minimal atmospheric water-vapour gradient. In practice, Arctic tundra is awkward for C-band InSAR because summer vegetation growth and soil moisture change decorrelate the signal rapidly. Temporal baselines longer than 12 days in summer often yield coherence below 0.3, which is too low for reliable phase unwrapping.
ALOS-2 PALSAR-2 in L-band partially solves this. The longer wavelength maintains coherence over vegetated surfaces and through shallow snow, and 42-day repeat pairs have been used successfully to map subsidence bowls of 5 to 30 cm around expanding lakes in the Lena Delta. The trade-off is spatial coverage: ALOS-2 is a commercial system with limited free-access data, and tasking costs limit systematic monitoring to priority areas.
Time-series InSAR methods, particularly SBAS (Small Baseline Subset) applied to Sentinel-1 stacks, improve the signal-to-noise ratio by averaging across many interferograms. Published results from the Tibetan Plateau and Siberia show detection of seasonal frost-heave and thaw-settlement cycles of 2 to 5 cm amplitude, with cumulative subsidence trends of 1 to 3 cm per year in actively degrading areas. These figures are site-specific; yedoma with 80 percent ice content will subside far faster than mineral soils with 20 percent ice.
Methane: from lake area to emission estimate
Thermokarst lakes are disproportionate sources of atmospheric methane relative to their surface area. Ebullition (bubbling) from anaerobic decomposition of thawed organic matter can contribute more than 70 percent of total lake methane flux in some yedoma lakes, according to field studies published in journals including Nature and Global Biogeochemical Cycles. Satellite observation cannot directly measure ebullition flux; that requires floating chambers or eddy covariance towers.
What satellite data can do is constrain the emission estimate spatially. Lake area change maps from Landsat, combined with published emission factors per unit area (which vary by two orders of magnitude depending on lake depth, organic carbon content and water temperature), produce regional flux estimates with large but quantifiable uncertainty. The honest range for mean methane flux from thermokarst lakes in published literature runs from roughly 5 to over 200 mg CH₄ m⁻² day⁻¹. Any satellite-derived emission estimate should propagate this uncertainty explicitly rather than reporting a single number.
ICESat-2 lake-surface elevation adds a useful constraint. A lake that is deepening (falling surface elevation at stable area) is likely expanding its anoxic bottom layer and increasing methanogenic conditions. A lake that is draining shows rapid surface-elevation drop before area visibly decreases. Combining ICESat-2 tracks with Landsat area time series therefore gives a more complete picture of lake state than either sensor alone, though ICESat-2's 91-day repeat means many drainage events will be missed between passes.
Infrastructure risk and what the data cannot tell you
Pipelines, roads, airstrips and buildings on permafrost are designed to specific ground-stability assumptions. Thermokarst lake expansion within a few hundred metres of linear infrastructure is a material risk: lateral thaw can undermine foundations, and drainage events can cause sudden ground settlement. InSAR subsidence maps at 20 to 50 m posting can identify areas of accelerating ground motion adjacent to assets, giving operators months to years of warning before structural failure.
The limit is that InSAR measures surface displacement, not the depth or geometry of the thaw front. A 3 cm subsidence signal could represent shallow active-layer deepening (recoverable) or deep talik formation beneath a lake (not recoverable on engineering timescales). Distinguishing the two requires ground-truth borehole data or airborne electromagnetic surveys. Satellite data narrows the area requiring investigation; it does not replace it.
Satellize runs multi-epoch Landsat lake-area change analysis and Sentinel-1 SBAS subsidence mapping as part of its permafrost monitoring analytics suite, the same open-constellation approach used in its Tonga crop-estimation work. For Arctic infrastructure clients, the practical first step is a baseline analysis covering the asset corridor: lake inventory, area-change rate per lake, and subsidence anomaly flags within a defined buffer. That scopes the field investigation rather than replacing it.
Typical figures
| Lake area detection limit | ~0.001 km² (1 ha) with Landsat 30 m; lakes smaller than ~4 pixels are systematically undercounted |
| Spatial resolution (optical) | 30 m (Landsat OLI); 10 m (Sentinel-2, supplementary use) |
| Spatial resolution (SAR InSAR) | 5 x 20 m (Sentinel-1 IW mode); 3 m (ALOS-2 PALSAR-2 fine beam) |
| Revisit cadence | 8 days (Landsat 8+9 combined); 6 days (Sentinel-1 at high latitudes); 91 days (ICESat-2 exact repeat) |
| InSAR vertical displacement sensitivity | 1 to 2 cm per year (Sentinel-1 SBAS stack); larger uncertainty in vegetated or high-moisture conditions |
| ICESat-2 lake surface elevation accuracy | Better than 10 cm vertical under clear skies (ATL13 product) |
| Optical archive depth | 1984 to present (Landsat TM/ETM+/OLI); 1972 for MSS (coarser, 60 m) |
| SAR archive depth | 2014 to present (Sentinel-1); 2014 to present (ALOS-2) |
| Cloud cover limitation | Optical unusable under cloud; SAR and ICESat-2 unaffected by cloud, though ICESat-2 requires cloud-free atmosphere for lidar return |
| Delivery formats | GeoTIFF lake-area polygons, NetCDF subsidence time series, CSV lake inventory with change metrics, GIS-ready alert layers |
Analytics Satellize can run
| Multi-decadal lake area change map | MNDWI water-index thresholding on Landsat TM/ETM+/OLI annual composites; change detection between user-defined epochs | GeoPackage polygon layer per epoch with area, perimeter and change-rate attributes; summary statistics report |
| Drainage event detection alert | Automated lake-area comparison between consecutive Sentinel-1 backscatter scenes and Landsat acquisitions; threshold trigger on area loss exceeding defined percentage | Near-real-time alert (latency 24 to 72 hours from acquisition) with lake ID, drainage fraction and scene date |
| Surface subsidence time series | SBAS InSAR processing of Sentinel-1 SLC stack; atmospheric correction using ERA5 tropospheric delay estimates | NetCDF displacement time series at 20 m posting; annual subsidence rate raster; anomaly flags within user-defined infrastructure buffer |
| Lake surface elevation profile | ICESat-2 ATL13 inland water surface product extraction and cross-epoch differencing; co-registration with Landsat area polygons | CSV of lake elevation per ICESat-2 overpass; trend analysis report indicating deepening or drawdown |
| Regional methane flux estimate | Lake area change multiplied by literature emission-factor ranges stratified by lake type (yedoma vs. non-yedoma); uncertainty propagated from both area and flux-factor distributions | Gridded flux estimate with explicit uncertainty bounds; not a point estimate |
| Infrastructure risk screening report | Spatial intersection of lake expansion polygons and subsidence anomaly zones with asset corridor buffer; ranked risk scoring per asset segment | PDF report with ranked asset segments, supporting map layers, and recommended field-investigation priorities |
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.