Supraglacial lake formation and rapid drainage on ice sheets

Supraglacial lakes on the Greenland and Antarctic ice sheets can drain catastrophically through moulins, injecting meltwater to the bed and accelerating ice flow. Multispectral reflectance retrieves lake area and depth; ICESat-2 elevation data confirms drainage events that optical imagery alone can miss.

Sensors

Sentinel-2 MSI: 10 m resolution in visible bands (B2, B3, B4) and 20 m in red-edge and SWIR bands; 5-day revisit at mid-latitudes, somewhat better poleward with overlapping swaths. The blue and green bands are the primary inputs for radiative-transfer depth retrieval. Cloud cover over Greenland in summer is the main operational constraint.
Landsat 8/9 OLI: 30 m multispectral resolution with a 16-day single-satellite revisit, reduced to roughly 8 days when both satellites are combined. Coastal-aerosol band (Band 1, 443 nm) and blue band (Band 2, 482 nm) extend the depth-retrieval dynamic range for deeper, optically thick lakes. Archive back to 2013 (Landsat 8) enables multi-year trend analysis.
ICESat-2 ATLAS: Photon-counting lidar measuring surface elevation to within a few centimetres vertically along six 17 m diameter ground tracks, with an 91-day exact repeat and denser polar coverage due to orbital convergence. Detects the surface-elevation drawdown signature of a draining lake and can measure water-surface elevation directly, providing depth validation independent of optical assumptions.
MODIS Terra/Aqua: 250 m resolution in red and near-infrared bands, with daily to twice-daily coverage. Too coarse to resolve small lakes or retrieve depth accurately, but useful for monitoring the seasonal onset of large lake fields and for flagging rapid area changes between Sentinel-2 or Landsat acquisitions. Archive extends to 2000, giving the longest continuous record.

What a blue lake on white ice is actually telling you

Supraglacial lakes form when surface melt exceeds percolation capacity, pooling in topographic lows on the ice surface. On the Greenland ice sheet, they appear each May to August across a widening elevation band that has crept inland as summers have warmed. They matter for two reasons: they lower surface albedo, absorbing more solar radiation and accelerating local melt; and a fraction of them drain suddenly, routing large volumes of water to the ice-sheet bed in timescales of hours to a few days.

That rapid drainage is the dynamic concern. Basal water pressure spikes reduce friction between ice and bedrock, producing measurable speed-up events at the glacier surface. The 2006 North Lake drainage event on the western Greenland ice sheet, documented in published literature, transferred an estimated 44 million cubic metres of water to the bed in roughly 1.4 hours. Detecting which lakes are likely to drain, and catching drainage events as they happen, is therefore a meaningful early-warning problem rather than a purely academic one.

Depth from colour: the physics and the caveats

Pure water absorbs strongly in the red and near-infrared, and scatters in the blue-green. The ratio of reflectance in the blue band to the red band increases predictably with water depth up to a saturation point, typically around 6 to 8 metres for optically clear meltwater over a bright ice bottom. This is the basis for radiative-transfer inversion depth retrieval: fitting a physically based model of downwelling irradiance, water-column attenuation and bottom reflectance to the observed spectral signal.

Sentinel-2's 10 m blue and green bands make it the instrument of choice for spatial detail. Published studies using this approach report depth retrieval errors on the order of 0.5 to 1.5 m for lakes shallower than roughly 6 m, degrading for deeper water where the bottom signal is lost. Turbid meltwater, surface wind ripples and thin ice lids all introduce additional uncertainty. Landsat 9's coastal-aerosol band helps extend sensitivity in the shallow end of the depth range. Neither sensor penetrates ice, so a lake that has partially refrozen or is covered by a thin snow layer will be underestimated or missed entirely. That is an honest limit of the optical method.

Catching the drain: time-series change detection

A draining lake leaves a clear optical signature: water area collapses between successive acquisitions, often accompanied by a residual dark wet patch or exposed ice. With Sentinel-2's 5-day revisit, a drainage event that occurs mid-interval may be confirmed only at the next clear-sky pass. MODIS's daily coverage can bracket the event more tightly in time, though at 250 m resolution it cannot resolve the spatial detail of the drainage scar.

ICESat-2 adds a dimension that optical sensors cannot provide. When an ICESat-2 ground track crosses a lake before and after drainage, the surface-elevation change gives a direct volume estimate without any optical depth assumption. The 91-day exact repeat limits opportunistic coverage, but the denser polar track spacing means that lakes on the western Greenland ablation zone are crossed more frequently than the nominal repeat suggests. Combining ICESat-2 elevation anomalies with Sentinel-2 area time series is the most defensible method for volume-change attribution.

Automated change detection on the time series uses spectral water indices (the Normalised Difference Water Index, NDWI, using green and near-infrared bands) thresholded against a dynamic ice-surface background. A lake that shrinks by more than a defined fraction of its area between two cloud-free acquisitions triggers a drainage flag. False positives arise from partial cloud shadow, which can mimic a dark ice surface. Quality filtering on scene-level cloud masks is non-negotiable.

What the archive reveals about long-term change

The Landsat record back to 2013 and MODIS back to 2000 allow analysis of interannual variability in lake area, count and distribution. Published work using these archives shows that the inland limit of supraglacial lake formation has migrated to higher elevations over the past two decades, consistent with surface-air temperature trends. Lake area in a given summer correlates with positive degree-day totals, which makes the satellite record a useful proxy for melt intensity when in-situ measurements are sparse.

Multi-year time series also reveal which topographic basins repeatedly host lakes, information that is directly relevant to assessing which parts of the ice sheet are most susceptible to future basal lubrication. That spatial persistence is detectable from the archive without any field campaign.

Operational limits worth stating plainly

Cloud cover is the dominant practical constraint. Greenland's western ablation zone is frequently cloud-free in summer, but the eastern and southern margins are not. A persistent cloud deck can produce a gap of two to three weeks in the Sentinel-2 record during the peak melt season, which is long enough to miss an entire lake-fill-and-drain cycle. SAR imagery from Sentinel-1 can detect open water under some conditions via low backscatter, but the contrast between wet ice and open water is ambiguous and depth retrieval from SAR alone is not operationally established.

Depth retrieval saturates at roughly 6 to 8 m for the optical method. Very large, deep lakes, which may hold the greatest volume, are the ones where optical estimates are least reliable. ICESat-2 track crossings provide ground truth but are not guaranteed for any specific lake at any specific time. Combining sensors reduces but does not eliminate these gaps. Any operational system should report retrieval confidence alongside the depth estimate, not just the number.

From detection to decision

For ice-sheet research programmes and polar agencies, the actionable outputs are a seasonal lake inventory with area and estimated volume, a drainage-event log with timing and confidence level, and a multi-year trend layer showing which basins are expanding their lake frequency. These products feed directly into ice-sheet models that require basal lubrication inputs.

Satellize runs this type of analytics pipeline on open constellations including Sentinel-2 and Landsat, with ICESat-2 elevation data integrated for validation passes. The methodology is the same radiative-transfer inversion and NDWI change-detection approach described here, applied operationally rather than as a one-off study. The Tonga crop-estimation programme is a different domain, but the underlying workflow, ingesting multitemporal open-satellite data and extracting quantitative geophysical parameters, is the same architecture. Clients with polar mandates can request a scoped analysis of a defined basin or drainage catchment rather than a full ice-sheet run.

Typical figures

Spatial resolution (depth retrieval)	10 m (Sentinel-2 B2/B3); 30 m (Landsat 8/9 OLI)
Revisit for change detection	5 days (Sentinel-2, cloud-permitting); 8 days combined Landsat 8+9; 1–2 days (MODIS, 250 m)
ICESat-2 elevation precision	~3 cm vertical, 17 m footprint diameter; 91-day exact repeat
Spectral bands used for depth retrieval	Blue (440–490 nm) and green (540–580 nm) primary; red (640–680 nm) for bottom contrast; coastal aerosol (443 nm, Landsat 9) for shallow extension
Depth retrieval range	0.5 m to ~6–8 m (optical saturation limit for clear meltwater over ice)
Depth retrieval uncertainty	±0.5–1.5 m for lakes shallower than 6 m; degrades significantly beyond saturation depth
Minimum detectable lake area	~100 m² at 10 m resolution (Sentinel-2); ~900 m² at 30 m (Landsat)
Archive depth	Sentinel-2: 2015–present; Landsat 8/9: 2013–present; MODIS: 2000–present; ICESat-2: 2018–present
Coverage	Global; polar orbits provide denser revisit above 60°N/S
Delivery formats	GeoTIFF depth rasters, GeoJSON lake polygons, CSV drainage-event log, NetCDF time-series stacks

Analytics Satellize can run

Seasonal lake inventory	NDWI thresholding on Sentinel-2 and Landsat time series; per-scene cloud masking	GeoJSON polygon layer with area, centroid, date and cloud-confidence flag for each detected lake
Lake depth raster	Radiative-transfer inversion using blue-to-red band ratio calibrated against known bottom reflectance of bare glacier ice	GeoTIFF depth map per acquisition, with per-pixel retrieval confidence layer
Lake volume estimate	Depth raster integrated over lake-polygon area; ICESat-2 surface-elevation crossings used for independent validation where track coverage exists	CSV time series of estimated volume per lake basin, with uncertainty bounds
Rapid drainage event detection	Area change detection between successive cloud-free Sentinel-2 acquisitions; threshold set at >50% area loss within one revisit interval	Drainage-event alert with lake ID, date range, estimated volume lost and confidence level
ICESat-2 elevation anomaly extraction	Differencing of repeat ICESat-2 ATL06 land-ice elevation profiles over known lake basins to quantify surface drawdown	Elevation-change report per track crossing, flagged against optical lake inventory
Multi-year lake frequency map	Pixel-level counting of lake-presence observations across full Sentinel-2 and Landsat archive for defined study region	GeoTIFF persistence raster showing fraction of melt seasons each pixel hosted open water
Melt-season onset and lake-field extent summary	MODIS daily time series processed for first-detection date of water index anomaly above background	Annual report table with onset date, peak lake area, total drainage events and interannual comparison

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.