Glacial lake outburst flood hazard monitoring
Glacial lake outburst floods give little warning at ground level but leave a long satellite record of precursor signals. Multi-sensor monitoring of lake area, dam deformation and ice velocity can push that warning window from minutes to weeks.
Sensors
- Sentinel-2 MSI: 10 m resolution in visible and near-infrared bands; 5-day revisit at the equator, shorter at high latitudes. The normalised difference water index (NDWI) using bands 3 and 8 reliably delineates open water down to roughly 0.01 km² at this resolution. Persistent cloud at altitude frequently voids acquisitions for days or weeks at a stretch.
- Sentinel-1 SAR (C-band): 6-day repeat in Interferometric Wide Swath mode at 20 m ground range resolution. Open water appears as a near-zero backscatter surface in SAR amplitude, making lake detection cloud-independent. Repeat-pass InSAR from 6- or 12-day pairs resolves centimetre-scale surface displacement on moraine dams and adjacent ice, detecting subsidence or uplift that precedes breach.
- Landsat 8/9 OLI: 30 m multispectral resolution with a 16-day single-satellite revisit, extended to roughly 8 days when both satellites are used together. The archive extends to 1972 across the broader Landsat programme, making it the primary tool for reconstructing decades of lake area change and establishing pre-hazard baselines. Band 3 / Band 5 NDWI performs comparably to Sentinel-2 for lakes larger than a few hectares.
- ICEYE SAR (X-band): Commercially taskable at 1 m spotlight resolution with same-day or next-day revisit on request. X-band penetrates cloud and provides high-resolution amplitude imagery of dam surface texture and ice margin geometry when a specific lake requires urgent scrutiny. Tasking latency and cost make it a targeted supplement, not a systematic monitoring layer.
Why a lake that takes years to grow can drain in forty minutes
Proglacial lakes form when glacial retreat leaves a depression impounded by a moraine or ice dam. The Himalayas, Andes, Karakoram and Patagonian Icefields each host hundreds of such lakes, and their collective area has grown measurably over the past three decades as temperatures rise. The hazard is not the lake itself but the dam. Moraine dams are unconsolidated, often ice-cored, and fail through overtopping, internal piping or seismic triggering. When they go, the discharge can reach tens of thousands of cubic metres per second within minutes. The 1994 Lugge Tsho outburst in Bhutan killed 23 people; the 2013 Chorabari Lake outburst in Kedarnath, India contributed to a death toll in the thousands.
The satellite record matters because ground-based monitoring at these elevations is sparse to nonexistent. A lake that has expanded by 30 percent over two years, or a moraine dam showing centimetric subsidence, will not be noticed by anyone standing in the valley below. It will be noticed by a correctly configured satellite time series.
Reading lake area expansion from optical imagery
The NDWI, computed as (Green minus NIR) / (Green plus NIR), separates open water from ice and rock with reasonable confidence at Sentinel-2's 10 m scale. A time series of NDWI images, cloud-masked and composited over rolling 30-day windows, produces a lake-area curve. Acceleration in that curve, particularly after a warm season, is an operationally meaningful precursor signal. Published work on Himalayan lakes has used this approach to detect area increases of 5 to 10 percent per year as indicative of elevated risk.
The honest limit is cloud. High-altitude lake regions in the Hindu Kush-Himalaya can experience cloud cover exceeding 70 percent of days during the monsoon, which is precisely when meltwater input is highest. A 5-day Sentinel-2 revisit becomes an effective revisit of three to four weeks under those conditions. Landsat's longer archive compensates for this by providing a multi-decadal baseline against which any cloud-free acquisition can be placed in context, but it does not solve the operational gap during active hazard seasons.
What a floating roof gives away: SAR amplitude and InSAR on the dam
Sentinel-1 C-band SAR solves the cloud problem for lake detection. Open water, being specularly reflective at low incidence angles, returns almost no signal to the sensor. A lake therefore appears as a dark patch in amplitude imagery regardless of weather. The transition from bright (ice or moraine) to dark (water) can be tracked automatically across a 6-day time series. Area estimates from SAR amplitude agree with optical NDWI estimates to within 10 to 15 percent for lakes larger than roughly 0.1 km², with the discrepancy driven mainly by wind roughening of the water surface and ice floes.
InSAR adds a second, more sensitive diagnostic. By comparing the phase of two SAR acquisitions taken 6 or 12 days apart, it is possible to detect surface displacement on the moraine dam itself at centimetre precision. Subsidence of the dam crest, or lateral movement of the ice core, has been documented as a precursor to breach in several well-studied cases. The practical difficulty is that InSAR coherence degrades over vegetated or snow-covered surfaces; bare moraine performs well, but a dam partially covered by debris or seasonal snow may produce noisy interferograms. Stacking multiple pairs using time-series methods such as SBAS or PS-InSAR improves the signal-to-noise ratio substantially.
Ice velocity mapping from offset tracking or feature correlation between successive SAR acquisitions adds a third layer. Accelerating glacier flow toward the lake margin can indicate that the ice dam is thinning or that a surge is underway, both of which elevate breach probability.
Combining signals: what a monitoring framework actually looks like
No single sensor is sufficient. The operational architecture that makes sense for a high-risk lake combines a Sentinel-1 amplitude time series as the baseline detection layer (cloud-independent, 6-day cadence, automated), Sentinel-2 or Landsat optical composites for area quantification and colour-based ice-versus-water discrimination when skies permit, and InSAR pairs processed on a rolling basis to track dam and ice-margin deformation. ICEYE or comparable commercial SAR is reserved for urgent tasking when a precursor threshold is crossed.
Alert thresholds are necessarily empirical. A lake area increase of more than 15 percent in a single melt season, combined with detected dam subsidence exceeding 5 cm over a 30-day InSAR stack, would constitute a credible multi-sensor trigger in any defensible monitoring protocol. Those numbers are not universal; they need calibrating against the specific lake geometry, dam type and downstream exposure. Getting that calibration wrong in the conservative direction means alert fatigue. Getting it wrong in the other direction means no warning at all.
What satellite data cannot do here
Satellite monitoring cannot determine internal dam structure. Whether a moraine is ice-cored, how saturated it is, or whether piping channels are developing below the surface are questions that require ground investigation or, at best, inference from surface deformation patterns. Satellite data also cannot provide sub-hourly warning once a breach has initiated; by the time a SAR acquisition captures a partially drained lake, the flood wave may already be in the valley. The satellite contribution is pre-event hazard characterisation, not real-time breach detection.
Spatial resolution is a genuine constraint for small lakes. Lakes smaller than roughly 0.05 km² are at or below the reliable detection limit of Sentinel-2 and Sentinel-1 in standard acquisition modes. Some of the most dangerous lakes in the Karakoram are in this size range. ICEYE spotlight imagery at 1 m resolution can resolve them, but only if tasked deliberately.
Satellize runs lake-area and InSAR deformation analytics on open Sentinel and Landsat archives for government clients requiring systematic basin-scale monitoring, with commercial SAR tasking added when a specific lake crosses a precursor threshold. The workflow is structurally similar to the multi-temporal change detection underlying the Tonga crop-estimation programme, adapted for cryospheric rather than agricultural signals.
Typical figures
| Optical spatial resolution | 10 m (Sentinel-2 MSI); 30 m (Landsat 8/9 OLI) |
| SAR spatial resolution | 20 m ground range (Sentinel-1 IW); 1 m spotlight (ICEYE) |
| Optical revisit | 5 days (Sentinel-2 two-satellite); 8 days (Landsat 8+9 combined) |
| SAR revisit | 6 days (Sentinel-1 single pass); same-day to next-day on tasking (ICEYE) |
| InSAR displacement sensitivity | ~5 mm per interferogram in line-of-sight; improved to ~1 mm with SBAS/PS stacking |
| Minimum detectable lake area (optical) | ~0.01 km² (Sentinel-2 NDWI); ~0.05 km² (Landsat OLI) |
| Minimum detectable lake area (SAR amplitude) | ~0.1 km² reliably; smaller lakes detectable with ICEYE spotlight |
| Archive depth | Sentinel-1 from 2014; Sentinel-2 from 2015; Landsat from 1972 |
| Cloud impact on optical | Effective revisit can degrade to 3 to 4 weeks during monsoon at high altitude |
| Delivery formats | GeoTIFF lake-extent polygons, displacement rasters, time-series CSV, GIS-ready vector alerts |
Analytics Satellize can run
| Lake area time series | NDWI thresholding on cloud-masked Sentinel-2 and Landsat composites; multi-temporal change detection | Monthly GeoTIFF and polygon layer with area in km², trend chart, anomaly flag when seasonal growth rate exceeds calibrated threshold |
| SAR-derived lake extent | Amplitude thresholding on Sentinel-1 IW SLC or GRD; water/non-water binary classification | 6-day cadence vector polygon; cloud-independent complement to optical time series; integrated into same area-trend dashboard |
| Moraine dam deformation map | Sentinel-1 repeat-pass InSAR; SBAS time-series inversion for cumulative displacement | Displacement raster (mm in line-of-sight) per 12-day epoch; subsidence alert if threshold crossed; delivered as GeoTIFF and summary report |
| Ice velocity and flow acceleration | SAR offset tracking or normalised cross-correlation between successive Sentinel-1 acquisitions | Velocity vector field (m/day) at 20 m posting; acceleration flag for ice margins adjacent to monitored lakes |
| Multi-sensor precursor score | Rule-based fusion of lake area trend, dam deformation signal and ice velocity anomaly; thresholds calibrated per lake | Weekly hazard-level indicator (low / elevated / high) per monitored lake, delivered as JSON feed or PDF briefing |
| Historical baseline reconstruction | Landsat archive NDWI time series from 1984 onward; linear and breakpoint trend analysis | Decadal lake area chart per site; identification of lakes with statistically significant accelerating growth |
| Urgent high-resolution acquisition | Commercial ICEYE SAR tasking triggered by precursor threshold breach; 1 m spotlight amplitude analysis | Same-day or next-day GeoTIFF with dam surface texture and lake margin geometry; analyst annotation of structural features |
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.