Tailings dam supernatant pond extent and freeboard estimation
SAR backscatter and optical water indices map the supernatant pond on tailings storage facilities at 6-day repeat, flagging rapid expansion before it reaches the embankment crest. The method supports risk triage, not survey-grade freeboard measurement.
Sensors
- Sentinel-1 SAR (C-band): 10 m ground range resolution in IW mode, 6-day repeat at mid-latitudes, cloud-independent. Open water returns near-zero backscatter (typically below -15 dB), making pond boundaries detectable even under heavy cloud or at night. The 6-day cadence is fast enough to catch rapid pond expansion events but not continuous.
- Sentinel-2 MSI: 10 m visible and NIR bands, 20 m SWIR, 5-day repeat (with both satellites). NDWI and MNDWI indices delineate the pond edge at sub-pixel accuracy in clear conditions. Cloud cover is the hard constraint: during wet seasons, consecutive usable acquisitions may be weeks apart over some sites.
- Planet SuperDove: 3 m resolution, near-daily revisit, 8 spectral bands including NIR. Useful for resolving fine pond-edge geometry and detecting thin beach strips between water and crest. Requires a commercial licence; archive depth varies by site coverage history.
- COSMO-SkyMed Second Generation (X-band SAR): Spotlight mode delivers 1 m resolution imagery, sufficient to resolve narrow freeboard strips and embankment geometry. Tasked commercially; X-band is more sensitive to surface roughness than C-band, which can sharpen the water-land boundary but also introduces ambiguity on wind-roughened pond surfaces.
What open water looks like to a radar
Synthetic aperture radar measures how much microwave energy a surface scatters back toward the sensor. Calm open water acts as a specular reflector: almost nothing returns. In Sentinel-1 C-band imagery processed to sigma-naught, a supernatant pond typically registers below -15 dB, while surrounding tailings material, damp or dry, scatters considerably more. That contrast is the detection mechanism, and it works regardless of cloud, smoke or darkness.
The practical floor is wind. A pond surface roughened by gusts above roughly 3 to 4 m/s begins to scatter more energy back, raising its apparent backscatter and blurring the boundary with the surrounding beach. At Sentinel-1's 10 m pixel size, a narrow beach strip of perhaps 5 to 10 m between the pond edge and the embankment crest can be missed entirely, or confused with wet tailings. This is the core ambiguity the method cannot resolve on its own.
Optical indices fill the gap when the sky cooperates
The Normalised Difference Water Index (NDWI, McFeeters 1996) and its SWIR variant MNDWI use the strong absorption of liquid water in the near-infrared and short-wave infrared to separate water from land. At Sentinel-2's 10 m visible/NIR resolution, the pond boundary can be mapped with reasonable confidence in a single clear-sky pass. Planet SuperDove's 3 m pixels resolve the beach strip more precisely, which matters when the question is whether 2 metres of freeboard remain or 20.
The catch is cloud. Many high-risk tailings facilities sit in tropical or sub-tropical climates where the wet season, the period of maximum pond risk, is also the period of minimum optical availability. A site in the Andes or central Africa may yield only two or three usable Sentinel-2 acquisitions across a four-month wet season. SAR is not a fallback in that scenario; it is the primary sensor.
Estimating freeboard without a survey pole
Freeboard, the vertical distance between the pond surface and the embankment crest, cannot be read directly from a flat satellite image. What remote sensing provides is pond surface area. Converting area to a freeboard proxy requires a bathymetric or storage curve for the facility: the relationship between water volume (or pond area) and the elevation of the water surface. Where operators have shared that curve, or where a digital elevation model of sufficient quality exists, satellite-derived pond area can be translated into an approximate water surface elevation and hence an estimated freeboard.
The word approximate carries weight. A 30 cm freeboard threshold, which some national regulations treat as a trigger for emergency action, is well below the vertical accuracy of most freely available DEMs over embankment crests. Shuttle Radar Topography Mission data has a stated absolute vertical accuracy of around 16 m at 90 percent confidence. Even commercial stereo DEMs rarely achieve better than 0.5 to 1 m over complex earthwork geometry. The satellite method is therefore a rapid-screening and trend-monitoring tool. It flags that the pond has grown by 15 percent in 12 days. It does not certify that freeboard is 0.4 m.
Change detection is the operational product
The most defensible use of the method is not absolute pond area on a given date but the rate of change across a time series. A pond that expands by 20 percent of its surface area within a single 6-day Sentinel-1 interval, during a period of heavy rainfall, is a different risk profile from one that has remained stable for three months. Automated thresholding on the SAR backscatter stack, combined with a rainfall anomaly layer from ERA5 reanalysis or GPM IMERG, allows a monitoring system to distinguish rain-driven expansion from anomalous inflow or internal drainage failure.
Sentinel-1's archive runs back to 2014 for many sites, which is long enough to establish a seasonal baseline: how large does the pond typically grow in a wet season at this facility? Departures from that baseline, rather than absolute area, are the alert signal. This approach also reduces false positives from seasonal variation that would otherwise trigger unnecessary inspections.
What this method will not tell you
Several failure modes fall outside the method's scope. Embankment piping, liquefaction and internal erosion are subsurface processes that produce no surface water signature until breach is imminent or complete. Seepage anomalies at the toe are a separate detection problem covered by thermal and moisture mapping methods. The satellite pond-extent method is specifically about the surface water body and its proximity to the crest.
Facility geometry also matters. A narrow, elongated pond on a valley-fill impoundment may have a very different SAR signature from a broad upstream-construction pond on flat terrain. Mixed pixels at 10 m can misclassify wet tailings beach as open water, particularly in the SWIR bands where both surfaces absorb strongly. Manual review of any automated alert against the raw imagery remains necessary before mobilising a ground team.
Satellize runs this analysis as part of its broader satellite-data analytics work, applying the same SAR and optical pipeline used in its Tonga crop-estimation programme to industrial monitoring contexts. Clients receive a GIS layer and a structured alert feed rather than raw imagery.
Putting a monitoring programme together
A practical monitoring design for a high-consequence tailings facility combines Sentinel-1 as the backbone (free, 6-day, cloud-independent) with Sentinel-2 for geometric confirmation on clear days and commercial tasking of COSMO-SkyMed or Planet when an alert fires and rapid high-resolution confirmation is needed. The cost structure is therefore tiered: most monitoring runs on open data, and commercial capacity is reserved for events.
Latency is manageable. Sentinel-1 data is typically available in the Copernicus Data Space within a few hours of acquisition. An automated processing chain can deliver a pond-extent update and comparison against the previous pass within the same working day. That is fast enough to support the kind of operational decision, call the dam safety engineer, mobilise inspection, notify the regulator, that a freeboard-risk event demands.
Typical figures
| Primary SAR resolution | 10 m (Sentinel-1 IW mode, ground range) |
| High-resolution SAR option | 1 m (COSMO-SkyMed SG Spotlight, tasked commercially) |
| Optical resolution | 3 m (Planet SuperDove) to 10 m (Sentinel-2 visible/NIR) |
| Revisit (SAR, cloud-independent) | 6 days at mid-latitudes (Sentinel-1 A+B); shorter with commercial tasking |
| Revisit (optical, clear sky only) | 5 days (Sentinel-2 A+B); near-daily (Planet SuperDove) |
| Spectral bands used | C-band SAR (5.4 GHz); Green, NIR, SWIR1, SWIR2 for NDWI/MNDWI |
| Minimum detectable pond expansion | Approximately one to two Sentinel-1 pixels (10–20 m) in favourable wind conditions; coarser in wind-roughened conditions |
| Vertical freeboard accuracy | Not survey-grade; dependent on facility storage curve and DEM quality; indicative only |
| Archive depth | Sentinel-1: 2014 to present for most sites; Sentinel-2: 2015 to present |
| Delivery formats | GeoTIFF pond-extent masks, GeoJSON change polygons, structured alert feed, PDF inspection briefing |
Analytics Satellize can run
| Pond surface area time series | SAR backscatter thresholding (Otsu or region-growing) on Sentinel-1 sigma-naught stack; NDWI/MNDWI on Sentinel-2 and Planet for optical confirmation | Monthly GeoTIFF stack with pond-extent polygons and area statistics in CSV |
| Rapid pond expansion alert | Automated change detection comparing current SAR-derived pond area against rolling 90-day baseline; anomaly flagged when expansion exceeds configurable threshold (e.g. 10 percent in one 6-day interval) | Email or API alert with pond-extent overlay, rainfall anomaly context from GPM IMERG, and link to raw imagery |
| Freeboard proxy estimate | Pond area converted to approximate water surface elevation via operator-supplied or satellite-derived storage curve; compared against embankment crest elevation from available DEM | Tabular freeboard estimate with explicit uncertainty range; flagged as indicative, not survey-grade |
| Seasonal baseline and exceedance report | Percentile analysis of Sentinel-1 pond-area time series across available archive; current season compared against historical wet-season envelope | Annual wet-season monitoring report with exceedance plots and annotated imagery |
| High-resolution crest proximity assessment | Planet SuperDove or COSMO-SkyMed SG imagery processed to resolve beach-strip width between pond edge and embankment crest following an alert event | Single-scene GeoTIFF with manually verified pond boundary and annotated crest proximity measurement |
| Multi-facility portfolio dashboard | Parallel SAR processing across a portfolio of named facilities; status classified as stable, watch or alert based on area-change thresholds | Weekly GIS layer and tabular status summary across all monitored sites |
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.