Sensors
- Sentinel-2 MSI: 10 m resolution in visible and near-infrared bands; 5-day revisit at the equator with two satellites. The red-edge bands (B5, B6, B7 at 20 m) are the primary source for NDVI and vegetation stress indices. Band 3 (green) and Band 4 (red) support turbidity proxies in river reaches. Free archive from 2015.
- Landsat-9 OLI: 30 m multispectral resolution; 16-day revisit. OLI's coastal/aerosol band (Band 1, 443 nm) and green band are used in published turbidity retrieval algorithms for inland waters. Paired with Landsat-8, effective revisit shortens to 8 days. Archive continuity from 1972 across the Landsat series is irreplaceable for long-term deforestation baselines.
- Planet SuperDove: 3 m resolution; near-daily revisit over most tropical latitudes. Eight spectral bands including two red-edge channels. Useful for resolving individual pit boundaries and confirming active workings flagged by coarser sensors. Commercial licence required; archive depth varies by site.
- Sentinel-1 SAR: C-band SAR at 10 m (IW mode); 6-day revisit per satellite, 12-day per platform. Cloud-penetrating, which matters enormously in equatorial regions where optical revisit is routinely blocked for weeks. Backscatter change detects new clearing and ponded water in pit areas even under persistent cloud cover.
Why the proxy chain works, and where it breaks
Mercury used in ASGM amalgamation enters river systems in two forms: dissolved methylmercury and particle-bound inorganic mercury attached to fine sediment. Neither is spectrally distinct from orbit at any currently operational sensor resolution. What satellites can see is the sediment itself. Turbid plumes in rivers downstream of active pits carry suspended solids concentrations that correlate, in published Amazon basin studies, with elevated mercury loads, because the same disturbed soils that cloud the water also carry the contamination.
The proxy chain has a real weakness: turbidity is not a direct mercury measurement. A river can be turbid from natural erosion or from legal dredging with no mercury involvement at all. The method only becomes meaningful when turbidity anomalies are co-located with mapped active ASGM pits and persist across multiple revisits. Treat the output as a spatial prioritisation tool for field sampling, not as a contamination certificate.
What a stressed riparian canopy gives away
Riparian vegetation within roughly 100 to 500 metres of contaminated river margins shows measurable NDVI suppression in published studies from the Tapajós and Madeira basins. Sentinel-2's red-edge bands are sensitive to chlorophyll reduction before visible yellowing appears, giving a lead time of weeks over simple RGB inspection. The signal is subtle: NDVI anomalies in mercury-stressed vegetation typically fall in the range of 0.05 to 0.15 below local seasonal baselines, which is within Sentinel-2's detection capability but easily confused with drought stress, selective logging or natural senescence.
Separating mercury stress from other causes requires a multi-temporal baseline of at least two to three years and careful masking of logged areas and agricultural clearings. The vegetation stress layer is therefore a supporting indicator, not a standalone diagnostic. Its value is in narrowing the spatial search: reaches where both turbidity anomaly and canopy stress co-occur downstream of a confirmed pit cluster are the highest-priority candidates for water sampling.
Mapping active pits: the detection logic
Active ASGM pits have a recognisable spectral and structural signature. Bare or ponded ground in otherwise forested terrain produces strong contrast in near-infrared reflectance. Ponded pit water is typically very dark in NIR and bright in shortwave infrared relative to surrounding vegetation. At Sentinel-2's 10 m resolution, pits larger than roughly 20 to 30 metres across are reliably detectable; smaller workings may be missed or merged with neighbours.
Sentinel-1 SAR adds a cloud-independent confirmation layer. Flooded pit surfaces produce specular reflection, lowering backscatter sharply relative to forest. Change detection between two SAR acquisitions separated by 12 to 24 days can identify new clearings that opened between optical windows. In the Venezuelan and Brazilian Amazon, where cloud cover exceeds 70 percent of days in the wet season, SAR is not supplementary; it is the primary detection mechanism for roughly half the year.
Planet SuperDove at 3 m resolves individual pit morphology, distinguishing active high-pressure hose workings (monitor pits) from abandoned waterlogged depressions. That distinction matters for assessing current mercury use, since abandoned pits may still contribute turbidity through bank erosion without active amalgamation.
Turbidity retrieval from multispectral imagery
Published turbidity retrieval algorithms for inland tropical rivers use the ratio of green to near-infrared reflectance, or the red band alone as a single-band proxy for total suspended solids. Sentinel-2 Band 3 (560 nm) and Band 4 (665 nm) are the most commonly applied. Landsat-9 OLI Band 3 (562 nm) and Band 4 (655 nm) serve the same role with coarser resolution but a longer archive. Calibration against field turbidity measurements (in NTU or mg/L suspended solids) is necessary for any quantitative output; without site-specific calibration, the indices are comparative rather than absolute.
Cloud shadow is the principal artefact source. Shadows lower surface reflectance in ways that mimic clear, low-turbidity water, producing false negatives in contamination-adjacent reaches. Rigorous cloud and shadow masking, followed by compositing across multiple acquisitions, is standard practice and reduces but does not eliminate the problem. Honest reporting of cloud-affected observation gaps is part of any credible deliverable.
Practical limits and what field teams need to know
The full proxy chain, pit detection plus turbidity anomaly plus vegetation stress, narrows a large river basin to a set of candidate reaches. It does not produce a mercury concentration map. Field teams deploying water sampling kits or portable XRF instruments still need to visit those reaches; the satellite layer tells them where to go first, not what they will find.
Archive depth is a genuine asset here. Sentinel-2 coverage from 2015 and Landsat from the 1970s allow analysts to establish pre-mining baselines and measure the rate of pit-cluster expansion alongside the propagation of turbidity anomalies downstream. That temporal narrative is often the most persuasive element for regulators and prosecutors. Satellize has applied similar multi-index temporal analysis in its crop-estimation work for the Kingdom of Tonga, where separating signal from seasonal noise in vegetation indices is an analogous methodological challenge.
One further limit: rivers with naturally high suspended sediment loads, such as whitewater Amazonian tributaries, have a high background turbidity that compresses the detectable anomaly from mining activity. The method performs better on clearwater and blackwater rivers, where the contrast between baseline and ASGM-affected reaches is larger.
Turning a proxy map into a monitoring programme
A useful monitoring programme runs on a quarterly cadence for the turbidity and vegetation indices, with monthly SAR-based pit detection during the wet season when optical data is unreliable. Alerts trigger when a new pit cluster appears within a defined buffer of a river reach that was previously clean, or when turbidity in a monitored reach exceeds a defined threshold relative to its seasonal baseline.
The output that regulators find most actionable is a ranked list of river reaches with associated evidence layers: the pit cluster map, the turbidity time series, the vegetation stress anomaly, and the cloud-gap record showing which periods lacked observation. That transparency about coverage gaps is what separates a credible monitoring product from one that overstates its certainty.
Typical figures
| Optical spatial resolution (pit detection) | 3 m (Planet SuperDove) to 30 m (Landsat-9 OLI); Sentinel-2 at 10 m is the operational default |
| SAR spatial resolution | 10 m (Sentinel-1 IW mode) |
| Optical revisit (cloud-free) | 5 days (Sentinel-2 two-satellite constellation); effectively longer in wet-season tropics due to persistent cloud |
| SAR revisit | 6 days per Sentinel-1 satellite; 12 days per platform; cloud-independent |
| Minimum detectable pit size | Approximately 20-30 m diameter at Sentinel-2 10 m resolution; smaller pits require Planet SuperDove tasking |
| Spectral bands for turbidity proxy | Green (560 nm) and red (665 nm) on Sentinel-2; equivalent OLI bands on Landsat-9 |
| Spectral bands for vegetation stress | Sentinel-2 red-edge B5 (705 nm), B6 (740 nm), B7 (783 nm) at 20 m resolution |
| Archive depth | Sentinel-2 from 2015; Landsat series from 1972; Sentinel-1 from 2014 |
| Processing latency (operational monitoring) | Typically 24-72 hours after satellite acquisition for automated index products |
| Delivery formats | GeoTIFF rasters, GeoJSON vector pit boundaries, CSV turbidity time series, PDF quarterly reports |
Analytics Satellize can run
| Active ASGM pit cluster map | Spectral change detection (NIR/SWIR contrast against forest baseline) combined with Sentinel-1 SAR backscatter change; binary classification with confidence score | Monthly GeoJSON polygon layer with pit area, first-detection date, and activity status |
| Downstream turbidity anomaly layer | Green/red band ratio turbidity index computed per river reach; anomaly scored against 3-year seasonal baseline; cloud-gap periods flagged explicitly | Quarterly GeoTIFF and river-reach ranked CSV, with observation-gap record attached |
| Riparian NDVI stress index | Red-edge NDVI and red-edge chlorophyll index (CIre) from Sentinel-2; anomaly detection against pixel-level seasonal baseline; masking of logged and agricultural areas | Quarterly GeoTIFF with stress magnitude and spatial extent statistics per river reach |
| Combined proxy contamination risk score | Spatial overlay of pit proximity, turbidity anomaly, and vegetation stress; weighted scoring per river reach segment; published co-location methodology from Amazon basin literature | Ranked river-reach priority list for field sampling, delivered as GIS layer and PDF summary |
| Pit expansion rate time series | Multi-date pit polygon differencing from Sentinel-2 and Landsat archive; area change per quarter expressed as hectares per month | Annual trend report with per-site expansion charts and archive imagery strips |
| Wet-season SAR activity alert | Sentinel-1 coherence change and backscatter anomaly detection for new clearing events during periods of optical cloud cover exceeding 70 percent | Near-real-time alert (within 48 hours of SAR acquisition) with polygon and confidence flag |
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.