Acid mine drainage detection in receiving water bodies
Multispectral indices sensitive to iron-hydroxide precipitation, turbidity and pH-correlated colour shifts can flag acid mine drainage in rivers and lakes at 10–30 m resolution. Detection is indirect: satellites see optical proxies, not pH meters.
Sensors
- Sentinel-2 MSI: 10 m resolution in visible and NIR bands (B2–B8A), 5-day revisit at mid-latitudes with two satellites. Band 4 (red, 665 nm) and Band 3 (green, 560 nm) respond strongly to ferric iron and suspended sediment. The red-edge bands (B5–B7, 705–783 nm) add sensitivity to turbidity gradients. Cloud cover is the principal operational constraint.
- Landsat 8/9 OLI: 30 m resolution, 16-day revisit per satellite (8-day combined). The archive extends to 1972 (Landsat 1 MSS), giving roughly 50 years of comparable imagery for attribution of degradation onset. OLI's coastal-aerosol band (443 nm) helps distinguish iron staining from suspended mineral sediment.
- Planet SuperDove: 3–5 m resolution, daily revisit over most land areas. Eight spectral bands include a dedicated red-edge channel. Useful for resolving narrow river reaches (under 10 m wide) where Sentinel-2 pixels mix water and bank. Requires a commercial licence; archive depth is shorter than Landsat.
- NASA PACE OCI: Launched 2024. Hyperspectral ocean-colour instrument covering 340–890 nm at roughly 1 km resolution and daily global coverage. Not useful for narrow rivers, but provides high-spectral-resolution retrieval of dissolved organic matter and iron in larger water bodies, potentially improving discrimination between AMD iron and natural humic colouration.
What a rusted river is actually telling you
Acid mine drainage forms when sulphide minerals, principally pyrite, are exposed to water and oxygen during mining or in abandoned workings. The resulting sulphuric acid dissolves iron and heavy metals. When that acidic, iron-rich water reaches a receiving stream and its pH rises slightly, ferric iron precipitates as iron-hydroxide minerals, most visibly schwertmannite and goethite. The result is the characteristic ochre or rust staining visible in rivers draining many coal, copper and gold districts worldwide.
Satellites do not measure pH or dissolved metal concentrations directly. What they measure is the optical consequence: a shift in water-leaving reflectance driven by the strong absorption and scattering properties of ferric iron minerals in suspension. The red and green bands respond most clearly. A high ratio of red-band to green-band reflectance in a water pixel is a reliable qualitative indicator of iron-hydroxide loading, provided the analyst controls for suspended mineral sediment from non-AMD sources, algal bloom and shallow-water substrate effects. That last caveat matters: a sandy riverbed can produce a similar spectral signature in very shallow water.
The spectral indices that do the work
Several published indices have been validated against in-situ AMD measurements. The Iron Band Ratio (IBR), computed as the ratio of red reflectance to blue or green reflectance, correlates with ferric iron concentration in peer-reviewed studies using both Landsat and Sentinel-2 data. The Normalised Difference Turbidity Index (NDTI), using red and green bands, captures total suspended solids, which co-vary with AMD plume extent even when iron staining is partially masked by high sediment loads. For Sentinel-2 specifically, the red-edge bands (B5 at 705 nm, B6 at 740 nm) improve discrimination between clear, turbid and iron-rich water classes in shallow reaches.
Atmospheric correction is non-negotiable. Water-leaving reflectance in the visible is tiny, often under 5 percent, so even small errors in aerosol correction produce artefacts that mimic or mask AMD signatures. The Sen2Cor processor (ESA) and ACOLITE (Royal Belgian Institute of Natural Sciences) are the standard open-source tools for Sentinel-2 water-quality work. ACOLITE in particular was designed for inland and coastal water bodies and handles the adjacency effect from bright riverbanks better than generic land-surface processors.
pH itself is not retrievable from current multispectral data. Highly acidic water (pH below roughly 4) tends to carry more dissolved ferric iron and therefore shows stronger red-band reflectance, but the relationship is not monotonic and varies with catchment geology. Satellite data can flag a plume; it cannot replace a field pH meter or ICP-MS analysis for regulatory compliance.
Forty years of archive as a forensic tool
The Landsat archive is uniquely useful for AMD attribution. A regulator or legal team asking when a river first showed signs of iron staining can, in principle, walk back through Landsat imagery to the 1980s and identify the season and approximate year when the spectral signature appeared or intensified. This has real value in jurisdictions where mine closure liability is contested: the archive provides an independent, satellite-derived timeline that is difficult to dispute.
The practical limit is cloud cover and the 16-day revisit. In humid tropical regions, cloud-free composites over a single dry season may contain only two or three usable scenes per year. Combining Landsat and Sentinel-2 into a harmonised time series, as the HLS (Harmonised Landsat Sentinel-2) product from NASA does, roughly doubles usable observations. Even so, a contamination event lasting only a few days after a storm-driven tailings release may fall entirely between cloud-free acquisitions.
Resolution floors and the narrow-river problem
Sentinel-2's 10 m pixels are adequate for rivers wider than roughly 20–30 m, allowing at least one or two pure water pixels per cross-section. Below that width, mixed pixels dominate and iron indices become unreliable. Many AMD-affected headwater streams in mountainous mining districts are 5–15 m wide. Planet SuperDove at 3–5 m resolution resolves these reaches, but introduces its own complications: the shorter archive, the need for a commercial licence, and the fact that eight-band SuperDove data requires different atmospheric correction workflows than Sentinel-2.
Lakes and reservoirs receiving AMD are easier targets. A water body of even a few hectares presents many pure water pixels to Sentinel-2, and temporal compositing over weeks can suppress cloud gaps. The challenge there is distinguishing AMD iron from naturally humic, peat-draining water, which also shows elevated red-to-green ratios. PACE OCI's full visible spectrum may eventually allow this discrimination through the shape of the iron absorption feature near 490 nm, but operational lake-scale AMD retrieval with PACE is not yet a mature workflow.
Building a monitoring programme, not a one-off map
A single image showing orange water confirms a problem exists. A time series shows whether it is getting worse, whether it tracks rainfall events (suggesting active drainage) or persists through dry periods (suggesting groundwater-fed seepage), and whether a remediation intervention is working. The analytical value is in the trend, not the snapshot.
A practical monitoring programme combines automated Sentinel-2 processing at every cloud-free acquisition, a Landsat historical baseline going back to the earliest available imagery, and flagged alerts when the iron index in defined river-reach polygons exceeds a threshold calibrated against at least one in-situ sampling campaign. That calibration step is not optional. Without at least a handful of field measurements linking spectral index values to actual iron concentrations or pH readings at the site in question, the satellite output is a relative indicator, not a quantitative one. Satellize structures AMD monitoring programmes around exactly this workflow, including support for the in-situ calibration design, drawing on the same open-constellation processing pipeline used in its Tonga crop-estimation work.
Regulators and mining companies have different needs from the same data. A regulator wants evidence of exceedance and a documented chain of custody for the imagery. A mine operator wants early warning before a plume reaches a compliance monitoring point. Both uses are legitimate; the delivery format and alert thresholds differ.
What satellite data cannot do here
Satellite AMD detection is a screening and surveillance tool. It cannot replace in-situ water chemistry for regulatory compliance, cannot measure dissolved metal species (arsenic, cadmium, lead) that carry no visible optical signature, and cannot detect AMD that has not yet reached the surface water system. Groundwater contamination, the most insidious long-term consequence of many legacy mine sites, is entirely invisible to optical sensors.
Cloud cover in wet tropical regions can reduce usable Sentinel-2 observations to fewer than ten per year over some sites. SAR (synthetic aperture radar) from Sentinel-1 can detect surface water extent through cloud but carries no water-chemistry information. The combination of SAR for flood-extent mapping and optical for water-quality indexing is complementary, not redundant, but it does not solve the cloud problem for spectral retrieval. Buyers should enter AMD monitoring with realistic expectations: satellite data narrows the search space and provides continuous spatial coverage that no field programme can match, but it works best as the first tier of a layered monitoring system, not as a standalone compliance instrument.
Typical figures
| Spatial resolution (primary) | 10 m (Sentinel-2 visible/NIR); 30 m (Landsat 8/9 OLI) |
| Spatial resolution (high-res option) | 3–5 m (Planet SuperDove, commercial licence required) |
| Revisit frequency | 5 days (Sentinel-2, two satellites); 8 days combined (Landsat 8+9); daily (Planet SuperDove) |
| Key spectral bands | Blue 443–490 nm, Green 560 nm, Red 665 nm, Red-edge 705–783 nm, NIR 842 nm (Sentinel-2 band designations) |
| Minimum detectable river width | ~20–30 m for reliable pure-water pixels (Sentinel-2); ~6–10 m (Planet SuperDove) |
| Archive depth | Landsat: ~1972–present; Sentinel-2: 2015–present; Planet SuperDove: ~2017–present |
| Processing latency (open data) | Sentinel-2 L2A available within 3–6 hours of acquisition via Copernicus Data Space |
| Cloud cover constraint | Optical retrieval fails under cloud; humid tropical sites may yield fewer than 10 usable scenes per year |
| Delivery formats | GeoTIFF raster indices, vector alert polygons, time-series CSV, PDF monitoring report |
| In-situ calibration requirement | Minimum 1 field campaign recommended to convert spectral index to site-specific iron/pH proxies |
Analytics Satellize can run
| Iron-index map per acquisition | Iron Band Ratio (red/green or red/blue reflectance) applied to atmospherically corrected Sentinel-2 or Landsat imagery using ACOLITE or Sen2Cor | GeoTIFF raster classified into low/moderate/high iron-index zones, updated at every cloud-free pass |
| Plume extent polygon | Thresholded water-mask (NDWI) intersected with iron-index raster to delineate the downstream boundary of detectable AMD influence | Vector shapefile or GeoJSON with area and linear extent, timestamped for regulatory reporting |
| Historical baseline and degradation timeline | Landsat archive time series (1980s–present) with annual cloud-free composites; change-point detection on per-reach median iron index | PDF report with annotated timeline, candidate onset date range, and representative imagery for each period |
| Exceedance alert | Automated threshold monitoring on iron index within user-defined river-reach polygons; alert triggered when index exceeds calibrated percentile | Email or API alert with scene date, affected reach ID, index value and thumbnail image |
| Turbidity trend analysis | NDTI time series over defined water polygons, decomposed by season to separate AMD signal from natural flood-driven turbidity | Time-series chart and CSV with seasonal decomposition, suitable for inclusion in environmental impact assessments |
| Remediation effectiveness tracking | Pre/post comparison of iron-index distributions in treated reaches against untreated control reaches, using paired Sentinel-2 acquisitions | Quarterly monitoring report with statistical summary of index change and spatial maps |
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.