Mine tailings and acid rock drainage mapping from spectral data
Iron-sulphate and iron-hydroxide minerals produced by acid mine drainage carry diagnostic spectral signatures that satellite sensors can read from orbit. Hyperspectral instruments resolve individual mineral species; multispectral sensors map gross contamination zones. Neither replaces field sampling, but both catch what ground inspection misses.
Sensors
- PRISMA (ASI): Hyperspectral pushbroom sensor covering 400–2500 nm in 239 contiguous bands at roughly 30 m spatial resolution and 12-bit radiometric depth. Revisit is approximately 29 days at nadir but can be tasked. Resolves the 900 nm Fe3+ absorption and the 2200 nm Al-OH/Fe-OH overtone features well enough to discriminate jarosite from goethite and ferrihydrite at sub-pixel level via spectral unmixing.
- EMIT (NASA/JPL, ISS): Imaging spectrometer covering 380–2500 nm at approximately 60 m ground sampling distance, optimised for mineral mapping. Not a tasked sensor; coverage depends on ISS orbital track. Published mineral identification maps are openly available. At 60 m, mixed-pixel effects at tailings margins are significant, but spectral unmixing recovers fractional abundances of iron-sulphate assemblages.
- Sentinel-2 MSI: Multispectral sensor with 13 bands from 443 to 2190 nm. The 20 m SWIR bands (B11 at 1610 nm, B12 at 2190 nm) and the red-edge bands give enough information to map gross iron-oxide and clay-mineral zones but cannot separate jarosite from goethite reliably because the diagnostic absorption near 900 nm sits between bands. Five-day revisit (two-satellite constellation) makes it the best tool for change detection at known sites.
- Landsat 8/9 OLI: Six reflective bands from 435 to 2294 nm at 30 m. The OLI SWIR-1 (1566–1651 nm) and SWIR-2 (2107–2294 nm) bands support iron-oxide ratio indices similar to Sentinel-2. Landsat's archive back to 1972 (TM/ETM+/OLI) is the primary source for long-term tailings expansion and drainage plume history. Revisit is 16 days per satellite, 8 days combined.
What the chemistry does to the spectrum
When pyrite and other iron-sulphide minerals in mine waste oxidise in the presence of water and oxygen, they produce sulphuric acid and a cascade of secondary iron minerals. Jarosite (KFe3(SO4)2(OH)6) forms in strongly acidic, oxidising conditions. As pH rises, jarosite dissolves into goethite and ferrihydrite. Each mineral has a distinct electronic and vibrational fingerprint in reflected sunlight.
Jarosite absorbs strongly near 430 nm and 900 nm due to Fe3+ crystal-field transitions, and shows a sulphate overtone near 2270 nm. Goethite has a broader Fe3+ absorption centred near 900 nm but lacks the sulphate feature. Ferrihydrite is spectrally broader still, with weaker features. A hyperspectral sensor with contiguous bands through the near-infrared and shortwave-infrared can separate these three species by fitting reference library spectra to observed reflectance. A multispectral sensor cannot, because the critical 900 nm region falls between its discrete bands.
Moisture complicates everything
Water absorption bands at 1400 nm and 1900 nm suppress reflectance across the SWIR, and liquid water in pore spaces broadens and shifts iron-mineral features. A wet tailings surface and a dry one containing the same jarosite concentration can look measurably different to a sensor. This is not a marginal effect: published laboratory studies show that increasing gravimetric moisture content from 5% to 25% can reduce the apparent depth of the 900 nm Fe3+ absorption by 30–50%.
The practical consequence is that acquisitions taken after rainfall are less reliable for mineral identification. Analysts should cross-reference acquisition date against precipitation records and flag scenes collected within 48 hours of significant rain. Thermal-infrared data from Landsat can help estimate surface moisture state as a secondary check. This is one reason that time-series approaches, using multiple dry-season acquisitions, outperform single-date retrievals.
Spectral unmixing at the margins where contamination matters most
The boundary between a tailings facility and surrounding soil or vegetation is exactly where regulators need precision, and it is where mixed pixels create the most ambiguity. At 30 m resolution, a single pixel at a drainage channel margin may contain tailings material, native soil, sparse vegetation, and standing water simultaneously. The observed spectrum is a mixture of all four.
Linear spectral unmixing decomposes the mixed pixel into fractional contributions from a set of endmember spectra, typically derived from spectrally pure pixels within the scene or from a laboratory library. For acid drainage mapping, the standard endmember set includes jarosite, goethite, ferrihydrite, dry soil, green vegetation, and water. The output is a fractional abundance map for each mineral, not a hard classification. That distinction matters for enforcement: a pixel that is 40% jarosite by area is a different regulatory finding from one that is 5%. EMIT's published surface mineralogy products use a variant of this approach.
What multispectral sensors can and cannot do
Sentinel-2 and Landsat remain genuinely useful despite their spectral limitations. Band ratio indices, such as the iron oxide ratio (Red / Blue) and the clay mineral ratio (SWIR-1 / SWIR-2), reliably highlight zones of elevated iron-oxide and sulphate mineralisation at the landscape scale. They are fast to compute, freely available globally, and backed by decades of published validation in mine-affected catchments.
The honest limit is species discrimination. If a regulator needs to know whether a drainage plume contains jarosite (indicating active acid generation) or goethite (indicating older, partially neutralised drainage), Sentinel-2 cannot answer that question with confidence. Hyperspectral data is required. The practical workflow is to use Sentinel-2 for site screening and change detection across a large portfolio of facilities, then task PRISMA or request EMIT coverage for sites that show anomalous change or require enforcement-grade characterisation.
Limits the buyer should know before commissioning
Vegetation cover is the most common failure mode. A tailings drainage channel that has been colonised by iron-tolerant grasses will show a mixed spectrum dominated by chlorophyll, masking the underlying mineralogy. Canopy closure above roughly 60–70% makes mineral detection from orbit unreliable regardless of sensor. Underground drainage, subsurface seepage, and contamination that has not yet reached the surface are invisible to any passive optical system.
Cloud cover is a persistent constraint in humid tropical mining regions, particularly in central Africa and Southeast Asia. A 29-day PRISMA revisit becomes a 90-day effective revisit when cloud-free acquisition probability is factored in. SAR sensors can observe through cloud but carry no mineralogical information; they contribute to the workflow only for detecting surface water extent and tailings pond geometry, not chemistry.
Satellize runs spectral unmixing pipelines on PRISMA and EMIT data alongside Sentinel-2 change-detection stacks for mining-sector clients. The Tonga crop-estimation programme demonstrated the team's approach to sub-pixel fractional cover retrieval, which uses the same methodological foundation as mineral unmixing.
From spectral map to enforcement record
A mineral abundance map is not itself an enforcement document. The path from satellite output to regulatory action requires field verification of at least a sample of flagged pixels, pH and metal-concentration measurements in drainage water, and a chain-of-custody record for the data. Satellite analysis narrows the field inspection effort from a facility-wide survey to targeted sampling at the locations the spectral data flags as anomalous. That is where the practical value sits.
For long-term compliance monitoring, the most defensible approach is a baseline mineral map produced from hyperspectral data at licence commencement, followed by quarterly Sentinel-2 change-detection alerts that trigger a PRISMA or EMIT re-acquisition when a threshold change is detected. The archive depth of Landsat (OLI back to 2013, ETM+ and TM further back) allows retrospective baseline construction for facilities that predate modern hyperspectral availability.
Typical figures
| Spatial resolution (hyperspectral) | 30 m (PRISMA), 60 m (EMIT) |
| Spatial resolution (multispectral) | 10–20 m (Sentinel-2 MSI), 30 m (Landsat 8/9 OLI) |
| Revisit (Sentinel-2 constellation) | 5 days at equator; shorter at higher latitudes |
| Revisit (PRISMA, tasked) | Approximately 29 days nadir; tasking can improve frequency |
| Spectral range for mineral ID | 400–2500 nm (PRISMA, EMIT); 443–2190 nm (Sentinel-2) |
| Key diagnostic absorptions | ~430 nm, ~900 nm (Fe3+ crystal field); ~2270 nm (sulphate overtone, jarosite) |
| Minimum detectable mineralised area (hyperspectral) | Sub-pixel fractional abundance detectable via unmixing; pure-mineral pixel requires ~30 m clear exposure |
| Landsat archive depth | OLI from 2013; ETM+ from 1999; TM from 1982 |
| Cloud sensitivity | All passive optical sensors: cloud-obscured acquisitions unusable; SAR provides geometry only, not chemistry |
| Delivery formats | GeoTIFF fractional abundance maps, GeoPackage mineral classification polygons, CSV change-detection time series |
Analytics Satellize can run
| Mineral species abundance map | Linear spectral unmixing against laboratory or scene-derived endmember library (PRISMA or EMIT data) | GeoTIFF raster with per-pixel fractional abundance for jarosite, goethite, ferrihydrite, soil, and vegetation |
| Gross iron-oxide zone map | Band ratio indices (iron oxide ratio, clay mineral ratio) applied to Sentinel-2 or Landsat OLI | GIS polygon layer of anomalous iron-mineralisation zones, ranked by index magnitude |
| Tailings pond boundary and expansion history | Multi-date NDWI and SWIR thresholding on Landsat archive stack | Annual boundary polygons from 2013 to present with area-change statistics in tabular report |
| Drainage plume extent and trajectory | Spectral mixture analysis combined with digital elevation model flow routing | Vector plume polygons with upstream source attribution and downstream receptor distances |
| Quarterly change-detection alert | Sentinel-2 time-series anomaly detection against site-specific baseline; threshold-triggered PRISMA tasking recommendation | Automated alert report with flagged pixel locations, acquisition date, and recommended field verification coordinates |
| Moisture-corrected mineral confidence layer | Precipitation-record cross-referencing and Landsat thermal moisture proxy applied before spectral classification | Mineral map with per-pixel confidence flag indicating acquisition moisture conditions |
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.