Rare-earth element tailings spectral characterisation
Hyperspectral imaging identifies REE-bearing mineral phases at tailings surfaces through diagnostic absorption features across 400–2500 nm. PRISMA and EMIT now make this possible from orbit, without a drill.
Sensors
- ASI PRISMA: Italian Space Agency hyperspectral mission; 30 m spatial resolution, 400–2505 nm range, 239 spectral bands, ~6.5 nm sampling interval. Revisit approximately 29 days at nadir but taskable for priority acquisitions. Archive open via ASI portal. Sufficient spectral resolution to resolve carbonate and phosphate absorption features diagnostic of bastnäsite and monazite.
- EMIT (VSWIR imaging spectrometer, ISS): NASA JPL instrument on the International Space Station; 60 m ground sampling distance, 380–2500 nm, ~7.4 nm spectral sampling. Designed for surface mineralogy; data freely distributed via NASA Earthdata. ISS orbit limits latitudinal coverage to roughly 51.6° N/S and revisit is irregular, typically days to weeks depending on ISS ground track.
- AVIRIS-NG (airborne): NASA Jet Propulsion Laboratory airborne spectrometer; 5–8 nm spectral sampling across 380–2510 nm, spatial resolution adjustable from sub-metre to ~5 m depending on flight altitude. Gold standard for mineral mapping but requires dedicated flight campaigns, limiting routine monitoring. Data from published campaigns available via NASA Earthdata.
- Sentinel-2 MSI: 13 multispectral bands, 10–60 m resolution, 5-day revisit at mid-latitudes. Lacks the spectral density to resolve individual REE mineral phases directly, but shortwave-infrared bands (1610 nm, 2190 nm) can serve as proxy indicators of carbonate and clay mineralogy, flagging areas for hyperspectral follow-up. Free and globally archived.
What the absorption features actually reveal
Rare-earth element minerals produce distinctive electronic and vibrational absorption features in the visible-to-shortwave-infrared range. Bastnäsite, a fluorocarbonate, shows carbonate absorptions near 2320 nm and 2530 nm alongside rare-earth electronic transitions in the 740–800 nm region. Monazite, a phosphate, and xenotime, an yttrium phosphate, have overlapping but distinguishable features, particularly in the 400–750 nm range where 4f electron transitions of neodymium, samarium and other lanthanides produce narrow absorption bands. These features are narrow enough, typically 10–30 nm wide, that multispectral sensors cannot resolve them. A spectrometer sampling every 6–10 nm can.
At tailings surfaces the signal is complicated by grain size, moisture, surface crusting and the presence of gangue minerals such as calcite, dolomite, barite and iron oxides. Calcite alone produces a strong carbonate absorption near 2340 nm that can mask or mimic bastnäsite. Careful spectral unmixing, drawing on reference libraries such as the USGS Spectral Library (splib07), is essential. The method does not give you a grade in parts per million. It gives you a mineral phase map of the exposed surface, which is a different and often more actionable piece of information.
The public record: Bayan Obo and Mountain Pass
Published hyperspectral studies of Bayan Obo, the world's largest REE deposit in Inner Mongolia, and Mountain Pass in California's Mojave Desert provide the empirical foundation for this method. Work using AVIRIS and field spectrometry at Mountain Pass has demonstrated that bastnäsite can be mapped at the surface through its carbonate and rare-earth electronic features, and that spectral mixture analysis separates it from co-occurring calcite and barite with reasonable confidence when the mineral is a significant surface constituent. Bayan Obo studies have similarly shown that the Fe-REE mineralogy produces detectable spectral signatures in SWIR imagery, though the complexity of the ore system, which contains over 70 mineral species, means that confident phase identification requires ground-truth validation.
These studies establish the method's honest limits. Detection is surface-only: material buried under even a few centimetres of dust or water is invisible. Mineral phases present at less than roughly 10–20% areal abundance in a pixel are difficult to unmix reliably without high signal-to-noise data. Wet or ponded tailings surfaces suppress reflectance across the SWIR and can render the diagnostic features undetectable. Cloud cover, which affects all passive optical sensors, adds further gaps in time series.
PRISMA and EMIT: what orbit now offers
Until PRISMA's launch in 2019 and EMIT's deployment to the ISS in 2022, spaceborne hyperspectral data at useful spectral resolution was scarce and largely experimental. Both instruments now provide freely accessible data that can be applied to tailings characterisation without a flight campaign. EMIT in particular was designed to map surface mineralogy globally, and its published mineral maps already include carbonate and phyllosilicate distributions at 60 m resolution across arid and semi-arid regions where many REE operations sit.
The practical workflow starts with atmospheric correction, which is non-trivial and has a large effect on the fidelity of subtle absorption features. Both ASI and NASA provide corrected surface reflectance products, but independent validation against field spectrometry is advisable before drawing operational conclusions. After correction, spectral angle mapping or constrained linear unmixing against a reference library identifies candidate REE-bearing pixels. Those pixels are then ranked by confidence and cross-checked against Sentinel-2 time series to screen for surface moisture or disturbance that might compromise the hyperspectral acquisition date.
Where the method earns its keep operationally
The most immediate application is secondary resource assessment: understanding whether a legacy tailings impoundment contains economically interesting concentrations of REE minerals that were discarded when primary processing was optimised for a different target. Several jurisdictions are actively reviewing tailings inventories for exactly this reason, given supply-chain pressure on REE availability. A hyperspectral surface map, even at 30–60 m, gives a rapid first-pass spatial picture of where REE-bearing phases are concentrated, guiding physical sampling rather than replacing it.
A second application is regulatory and environmental monitoring. Windblown REE tailings dust is a recognised exposure concern. Mapping the spatial distribution of REE mineral phases at the surface helps prioritise which areas of an impoundment need dust suppression or capping. This is a use case where periodic revisit, rather than a single snapshot, matters: seasonal drying cycles expose different surface layers, and a time series of hyperspectral acquisitions can track which areas become active dust sources.
Satellize can run atmospheric correction pipelines, spectral unmixing and mineral phase mapping on PRISMA and EMIT acquisitions as a delivered GIS product, drawing on the same open-data infrastructure that supports the Tonga crop-estimation programme. The output is a classified raster with associated confidence layers, not a grade estimate.
Honest limits and what fills the gaps
No spaceborne hyperspectral sensor currently operating offers better than 30 m spatial resolution. At that scale, a single pixel covers 900 square metres, and mineral phase maps represent areal averages rather than point measurements. Sub-pixel unmixing helps, but the result is a fractional abundance estimate, not a map of individual mineral grains. For detailed characterisation of a specific tailings cell, airborne AVIRIS-NG or a commercial hyperspectral drone survey will outperform anything in orbit.
Cloud cover remains the fundamental constraint for passive optical methods in humid tropical or high-latitude settings. Many REE operations, including those in southern China and parts of Brazil, sit in regions where cloud-free SWIR acquisitions are rare. In those environments, hyperspectral characterisation may require careful selection of dry-season acquisition windows, and even then a multi-year archive may yield only a handful of usable scenes. SAR-based methods, covered in sibling pages on deformation and surface change, are not subject to this constraint but cannot detect mineralogy. The two method families are complementary, not interchangeable.
Typical figures
| Spatial resolution (spaceborne) | 30 m (PRISMA), 60 m (EMIT) |
| Spatial resolution (airborne) | 0.5–5 m typical (AVIRIS-NG, altitude-dependent) |
| Spectral range | 380–2510 nm (VSWIR); key REE features at 400–800 nm and 2300–2500 nm |
| Spectral sampling interval | ~6.5 nm (PRISMA), ~7.4 nm (EMIT), ~5–8 nm (AVIRIS-NG) |
| Revisit (spaceborne) | ~29 days nadir (PRISMA, taskable); irregular days-to-weeks (EMIT/ISS orbit) |
| Minimum detectable mineral abundance (per pixel) | Approximately 10–20% areal fraction for confident unmixing; lower with high SNR data and strong spectral contrast |
| Cloud sensitivity | Total occlusion; wet surface suppresses SWIR features even under clear sky |
| Archive depth | PRISMA from 2019; EMIT from 2022; AVIRIS campaign data from 1990s via NASA Earthdata |
| Delivery format | Classified mineral phase raster (GeoTIFF), fractional abundance layers, confidence mask, change overlay |
| Detection depth | Surface only; signal from material buried under a few centimetres of dust or water is not retrievable |
Analytics Satellize can run
| REE mineral phase classification map | Spectral angle mapping and constrained linear unmixing against USGS Spectral Library reference spectra | Classified GeoTIFF with per-class confidence layer, delivered per acquisition |
| Bastnäsite / monazite / xenotime fractional abundance surface | Sub-pixel spectral unmixing using atmospherically corrected PRISMA or EMIT surface reflectance | Multi-band GeoTIFF of fractional abundance per mineral endmember |
| Gangue interference assessment | Calcite, dolomite and iron oxide mapping to flag pixels where co-occurrence complicates REE phase identification | Annotation layer overlaid on phase map, with flagged zones of ambiguity |
| Dust-risk surface prioritisation | Intersection of REE mineral phase map with surface dryness index derived from SWIR reflectance ratio | Ranked zone report (PDF and GIS layer) indicating highest-priority areas for dust suppression |
| Seasonal surface change time series | Multi-date PRISMA or Sentinel-2 SWIR composite differencing to track exposure of REE-bearing layers across wet and dry seasons | Annotated time series chart and change raster stack |
| Sentinel-2 proxy screening layer | SWIR band ratio (B11/B12) and carbonate index applied to Sentinel-2 archive to flag candidate areas ahead of hyperspectral tasking | Screening GIS layer with recommended hyperspectral acquisition priority zones |
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.