Sensors
- ASI PRISMA: Italian Space Agency hyperspectral satellite; 30 m spatial resolution across 239 contiguous bands from 400 to 2500 nm (VNIR and SWIR); revisit roughly 29 days at nadir but programmable for off-nadir tasking. The SWIR coverage to 2500 nm is the critical range for carbonate, sulphate, phyllosilicate and iron-oxide discrimination.
- NASA EMIT (ISS): Earth Surface Mineral Dust Source Investigation instrument on the International Space Station; 60 m spatial resolution, 285 bands from 380 to 2500 nm. Designed specifically for arid mineral dust characterisation; non-sun-synchronous ISS orbit gives variable but frequent revisit at mid-latitudes. Level-2 mineral-identification products are publicly released.
- HyMap (airborne): Airborne hyperspectral scanner; 126 bands from 450 to 2480 nm; spatial resolution adjustable from roughly 2 m to 10 m depending on flight altitude. No revisit constraint; flown on demand. The resolution advantage over satellite sensors makes it the reference standard for sub-dump-bench mineral mapping and validation of satellite retrievals.
- DESIS (ISS): DLR/Teledyne MEMS-based hyperspectral imager on the ISS; 2.55 nm spectral sampling across 400 to 1000 nm, 30 m ground sampling. Covers only the VNIR range, so it resolves iron oxides and some clay overtones but cannot reach the diagnostic SWIR absorption features of sulphates and carbonates. Useful as a complement, not a substitute.
Why mineralogy matters for dust attribution
Windblown dust from a mine site is not a single substance. A tailings storage facility can expose jarosite, gypsum, kaolinite, smectite, iron oxides and residual sulphide minerals within a few hundred metres of each other, each with a distinct shortwave-infrared absorption signature. Haul-road dust is typically dominated by the country rock being crushed underwheel, which may be entirely different from the tailings mineralogy. That distinction is what makes spectral fingerprinting useful: a dust sample collected at a receptor site, or a plume imaged mid-flight, can be compared against a library of source spectra to attribute proportional contributions.
Regulatory dust-attribution studies have historically relied on chemical tracer elements measured at receptor stations, a method that is accurate but slow and spatially sparse. Hyperspectral mapping of the source areas provides the spatial complement: a continuous mineralogical map of every exposed surface that could be contributing, which can then be used as the source-term input for atmospheric dispersion models. The two approaches are most powerful together.
What a tailings surface gives away in SWIR
The shortwave-infrared region from roughly 1900 to 2500 nm is where the diagnostic action happens. Sulphates such as gypsum and jarosite show characteristic absorption doublets near 2200 nm and 1750 nm respectively. Phyllosilicates, including kaolinite and illite, have Al-OH absorptions near 2200 nm that are distinguishable from sulphates by their shape and position. Carbonates absorb near 2320 to 2340 nm. Iron oxides, including haematite and goethite, are better resolved in the visible and near-infrared through charge-transfer and crystal-field absorptions, but their SWIR behaviour still contributes to a full spectral fit.
Spectral unmixing extracts sub-pixel abundance fractions for each endmember mineral. At PRISMA's 30 m resolution, a single pixel over a tailings beach may contain a mixture of jarosite, gypsum and bare quartz. The unmixing algorithm, typically a constrained linear mixture model or a more sophisticated non-negative matrix factorisation, decomposes that pixel into fractional contributions from a spectral library. The output is not a classified map with hard boundaries but a set of abundance layers, one per mineral, each expressing the proportion of that mineral's signal in each pixel. That is the form most useful for dispersion modelling.
EMIT's public mineral products and their practical limits
EMIT was designed from the outset to answer exactly this question at global scale. Its Level-2B mineralogy product, released publicly through NASA Earthdata, provides per-pixel dominant mineral identifications and spectral fit scores across arid and semi-arid land surfaces. For mine-site work, the 60 m resolution is coarse enough that a narrow tailings beach or a single waste-dump bench may occupy only a handful of pixels. The spectral quality is high, but spatial detail is limited.
The honest trade-off is this: EMIT gives you a defensible, peer-reviewed mineral map of the whole facility at no acquisition cost, with archive depth from 2022 onwards. PRISMA gives you 30 m resolution and the full 400 to 2500 nm range, but requires tasking and has a narrower swath of 30 km. HyMap gives you 2 to 5 m resolution and the ability to map individual bench faces, but at a cost and logistics burden that makes it suitable for periodic validation campaigns rather than routine monitoring. Cloud cover is a hard constraint for all three: a dust event occurring under overcast skies cannot be spectrally attributed from optical data collected at that moment, though pre-event mineralogical maps remain valid as source-term inputs.
Building the spectral library: the step most projects underestimate
The quality of an unmixing result is bounded by the quality of the spectral library used as reference. A generic library drawn from the USGS Spectral Library or ASTER spectral library will capture the major minerals but may miss site-specific mixed phases, weathering rinds or processing-plant residues that have no clean laboratory analogue. The standard practice for a credible attribution study is to collect field spectra with a portable spectrometer (ASD FieldSpec or similar) at representative locations across each source zone, then use those site-specific endmembers to anchor the unmixing.
This field-to-satellite calibration step also resolves a persistent ambiguity: kaolinite and alunite have overlapping absorption features near 2200 nm that can be confused at coarser spectral sampling. A field campaign that confirms which mineral is actually present at a given location allows the analyst to constrain the library and reduce false attributions. Skipping this step and relying entirely on satellite-derived spectra is technically possible but produces results that are harder to defend in a regulatory context.
From mineral map to dispersion model input
A mineral abundance map is not itself a dust emission estimate. Converting it to a source term requires two additional inputs: surface erodibility (a function of particle size, moisture and crust state, none of which are directly readable from hyperspectral data) and wind climatology. The mineral map contributes the composition of what is emitted, not the flux. Dispersion modellers typically parameterise each source zone by its dominant mineral assemblage and assign emission factors from wind-tunnel studies or published empirical relationships for tailings materials.
The practical deliverable for a regulatory submission is a georeferenced polygon layer of source zones, each labelled with its dominant mineral assemblage and fractional abundances, accompanied by the spectral fit statistics that support the attribution. Some jurisdictions now accept hyperspectral source mapping as a primary line of evidence in dust-attribution disputes; others treat it as supporting information alongside receptor chemical data. Satellize's analytics workflow can produce this layer from EMIT or PRISMA data, with the spectral library and unmixing parameters documented for audit. The Kingdom of Tonga crop-estimation programme is a different domain entirely, but the same principle of rigorous source documentation applies: a result that cannot be reproduced from its inputs is not useful to a regulator.
Honest limits and where the method breaks down
Spectral unmixing assumes that the minerals present at the surface are the minerals contributing to dust. That assumption fails when a thin mobile layer of fine particles sits on top of a coarser, spectrally different substrate: the sensor reads the substrate, but the wind mobilises the surface fines. Moisture complicates matters further. Wet tailings surfaces suppress reflectance and shift absorption features, making mineral identification unreliable until the surface dries. Any mineral map produced after rainfall should be treated with caution until validated against field observations.
Vegetation cover, even sparse, masks the underlying mineral signal. A tailings surface with more than roughly 15 to 20 percent fractional cover of dry vegetation becomes difficult to unmix reliably without a vegetation endmember that is well characterised. Dark minerals, including some sulphides, have low reflectance across the full SWIR range and produce weak spectral contrast, making them difficult to identify from abundance alone. These are not reasons to avoid the method; they are reasons to document its assumptions clearly in any report submitted to a regulator or used in litigation.
Typical figures
| Spatial resolution (satellite) | 30 m (PRISMA, DESIS); 60 m (EMIT) |
| Spatial resolution (airborne) | 2 to 10 m (HyMap, altitude-dependent) |
| Spectral range | 400 to 2500 nm (PRISMA, EMIT, HyMap); 400 to 1000 nm (DESIS, VNIR only) |
| Spectral bands | 239 bands (PRISMA); 285 bands (EMIT); 126 bands (HyMap); ~235 bands (DESIS) |
| Revisit (satellite) | ~29 days nadir (PRISMA, off-nadir tasking improves this); variable, frequent at mid-latitudes (EMIT/DESIS on ISS) |
| Key diagnostic spectral windows | 1900 to 2500 nm (sulphates, phyllosilicates, carbonates); 400 to 1000 nm (iron oxides) |
| Minimum mappable source zone | Approximately 1 to 4 ha at 30 m resolution; sub-hectare with HyMap |
| Cloud constraint | Optical only; cloud cover renders acquisition unusable; pre-event maps remain valid as source terms |
| EMIT archive depth | From mid-2022 (ISS deployment); publicly available via NASA Earthdata |
| Deliverable formats | GeoTIFF mineral abundance layers, georeferenced polygon source-zone GIS layer, spectral library files, PDF attribution report |
Analytics Satellize can run
| Mineral abundance map of tailings and waste-dump surfaces | Constrained linear spectral unmixing against a site-calibrated or published spectral library (USGS, ASTER); applied to PRISMA or EMIT Level-2 reflectance | Multi-band GeoTIFF with one layer per mineral endmember (fractional abundance 0 to 1), plus spectral fit residual layer |
| Source-zone polygon layer for dispersion modelling | Segmentation of abundance maps by dominant mineral assemblage; zones labelled with composition and fit statistics | GeoPackage or Shapefile with attribute table; compatible with standard dispersion model input formats (e.g. AERMOD, CALPUFF source-term tables) |
| Change detection in exposed mineral surface area | Multi-date comparison of mineral abundance layers to identify newly exposed or re-vegetated source zones between acquisition epochs | Change GeoTIFF and summary table showing area (ha) gained or lost per mineral class per period |
| Spectral endmember extraction and site-specific library | Automated endmember extraction (e.g. SMACC, N-FINDR) from hyperspectral scene, validated against field ASD spectra where available | ENVI-format spectral library file with metadata; used as audit-ready input to unmixing runs |
| Regulatory dust-attribution evidence package | Integration of mineral source map with receptor chemical data and published dispersion modelling; attribution fractions estimated per source zone | PDF technical report with methodology, uncertainty bounds, spectral fit statistics and GIS annexes suitable for regulatory submission |
| Iron-oxide and sulphate weathering index for tailings surface | Band-ratio and spectral-feature-depth indices (e.g. jarosite index, clay index) applied to SWIR data; tracks oxidation state of exposed sulphidic tailings | Time-series GeoTIFF stack and trend chart; quarterly or per-acquisition cadence |
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.