Short answer

Remote sensing screens for mineralisation by mapping the spectral fingerprints of alteration — the iron oxides, clays, micas, and carbonates that hydrothermal systems leave around an ore body — across large, often roadless areas, then ranking where to spend scarce field and drilling budget. The sensor of choice for most free-data alteration work is ASTER, because its shortwave-infrared (SWIR) bands resolve the diagnostic mineral absorptions; Sentinel-2 adds resolution and revisit, and Landsat adds decades of archive. None of it confirms ore. The output is a set of validated candidate targets, with cloud, vegetation, and confidence handled honestly.

Why alteration, not the ore itself

You almost never see the deposit directly from orbit. What you can see is the halo: mineralogical changes in the host rock produced by the same fluids that formed the deposit. Porphyry and epithermal systems, for instance, develop zoned alteration — potassic, phyllic (sericite/clay), argillic, propylitic — and several of those minerals have sharp, diagnostic absorption features in the visible-to-SWIR range (roughly 0.4–2.5 µm).

  • Iron oxides (hematite, goethite, jarosite) absorb strongly in the blue and reflect in the red, producing the gossan/limonite signal.
  • Al-OH clays and micas (kaolinite, illite, sericite, alunite) have diagnostic absorptions near 2.16–2.21 µm.
  • Mg-OH and carbonates (chlorite, epidote, calcite, dolomite) absorb near 2.30–2.35 µm.

Mapping where these features are strong, and where they cluster along structures, is the core of optical exploration screening.

Choosing the sensor

The choice is driven by spectral coverage first, then resolution and revisit.

ASTER is the workhorse for alteration on a free budget. It carries 3 VNIR bands at 15 m, 6 SWIR bands at 30 m, and 5 thermal bands at 90 m. Those six SWIR bands are the key asset: they straddle the 2.1–2.4 µm region finely enough to separate Al-OH clays from Mg-OH/carbonate minerals — something broadband sensors cannot do. Caveat: the ASTER SWIR detectors failed in April 2008, so for SWIR-based alteration work you must use pre-2008 scenes.

Sentinel-2 (MSI) offers 13 bands, with 10 m in the visible/NIR, 20 m in the red-edge and SWIR, and a 5-day revisit from the twin satellites. But it has only two broad SWIR bands (B11 ~1.61 µm, B12 ~2.19 µm), so it detects "there is clay/hydroxyl alteration here" but cannot discriminate clay species the way ASTER can. Its strengths are resolution, revisit, and free, analysis-ready Level-2A surface reflectance.

Landsat 8/9 (OLI) gives 30 m multispectral with two SWIR bands, and — critically — a continuous archive back to the 1980s (with earlier TM/ETM+). Use it for change context, vegetation and disturbance history, and regional reconnaissance.

For genuine mineral discrimination, hyperspectral sensors (airborne, or spaceborne like EnMAP and the EMIT instrument on the ISS) resolve hundreds of narrow bands and can identify specific minerals — but availability and cost are different constraints. For free regional screening, ASTER + Sentinel-2 is a strong default pairing.

Worked band ratios

Band ratios suppress topographic shading (the illumination cancels in the ratio) and emphasise spectral slope. They are crude relative to spectral matching, but fast and effective for first-pass screening.

Landsat 8/9 OLI:

  • Iron oxide / ferric iron: B4 / B2 (red / blue) — high values over gossans and limonite.
  • Clay / hydroxyl (phyllic–argillic): B6 / B7 (SWIR1 / SWIR2).
  • Ferrous iron: B6 / B5 (SWIR1 / NIR).

ASTER (VNIR/SWIR):

  • Ferric iron: B2 / B1.
  • Al-OH clay (kaolinite/illite/alunite): B4 / B6 — the classic argillic indicator.
  • Mg-OH / carbonate (propylitic): (B6 + B9) / (B7 + B8) or B5 / B8.

A practical "Crósta"-style approach combines several ratios into an RGB composite or runs a principal component analysis to isolate the alteration signal from background. In QGIS, you compute these in the Raster Calculator or with gdal_calc.py; on Sentinel-2's broad SWIR, a simple B11/B12 highlights generic hydroxyl alteration.

gdal_calc.py -A L8_B4.tif -B L8_B2.tif \
  --outfile=iron_oxide_ratio.tif --calc="A.astype(float)/B" --NoDataValue=0

A screening workflow

  1. Define the target system. Porphyry, epithermal, IOCG, and orogenic gold leave different alteration signatures; the ratios and minerals you prioritise depend on the model. Start from the geology, not the imagery.
  2. Acquire and correct. Use surface-reflectance products (Sentinel-2 L2A, ASTER AST_07 surface reflectance / on-demand corrected). Atmospheric correction matters because band ratios at 2 µm are sensitive to residual atmospheric effects.
  3. Mask aggressively. Cloud, cloud shadow, water, snow, and dense vegetation all mimic or swamp alteration signals. Sentinel-2 ships a scene classification (SCL) and cloud probability layer; use them. Dense canopy is the hard limit — optical alteration mapping works best in arid, exposed terrain.
  4. Compute indicators. Run the relevant band ratios and/or a PCA; threshold to highlight anomalies.
  5. Overlay structure and geology. Lineaments from a DEM (hillshade, slope) and existing geological maps; mineralisation favours structural intersections and contacts. Reproject everything to a common projected CRS (the appropriate UTM zone) before overlay.
  6. Rank and document. Produce a ranked target list with a confidence note per anomaly, distinguishing strong multi-indicator clusters from single-ratio noise.

Common pitfalls and why they happen

  • Reading false-colour composites as lithology. A red blob is a spectral response, not a rock unit. Vegetation, iron-stained soils, and even some man-made surfaces can mimic alteration. The signal is a hypothesis until validated.
  • Using post-2008 ASTER for SWIR work. The SWIR detectors are dead after April 2008; those bands return garbage in later scenes.
  • Skipping atmospheric correction or comparing mixed dates. Different acquisition dates carry different atmosphere, sun angle, soil moisture, and phenology; comparing raw DN values across them produces artefacts that look like change.
  • Working in vegetated terrain. Canopy hides the ground spectrum. If vegetation cover is high, optical alteration mapping has limited power and other methods (geophysics, geochemistry) should lead.
  • Confusing detection with discrimination. Sentinel-2 detects hydroxyl alteration broadly; it does not tell kaolinite from illite. Claiming species-level mapping from broadband data overstates the result.

QA and validation

Before a screening layer informs a field campaign: confirm inputs are atmospherically corrected and the correction method is recorded; verify cloud/shadow/vegetation/water masks are applied and visualised; check that all layers share one projected CRS and align at the edges; cross-validate anomalies against any existing geological maps, geochemistry, or known occurrences; and label every target with a confidence level and the indicators that support it. Treat anything unverified as a ground-truthing target, not a result.

Bathyl perspective

We use remote sensing to point the field team at the right ground, not to replace it. A defensible screening deliverable says clearly what the sensor can resolve (broad hydroxyl alteration on Sentinel-2, clay species on ASTER SWIR), what was masked, and which anomalies are strong multi-indicator clusters versus single-ratio noise — so the geologist decides where to walk with eyes open.

Related reading

Sources