The short answer

Hydrothermal alteration mapping from satellite data works because the minerals produced when hot, chemically active fluids react with host rock — iron oxides, clays, sericite, alunite, chlorite, epidote, carbonates — each have diagnostic absorption features in the visible-near-infrared (VNIR) and shortwave infrared (SWIR) parts of the spectrum. By measuring reflectance in carefully chosen bands and computing ratios or spectral indices, you can map where those minerals are concentrated at the surface. The catch is that band ratios point to mineral groups, not specific minerals, and only with adequate SWIR resolution. ASTER remains the most accessible multispectral sensor for this, Sentinel-2 and Landsat handle iron oxides and broad clay screening, and field or hyperspectral confirmation is mandatory before any anomaly drives a decision.

Why alteration shows up in the spectrum

Alteration assemblages zone outward from a hydrothermal source. A porphyry system, for example, often grades from a potassic core through phyllic (quartz-sericite-pyrite), argillic (kaolinite-illite-smectite), to a peripheral propylitic halo (chlorite-epidote-calcite). Epithermal systems add advanced argillic alteration (alunite, pyrophyllite, dickite). Each assemblage carries molecular bonds that absorb light at predictable wavelengths:

  • Al-OH bonds (sericite, muscovite, illite, kaolinite, alunite) absorb near 2.20 micrometres, with the exact position shifting between 2.16 and 2.21 micrometres depending on mineral and crystallinity.
  • Mg-OH and CO3 bonds (chlorite, epidote, calcite, dolomite — propylitic) absorb near 2.30-2.35 micrometres.
  • Ferric iron (Fe3+) in hematite, goethite and jarosite produces strong absorption below ~0.55 micrometres and a charge-transfer feature, giving the red-brown reflectance signature of gossans.

A sensor only "sees" alteration if it has bands placed on and off these features. That single fact determines which satellite you reach for.

Sensor capabilities, honestly

ASTER carries three VNIR bands at 15 m, six SWIR bands at 30 m positioned across 2.145-2.430 micrometres, and five thermal infrared bands at 90 m. Those narrow SWIR bands are why ASTER became the default multispectral alteration tool: B5, B6, B7, B8 straddle the Al-OH and Mg-OH/carbonate regions well enough to separate argillic, phyllic and propylitic groups. Note that the ASTER SWIR detector failed in April 2008, so any post-2008 scene has unusable SWIR — you must source pre-2008 acquisitions for SWIR work.

Sentinel-2 offers B11 (1.61 micrometres) and B12 (2.19 micrometres) SWIR bands at 20 m, plus 10 m VNIR. Two broad SWIR bands cannot resolve the 2.20 vs 2.30 micrometre distinction, so Sentinel-2 is good for ferric-iron and broad clay/gossan screening, not specific assemblage mapping. Its advantage is 5-day revisit and 10-20 m resolution for detailed iron mapping.

Landsat 8/9 OLI is similar in concept (SWIR1 1.61, SWIR2 2.20 micrometres, both 30 m) with a longer archive for historical context. WorldView-3 (eight SWIR bands at 3.7 m) and airborne hyperspectral (AVIRIS, HyMap) approach laboratory-grade mineral discrimination but cost far more.

A worked ASTER workflow

Assume a pre-2008 ASTER L1T scene over an arid prospect with sparse vegetation (ideal conditions — alteration mapping fails badly under canopy).

1. Pre-process. Convert DN to radiance, then to surface reflectance. Apply atmospheric correction; a relative method like Internal Average Relative Reflectance (IARR) or a flat-field correction is common when no atmospheric profile is available, otherwise FLAASH/ATCOR. Reproject to the local UTM zone, e.g.

gdalwarp -t_srs EPSG:32740 -r bilinear aster_swir.tif aster_swir_utm.tif

2. Crosstalk and registration. Apply the ASTER SWIR crosstalk correction (a known band-to-band light leakage) before computing ratios, and co-register VNIR (15 m) and SWIR (30 m) so multi-band indices line up.

3. Compute band ratios. Standard, widely cited ASTER ratios (treat as screening proxies):

  • Al-OH / phyllic-argillic (sericite, kaolinite): (B5 + B7) / B6
  • Alunite / advanced argillic: B4 / B6
  • Propylitic (Mg-OH, chlorite-epidote-carbonate): B7 / B8 or (B6 + B9) / (B7 + B8)
  • Ferric iron: B4 / B3; ferrous iron: B5 / B4
gdal_calc.py -A b5.tif -B b7.tif -C b6.tif \
  --calc="(A.astype(float)+B)/C" --outfile=aloh_ratio.tif --NoDataValue=0

4. Composite and stretch. Build an RGB of three ratios (e.g. Al-OH, ferric iron, propylitic) and apply a 2%-98% linear stretch so anomalies stand out without saturating. In QGIS this is the layer's Min/Max value settings → Cumulative count cut.

5. Mask confounders. Build an NDVI mask (NDVI > ~0.25) to drop vegetation, mask water, and flag deep topographic shadow from a hillshade. Vegetation Al-OH-like leaf-water absorptions and dry salt/clay playas are the two classic false positives.

For more rigorous separation, run a Spectral Angle Mapper (SAM) or matched-filter classification against a library of endmembers (USGS Spectral Library) instead of, or alongside, ratios — SAM is comparatively insensitive to illumination and gives per-pixel mineral-class membership.

Validation before anyone trusts it

An alteration raster is a hypothesis. Validate it like one:

  • Spectral ground-truth. Collect field spectra with an ASD/TerraSpec at anomaly centres and backgrounds; confirm the 2.20 vs 2.30 micrometre feature positions match the predicted assemblage. XRD or short-wave infrared spectroscopy on hand samples closes the loop.
  • Geological cross-check. Overlay the published geology, structural lineaments and any known prospects. Real alteration usually has a structural or lithological logic; a blob with none is suspect.
  • Confounder audit. Re-check each strong anomaly against vegetation, evaporite/playa surfaces, anthropogenic features (roads, tailings) and shadow. Inspect the edges of the scene and the steep slopes, not just the clean centre.
  • Repeatability. Document the scene ID, acquisition date, correction method, and exact ratio formulas so a colleague can regenerate the layer.

Common pitfalls and why they happen

  • Reading ratios as minerals. A high (B5+B7)/B6 means "Al-OH-rich pixel," which could be sericite, illite, kaolinite or muscovite — and the geology, not the ratio, decides which. The ratio collapses several spectrally similar minerals into one.
  • Using post-2008 ASTER SWIR. The SWIR detector is dead; the bands return noise that looks like structure. Always check the acquisition date.
  • Skipping atmospheric correction across scenes. Comparing two dates without consistent correction conflates atmosphere and surface change, especially in the iron region where atmospheric scattering is strongest.
  • Mapping under vegetation. Canopy reflectance dominates and the leaf-water absorption near 2.2 micrometres mimics clay; closed-canopy terrain is largely off-limits to optical alteration mapping.
  • Ignoring crosstalk. Uncorrected ASTER SWIR crosstalk biases ratios systematically and produces spurious regional patterns.

Bathyl perspective

We treat an alteration layer as one input into a target model, weighted alongside structure, geochemistry and geophysics — never as a standalone target generator. The deliverable that survives review is the one that states its sensor, acquisition date, correction chain and validation status, so a geologist can judge how far to trust each anomaly before committing field or drill budget.

Related reading

Sources