The short answer

Shortwave infrared (SWIR), roughly 1000-2500 nm, is the most diagnostic spectral region for geological remote sensing because that is where the molecular bonds in alteration minerals — Al-OH, Mg-OH, carbonate, sulphate and water — produce sharp, mineral-specific absorption features. The single most important feature for exploration is the Al-OH absorption near 2200 nm, which fingerprints clays and white micas associated with hydrothermal alteration. How well you can use the SWIR depends entirely on how many bands your sensor places there: two broad bands (Landsat, Sentinel-2) screen for alteration; six bands (ASTER) separate mineral groups; hundreds of bands (hyperspectral) identify species and even composition.

The physics: why SWIR sees minerals visible light cannot

In the visible and near-infrared, mineral colour is dominated by electronic processes in transition metals — chiefly iron, which gives ferric oxides their broad reddish absorptions. Those are useful but not very specific. The SWIR is different: it is governed by vibrational overtones and combination tones of molecular bonds, and these are narrow and diagnostic. The key bond groups and their absorption positions:

  • Al-OH (kaolinite, illite, montmorillonite, muscovite/sericite, pyrophyllite, alunite): a strong feature near 2160-2210 nm. The exact position and shape distinguish, for example, kaolinite (doublet near 2160/2200 nm) from white mica (single feature whose wavelength shifts with composition).
  • Mg-OH and Fe-OH (chlorite, biotite, epidote, amphibole, serpentine): features near 2250 nm and 2300-2360 nm.
  • Carbonate (calcite, dolomite, ankerite): a feature near 2300-2340 nm, with position shifting between calcite and dolomite.
  • Sulphate (gypsum, alunite, jarosite): OH/water features near 1750, 2200 and shorter wavelengths.
  • Water and hydroxyl generally: broad bands near 1400 nm and 1900 nm (atmospheric water vapour also absorbs strongly there, which is why these regions are noisy).

Because these features are tied to specific chemistry, a well-sampled SWIR spectrum is effectively a mineralogical fingerprint. That is why SWIR underpins alteration mapping for porphyry, epithermal and many other deposit styles, where the alteration mineralogy maps the plumbing of the hydrothermal system.

How many SWIR bands you have changes everything

The geological value of a sensor in the SWIR scales with its spectral sampling:

  • Landsat 8/9 OLI: two SWIR bands, B6 (~1610 nm) and B7 (~2200 nm). Enough to flag clay/hydroxyl alteration via the B6/B7 clay ratio, not enough to name the mineral.
  • Sentinel-2 MSI: two SWIR bands, B11 (~1610 nm) and B12 (~2190 nm) at 20 m. Similar capability to Landsat at finer resolution and faster revisit.
  • ASTER: six SWIR bands between ~1600 and 2430 nm (bands 4-9). This sampling lets you separate Al-OH, Mg-OH/carbonate and other groups with band ratios and indices, which is why ASTER was the standard spaceborne mineral mapper. Note ASTER's SWIR detector failed in 2008, so only the archive before then carries usable SWIR.
  • Hyperspectral (EMIT, PRISMA, EnMAP, airborne HyMap/AVIRIS): hundreds of contiguous narrow bands. These resolve the exact wavelength of the 2200 nm feature, so you can not only identify white mica but estimate its Tschermak (Al content) composition, a vector toward mineralisation in some systems.

A simple but instructive ASTER ratio set: Al-OH (clay) ~ band 5+7 over band 6, Mg-OH/carbonate ~ band 6+9 over band 8, ferric iron from VNIR ~ band 2/band 1. The principle is to place absorption bands in the denominator and shoulder bands in the numerator so the ratio peaks where the mineral absorbs.

A worked alteration-screening workflow

Goal: a fast clay-alteration screen over an arid prospect using Landsat 9.

  1. Get Level-2 surface reflectance (Landsat Collection 2 L2) so atmospheric effects are removed.

  2. Mask cloud, shadow and water using the QA_PIXEL bitmask band.

  3. Compute the clay ratio B6/B7:

    gdal_calc.py -A LC09_..._SR_B6.TIF -B LC09_..._SR_B7.TIF \
  --outfile clay_ratio.tif --calc="A.astype(float)/B" --type=Float32

  4. Stretch and threshold. Apply a 2-98% percentile stretch; high ratio values indicate stronger 2200 nm absorption (more Al-OH). Threshold conservatively and treat the result as a candidate mask.

  5. Cross-check with iron. Compute the ferric ratio B4/B2 to see whether the same ground is also iron-stained, which is common around gossans and oxidised caps.

  6. Overlay on a hillshade and geological map and select targets for ground follow-up. If higher confidence is needed, escalate to ASTER archive ratios or hyperspectral data.

The same logic transfers to Sentinel-2 (B11/B12) and ASTER (the six-band ratios), with increasing mineral specificity as bands increase.

Common pitfalls and why they happen

  • Atmospheric water masking the signal. The 1400 and 1900 nm regions are dominated by water-vapour absorption; mineral features there are unreliable from space. This is why spaceborne SWIR bands avoid those windows and why you must use atmospherically corrected (surface reflectance) products.
  • Vegetation overprint. Green and dry vegetation have their own SWIR features (cellulose, lignin, leaf water). A few percent canopy can swamp a subtle mineral signal, so mask vegetation (NDVI threshold) before interpreting.
  • Over-reading two-band sensors. A high Landsat or Sentinel-2 clay ratio means "hydroxyl alteration likely," not "kaolinite." Naming minerals requires ASTER-class or hyperspectral sampling.
  • Ignoring the ASTER 2008 SWIR failure. Scenes after April 2008 have no valid SWIR; using them for mineral ratios produces garbage.
  • Mixing reflectance scaling. Landsat C2 and Sentinel-2 L2A use scale factors and offsets; apply them before ratios are compared across scenes (ratios cancel a pure multiplicative scale but not an additive offset).

Validation and QA

Confirm you are using surface reflectance, not top-of-atmosphere; verify cloud and vegetation masks at full resolution; and where possible compare a sample of high-ratio pixels against field observations, drill logs, or a portable spectrometer / library spectrum (the USGS spectral library is the standard reference). Record the sensor, product level, acquisition date, scaling and the exact band-math so the alteration layer is reproducible. Always present the output as an evidence layer with stated confidence, never as a mineral map in itself.

Bathyl perspective

We treat SWIR as the analytical core of geological remote sensing and choose the sensor to match the question: free two-band data to screen large areas, ASTER archive or hyperspectral to characterise targets. Every SWIR product we deliver states what the band count can and cannot resolve.

Related reading

Sources