What can thermal infrared remote sensing reveal about geology?

Thermal infrared (8-12 um) is sensitive to silicate mineralogy through the reststrahlen bands, so it can map quartz-rich versus mafic rocks, silica content, and carbonates. It also captures thermal inertia differences that distinguish bedrock from loose cover.

Which thermal sensors are used for geological mapping?

ASTER, with five thermal bands between 8.1 and 11.6 um, is the workhorse for spaceborne lithological TIR. Landsat 8/9 TIRS provides two thermal bands at 100 m, and airborne sensors like MASTER and HyTES give higher spectral and spatial detail.

Why does emissivity matter more than temperature for lithology?

Surface temperature is driven by sun, weather, and time of day, which carry no mineralogical information. Emissivity is an intrinsic property of the rock surface, so lithological mapping requires separating emissivity from temperature before interpretation.

How do I process thermal bands for a silica index?

Start from at-sensor radiance, apply atmospheric correction, run temperature-emissivity separation to recover band emissivities, then compute ratio-based indices such as the ASTER quartz index from bands 10-12 against band 11.

Thermal Remote Sensing for Geology: TIR Bands & Methods

Short answer

Thermal remote sensing for geology uses the thermal infrared (TIR) window, roughly 8 to 12 micrometres, to map rocks by the way their constituent minerals emit and absorb long-wave radiation. Silicate minerals have diagnostic emissivity minima in this region — the reststrahlen bands — and the position of those minima shifts systematically from felsic (quartz-rich) to mafic compositions. That makes TIR uniquely able to discriminate silica content, something the visible/near-infrared and shortwave infrared cannot do well.

The crucial conceptual move is that lithology lives in emissivity, not in temperature. Two pixels at the same kinetic temperature can be different rocks, and the same rock can be hot at midday and cold at dawn. The whole TIR geology workflow is about separating the intrinsic emissivity signal from the temperature signal that the sun and weather impose.

The physics: reststrahlen and Planck

Every surface above absolute zero emits radiation described by the Planck function scaled by its spectral emissivity. In the 8-12 um window, Si-O bond stretching produces strong emissivity lows. As bulk silica content drops from quartz and feldspar (felsic) through to olivine and pyroxene (mafic), the wavelength of the emissivity minimum moves to longer wavelengths. By measuring the shape of the emissivity spectrum across several TIR bands you can estimate composition — carbonates, sulphates, and some clays also have distinct TIR features.

The signal arriving at the sensor is at-sensor radiance, which mixes three things: the surface's emitted radiance (temperature × emissivity), reflected downwelling sky radiance, and radiance added and absorbed by the atmosphere along the path. Untangling these is why TIR processing is more involved than a simple band ratio.

Sensors that matter for geology

ASTER (Terra): five TIR bands (band 10 ≈ 8.29 um, 11 ≈ 8.63 um, 12 ≈ 9.08 um, 13 ≈ 10.66 um, 14 ≈ 11.32 um) at 90 m. Five bands is enough to resolve emissivity-spectrum shape, which is why ASTER underpins most published spaceborne lithological TIR mapping. Its SWIR detectors failed in 2008, but the TIR subsystem continued operating.
Landsat 8/9 TIRS: two thermal bands (band 10 ≈ 10.9 um, band 11 ≈ 12.0 um) acquired at 100 m and resampled to 30 m. Two bands cannot resolve the silica trend on their own, but TIRS is excellent for thermal-inertia and surface-temperature context and has a long-running archive.
Airborne MASTER and HyTES: many narrow TIR channels at metre-scale resolution; the standard for detailed mineral mapping where a campaign budget exists.
ECOSTRESS (ISS): higher-frequency TIR temperature observations, more often used for thermal inertia and evapotranspiration than direct lithology.

Temperature-emissivity separation

Because n thermal bands give n radiance measurements but you want n emissivities plus one temperature (n+1 unknowns), the problem is underdetermined and needs an assumption to close it. Standard methods include:

Reference channel / emissivity normalisation: fix one band's emissivity to a constant (often 0.96-0.99) and solve the rest. Simple, but biased if the assumed value is wrong.
ASTER Temperature/Emissivity Separation (TES): uses an empirical relationship between the minimum emissivity and the spectral contrast (the min-max difference) to add the missing constraint. The standard ASTER surface-emissivity product (AST_05) and surface-kinetic-temperature product (AST_08) are produced this way and are usually the right starting point rather than reprocessing L1 radiance yourself.

Skipping TES — interpreting raw radiance or brightness temperature as if it were mineralogy — is the single most common error, because it confounds "this pixel is warm" with "this pixel is quartz-rich."

Worked example: an ASTER silica index

A widely used approach maps silica content from ASTER TIR emissivities. Conceptually:

Obtain emissivity bands. Download AST_05 (surface emissivity, bands 10-14) rather than starting from L1B, unless you have a reason to run your own atmospheric correction and TES.

Reproject and align. Put the scene in a projected CRS and stack the five bands:

gdalwarp -t_srs EPSG:32643 ast05_b10.tif ... b10_utm.tif   # repeat per band
gdal_merge.py -separate -o aster_tir_stack.tif b10_utm.tif b11_utm.tif b12_utm.tif b13_utm.tif b14_utm.tif

Compute ratio indices in a raster calculator. A common quartz index exploits the band-11 emissivity low against the average of neighbouring bands, for example (b10 + b12) / (2 × b11); higher values flag quartz-rich, felsic surfaces. A mafic/silica index such as b13 / b14 tracks the shift toward longer-wavelength minima in more mafic rocks. Exact coefficients vary by author and terrain, so calibrate against known outcrops.
Mask confounders. Vegetation, water, snow, and clouds have their own TIR behaviour and must be masked (use a coincident NDVI from the VNIR bands and a cloud mask) before the index is interpreted.
Validate against the geological map and field spectra. Treat the index as a candidate-anomaly layer, then confirm with field checks or library spectra.

Thermal inertia: a second, complementary signal

Beyond emissivity, the rate at which a surface heats and cools — thermal inertia — separates competent bedrock (high inertia, small day-night temperature swing) from dry, loose, or porous cover (low inertia, large swing). Apparent Thermal Inertia (ATI) is approximated from albedo and the day-minus-night temperature difference, which is why paired dawn/afternoon acquisitions are valuable. ATI is useful for distinguishing outcrop from regolith and for mapping surficial cover, but it is sensitive to soil moisture, so post-rain scenes are unreliable.

Common pitfalls and why they happen

Interpreting temperature as lithology. Brightness temperature is dominated by illumination and time of day; without TES it carries no clean mineral signal.
Ignoring the atmosphere. Water vapour absorbs strongly in the TIR window, so an uncorrected scene mixes ground and air. Use atmospherically corrected products or a radiative-transfer correction.
Comparing scenes from different times. A morning ASTER scene and an afternoon one differ in temperature and shadow even over identical rock. Match acquisition time, season, and sun geometry before differencing.
Pushing two-band sensors too far. Landsat TIRS cannot resolve the silica trend; do not derive lithology from two bands when five are needed.
Slope and shadow effects. Topography changes both heating and viewing geometry; in rugged terrain, combine TIR with a DEM and treat shaded, cold north-facing slopes cautiously.

Validation and QA

Cross-check every TIR anomaly against an independent layer: the published geological map, a SWIR mineral product (for example ASTER alteration ratios), and, where possible, field or laboratory emissivity spectra. Confirm the emissivity values are physically plausible (rocks typically 0.85-0.98 in this window); values above 1.0 indicate a processing error. Inspect the histogram per band for sensor striping, and verify the cloud and vegetation masks actually removed those pixels before you trust an index map.

Bathyl perspective

We use thermal infrared as the compositional layer that VNIR and SWIR cannot supply — the silica axis from felsic to mafic — and we keep emissivity and temperature strictly separated in the products we deliver. Each anomaly is presented as a candidate to be confirmed against geology and field evidence, with acquisition time, atmospheric treatment, and masking documented so the result can be audited.

Thermal Remote Sensing for Geology