The short answer

Remote sensing supports geological mapping in three ways: it discriminates lithology and alteration through how rocks reflect VNIR and SWIR light, it reveals structure — faults, folds, lineaments — through topographic expression in DEMs and radar, and it provides regional context that targets where field effort pays off. Sentinel-2 and Landsat handle broad lithological and iron/clay screening, ASTER adds SWIR detail for alteration, and DEMs (Copernicus, SRTM, lidar) plus SAR carry the structural signal. None of it replaces the hammer: a satellite shows where contacts and structures are, but stratigraphy, kinematics and age still come from the field.

Lithology: why rocks separate spectrally

Rocks and their weathering products have characteristic reflectance shaped by mineralogy. Iron-bearing minerals absorb in the visible-NIR; clays, micas, carbonates and sulphate alteration minerals absorb in the SWIR (around 2.2-2.35 micrometres); mafic versus felsic rocks differ in overall brightness and iron content. Multispectral sensors sample this coarsely but usefully.

Useful composites and ratios:

  • Geology false-colour: SWIR-NIR-Green — Landsat 8/9 B7-B5-B3, Sentinel-2 B12-B8-B3. Separates bare rock and soil from vegetation and brings out lithological tone differences.
  • Iron oxide: Red/Blue — Landsat B4/B2; Sentinel-2 B4/B2.
  • Ferrous iron: SWIR/NIR — Landsat B6/B5.
  • Clay / hydroxyl: SWIR1/SWIR2 — Landsat B6/B7; Sentinel-2 B11/B12.
  • ASTER alteration: Al-OH (B5+B7)/B6, propylitic (B7+B9)/B8 — narrow SWIR bands give far better mineral-group separation than Landsat/Sentinel-2.

Composite three ratios into an RGB to build an unsupervised "lithology enhancement," then refine with supervised classification (Random Forest, SVM) trained on field-verified polygons.

Structure: the DEM is your best geologist

Bedrock controls topography, so a good elevation model often reveals structure better than spectral data. Bedding traces, fold hinges, fault scarps and joint sets express as linear and curvilinear topographic features.

Multi-azimuth hillshade is the core technique — a single sun angle hides structures parallel to it, so generate several and combine:

gdaldem hillshade dem.tif hs_315.tif -az 315 -alt 45
gdaldem hillshade dem.tif hs_045.tif -az 45  -alt 45
gdaldem hillshade dem.tif hs_135.tif -az 135 -alt 45

Sources: SRTM 1-arc-second (~30 m), Copernicus GLO-30 (30 m, generally cleaner), ALOS PALSAR DEMs, and lidar (sub-metre) where available. Lidar's bare-earth filtering can expose fault scarps and bedding hidden under canopy that optical sensors never see.

Derive slope, aspect and curvature with gdaldem; steep, laterally continuous slope breaks frequently mark contacts or faults. For lineament work, interpret multi-azimuth hillshades manually first (the eye is excellent at this), then optionally run edge detection (Sobel/Canny) or a dedicated lineament-extraction tool, and plot the results on a rose diagram to read the dominant structural trends. Treat automated lineaments sceptically: roads, field boundaries and drainage produce false positives.

Radar (SAR) — Sentinel-1 C-band — adds structural sensitivity because side-looking illumination emphasises relief independent of cloud and sun, useful in persistently cloudy or vegetated terrain. InSAR coherence and amplitude can highlight fault-related fabric.

A worked regional mapping workflow

  1. Assemble and align inputs. Sentinel-2 L2A, an ASTER pre-2008 scene if alteration matters, a Copernicus 30 m DEM, and Sentinel-1. Reproject everything to one projected CRS, e.g.

    gdalwarp -t_srs EPSG:32737 -r bilinear sentinel2_l2a.tif s2_utm37s.tif

  2. Mask vegetation and cloud. NDVI mask and the S2 SCL band — dense vegetation defeats optical geology, so know your exposed fraction up front.

  3. Spectral enhancement. Build the SWIR-NIR-Green composite plus iron and clay ratio RGBs; add ASTER alteration ratios.

  4. Structural enhancement. Multi-azimuth hillshades, slope and a Sentinel-1 amplitude layer.

  5. Interpret. Digitise lithological domains and structural lineaments at a fixed mapping scale (state it — 1:50,000 interpretation cannot be used at 1:5,000).

  6. Classify (optional). Train a Random Forest on field-checked samples to extend interpretation across the scene.

  7. Validate in the field. Ground-truth contacts, sample lithologies, measure orientations, and adjust the map.

Validation before the map is "geology"

A remote interpretation is a hypothesis map. Validate it:

  • Field traverses across interpreted contacts and lineaments; confirm lithology, measure strike/dip, establish kinematics on faults.
  • Existing maps and literature — overlay published geology; large disagreements need explanation, not silent overwriting.
  • Accuracy assessment for any classification — a confusion matrix and kappa against independent field points, reported, not assumed.
  • Scale honesty — record the interpretation scale and source resolution in the metadata and legend.

Common pitfalls and why they happen

  • Calling tone "lithology." Reflectance varies with weathering, lichen, soil and shadow, not just bedrock; without field control, a colour boundary is not a contact.
  • Mapping under vegetation. Optical geology needs exposure; in vegetated terrain, lean on DEM/radar structure and accept that lithology is largely inferred.
  • Trusting automated lineaments. Anthropogenic and drainage features mimic structure; unfiltered lineament maps overstate faulting.
  • Single-azimuth hillshade. Structures parallel to the light vanish; one hillshade systematically under-maps one orientation.
  • Mixing CRS or DEM datums. Misaligned spectral and elevation layers shift contacts off topography and corrupt overlays.
  • Post-2008 ASTER SWIR. The SWIR detector failed; those bands are noise for alteration work.

QA and validation summary

Confirm every layer shares CRS and datum; verify masks on cloud edges and dense vegetation; report classification accuracy with a confusion matrix; ground-truth a representative sample of contacts and structures; and state the mapping scale on the product. Keep intermediate enhancements so a reviewer can see how each interpreted line was derived.

Bathyl perspective

We use remote sensing to make fieldwork sharper, not to skip it — the satellite tells the geologist where the contacts and structures probably are, and the field confirms what they actually are. The deliverable that earns trust records its sensors, DEM source, mapping scale and validation, so the next user knows exactly how far the interpretation can be pushed before a hammer is required.

Related reading

Sources