Short answer

NDVI (the Normalized Difference Vegetation Index) measures how green and photosynthetically active a surface is — and that is all it directly measures. It does not see rock, mineralogy, or structure. Its value in geology is entirely indirect: where the underlying substrate controls vegetation through drainage, soil depth, or chemistry, an NDVI pattern can trace a geological boundary, and where the ground is bare, NDVI doubles as a useful bare-ground mask. The honest framing is that NDVI is a geobotanical proxy and a masking tool, not a geological mapping product, and most of the time its signal is dominated by climate and land use rather than geology.

What NDVI actually is

NDVI exploits a single robust fact: healthy green vegetation reflects strongly in the near-infrared (NIR) and absorbs strongly in the red, because chlorophyll absorbs red light while leaf internal structure scatters NIR. The index normalises that contrast:

NDVI = (NIR - Red) / (NIR + Red)

For Sentinel-2 this is (B8 - B4) / (B8 + B4); for Landsat 8/9 OLI it is (B5 - B4) / (B5 + B4). Output ranges from -1 to 1:

  • < 0 — water, snow, cloud.
  • ~0 to 0.2 — bare soil, rock, sparse cover. This is the band that matters most to a geologist, because it is where the ground shows through.
  • 0.2 to 0.5 — grassland, shrubs, stressed or sparse vegetation.
  • > 0.5 — dense, vigorous canopy.

In QGIS the Raster Calculator computes it directly; with GDAL:

gdal_calc.py -A B8.tif -B B4.tif \
  --calc="(A.astype(float)-B)/(A+B+1e-10)" \
  --outfile=ndvi.tif --NoDataValue=-9999

The +1e-10 guards against divide-by-zero where NIR and red are both zero (nodata, deep shadow).

Where NDVI genuinely helps geology

1. Geobotanical mapping. Plants respond to their substrate. Limestone drains fast and supports calcicole communities; serpentinite soils are toxic to many plants and carry sparse, distinctive flora; clay-rich units hold water; a fault zone may channel groundwater and support a line of greener vegetation across an otherwise dry slope. Where these effects are strong, an NDVI boundary can coincide with a lithological contact or a structure even though NDVI never imaged the rock. The classic, defensible use is detecting linear vegetation anomalies that follow faults or dykes.

2. Bare-ground masking — the workhorse use. Before any direct geological remote sensing (iron-oxide ratios, SWIR clay mapping), you need to remove vegetated pixels, because canopy spectra swamp mineral spectra. An NDVI threshold (commonly masking out NDVI > ~0.2-0.3) isolates exposed ground. This is probably NDVI's most valuable contribution to a geology workflow: not as a map, but as the gate that makes the mineral-mapping bands trustworthy.

3. Change and seasonal context. A multi-date NDVI series shows where vegetation is stressed or where bare ground expands — useful for monitoring erosion, mine rehabilitation, or seasonal moisture that affects other interpretations.

Where NDVI misleads

The dominant controls on NDVI are climate, season, land use, and soil moisture — not geology. In most landscapes those factors produce far larger NDVI variation than any substrate effect, and the geological signal, if present, is buried in the noise. Specific failure modes:

  • Saturation. Over dense canopy NDVI flattens near its maximum and stops responding, so it cannot distinguish a forest on granite from a forest on schist.
  • Seasonality. The same field reads 0.7 in spring and 0.2 after harvest. Comparing dates without controlling for phenology produces "change" that is purely calendar-driven.
  • Soil background and moisture. Over sparse cover the soil colour and wetness dominate; a recent rain shifts NDVI without any change in vegetation. This is why soil-adjusted variants (SAVI, with its L soil-brightness term) exist.
  • Land use. Field boundaries, irrigation, and logging draw crisp NDVI edges that look like geology but are entirely anthropogenic.
  • Atmosphere and cloud. Haze and thin cloud depress NIR; under cloud NDVI is meaningless. Use surface-reflectance (Sentinel-2 L2A) inputs and a cloud mask.

The cumulative lesson: an NDVI boundary is a hypothesis about the surface, never a confirmed geological line.

A defensible workflow

To use NDVI for geology without overreaching:

  1. Start with a surface-reflectance, cloud-masked scene (Sentinel-2 L2A), reprojected to the project CRS.
  2. Compute NDVI and inspect the histogram to set sensible class breaks for your landscape, rather than using textbook thresholds blindly.
  3. For bare-ground work, threshold and use the mask to gate iron-oxide and SWIR mineral products — this is the high-confidence application.
  4. For geobotanical work, look specifically for linear or sharply-bounded anomalies that align with mapped faults, contacts, or drainage, and ignore broad gradients that track climate or land use.
  5. Corroborate every candidate against the geological map, a DEM (does the anomaly follow a structural lineament?), and ideally direct imagery or field checks.
  6. Use a multi-date composite or median to suppress one-off seasonal and atmospheric artefacts before drawing conclusions.

QA and validation

  • Confirm surface-reflectance inputs and a cloud mask — NDVI on top-of-atmosphere or cloudy data is unreliable.
  • Check the date(s). Note season and recent rainfall; a single-date NDVI carries a phenological fingerprint.
  • Cross-reference anomalies with a DEM and the geological map. A real geobotanical signal usually has a structural explanation.
  • Watch for land-use edges — straight lines and right angles are almost always human, not geological.
  • Quantify, do not eyeball. Report the threshold and date used so the interpretation is reproducible.

Common pitfalls and why they happen

  • Treating an NDVI image as a lithology map. NDVI measures vegetation; rock type is inferred at best, so any direct read-off is unfounded.
  • Comparing seasons as if they were geology. Phenology dwarfs substrate effects, so uncontrolled multi-date NDVI invents change.
  • Trusting NDVI over dense canopy. Saturation erases the very contrast you hoped to see.
  • Skipping the cloud mask. Haze suppresses NIR and fabricates low-NDVI "anomalies."
  • Confusing field boundaries with contacts. Land-use edges are crisp and linear, exactly like the structural signal people hope to find.

Bathyl perspective

We use NDVI mostly as plumbing — the bare-ground mask that makes our mineral-mapping layers honest — and only cautiously as a geobotanical clue where vegetation clearly tracks structure. When we report an NDVI-derived geological inference, we state the date, the threshold, and the corroborating evidence, because an index that cannot see rock should never be presented as if it could.

Related reading

Sources