The short answer

Remote sensing is a powerful reconnaissance tool for geology, but it has hard limits that no processing can remove. It senses only the surface skin, so subsurface lithology, rock age, stratigraphic order and fault kinematics are out of reach. Vegetation, soil and regolith mask the bedrock signal. Spatial and spectral resolution cap the smallest feature and the finest mineral distinction you can make. Atmosphere, illumination and mixed pixels add ambiguity. Understanding these limits is not pessimism — it is what lets you use remote sensing for what it is good at (regional context, structure, alteration screening) and send the field team to resolve what it cannot.

Limit 1: it only sees the surface

Every optical and most radar measurements respond to roughly the top millimetres to centimetres of the surface. That surface is often not fresh bedrock — it is weathered rind, desert varnish, lichen, soil, or transported cover. So:

  • No subsurface lithology. What lies beneath a soil cap, an alluvial fan or a lava flow is invisible. This is why mineral exploration still needs geophysics (magnetics, gravity, IP) and drilling.
  • No depth or thickness. Optical imagery gives no direct stratigraphic thickness; you infer it from dip and outcrop width, with large error.
  • Weathering disguises composition. A spectrally "iron-rich" surface may be a thin oxide film over unrelated bedrock.

The consequence: a remote map is a map of surface expression, and surface expression is a sometimes-faithful, sometimes-deceptive proxy for geology.

Limit 2: vegetation masks the signal

Optical sensors measure light reflected from the top of the canopy. Vegetation reflectance — strong NIR, chlorophyll absorption — dominates the pixel, and leaf-water absorption near 2.2 micrometres even mimics clay alteration. Practically:

  • Optical geology works well in arid, exposed terrain (deserts, alpine, badlands).
  • In savanna or partial cover, lithology becomes inferential and alteration mapping degrades sharply.
  • In closed-canopy forest, optical lithology is effectively unavailable; you fall back on DEM/radar structure.

There is no spectral trick to undo this; you cannot recover a signal the sensor never received. The honest move is to map and report the exposed-rock fraction and treat vegetated areas as low-confidence.

Limit 3: resolution sets the smallest mappable feature

Spatial resolution caps detail. A 30 m Landsat pixel averages all surface material within 900 m²; a 10-20 m Sentinel-2 pixel is finer but still blurs:

  • Thin beds, dykes, small plugs and narrow fault zones below the pixel are smeared into neighbours (the mixed-pixel problem) or vanish.
  • Resolution dictates the legitimate interpretation scale: a 30 m product cannot honestly support 1:5,000 mapping. State the scale and stick to it.

Spectral resolution caps mineralogy. Broad bands cannot resolve features that narrow ones can:

  • Landsat/Sentinel-2 have one or two coarse SWIR bands — fine for iron and broad clay, useless for separating sericite from kaolinite from alunite.
  • ASTER's narrow SWIR bands distinguish alteration groups but still not individual minerals.
  • Only hyperspectral (hundreds of narrow bands: AVIRIS, HyMap, EnMAP, PRISMA) approaches mineral-specific identification — and at much higher cost and complexity.

Radiometric and temporal limits matter too: low dynamic range loses subtle contrasts, and revisit/cloud constraints limit getting a clean scene at all in wet climates.

Limit 4: spectral ambiguity and confounders

Even with exposed rock, reflectance is not a unique fingerprint of lithology:

  • Different rocks, same spectrum. Two unrelated units can be spectrally indistinguishable; only field relationships separate them.
  • Same rock, different spectrum. Weathering state, grain size, moisture, shadow and varnish change the reflectance of one unit across a scene.
  • False positives. Dry salt pans and playas mimic clay/alteration; red soils mimic iron oxide; anthropogenic surfaces (roads, tailings, buildings) generate spurious anomalies.
  • Illumination and topography. Slope and aspect change apparent brightness; uncorrected, north- and south-facing slopes of the same rock look like different units. Topographic correction helps but does not fully resolve deep shadow.

Limit 5: what only the field can resolve

Some properties are simply not encoded in surface reflectance or topography:

  • Age and stratigraphic order — a function of geological history, requiring field superposition, fossils or geochronology. Two rocks of vastly different age can look identical.
  • Fault kinematics — imagery shows a lineament, not whether it is normal, reverse or strike-slip, nor its offset; that needs slickensides, offset markers, field measurement.
  • Detailed petrology and geochemistry — mineral assemblage, texture, protolith — needs hand samples, thin sections, XRD/XRF.
  • Three-dimensional geometry — dip, fold plunge and subsurface structure are inferred at best from 2D imagery.

A realistic example

Map a partly vegetated greenstone belt with Sentinel-2 plus a Copernicus 30 m DEM. You can reliably extract major lineaments from multi-azimuth hillshade, separate broad bare-rock domains in a SWIR-NIR-Green composite, and flag iron-oxide gossan candidates. You cannot from imagery alone confirm a lineament is a thrust rather than a joint set, distinguish two spectrally similar mafic units, see mineralisation under the laterite cap, or assign ages to the domains. The map's correct status is "structural and lithological interpretation, 1:50,000, optical + DEM, field verification pending" — and the field campaign is designed precisely to close those gaps.

Working within the limits

  • Match sensor to question. Iron/broad clay → Sentinel-2/Landsat; alteration groups → ASTER SWIR (pre-2008); mineral-specific → hyperspectral; structure under cloud/canopy → SAR and lidar.
  • Fuse modalities. Spectral + DEM + radar + geophysics together beat any single source; each covers another's blind spot.
  • State confidence spatially. Map exposed-rock fraction and mark vegetated/shadowed areas as low-confidence rather than mapping through them.
  • Validate the assumptions. A confusion matrix and field traverses turn an interpretation into a measured product (see the ground-truth checklist).

Common pitfalls and why they happen

  • Mapping through vegetation as if it were rock. The canopy signal is mistaken for surface geology, producing confident nonsense.
  • Over-precise interpretation. Drawing 1:5,000 detail on a 30 m product implies accuracy the data cannot support.
  • Spectral classes read as lithology. Reflectance similarity is taken for geological identity, ignoring weathering and confounders.
  • Ignoring topographic illumination. Slope-driven brightness is mistaken for compositional change.
  • No field plan. The interpretation is treated as final, so its inherent ambiguities are never resolved.

Bathyl perspective

Knowing what remote sensing cannot do is what makes the part it can do trustworthy. We deliver interpretations that explicitly flag their limits — exposure, resolution, scale, confidence — and pair them with a field plan that targets exactly the questions the imagery left open, so the satellite reduces field cost without overstating what it knows.

Related reading

Sources