Short answer

In mineral exploration, GIS is the integration layer that brings geology, structure, geochemistry, geophysics, and remote sensing into one co-registered spatial database, then combines those evidence layers into a prospectivity model that ranks ground for drilling and field follow-up. Done well it compresses years of multidisciplinary data into a defensible map of where to look next — and keeps the underlying interpretation auditable. The exploration value is not the pretty geological map; it is the ability to query, weight, and overlay independent datasets that each carry a piece of the mineral-system signal.

The mineral systems framing

Modern exploration GIS is organised around the mineral systems concept rather than isolated indicators. A mineral system needs a metal source, an energy/fluid drive, pathways (structures), a trap or depositional site, and a preservation history. Each of those translates into mappable proxies:

  • Source / fertility — favourable lithologies, intrusive bodies, regional geochemistry.
  • Pathways — faults, shear zones, and lineaments interpreted from geophysics and DEMs.
  • Trap / deposition — lithological contacts, fold hinges, permeability contrasts.
  • Alteration footprint — hydrothermal mineral assemblages mapped from spectral remote sensing.

Building the GIS around these criteria, rather than around whatever datasets happen to exist, keeps the model interpretable and forces you to be explicit about the deposit model you are targeting (porphyry Cu, orogenic Au, VMS, etc.).

The core data layers

Geology and structure

A structured digital geological map is the backbone. The USGS GeMS (Geologic Map Schema) defines a standard geodatabase model where map units, contacts, faults, descriptions, ages, and data sources are stored as linked feature classes and tables, not as styling. The point is that "what kind of contact," "how confident," and "which survey it came from" are attributes you can query — not colours you can only look at. Faults and contacts become the structural pathway and trap layers; rasterise them and run a proximity analysis (gdal:proximity, or ArcGIS Euclidean Distance) to produce continuous "distance-to-structure" evidence.

Geochemistry

Soil, stream-sediment, and rock-chip assays are point data. Workflow: QA the data (detection limits, censored values, units — ppm vs ppb), reduce closure effects with a log or centred-log-ratio transform, identify anomalies (often >95th percentile or mean + 2σ per element), and interpolate cautiously (IDW or kriging) only where sampling density justifies it. Multi-element ratios and pathfinder elements (As, Sb, Bi for Au; Mo for porphyry) often outperform single elements. Keep the interpolated surface clearly flagged as derived.

Geophysics

Magnetics, radiometrics, and gravity grids are continuous rasters that reveal structure and lithology beneath cover. Derivatives — analytic signal, tilt derivative, first vertical derivative of magnetics — sharpen contacts and faults; radiometric K/Th ratios can flag potassic alteration. These are reprojected and resampled onto the common grid like any other raster.

Remote sensing for alteration

Spectral data maps the hydrothermal alteration halo directly:

  • ASTER is the workhorse for alteration because its six SWIR bands resolve Al-OH (clay/sericite), Mg-OH, and carbonate features. Classic band ratios: (4+6)/5 for Al-OH (argillic/phyllic), (6+9)/(7+8) for carbonate, 2/1 for ferric iron.
  • Sentinel-2 (10–20 m, free, frequent revisit) and Landsat 8/9 OLI give broad reconnaissance. Iron-oxide ratio ≈ B4/B2 (Landsat 4/2); a hydroxyl/clay ratio on Landsat 8 ≈ B6/B7.
  • Process in QGIS with the raster calculator or in Google Earth Engine for large areas, masking vegetation (NDVI threshold) and cloud first, since dense vegetation swamps the mineral signal.

Building the prospectivity model

Once every evidence layer is reprojected to one CRS and resampled to a shared grid (e.g. gdalwarp -t_srs EPSG:32733 -tr 30 30 -r bilinear), combine them. Two families of method:

Knowledge-driven (when you have few or no known deposits):

  • Fuzzy logic — rescale each evidence layer to 0–1 membership ("favourability"), then combine with fuzzy operators (fuzzy OR/AND/gamma). Intuitive and transparent.
  • Index overlay / AHP — assign expert weights to classes and sum.

Data-driven (when you have a training set of known occurrences):

  • Weights-of-evidence — Bayesian; computes positive/negative weights per layer from the spatial association with known deposits, with a contrast statistic that tests each layer's significance.
  • Logistic regression / machine learning (random forest, etc.) — trained on deposit and non-deposit sites.

Whichever method, classify the output into relative prospectivity classes and validate against withheld known occurrences. Report the method, the weights, and the deposit model assumed — a prospectivity map without those is uninterpretable.

Worked example: orogenic-gold reconnaissance

  1. Assemble: 1:100,000 geology (GeMS), interpreted faults from aeromagnetics, stream-sediment Au+As assays, ASTER Al-OH ratio for sericite alteration.
  2. Reproject all to EPSG:32733, resample to 30 m, identical extent.
  3. Evidence layers: distance-to-fault (proximity), favourable-lithology buffer, Au-As anomaly surface (>95th percentile), ASTER Al-OH high.
  4. Fuzzy-membership each to 0–1; combine with a fuzzy gamma (γ ≈ 0.85) operator.
  5. Classify into five prospectivity classes; overlay the three known historic workings.
  6. If most known workings fall in the top two classes, prioritise those zones for soil grids and mapping; treat the rest as untested, not barren.

Common pitfalls and why they happen

  • Misregistered layers — combining datasets in different CRSs or grids smears every overlay. Co-register first; this is the most frequent and most damaging error.
  • Treating interpolated geochemistry as truth — kriging across sparse, clustered samples invents anomalies. Honour sample density and flag derived surfaces.
  • Ignoring closure in geochemistry — raw percentages/ppm are compositional; skipping log/CLR transforms distorts correlations.
  • Vegetation contamination in spectral ratios — un-masked vegetation mimics or hides alteration signatures.
  • Circular validation — using the same deposits to both weight and validate a data-driven model inflates apparent performance.
  • Map colours as the data model — digitising a scan into polygons without preserving source scale, contact confidence, or lineage loses the very attributes that make the map exploration-grade.

QA and validation checklist

  • All layers share one projected CRS, pixel size, and extent.
  • Geochemistry QA'd for detection limits, units, and compositional transform.
  • Spectral products vegetation- and cloud-masked; ratios documented.
  • Geology stored with units, contacts, faults, confidence, and source in a GeMS-style schema.
  • Prospectivity method, weights, and target deposit model documented.
  • Model validated against withheld known occurrences, not the training set.

Bathyl perspective

We build exploration GIS so the interpretation survives the project — every evidence layer traceable to its source, every weight justifiable, every derived surface labelled as derived. A prospectivity map is a hypothesis ranking, not a discovery; its job is to make the next dollar of fieldwork land on the highest-confidence ground while keeping the geological reasoning open to inspection and revision.

Related reading

Sources