Short answer

A geological map is not a coloured picture — it is a small relational dataset. The three layers no geological map can omit are: a polygon layer of map units (lithostratigraphic or lithologic polygons), a line layer of contacts and faults, and a point layer of structural measurements (strike/dip, foliation, lineation). The USGS GeMS schema names these MapUnitPolys, ContactsAndFaults, and OrientationPoints, and pairs them with a DescriptionOfMapUnits table and a Glossary. Around that core you add base/context layers (DEM, imagery, hydrology, boundaries, access), evidence layers (field stations, samples, boreholes, geophysics, remote-sensing indices), and decision layers (constraints, targets, and confidence). The point of structuring them this way is that a reader can separate what was observed from what was interpreted, and reuse the result rather than trace it off a JPEG.

The interpretation core

Map units — MapUnitPolys

Polygons that completely tile the map area, each carrying a map-unit label that keys into the unit description. Each polygon should attribute at least: the unit identifier (MapUnit), an identity/confidence flag, and a data-source reference. The unit identifier is a foreign key into the DescriptionOfMapUnits (DMU) table, which holds the full name, age, lithology, rank (formation, member), and the legend ordering. Modelling units this way means a query like "show all Cretaceous volcanic units" is a single attribute filter, not a manual colour hunt.

Contacts and faults — ContactsAndFaults

A single line layer carrying all boundary linework — depositional/intrusive contacts and faults alike — distinguished by a type attribute, plus an existence confidence (is the feature real?) and a locational accuracy (how well is it positioned?). Faults additionally need geometry of the fault plane where known. Keeping contacts and faults in one topologically clean line layer is deliberate: the polygon boundaries of MapUnitPolys must coincide exactly with these lines, so they share edges and are validated together.

Structural measurements — OrientationPoints

Point observations of orientation — bedding strike/dip, foliation, joints, fold axes, lineations — each with the measurement type, azimuth/inclination values, and a station/source reference. In GIS these drive data-defined structural symbols: a strike-and-dip tick rotated by the azimuth attribute and labelled with the dip value, rendered straight from the table rather than hand-placed.

The supporting layers

Base / context — the frame interpretation sits on:

  • DEM (and its derivatives: hillshade, slope, aspect) for relief and structural grain.
  • Orthoimagery / satellite imagery for ground truth and lineament tracing.
  • Hydrography (streams, lakes), administrative/licence boundaries, roads/access, and topographic base for orientation and logistics.

Evidence — the observations that justify the interpretation:

  • Field stations (where a geologist stood and recorded).
  • Samples (geochemistry, petrology, geochronology) with lab IDs.
  • Boreholes/drillholes with collar locations and downhole data.
  • Geophysics (magnetics, gravity, radiometrics) as raster grids.
  • Remote-sensing products (e.g. Sentinel-2 alteration/mineral indices).

Decision — what the map is for:

  • Constraints (no-go, environmental, tenure), targets/opportunities, and explicit confidence layers.

Crucially, evidence and interpretation are separate layers. The value of a geological map is that someone can switch off the interpreted contacts, look at the sample points and imagery underneath, and judge whether the boundary was well-constrained or speculative.

Why GeMS exists

GeMS is the USGS standard format for the digital publication and archiving of geologic maps, adopted by the National Cooperative Geologic Mapping Program. It standardises the feature classes above, enforces controlled vocabularies (so "fault, normal" means the same thing across maps), requires a DescriptionOfMapUnits and Glossary, and — importantly — bakes existence confidence and locational accuracy into the schema so uncertainty is a first-class attribute rather than an afterthought. Even if you are not publishing through USGS, the GeMS data model is a sound template: it forces you to record which feature, how confident, how accurately located, and from what source. The National Geologic Map Database (NGMDB) is the companion catalogue that makes such maps discoverable and persistent.

Worked example: a minimal but complete stack

For a 1:25,000 mapping project, a defensible layer set in a single GeoPackage might be:

project.gpkg
  MapUnitPolys        (polygons; MapUnit, IdentityConfidence, DataSourceID)
  ContactsAndFaults   (lines; Type, ExistenceConfidence, LocationConfidenceMeters, DataSourceID)
  OrientationPoints   (points; Type, Azimuth, Inclination, StationID)
  Stations            (points; StationID, observer, date)
  Samples             (points; SampleID, type, lab ref)
  DataSources         (table; source citations keyed by DataSourceID)
  DescriptionOfMapUnits (table; MapUnit, Name, Age, Lithology, rank, legend order)
  base/  DEM, hillshade, orthoimage  (rasters or references)

Build topology so MapUnitPolys boundaries snap to ContactsAndFaults (in QGIS via the Topology Checker and snapping; in ArcGIS via geodatabase topology rules). Symbolise units via a join from MapUnitPolys.MapUnit to the DMU legend order, and structural points via data-defined rotation on Azimuth. Reproject the whole project to a single metric CRS (e.g. the relevant UTM zone) so distances and areas are valid.

Modelling uncertainty as data, not as a caveat

The single feature that separates a professional geological dataset from a tidy drawing is that uncertainty is attributed, not relegated to a marginal note. GeMS encodes this in two distinct fields that beginners often conflate:

  • Existence confidence — does this feature exist at all? A contact you walked and a contact you inferred from a tonal break on imagery are both real linework, but one is "certain" and the other "questionable." This drives the dashed-vs-solid cartographic convention, but as an attribute it also lets a downstream user filter to only well-established features.
  • Locational accuracy (e.g. LocationConfidenceMeters) — how well is it positioned? A contact mapped at 1:5,000 with RTK GPS might be accurate to a few metres; one traced from a 1:250,000 legacy sheet might be accurate to hundreds of metres. Recording this in metres lets you, for example, buffer faults by their own locational uncertainty before a setback analysis instead of treating every line as a sharp truth.

Provenance is the third pillar: every feature should reference a DataSourceID that resolves to a citation in a DataSources table. When a reviewer asks "where did this fault come from," the answer is a join, not a memory. Together, existence confidence, locational accuracy, and source turn the map from an assertion into evidence you can interrogate — which is exactly what makes it safe to reuse for targeting, hazard screening, or regulatory review years later.

Common pitfalls and why they happen

  • Colour as the data model. A polygon fill is styling, not an attribute; if the unit lives only in the symbology, you cannot query or re-map it. Store the unit code in a field keyed to the DMU.
  • Contacts and unit boundaries that do not coincide. Drawing polygons and lines independently leaves slivers and gaps. Share edges with snapping/topology so boundaries and contacts are the same geometry.
  • No confidence or source fields. Without ExistenceConfidence, LocationConfidenceMeters, and DataSourceID, a future user cannot tell a surveyed contact from an inferred one. Make uncertainty and provenance mandatory attributes (the GeMS way).
  • Mixing scales silently. Combining a 1:250,000 regional map with 1:10,000 detail without flagging it implies false precision. Record source scale per feature.
  • Digitising a scan without preserving uncertainty. Tracing a raster map discards the original's accuracy information. Capture source scale and confidence as you digitise.
  • Publishing a flat image. A PNG cannot be queried, joined, or updated. Deliver structured layers (GeoPackage/GeMS) with the image only as a cartographic export.

QA before you trust the map

  • Run topology checks: no gaps/overlaps in MapUnitPolys, no dangles where contacts should close, unit boundaries coincident with ContactsAndFaults.
  • Confirm every MapUnit value resolves to a DMU entry and every feature has a DataSourceID.
  • Verify CRS and units are consistent across all layers and appropriate for measurement.
  • Check that confidence/locational-accuracy fields are populated, not left default.
  • Validate that evidence layers (stations, samples) actually support the drawn interpretation in a few test areas — toggle interpretation off and look at the evidence.

Bathyl perspective

We build geological maps as inspectable systems: observation, interpretation, context, and confidence in separate, topologically clean layers, with provenance and uncertainty modelled as attributes rather than implied by colour. That structure keeps the science careful while making the map queryable, comparable, and safe to hand to the next analyst or stakeholder.

Related reading

Sources