Skip to main content

Turn results into GIS data

Once the candidate names have been checked, the AI table can become ordinary GIS data. The important step is not the file format. It is keeping enough provenance to show where each record came from and how its geometry was obtained.

A practical table

A small place-name dataset could use these fields:

FieldPurpose
record_idstable ID for your working record
place_as_writtensource spelling, unchanged
matched_namechecked or normalised name where applicable
feature_typeriver, lake, locality, maunga, other
source_documentreport or document title
source_pagepage or section
source_contextshort evidence note
name_sourceGazetteer, local source, historical record, other
name_source_idGazetteer Name ID or local identifier
review_statusunreviewed, matched, unresolved, rejected, other
review_noteswhy the match was made
latitudeonly from an identified source
longitudeonly from an identified source
geometry_sourcewhere the coordinates or shape came from
sensitivitypublic, internal, restricted or your local vocabulary

A wider internal dataset may include reviewer, review date, AI model, embedding model, OCR quality and authority/approval fields.

CSV first

CSV is useful because it is simple to inspect and edit. Save text as UTF-8 so macrons are preserved.

A reviewed CSV can be loaded directly into QGIS when it has latitude and longitude fields. For Aotearoa data, keep the original geographic coordinates and set the layer to WGS 84 / EPSG:4326 when that is the coordinate system supplied by the source.

Do not guess the CRS. Record the coordinate system supplied by the geographic source.

See QGIS getting started for the GIS steps rather than repeating the full application guide here.

GeoJSON

GeoJSON is useful for web mapping and data exchange. Convert to GeoJSON only after the coordinate source and geometry type are known.

A point feature should carry the provenance attributes with the geometry. Do not reduce the record to only name, lat and lon.

Example structure:

{
"type": "Feature",
"properties": {
"record_id": "DEMO-001",
"place_as_written": "example source wording",
"matched_name": "checked name",
"source_document": "public source",
"source_page": "123",
"name_source": "New Zealand Gazetteer",
"name_source_id": "example-id",
"review_status": "matched_public_source"
},
"geometry": {
"type": "Point",
"coordinates": [174.0, -41.0]
}
}

The coordinates above are syntax examples only, not a demonstration place. Use the downloadable worked dataset on the complete example page for genuine public locations with traceable geographic sources.

KML and Google Earth

KML works well for a simple wānanga demonstration because a reviewed table can become named placemarks that open in Google Earth.

Keep the description field useful. Include the source document, page, review status and geographic source so a person clicking the placemark can understand why it exists.

For Google Earth basics, use the existing Google Earth guide.

QGIS workflow

A simple route is:

  1. Export the reviewed table as UTF-8 CSV.
  2. In QGIS, add a delimited text layer.
  3. Select the longitude field for X and latitude for Y.
  4. Set the correct source CRS, commonly EPSG:4326 for Gazetteer latitude/longitude.
  5. Inspect every point against a trusted basemap or reference layer.
  6. Save the layer to GeoPackage for ongoing work.
  7. Keep unresolved rows in the table even if they have no geometry.

A missing geometry is better than a fabricated point.

Do not discard unresolved records

An unresolved place-name record can still be valuable research data. It may later be matched using a historical plan, Māori Land Court material, local knowledge or a better source.

Keep it in a non-spatial table with its evidence and review notes. GIS does not require every useful record to become a point immediately.

AI can help format, not author the geography

It is reasonable to ask a local model to reformat a reviewed table, generate a CSV header, suggest a data dictionary or convert valid records into GeoJSON syntax.

It is not reasonable to ask it to fill blank coordinates from memory.

A useful prompt is:

Convert only the reviewed rows below into CSV.
Preserve every source and provenance field.
Do not add coordinates, names or values that are absent.
Leave blank values blank.
Return UTF-8 text and preserve macrons.

Publishing comes later

A correct point can still be inappropriate to publish. Before creating a public map, review Publishing and sharing, Sensitive locations and the relevant governance process for the information.

For public data discovery, see the MāoriGIS.nz data catalogue.

Next, work through the complete example.

Last reviewed: 22 August 2026

Keep the source attached to the feature

For Treaty, ingoa wāhi and whenua research, a GIS point or polygon should retain enough provenance to return to the page, record or map from which it came. Use Provenance and source checking and Archives to map when designing the output fields.