Skip to main content

Inference risk

GIS is built to combine information. That means a map can reveal something that none of its individual layers says explicitly.

For iwi, hapū, marae and taiao teams this matters when ordinary public or working data can be joined to make a whenua relationship, likely site, access pattern or other place-based information much easier to infer.

A simple whenua example​

Imagine a research project contains:

  • an old ingoa wāhi from a historical report
  • a Māori land block reference
  • a broad area from an old map
  • modern aerial imagery
  • LiDAR terrain
  • a road or track layer

No one layer gives an exact location. Together they may narrow the likely place considerably.

Conceptual flow

That is not necessarily a problem. It is simply a property of spatial analysis that is useful to recognise when deciding what version of a result belongs in a research workspace, a hui map, a partner extract or a public web map.

Derived layers can reveal more than their inputs​

Common GIS operations can create new information:

  • buffers
  • spatial joins
  • nearest-feature analysis
  • suitability modelling
  • network analysis
  • clustering
  • raster classification
  • change detection
  • image recognition
  • machine-learning predictions

For example, a taiao team may combine restoration observations, slope and catchment data to identify likely erosion pressure. A heritage researcher may combine an old name, terrain and historical imagery to narrow a possible kāinga location. The derived result is a new interpretation and should be labelled as such.

See Authoritative for what? and Provenance and source checking.

Remote sensing changes what can be observed​

Satellite imagery, aerial photographs and LiDAR can show features without anybody entering the whenua to collect them directly.

That can be extremely useful for:

  • wetland change
  • river movement
  • slips and erosion
  • pā landscapes
  • old tracks
  • vegetation change
  • forestry and land-use history
  • access and terrain investigation

It also means that leaving a coordinate out of one dataset does not make the wider landscape impossible to investigate. Imagery and LiDAR and elevation explain the practical capabilities and limitations of those sources.

AI changes the effort required​

Computer vision and multimodal AI can inspect imagery, photographs and text collections much faster than a person working item by item. A model may infer location from landscape, buildings, signs, vegetation or text even when a photograph contains no EXIF coordinate.

For historical and Treaty research, AI can also connect repeated names, descriptions and coordinates across many documents. That can be useful, but the source documents remain the evidence and the AI result remains a candidate or interpretation until checked.

See Local AI and Practical AI applications.

Check combinations, not only individual layers​

When preparing a wider shared or public version of a map, it can be useful to look at the complete combination rather than reviewing each layer separately.

A practical check is:

  1. identify what the audience needs to understand
  2. turn on the ordinary public layers likely to sit beside the map
  3. compare the combined result with imagery and terrain
  4. check whether attributes, labels, popups or metadata reveal additional context
  5. identify any derived interpretation clearly
  6. choose the geometry and fields that make sense for that version

This is ordinary map and information design. A working research layer can remain detailed while a public explanatory map carries a different level of detail.

Inference is not certainty​

An inferred location or relationship can be plausible and still be wrong.

Several roads may fit an old description. A historical name may have moved between maps. Terrain may suggest a likely route without proving it was used. AI may confidently connect records that refer to different places.

Keep the difference visible between:

  • observed or source data
  • mapped relationships
  • derived analysis
  • historical interpretation
  • hypothesis
  • uncertain location

That distinction is particularly useful for whenua, ingoa wāhi and pā research where the evidence chain may span maps, court records, kōrero, imagery and terrain.

Sources​

Last reviewed: 26 August 2026

Last updated on