The Problem with Treating Derived Data as Original Evidence

A derived dataset can look every bit as concrete as the source it came from.

A buffer and an isochrone are polygons, a hotspot surface is a raster, and a joined table contains values. Once the result is stored, symbolised and shared, its origin can disappear behind ordinary GIS geometry.

That is where a subtle mistake begins: treating a useful analytical result as though it were an original observation.

Geometry does not reveal how it was produced

Consider two polygons.

One is the surveyed boundary of a protected area, while the other is a 30-minute travel-time contour generated from a road network and routing profile.

At the rendering level, both are polygons that can be filled, intersected, measured and exported, but they make radically different claims.

The protected-area polygon represents a boundary established or recorded by some authority. The isochrone represents the output of a model: network data, speed assumptions, barriers, routing rules, origin point and computation date.

If both are stored without provenance, their shared geometry type can encourage a false equivalence. The map stops telling us which features were observed, declared, measured or modelled.

Derivation adds assumptions

Every transformation introduces choices.

A centroid reduces an area to a representative point. A spatial join assigns attributes according to a relationship. An interpolation estimates values where no observation occurred. A boundary harmonisation redistributes information into a common geography. AI-assisted extraction turns evidence into proposed features through interpretation.

The result may be exactly what the analysis needs, but it does not inherit the epistemic status of its inputs.

A derived layer is not weaker merely because it is derived; indeed, many of the most useful geographic products are derived. The problem is losing the distinction.

Reuse amplifies provenance loss

Inside the original analysis, everyone may know what the layer means. The filename service_area_30min_v2 might be enough for the person who created it.

Then the layer is reused.

It is added to another project, exported, renamed, clipped and published. The next analyst sees a polygon called accessible_area, but the routing profile is no longer obvious, the road-network date is forgotten and seasonal closures were never represented. The contour begins to function as though it described a directly observed fact about access.

This is how context decays: not through one dramatic error, but through repeated reuse of a result whose derivation was never attached to it.

Derived status should be first-class metadata

The remedy is not to burden every layer with a laboratory notebook. It is to preserve the few facts that allow a future user to reconstruct the claim.

A derived result should be able to answer:

What inputs produced it?
Which versions of those inputs were used?
What operation or model transformed them?
Which parameters materially affect interpretation?
When was the result produced?
What uncertainty or limitation should survive reuse?

Provenance standards such as W3C PROV formalise the broader idea that outputs are entities generated by activities using other entities. A GIS product does not need to expose the full ontology to every user. It does benefit from respecting the relationship.

Original evidence also has provenance

Calling something “original” should not make it mysterious either.

Surveyed boundaries, field observations and official registries still come from methods, institutions and dates. They can contain error and interpretation. The useful distinction is not pure versus impure data. It is between different kinds of claim and different distances from the underlying observation or authority.

A derived result deserves confidence when its method and inputs support the intended use, while an original observation deserves confidence when its collection method and context do.

The mistake is flattening both into a generic category called “data”.

Maps should reveal the distance between evidence and conclusion

Some derived results can be labelled directly: “modelled 30-minute travel area”, “interpolated rainfall”, “estimated population”, “harmonised 2020 districts”. That small amount of language prevents the geometry from carrying more authority than it deserves.

For consequential analysis, the underlying metadata should remain available too.

This does not make the map hesitant. It makes the reasoning legible.

Derived data is valuable precisely because it lets us answer questions that raw observations cannot answer on their own. Preserving its derivation does not diminish that value; it tells the next user what kind of answer they are looking at.

References

W3C — PROV-O: The PROV Ontology