Skip to content

INDUSTRY FIELD GUIDE

Space research observation collection and processing metadata for AI

A space research program can publish several product levels from related observations.

By Fiund · Updated 2026-11-05

NAICS 927110 · Space Research and Technology

A space research program can publish several product levels from related observations. NASA Earthdata documents processing levels and collection metadata. An original dataset exercise can test whether software preserves which collection, version and processing basis produced a derived observation. NASA Earthdata processing levels · NASA Earthdata collection schema

Create a synthetic observation with two versions and an abstract derived image. Give one version an explicit processing level and leave that field missing on a second. Ask the model to cite the matching source and describe why another comparison is incomplete.

Use invented measurements or an approved open release with terms. Exclude controlled technology, nonpublic payload records, access details and sensitive target data.

Describe a synthetic space data lineage example

Original Fiund exercise

A derived image needs a collection-version and processing link

A fictional research collection uses invented observations and abstract processing labels. It includes no satellite instrument, payload, target or controlled technical data.

Fictional records supplied to the evaluator
RecordSourceWhat it records
SPC1Practice collection C version 1Observation O1 value 4 units, processing label L-A; derived image IMG1 explicitly names C/V1/O1.
SPC2Practice collection C version 2O1 value 5 units; processing label absent. No equivalence statement to V1 supplied.
SPC3Derived image IMG2Filename says newer image, but source collection/version fields empty. No source map links it to V2.

Question for the model

Which image has traceable processing metadata, and can the V1-to-V2 value difference or IMG2 filename establish a comparable processed change?

Show the reference answer and checks

SPC1 makes IMG1 traceable to C/V1/O1 and L-A. SPC2 supplies a V2 value but leaves processing metadata absent, so the numerical change from 4 to 5 cannot be characterized as a comparable processed change without that basis. SPC3's filename does not establish its collection or version. Preserve observation, processing and derivative lineage rather than assigning V2 from recency language alone.

  • Inferring processing equivalence from the same observation alias and unit.
  • Using a newer-image filename as the missing collection-version source map.

Fields to preserve

Collection, version and observation
Defines the source resource.
Processing label and basis
Makes comparison conditions explicit.
Derived-image source map
Links the output to its actual inputs.

Test another case

Supply IMG2's V1 lineage despite its newer filename. Require the model to follow the source map while leaving V2's missing processing label unresolved.

Use invented measurements and derivatives or separately approved open releases. Exclude controlled technology, payload records, access details and sensitive targets.

Download this fictional example and answer key (JSON)

Discuss a contribution

Describe the sources you control, the task they could support and what has already been checked. Use the existing-data assessment for an archive or the collection brief for a new recording project. The contributor agreement defines Fiund’s licensing authority and compensation. A guide or fictional exercise does not establish an accepted collection or a sale.

Primary sources and further reading