INDUSTRY FIELD GUIDE
Conservation observation and validation records for AI
A volunteer’s observation and an expert’s later identification have different evidence.
NAICS 813312 · Environment, Conservation and Wildlife Organizations
A volunteer’s observation and an expert’s later identification have different evidence. EPA’s participatory science quality handbook provides context for planning collection and review. A conservation group can test how well a model preserves the distinction with cleared or fictional examples. EPA participatory science quality handbook
Use original training images of common subjects or independently drawn examples. Include two observation records with similar timestamps and a later expert label tied to only one. Ask the model to preserve the other record’s uncertainty and cite the reviewed image. Give a separate score for matching identifiers and species labels.
Obtain photographer permission and document volunteer terms. Remove precise location, hidden image metadata and sensitive wildlife coordinates from the first sample.
Describe a conservation observation sample
Original Fiund exercise
An expert label validates the named observation, not every similar image
Original drawings of common abstract plant forms replace wildlife photographs. Dummy observations and expert labels test source attribution without locations or volunteer identities.
| Record | Source | What it records |
|---|---|---|
| COB1 | Practice observations | OBS-A/image IA and OBS-B/image IB captured at the same artificial timestamp. Initial labels both uncertain. |
| COB2 | Expert review | Explicitly names OBS-A/IA and assigns practice class Leaf-X. Does not review IB. |
| COB3 | Unverified summary | Says Leaf-X seen in the session, without observation ID. IB resembles IA but has no reviewed assignment. |
Question for the model
Which observation has expert validation, and can the session summary or visual resemblance establish the same class for OBS-B?
Show the reference answer and checks
COB2 validates the Leaf-X assignment for OBS-A and image IA only. COB1 keeps OBS-B/IB separately identified, and COB3's session-wide wording supplies no observation-specific expert link. Similar appearance or a shared timestamp does not validate IB. Report A's reviewed class and B's remaining uncertainty, distinguishing a model's possible suggestion from the expert-confirmed source label.
- Propagating an expert identification across every image from the same session.
- Treating a visually similar unreviewed subject as independently validated.
Fields to preserve
- Observation-to-image key
- Defines which source received the label.
- Initial versus expert assignment
- Preserves evidence stages.
- Review scope and uncertainty
- Keeps unreviewed observations unresolved.
Test another case
Give OBS-B a later expert label Leaf-Y. Require the model to retain separate classifications despite the shared timestamp and similar drawings.
Use original images or independently drawn examples. Exclude volunteers, precise locations, hidden geotags and sensitive wildlife coordinates.
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.