Biodiversity & species data

GBIF holds billions of occurrence records: a specimen in a drawer, a photograph on a phone, a detection on a camera trap, all in one schema. It is an extraordinary resource and it answers a narrower question than it appears to, because every record is a presence and there is no record anywhere that says "I looked here and found nothing".

That single missing field shapes everything downstream. Absence has to be assumed, so distribution models compare the records against a background the analyst chooses — and that choice moves the result more than the algorithm does. Richness per cell tracks the number of records almost perfectly, so a richness map is an effort map until it is corrected. Coordinates carry an uncertainty radius that is missing a third of the time and, where present, is often a round number of miles. And a convex hull over uncleaned records can cover a quarter of the planet because one specimen was catalogued at the museum that holds it.

The concepts below set out what a record is, how it is biased and what the metrics measure; the how-to guides download, clean, thin, grid, rarefy, model, map and join real records against real climate and protected-area layers; and the fixes are the five ways these analyses go wrong in public.

Learning paths run through this topic too — see the end of this page.