About GeoFauna

Standard web mapping usually relies on Web Mercator, which introduces extreme area distortion near the poles (making Greenland look as large as Africa).

GeoFauna adopts the Robinson Projection (formulated by Arthur H. Robinson in 1963 and adopted by the National Geographic Society from 1988 to 1998). It is a pseudocylindrical projection designed to visually balance area and angular distortions across the global terrestrial biosphere.

When a player submits a painted distribution mask, GeoFauna rasterizes the prediction onto a 1° × 1° geodesic sphere grid. To respect spherical geometry, each cell area at latitude φ is weighted by its true surface area cos(φ).

The final score is 85% spherical IoU and 15% a symmetric nearest-miss measure, computed from both overpainted and missed cells with a 2,500 km cap. Dice/F1, precision, recall, mean miss distance, and guess-size ratio are reported separately so players can see whether their error was too broad, too narrow, or geographically displaced. An exact mask earns 1,000 points and Rank S.

A compact endemic range and a continent-spanning one cannot fairly share one curve, so every species also carries a difficulty calibration computed at build time from its own truth mask: a baseline anchor (the best trivial guess — a bounding box, a filled continent, or a centroid disc) and an attainable anchor (the truth mask expanded by one fine-brush halo). Raw IoU is remapped through those two anchors before it becomes a score, so beating the best trivial guess lands around the middle of Rank C and reaching the attainable ceiling reaches Rank S — for every species, not just the large, easy-to-outline ones.

Scientific honesty matters more than a prestigious source list. The playable species masks in this release are occurrence-derived extents: built from GBIF occurrence density with a documented, reproducible pipeline, not hand-drawn boxes. They are useful for a broad biogeography game, but they are not IUCN range maps, official occurrence maps, estimates of occupied area, or projections of future range.

• GBIF (Global Biodiversity Information Facility): the primary evidence source for playable ranges. Each catalogue species is resolved to its exact GBIF backbone taxon, then its occurrence density is aggregated at 1° resolution — filtered server-side to exclude fossil and captive/cultivated records, weighted toward observations from 1970 onward, with an all-years fallback for sparsely recorded taxa.
• From density to a mask: cells are kept only above a presence threshold, trimmed of likely vagrant records, cleaned with morphological closing and small-blob pruning, and clipped to the species' land, ocean, or coastal realm. A small number of species carry a declarative editorial override — most often for an introduced population that occurrence density alone can't distinguish from a native one — and every override ships with a written justification note.
• Licence-filtered GBIF evidence snapshot: a separate, dated, reproducible 1° evidence footprint drawn only from CC0 and CC BY occurrence records — no CC BY-NC, no coordinate uncertainty above 25 km, no fossil or introduced-flagged records. This open-licence-only snapshot is what automated QA gates check every playable mask against before it ships, and every retained publisher dataset is exposed on the species page.
• iNaturalist: linked per species as a citizen-science observation registry. It is a valuable discovery and validation resource, not the source of the game's ground-truth mask.
• IUCN Red List: referenced only for extinction-risk categories, not for range data. Its assessor-produced distribution maps have their own terms of use and are not bundled here.
• Natural Earth 1:110m: the public-domain vector land and boundary layer rendered beneath the game.
• Wikimedia Commons: High-resolution wildlife photography under open Creative Commons licenses (CC BY, CC BY-SA, Public Domain).

GBIF's density aggregates cannot be filtered by licence on the server, so they do include some CC BY-NC records. GeoFauna's answer is disclosure, not omission: every species page states the percentage of its underlying records published under each licence, the playable output is a per-cell presence fact rather than a redistribution of any individual record, and every mask is independently verified against the open-licence-only evidence snapshot above. Where a fully licence-filtered mask is needed, a species can instead be rebuilt from a credentialed GBIF SQL download, which excludes non-open records server-side and produces a citable DOI stored in that species' provenance record.

Personal statistics — streaks, grades, history — live only in this browser's local storage. They are never sent anywhere and disappear if you clear your site data.

An optional, anonymous daily community histogram can show how today's expedition went for everyone. When enabled, it records nothing but a score-band counter for the day — no account, no device identifier, no per-player record of any kind. The game is fully playable, with no functionality lost, when this feature is disabled or not configured.