The iNaturalist Geomodel

I'm pretty excited about this.

Until yesterday, iNaturalist, a platform for recording and identifying observations of organisms & citizen science, used a "Seen Nearby" rubric when selecting which predictions from its computer vision to show to a user:

> What does "Seen Nearby" mean?

> The “Seen Nearby” label on the computer vision suggestions indicates that there is a Research Grade observation, or an observation that would be research grade if it wasn't captive, of that taxon that is:

> - within nine 1-degree grid cells in around the observation's coordinates and

> - observed around that time of year (in a three calendar month range, in any year).

https://web.archive.org/web/20230919033734/https://www.inatu...

I'd discussed how it then worked in a thread seven days ago with another user who was using the CV model separately from iNaturalist: https://news.ycombinator.com/item?id=37515996

Prediction maps for all of the species included are available on the iNaturalist website using the associated taxon number,

e.g. https://www.inaturalist.org/geo_model/63311/explain

where 63311 = Sagebrush Mariposa Lily Calochortus macrocarpus)

cf. https://www.inaturalist.org/taxa/63311-Calochortus-macrocarp...

The update changed this to a model which tries to use the often sparse data to estimate where a given species (or taxon) is expected to be observed, going beyond simply having adjacent observations. The announcement & ensuing discussion in the comments does a thorough job of explaining how this works. There are also two relevant papers posted to arXiv by iNaturalist's research collaborators that explain how the new model works:

Spatial Implicit Neural Representations for Global-Scale Species Mapping (2023) https://arxiv.org/abs/2306.02564

Presence-Only Geographical Priors for Fine-Grained Image Classification (2019) https://arxiv.org/abs/1906.05272