PS 8-73 - Losing the needle in the haystack: Is increasing reliance on crowd-sourced plant observations missing rare species?

Background/Question/Methods

Crowdsourced biodiversity information is growing rapidly, with huge potential to provide low-cost data for research and conservation. However, image recognition tools used on these platforms are not fool-proof, causing a potential bias in the coverage of species associated with whether they are common or rare. In addition, bias could also be related to membership in certain taxonomic groups, since with traditional modes of identification, key diagnostic traits are compared under microscopes, and it is unlikely these traits are recorded by observers who upload their images to crowdsourced apps.
If substantial bias is present in the data-outputs of community science apps related to the collection and identification of natural history observations, knowing the nature of that bias would be informative for how best to treat and use that data. To this end, we chose as a case study all iNaturalist ‘Research Grade’ observations for vascular plants in Canada. We considered how membership in different taxonomic groups and legal status of species (a proxy for commonness) are associated with the difficulty in correctly identifying an observation to species. Furthermore, we examined how rarity might result in unequal coverage of species observations across the dataset.

Results/Conclusions

A comparison of families from iNaturalist ‘Research Grade’ data showed specimens identified only to genus belong to a wide range of vascular plant families, ranging from non-flowering plants to different monocot and eudicot families. In addition, we observed similar divergent results when observations were grouped based on their order membership. We also found that the representation of rare species in the iNaturalist ‘Research Grade’ observations was similarly from diverse taxonomic groups. Finally, our preliminary results indicate that the number of observations per species is less for species with conservation designations compared to species that are not of conservation concern, comprising less than 1% of observations. In summary, our study suggests that rare species are as equally represented as those that are not rare in iNaturalist ‘Research Grade’ observations, but there is a difference in the number of observations per species when comparing the two groups. Further work will focus on examining if plant identification apps preferentially identify common species, as well as the potential role of these tools in plant poaching.