Large-scale field collections of entomological specimens are necessary tools for evaluating continent- and global-scale trends in biodiversity and agroecology. However, the utility of these collections is limited by reliance on specialized and increasingly scarce taxonomic expertise. Ecdysis Foundation, a leading voice in the regenerative agriculture movement, performs broad-spectrum ecological monitoring on agricultural lands across North America for the ongoing 1000 Farms Initiative. This large experiment involves processing thousands of entomological samples (sweep nets and quadrats), and hundreds of thousands of individual arthropod specimens. To facilitate this high volume of data, we have developed an innovative software application called BugBox, which uses artificial intelligence to rapidly classify specimen images to the morphospecies level. The AI is currently trained on an internal database of ~30,000 images, representing approximately 300 morphospecies of insects, spiders and myriapods. BugBox is currently able to classify ~60% of images to the correct morphospecies. On a test set of samples collected from wheat fields in Washington state, diversity indices calculated from AI classifications showed a strong correlation with indices calculated from human identifications (R2 = 0.833 for Shannon’s H, R2 = 0.834 for Simpson’s (1-D)), and the same conclusions about on-farm treatment effects (regenerative vs conventional agricultural practices) were reached in both cases. This indicates that, while AI models cannot replace human expertise, classification data generated by BugBox’s AI can still be used to provide rapid and meaningful biodiversity assessments and enhance the ability of ecological research labs to collect and process large biodiversity datasets.
.png)
.jpg "Jonathan G. Lundgren, Dr. photo")