Can AI Unlock the Hidden Data in Biological Collections?

What’s the question? How can AI-enabled image analysis turn physical specimens into large-scale trait data for ecological research?

The big picture: Biological collections hold vast amounts of information about how organisms vary across space, time and environments, but much of that information has been difficult to use at scale because extracting traits from physical specimens is slow and labor-intensive. AI-enabled image analysis can help unlock that information, turning carefully curated specimens into large, reusable trait datasets that make new kinds of biodiversity and ecological research possible.

NEON Biorepository samples used:

Researchers used carabid (ground beetle) samples from the NEON Biorepository, including:

• Pinned carabid specimens
• Ethanol-preserved carabid specimens from 2018 pitfall-trap collections

The pinned specimens included beetles from all of NEON’s 47 terrestrial sites, whereas the ethanol-preserved specimens represented 30 NEON sites, all together covering 518 species.

Researcher(s): Alyson East (under the supervision of PhD advisor Sydne Record at the University of Maine), Nathan Cain, S.M. Rayeed, Samuel Stevens, Tanya Berger-Wolf, Charles Stewart. Special thanks to Leah Cotton, Jacqueline Dominguez, and Michael Belitz for their support in preparing and photographing specimens at the NEON Biorepository and to Isadora Fluck and Benjamin Baiser for their support in photographing ethanol-preserved specimens. Special thanks also to the NSF funded Imageomics Institute for hosting Alyson East and Sydne Record to enable this interdisciplinary collaboration.

What they did and what they learned:

East and her collaborators developed a workflow to turn physical ground beetle specimens into analysis-ready trait data. They first established standardized methods for photographing specimens so the resulting images would work for computer vision, including consistent positioning, lighting, scale references, and metadata. The team then photographed NEON ground beetle specimens and used computer vision to detect individual beetles within group images, crop them into separate images and link each specimen back to its NEON data record.

Alyson East and Sydne Record hold up beetle collection trays at the NEON Biorepository. Photo courtesy of Isa Betancourt.

From there, the researchers used image-analysis models to measure elytra (wing cover) length, a reliable proxy for overall body size. Because every measurement remains associated with information about the specimen’s species name and collection location, including NEON site and plot, the team can compare body-size patterns across species and environmental gradients. The initial dataset paper includes more than 13,200 digitized specimens and validates image-based measurements against physical caliper measurements. The team has since expanded the number of images and automated trait measurements; the dataset now includes roughly 80,000 beetle measurements, to be published early next year.

“We’ve never been able to measure invertebrate traits at this scale before. These methods and this interdisciplinary collaboration open up new and novel data streams that allow us to answer foundational questions where we’ve been hampered by limited data.” – Alyson East, The University of Maine  

Those measurements are beginning to reveal patterns that would have been difficult to examine at this spatial extent and across so many individuals before. Beetle communities span a wide range of body sizes, with substantial overlap among some species. Within individual species, average body size can also shift among NEON sites and even among plots within the same site, raising questions about the relative influence of environmental conditions and competition on local trait distributions.

One of the clearest patterns so far is geographic: average beetle body size tends to be larger at lower latitudes and smaller farther north. This “inverse Bergmann’s” pattern runs opposite to the classic trend seen in many warm-blooded animals. East is still analyzing what drives these patterns, including the roles of growing-season constraints, environmental filtering, and competition. The larger goal is not simply to measure beetles faster, but to create the volume of standardized trait data needed to test long-standing ecological theories at a continental-scale.

Read more:  

• From collection trays to AI-ready data: An operational framework for automated batch entomological specimen processing (preprint: ARPHA, in review Biodiversity Data Journal)
• A continental-scale dataset of ground beetles with high-resolution images and validated morphological trait measurements (preprint: arXiv, accepted in Nature Scientific Data)
• BeetleFlow: An integrative deep learning pipeline for beetle image processing (preprint: arXiv)

This work was recognized with the Best NEON Presentation Award at the 2026 Ecological Society of America (ESA) Annual Meeting for the presentation, “A Use Case of AI-Enabled Macroecology: Unlocking Ecological Insights from Natural History Collections for Continental-Scale Trait-Based Research”