A continental-scale dataset of ground beetles with high-resolution images and validated morphological trait measurements
Scientific Data (Nature Portfolio), Data Descriptor
13,200+ ground beetles from 30 NEON sites, imaged and measured, so that trait extraction can be learned from images instead of calipers.
Global trait databases cover vertebrates and plants far better than invertebrates, which limits what ecologists can ask about diverse groups such as ground beetles. Body size alone links to dispersal, diet and response to warming, yet measuring it by hand does not scale.
We digitized specimens from the National Ecological Observatory Network across the continental United States and Hawaii, and added human measurements of elytra length and width. These measurements serve two purposes: they are trait data in their own right, and they are ground truth for training and testing automated trait extraction.
Data records
Hawaii-beetles
| Specimens | Pinned, NEON Pu'u Maka'ala site |
|---|---|
| Images | 162 group images, 1,614 individual crops at 6960 x 4640 px |
| Traits | 1,579 trait rows, 14 endemic species |
| License | CC BY 4.0, DOI 10.57967/hf/7272 |
2018-NEON-beetles
| Specimens | Ethanol-preserved, 2018 collections |
|---|---|
| Images | 577 group images, 11,654 individual specimens at 5568 x 3712 px |
| Traits | 39,064 elytra measurement records, segmentation splits |
| License | CC BY-SA 4.0, DOI 10.57967/hf/6890 |
Coverage
| Sites | 30 of 47 NEON terrestrial sites, in 14 of 20 ecoclimatic domains |
|---|---|
| Taxa | 79 species among vial specimens, led by Pterostichus, Carabus and Calathus; 14 Hawaiian species among pinned specimens |
| Resolution | Mean pixel density of 97 px/cm for vial images and 785 px/cm for pinned images |
| Code | Imageomics/carabidae_beetle_processing |
Validation
| Three annotators, vial images | Elytra length R² 0.94, RMSE 0.075 cm, on 234 images |
|---|---|
| Two observers, calipers | R² 0.94, RMSE 0.024 cm on pinned specimens |
| Image tool (TORAS) vs calipers | R² 0.977, RMSE 0.015 cm, bias 0.006 cm |
Elytra width for vial specimens is released with the data and kept out of the analyses, since vials often hide the landmarks.
What it is used for
- Species identification. Two of the four datasets in the fine-grained taxonomy benchmark come from this collection.
- Trait extraction. Paired images and measurements make it possible to train and test models that measure beetles automatically.
- Processing pipelines. The tray images drive BeetleFlow and the batch-imaging workflow for detection, segmentation and metadata validation.
Related work
From collection trays to AI-ready data: An operational framework for automated entomological specimen processing
Methods in Ecology and Evolution (under review)BeetleFlow: An Integrative Deep Learning Pipeline for Beetle Image Processing
NeurIPS 2025 Workshop on ImageomicsOptimizing Image Capture for Computer Vision-Powered Taxonomic Identification and Trait Recognition of Biodiversity Specimens
Methods in Ecology and Evolution, 2025
Cite
@article{rayeed2026continental,
title = {A continental-scale dataset of ground beetles with high-resolution images and validated morphological trait measurements},
author = {Rayeed, S. M. and Khurana, Mridul and East, Alyson and Fluck, Isadora E. and Campolongo, Elizabeth G. and Stevens, Samuel and Zarubiieva, Iuliia and Lowe, Scott C. and Denslow, Michael W. and Donoso, Evan D. and Wu, Jiaman and Ramirez, Michelle and Baiser, Benjamin and Stewart, Charles V. and Mabee, Paula and Berger-Wolf, Tanya and Karpatne, Anuj and Lapp, Hilmar and Guralnick, Robert P. and Taylor, Graham W. and Record, Sydne},
journal = {Scientific Data},
year = {2026},
doi = {10.1038/s41597-026-08120-3}
}