CWDBench: A Benchmark for Instance-Level Characterization of Coarse Woody Debris in Aerial Imagery

Coarse woody debris (CWD) is a key indicator of forest disturbance, carbon storage, and wildfire fuel loading. It has high relevance following extreme events, but traditional methods relying on manual CWD assessment on the ground are labor-intensive and not scalable. While the combination of unmanned aerial vehicle (UAV) and AI-enabled computer vision offer a promising alternative, there is a lack of public datasets to support development and benchmarking of frameworks for instance-level detection of fallen logs in aerial imagery. We introduce CWDBench, the first public benchmark for fallen-log detection in centimeter-resolution UAV orthomosaics, containing 10,399 human-verified instances annotated with both oriented bounding boxes and segmentation masks. Imagery are sourced from dedicated UAV surveys (2026) over two forest areas affected by Hurricane Helene (2024), providing co-registered RGB orthomosaics with LiDAR- and photogrammetry-derived surface, terrain, and canopy height models (CHM). Raw data available under: Cushman, K., Krassovski, M., Ambrozio Dias, P., Branoff, B., Ferrier, M., Kumar, J., Murphy, B., Rathore, S., Regmi, P., Santos, F. (2026). UAS imagery and lidar data (Levels 0 and 1) over forested sites affected by Hurricane Helene, April 2026. Oak Ridge National Laboratory. https://doi.org/10.13139/ORNLNCCS/17354438