Using UAV images and deep learning to enhance the mapping of deadwood in boreal forests

Abstract

Deadwood and decaying wood are the most important components for the biodiversity of boreal forests, and around a quarter of all flora and fauna in Finnish forests depend on them, with third of these species being red-listed. However, there is a severe lack of stand-level deadwood data in Finland, as the operational inventories either focus on the large-scale estimates or omit deadwood altogether. Unoccupied Aerial Vehicles (UAVs) are a cost-effective method for remotely mapping small objects, such as fallen deadwood, over compartment-level areas, as even the most spatially accurate commercial satellites and aerial photography provide 30 cm ground sampling distance, compared to less than 5 cm that is easily achievable with UAVs.

In this study, we utilized YOLOv8 by Ultralytics for detecting and segmenting standing and fallen deadwood instances from RGB UAV imagery. Our study consists of two geographically distinct study areas in Finland, Hiidenportti National Park and Evo. We manually annotated around 13 800 deadwood instances to be used as the training and validation data for the instance segmentation models. These annotations were also compared to field-measured deadwood data from Hiidenportti to assess the extent on how much of the deadwood can even be seen from UAV imagery. We also compared how the models perform on another area than from which its training dataset was from, and whether adding data from another areas to the training dataset improves the performance compared to training only with images from one location.

The best performing model achieved test set mask mAP50 score of 0.682 for Hiidenportti and 0.651 for Sudenpesänkangas datasets. For both areas, including imagery from the another area improved the instance segmentation metrics, whereas using data only from another site to train the models produced significantly worse results. While methods utilizing UAV imagery cannot completely replace traditional field work, they should still be considered as an additional tool for field campaigns.