Automated detection of boreal forest deadwood using RGB UAV imagery

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 it, with third of these species being red-listed. However, there is a severe lack of stand-level deadwood data in Finland and in boreal forest in general, as the operational inventories either focus on the large-scale estimates or omit deadwood altogether. Unoccupied Aerial Vehicles (UAVs) are the only method for remotely detection of small objects, such as fallen deadwood, as even the most spatially accurate commercial satellites and aerial photography provide 30 cm ground sampling distance, compared to less than 2 cm that is easily achievable with UAVs. Interesting question is that can high resolution UAV data replace reliably enough usually lacking deadwood field data? In this work, we utilized Mask R-CNN to detect individual standing and fallen deadwood instances from RGBUAV imagery. We manually annotated over 16 000 deadwood instances from three separate study sites to be used as the training and validation data, and also compared these data to field-measured deadwood data which consisted of 3200 standing and downed deadwood measured with RTK GPS. These three sites represents southern, middle and northern boreal forests. Our models achieved test set Average Precision (AP) of 0.341 for the same geographical area the models were trained on, and AP of 0.236 for geographically distinct area used only for testing. In addition to instance level deadwood maps, we also estimated stand-level deadwood characteristics, such as length and approximate total volume of fallen deadwood based on the annotated polygons. These stand-level features clearly show the borders of the conserved forest, and the volume estimations can distinguish the naturally formed deadwood hotspots from the areas with logging residue. The proposed method enables deadwood mapping for areas consisting of multiple forest compartments, thus complementing the traditional field work