Comparing field measurements and annotations as training data for UAV-based detection of standing dead trees in coniferous forest

Deadwood provides a habitat for large number of species and acts as significant carbon storage. Thus, deadwood mapping allows planning various actions related to biodiversity conservation forest management. Airborne laser scanning (ALS) an efficient means monitoring forested areas is thus potential method mapping. Machine learning (ML) can be used automating detection from ALS data. However, ML requires training data, collecting which laborious using field measurements. This study inspected the feasibility data manually annotated aerial images individual standing dead trees. The trees were mapped dataset collected with unmanned vehicle (UAV). was carried out by performing ML-based tree field-measured comparing performances models trained on these two datasets. found that improves performance due its higher availability compared For (height > 14 m), precision, recall, Cohen’s kappa score, Matthews correlation coefficient achieved best classifier 0.23, 0.48, 0.17, 0.20, respectively. In comparison, corresponding metrics 0.14, 0.52, 0.04, 0.13. Annotated not representative sample true population, small without crowns (snags) often identified images. identifying such challenging even when does bias results. general, results showed ALS-based should focus