Locating and identifying seed tree species for forest restoration in northern Thailand using an unmanned aerial vehicle

 

ABSTRACT: Rapid and reliable location of seed trees of required species, within forest, is essential, if global forest restoration targets are to be achieved, e.g. the Bonn Challenge (forest restored to 350 million ha by AD 2030). In dense forest, finding seed trees from the ground is laborious and inefficient, due to limited visibility and accessibility. In contrast, the use of quadcopters with high-resolution cameras, to view tree crowns from above, has become affordable and user-friendly. In this study, drone imagery, classical taxonomy (using leaf and crown characteristics) and image filtering were combined, to develop keys to distinguish 9 tree species, during monthly automated flights over regenerating evergreen forest in Chiang Mai Province, northern Thailand, from June 2018 to January 2019. Independent volunteer observers tested the keys’ reliability, using images from a second, similarly aged validation plot. Overall, identification accuracy exceeded 50% for seven of the target species and over 70% for four species. However, identifiability varied with season, with reliability peaking (often at 100%) for most species, during their most distinctive phenophases. Consequently, development and use of aerial tree-identification systems will depend on building up databases of tree species characteristics, visible from drones, and their seasonal variability.