Multi-temporal Pacific madrone leaf blight assessment with unoccupied aircraft systems

Abstract

Pacific madrone leaf blight (PMLB) is a contributing agent to the decline of Pacific madrone (Arbutus menziesii) trees. Multiple fungal pathogens cause PMLB, resulting in leaf spotting that can eventually kill leaves, increasing stress in individuals, and leaving them more susceptible to deadly cankers. Spores transmit via air and water droplets, particularly during wet Spring months. Unoccupied aircraft systems (UAS) technologies are in their relative infancy, but UAS are becoming more affordable and accessible. UAS promise increased efficiency in forest health monitoring applications, providing a safer aerial data collection method at a relatively-low cost when compared to occupied aircraft. In this study, we develop and present a UAS methodology to detect PMLB with a multispectral sensor. This methodology combines orthomosaic products derived from high-resolution (~4 cm) multirotor platform UAS multispectral imagery with machine learning and ground assessment of PMLB to classify visual presence of blight at the individual tree-level during multiple site revisits. The resulting model detected PMLB infection status of 29 field surveyed madrone trees with a kappa coefficient of , a balanced accuracy of 0.85, and a true positive rate of 0.92. The method presented here can be readily scaled to efficiently cover a much larger extent with a beyond-line-of-site capable UAS and minimal field sampling. The increased efficiency of this approach may be critical to characterizing PMLB in the near future as it is anticipated that PMLB prevalence will continue to increase as a result of climate change.