THE APPLICATION OF REMOTE SENSING TO MONITOR TREE PESTS AND DISEASES IN WELSH FORESTS

Abstract

Exacerbated by international trade and climate change, exotic pests and diseases are an omnipresent challenge for tree and forest management. Timely and accurate monitoring of the extent and severity of outbreaks is needed in order to enable intelligent and effective responses. Recent technological advancements have served to garner interest in the use of remote sensing techniques for the surveillance of such outbreaks. One such disease is Phytophthora ramorum which has affected larch trees across large swathes of the United Kingdom. Here, we explore the potential for airborne hyperspectral imagery to be applied to the task of mapping its extent and severity in a 530-hectare area of Gwydyr Forest, Wales,
UK. A workflow is presented which combines data from a single aircraft flight campaign with field gathered training samples for input into machine learning classification algorithms. Three algorithms: random forest, support vector machine and maximum likelihood are compared for accuracy and efficiency. The spectral response of disease symptoms is analysed alongside an assessment of the hyperspectral variables most sensitive to the disease. By using the hyperspectral wavebands with the highest variations in reflectance intensity across disease severity classes, we present a novel vegetation index, Normalised Larch Disease Index (NDLI), capable of detecting subtle changes in the physical and chemical structure of foliage associated with P. ramorum in larch trees. We found disease severity classes to have particularly high separability at the peak of the red edge reflectance region (751 nm) as well as the middle of the short wave infra region (1984 nm) suggesting that spectral variation associated with changes in leaf structure and desiccation are important in disease detection. A variable importance model found NDLI to be superior to other traditional and established vegetation indices for the detection of P.ramorum in larch trees whilst other chlorophyll based indices were shown to outperform anthocyanin and lignin based spectral variables. The classification algorithm with the highest overall accuracy (81.49%) was random forest which produced a map of disease incidence and severity with potential for use in decision making and disease management. This research contributes to our understanding of how remote sensing technology can be applied to the surveillance of a specific tree disease. The workflow may be tailored and applied to other established and emerging forest pathogens.

Date of Award	16 Jan 2023
Original language	English
Awarding Institution	Bangor University
Supervisor	James Walmsley (Supervisor) & Norman Dandy (Supervisor)