Leveraging Multispectral and Lidar Uav to Predict Individual Tree Health: A Case Study of Viscum Album in Scots Pine Forests

Abstract

The presence of mistletoe in pine stands has expanded in recent decades, currently threating Mediterranean forests. Mistletoe outbreaks can make the host trees more vulnerable to intense droughts, which are expected to increase due to climate change. We use multispectral (MS) and LiDAR UAV-derived data to determine Viscum album ssp. austriacum infestation levels at individual tree level in Scots pine (Pinus sylvestris L.) forests. First, spectral and structural differences between four infestation levels were assessed employing Kruskal-Wallis test and post hoc Dunn’s test for individual tree crowns. Second, machine learning classification algorithms were applied to evaluate infestation levels at the individual tree scale by comparing or combining UAV-derived datasets. Outcomes revealed significant differences between infestation levels in canopy cover and height based on LiDAR derived metrics. Significant changes in vegetation vigor were also found through spectral and textural metrics. Using two vegetation indices (CIRE and NDVI) an overall accuracy of 0.83 was achieved by applying SVM, while combining a spectral metric (NDRE) and a LiDAR metric (D0) resulted in 0.82 accuracy with SVM. Using only LiDAR variables, we obtained an accuracy of 0.64 with SVM and RF. This approach demonstrates their value for detecting and characterizing morphological changes in up to four levels of mistletoe infestation at individual trees in Mediterranean Scots pine forests, lending support to forest management monitoring.