Early-stage detection of bark beetle infested spruce forest stands using Sentinel-2 data and vegetation indices

Abstract: The European spruce bark beetle is an insect that is often referred to as a pest. Responsible for the destruction of over 150 million m3 of Norwegian spruce forest in Europe over the last 50 years makes this insect one of the major disturbances to the forest industry. With global warming at large, changes of the distribution of the bark beetle have emerged with large outbreaks now regularly occurring in northern Europe due to warmer and prolonged summer seasons. Sweden has since 2018 been affected by mass outbreaks that have destroyed over 27 million m3 of spruce forest. In order to mitigate the disturbance of forest and limit the spread of further attacks, it is important to detect and cut down infested trees at an early stage. In recent time, many studies have focused on early-stage bark beetle detection using remote sensing methods. Efforts to detect early-stage infestation, i.e., “green attacks” using vegetation indices (VI) on a pixel level, have found varying levels of success but have shown the potential of using VIs sensitivity of changes to biochemical leaf properties to detect early-stage infestation. Hence, the aim of this thesis was to study if the variability between pixels for a vegetation index on a forest stand scale change during a spruce bark beetle outbreak and to test if the variability can be used as an early indicator for bark beetle infestation. This was done by calculating the coefficient of variation between 2017 and the outbreak year 2018 from Sentinel-2 data for four different VIs (NDVI, NDWI, DRS, RDI) in 17 spruce forest stands where 10 stands were infested and 7 were healthy. The coefficient of variation was used to classify the stands into healthy and infested by computing the cumulative sum of each VI in each stand. The classification performance for each VI was evaluated using receiver operating characteristics graphs which were used to find the optimal threshold for each classification. The classification was done using the cumulative sum for two different timeframes, early stage (1st of May-1st of July 2018) and the whole bark beetle season (30th of April-30th of September) The results of the thesis, indicated that the variability within a forest stand does change during a bark beetle outbreak with increased variability over time within stands that have been attacked. It also showed that changes in variability have the potential to be used as an early indicator for bark beetle infestation, and the variability can be used to detect and classify individual forest stands that were infested at an early stage, i.e., green attack stage. It was found that NDWI was the most suitable index during the period May-July to detect infested forest stands. However, for the whole season NDVI and RDI also displayed potential as both were able to detect high rates of infested forest stands while limiting the misclassification of healthy ones. An infested forest stand could be detected as early as ca. the 29th of May. Meaning that the infestation is still in the green attack stage, in which mitigation is still possible to eliminate the spread of further bark beetle attacks.