18 September 2023
In European forests, the bark beetle has been a major source of insect outbreaks. Over the coming decades, the costs of insect outbreaks are projected to increase as the higher temperatures and more frequent extreme weather events caused by climate change provide conditions for increased breeding. For example, in 2018, the volume of affected conifer trees in France was already much higher than during the previous beetle infestation in 2003-2007.
Although higher temperatures and drought increase bark beetle infestations on a large-scale, local damage is often patchy. So the mapping of the spatial spread of the bark beetle infestation is an important, challenging problem in both forest management and ecological research. Traditionally, this mapping is performed manually, requiring labour-intense and time-consuming excursions into forests – a process which is poorly scalable to wider areas where it becomes necessary to create sample plots and extrapolate the data.
On the other hand, the increasing availability of remote sensing technologies enables image classification to be used as a key tool to (partially) automate the mapping of forest health status and systematically reduce the amount of fieldwork. For example, Copernicus’ Sentinel-2 instruments allow the measurement of multispectral bands, corresponding to the pigment in the leaves, to the cell structure of the plants, and to their water content.
In their recent paper, SWIFTT partners from University of Bari Aldo Moro, explore how machine learning methods can be used to map bark beetle infestation hotspots in Sentinel-2 images by leveraging spectral vegetation indices and performing self-training.
Another contribution of this study is the use of an eXplainable Artificial Intelligence (XAI) technique to strengthen the mapping results for exploring the effect of both spectral bands and spectral vegetation indices on decisions concerning forest patches with healthy and dying trees.
Source: G. Andresini, A. Appice and D. Malerba, "SILVIA: An eXplainable Framework to Map Bark Beetle Infestation in Sentinel-2 Images," in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, doi: 10.1109/JSTARS.2023.3312521.
