Detecting Spruce Bark Beetle Infestations with Satellite Imagery

Abstract: Sveaskog is Swedens largest forest owner, owning 14 percent of the Swedish forest lands. Recently, due to warmer and drier summers as a consequence of climate change, spruce bark beetles have caused damages at a massive scale. In 2019 Sogeti developed a promising first product for monitoring the vitality of large areas through the use of Sentinel-2 data by comparing images from the same month between two years, and the results from this first product where promising. To take the detection of bark beetle infestations to the next stage of development, supervised learning was used. Models where trained with a time-series of Sentinel-2 data in conjunction with national landcover data, ground level humidity data, and height data to predict segmentations that represented infestations. Target segmentations where created by clustering GPS points and tresholding a rasterized representation of the generated clusters. In total four different model architectures where tested (LSTM, GRU, 3D convolutional, and logistic regression) and then evaluated both quantitatively and qualitatively on a test set. It was found that the GRU based model was best able to identify bark beetle infestations.