Days with the storm winds had taken place in the beginning of January 2019. Therefore, a time series of Sentinel-1 images from December 2018 and January 2019 was used. Open forest data sources were also utilized as an auxiliary source of information, as were the forest use declarations of the area. The objective of the exercise was to build an approach to detect the damaged areas and to further guide operational forestry and forest harvesting.
Challenges of the interpretation work stem from the very nature of SAR data. SAR image observations include random scattering (speckle) that can lower the usability of image material, as do rain or snow conditions of the set time period. The time window for the image data can also show other changes that have come to pass in the region, so it cannot be too long to avoid appearance of, for example, conventional harvesting operations in the images.
In its core, detecting the wind damages is a classification task, where the objective is to separate the damaged observations (field plots or forest stands) from the undamaged ones. When building a classifier, the common steps include training material selection, model estimation – machine learning algorithms offer some alternatives – and model testing. After the classification model has been established, it is possible to apply it, that is, to run the classification for the target observations (forest stands or grid cells) covered by similar SAR image data.