Sentinel-2 images for detection of wind damage in forestry

Using of Remote sensing for the sake of Earth Observation is getting more and more popular as the number of satellites that are able to measure electromagnetic radiation with a higher spatial, temporal and radiometric resolution is considerably rising. Of all usage of Earth Observation, detection of disturbances caused by natural catastrophe such as wind, earthquake and fire is highly important. On 12th of August 2017, a storm hit South and South East of Finland, bringing harsh disturbances to the forest area in which Pine and Spruce were the main types of land cover. The study area in this region contained the extent of a sentinel-2 image that covered an area of 100 km by 100 km. Two sentinel-2 images from 11th of August 2017 and 5th of September 2017 were used to measure spectra behavior of existing features before and after storm in the region. Forest use notifications data, by which damaged stands were identified, and forest-stand dataset, with which stands that were not touched by the storm (undamaged stands) were characterized, were used as ground truth data. For change extraction, univariate image differencing was used using six different indices, namely EVI, NDVI, NDMI, SATVI, TCB, and TCG. Two main approaches were taken in this thesis, namely pixelwise and average based, where in the former individual pixels were extracted (from stands) and used for training the models while in the later average of pixels inside each stand was calculated and used for training. Results achieved by average-based showed a better performance in terms of user accuracy and stability of the results than pixelwise approach did