More information
Description
The ongoing InCoNaDa project is funded by the Norway Grants via the National Centre for Research and Development’, programme ‘Applied Research’, the POLNOR 2019 Call. The proposed Action is complementary with the ongoing InCoNaDa project. As part of the project the research team has developed the methodology for the land cover mapping and land cover change detection based on a time series of Sentinel-2 data. The algorithms were tested over the study area in Poland and Norway. With the proposed Action the algorithms will be verified over the selected regions in Ukraine, where many changes have
Summary of the results
The goal of the Action was to determine the extent of the forests, forest types, and to detect forest changes. The forest cover and forest type classification were performed for each Sentinel-2 granule individually for the Kyiv, Lviv and Zhytomyr regions. During independent verification, the accuracy of the final products was determined. In total, the overall accuracy (OA) for forest cover classification was more than 96% and Kappa coefficient was above 92%.The next classification of types of forest was made inside the forest range. The accuracy of the broadleaved and coniferous forests for different regions of Ukraine was even higher, with more than 98% and 94% of user’s (UA) and producer’s (PA) accuracy, respectively. The next objective was to detect forest changes on annual basis from 2020 to 2022. The overall accuracy of the change detection model was 98% OA and 97% Kappa for three Tree Cover Change (TCC) classes: 1 – no change, 2 – wood to non-wood, and 3 – burnt area. Independent verification was also carried out. The best accuracy was for class 1 – No change, from 99% to 100% UA. PA for this class was lower, from 77% to 86%.The lowest accuracy was for the class “3 – Burnt area” ranged from 9% to 93% for UA and from 80% to 100% for PA. The reason for very low UA for this class is similarity in spectral range between burnt areas and clear-cut or forest shadows. Moreover, the differences depend on the type of forests susceptible to fire occurring in a particular region.