Development of a forest clear-cut reference database for Continental Portugal at Sentinel-2 Resolution

Given Portugal’s initiative to establish a land cover monitoring system delivering annual products for its mainland territory, it becomes crucial to ensure an adequate representation of areas experiencing vegetation loss. The problem at hand consists of creating and updating, using Sentinel-2 imagery, a vegetation loss mask. Towards that end, the study aims to build a robust forest clear-cut reference database that will serve as a training data for a change detection algorithm, automating the creation of these maps. The final product will be a vectorbased land cover loss map, focusing on forest clear-cuts. The dataset utilized in this project was provided by the Direção Geral do Território (DGT) in collaboration with The Navigator Company (NVG). The data from the NVG was assumed to be incomplete due to their uncertainty on clear-cut dates as well as the assumption of the homogeneity of clear-cut dates for the sub-parcels not being true. Sentinel-2 time series data was integrated to estimate clear-cut dates, based on the NDVI. As the first approach at the sub-parcel level did not return accurate results, a pixel-level solution was explored, leading to a development of a semi-automatic method to estimated clear-cut dates at Sentinel-2 resolution. The pixel-based approach, utilizing the pyCCD tool, allows for a finer analysis by detection changes at pixel-level. A key aspect of this project was the visual analysis of the CCD algorithm's results by comparing them with Sentinel-2 time series, highlighting the importance of semi-automatic methods. The results obtained for Tile 29SNB indicate that this methodology can be successfully applied to other tiles in Portugal. Once complete for all tiles, this revised reference database will be the best and first comprehensive reference database for forest clear-cuts in Continental Portugal.