Forest disease dataset for France

This is a publicly available dataset for the problem of detecting forest damaged territory for semantic segmentation and for the use of machine learning algorithms within the Horizon SWIFTT project The study area located in the northwest of the territory of France and consists of 92 locations in the form of rectangles. The created data set consists of 92 separate directories, which are named with a corresponding unique identifier. Each such directory contains the next data: 1) Mask of disease forest in EPSG:3857 with 10 meters resolution using GDAL python library. Mask names for each location (geojson) have the following format: geojson_NN_mask.tif, where NN – the unique identifier of location. The mask contains 2 classes: 0 – non damaged (black), 1 – damaged (white). Each location has only one mask that corresponds to the date of forest damage identification (Féret et all, 2022). This date is indicated in the vector file with locations. 2) Forest type map 2018 (Giannetti et all, 2018) and 2022 (Kussul et all, 2022) in EPSG:3857 with 10 meters resolution for each location contains 4 classes: 0 – non forest, 1 – broadleaved (yellow), 2 – coniferous (green), 3 - mixed and have the following name-format - geojson_NN_forest_type_YYYY.tif, where YYYY – corresponding year. 3) Time series of cloud free available satellite data Sentinel-2 with all available bands (B1 – B12) in EPSG:3857 with 10 meters resolution from 2017 to 2021. To unify the data set, bands with a spatial resolution of more than 10 meters were warped to 10 meters. Satellite image names have the following format: geojson_NN-YYYYs-MMs-DDs-YYYYe-MMe-DDe-S2.tif where the first date (YYYYs-MMs-DDs) corresponds to start of period, the second (YYYYe-MMe-DDe) – the end of period within which the satellite composite was created.