A dataset of 10 000 Sentinel 2 Level-1C RGB satellite images, 512x512 pixels (minimum) at 10 m / pixel resolution, centered at random locations in Finland over the years 2015–2022, collected for machine learning purposes. There is a machine-downloadable mirror of the dataset at: https://a3s.fi/swift/v1/AUTH_ef7e4759a26645089a38de9633f52afd/sentinel2-l1c-rgb-finland-10k/ Each image is large enough to contain a 5120 m x 5120 m square in WGS 84 / UTM zone 35N coordinate system. Individual images come in their native coordinate system (depending on the coordinate system of the original satellite image tile) which is either WGS84 / UTM zone 34N or 35N and have a minimum size of 512x512 pixels at 10 m / pixel resolution in both their native coordinate system and in WGS84 / UTM zone 35N. The format is netCDF following the CF-1.9 convention. Location and time information is included with the images. The data was downloaded from the openEO platform of Copernicus Data Space Ecosystem without screening against cloudy images. Images containing missing data (typically at original satellite image swath edges) have been removed and replaced with valid images. The used approximate geographical area of Finland is based on free vector data from Natural Earth. https://www.naturalearthdata.com/downloads/10m-cultural-vectors/10m-admin-0-countries/ The procedure to simplify the geometry in QGIS was: Select country, Keep N biggest parts (20 parts), Multipart to singleparts, Simplify (tolerance 0.01 deg), save GeoJson at a precision of 2 decimal places. Source code for building the dataset is available at https://github.com/hamk-uas/sentinel2-l1c-random-finland. Contains Copernicus Sentinel data 2015–2022.

2024