20 m resolution geotifs derived from Ground Data The ground data were collected through using a combination of the following sources, using a broad search that stretched beyond our Areas of Interest: ● Habitat Map of Scotland (ground polygons)1 ● 2018 National Forest Inventory2 ● Ordnance Survey3 ● Global Forest Change v1.64 ● High resolution imagery5 In all cases the ground data were not used naively: we used a careful combination of at least two data sources to create each polygon, and checking against recent high resolution imagery to ensure each polygon was ‘pure’ (i.e. included only one class) and up to date (for example, if it was a forest polygon, the trees had not been cleared since the data were collected). Satellite remote sensing datasets used for mapping Optical Sentinel 2 (S2) (30/03/2019-10/11/2019) Radar Sentinel-1, descending and ascending (01/01/2019-31/12/2019) ALOS-PALSAR 2, 2018 annual composite Topography Shuttle Radar Topography Mission (SRTM, 2000) Process Extensive training datasets, and derived features from remote sensing data, to implement a complex set of tuned machine learning algorithms to produce a Prediction Model, and ultimately a prediction of a class for each pixel. Through the project duration the sophistication of the models used increased, increasing accuracy and efficiency. For commercial reasons the details of the final algorithms used will not be revealed here. EUNIS attribute information: Classification code O is not an official category from EUNIS classification. In the report "Bare ground O" is listed as including felled woodland, sediments, sand dune and unvegetated rock cliffs, ledges, shores and islets. Report: InIT - Space Intelligence AI for Good Phase 2 - Final Mapping Report - 26 Mar 2021_compressed (A3423939)