Rooftop-level photovoltaic segmentation for the Switzerland

The provided dataset contains data of photovoltaic installations on building rooftops. It provides the segment polygons and area as well as the tilt and orientation of the rooftop. It is tha basis for the pv production calculation published in the energiedashboard. To generating this dataset, 10cm resolution satellite images from swisstopo were used. A machine learning model was trained to segment the pv installations. The detected segments were then merged with the swissBUILDINGS3D dataset to get the tilt and orientation. Finally, the directory of buildings was mapped by coordinates to get the address and building identifier EGID. The dataset has following limitations: The mapping to the directory of buildings can be inaccurate. The satellite snapshot of swisstopo is not up-to-date. We indicate the image and year of the snapshot in the corresponding column The machine learning model can make mistakes (F1 score of 90% and intersection over union of 85%) Here is a quick column description: BDG_EGID: GWR building identifier address: Address from the mapped building address_easting: easting coordinate of the mapped building (LV95) address_northing: northing coordinate of the mapped building (LV95) district: district based on the building coordinate from swissBOUNDARIES tilt: tilt of the merged rooftop from swissbuildings3D (0°=horizontal, 90°=vertical) orientation: orientation of the merged rooftop from swissbuildings3D (0°=north, 90°=east, 180°=south, 270°=west) pv_area_m2: area of the polygon in m^2 geom: Polygon of the detected segment (LV95) snapshot_year: year the satellite image was taken satellite_image: link to the satellite image used for detection