GOES-East ABI Contrail Detection Dataset: 48 Days in 2024 (135°W-35°W, 0°N-50°N)

Sample dataset of 48 days of contrail detections in GOES-East ABI imagery, using a detection algorithm based on a U-NET CNN and Visual Transformer, and identifying individual instances through image segmentation and decomposition techniques. Domain: 135°W-35°W, 0°N-50°N Total Objects: 1.3 million Consists of two published components, with a third* component available upon request due to file size: (1.) A tabular contrail summary (.parquet), providing key information for each contrail instance: contrail_id: unique id for instance timestamp: GOES-East image timestamp originally_complex: True for instances obtained from a region of overlapping contrails through segmentation. False for simple standalone contrails centroid_lon, centroid_lat, endpoint1_lon, endpoint1_lat, endpoint2_lon, endpoint2_lat: location of centroid and endpoints centroid_scantime, endpoint1_scantime, endpoint2_scantime: estimated timestamps for when each point would be seen by the satellite given its scanning pattern (i.e., GOES-East image base time + scan time offset) orientation: orientation in degrees from north great_circle_length_km: length calculated as geodesic distance between endpoints n_pixels: number of pixels occupied in GOES-East image area_km2: area estimated from surface areas of occupied pixels width_km: average width estimated by dividing area_km2 by great_circle_length_km min_score, max_score, mean_score, std_score: algorithm detection statistics for occupied pixels contrail_centroids_all_dates.png shows plot of contrail centroids over the 48 day sample (2.) Individual contrail instances (.geojson) extracted from detection masks: contrail_id timestamp geometry: Multipoint latitude and longitude object (3*.) UPON REQUEST, Raw contrail detection masks: detection scores in three dimensions: GOES full disk frame time (every 10 minutes), GOES-East geostationary coordinates – x (units of metres), GOES-East geostationary coordinates – y (units of metres), In addition, carries key per-pixel metadata (e.g., lat-lon coordinates, scan time offsets from the image base time), metadata for the image itself (e.g., satellite source, satellite position, domain boundaries). Provided as a zarr archive per day, loadable via xarray.