GAF: a global alluvial fan map using deep learning

The Global Alluvial Fan (GAF) provides polygon boundaries for 608,354 mapped alluvial fans worldwide, including 759 fans manually delineated during the final quality review. The dataset was developed using the AFMask R-CNN instance segmentation model, integrating Landsat imagery and terrain information derived from ASTER GDEM. The dataset is organized into seven continent-specific ZIP archives: Africa, Antarctica, Asia, Europe, North America, Oceania, and South America. Each archive contains an Esri Shapefile in the WGS 84 geographic coordinate system (EPSG:4326). The attribute table includes unique identifiers, centroid latitude and longitude, model-derived confidence scores, prediction entropy, area, perimeter, elevation statistics, relief, mean slope, mean aspect, fan length and width, azimuth, sweep angle, concave-hull ratio, and climate classification. Model-derived confidence scores and prediction entropy may not be available for manually delineated fans.