Manure Spotter GeoPackage training files

This Zenodo record contains the vector training/validation polygons used by the manure_spotter workflow to train and evaluate a Random Forest classifier for detecting manure applications on snow-covered croplands from optical satellite imagery (Sentinel-2 and Landsat 8/9). What’s included (GeoPackage format, WGS84 / EPSG:4326, MultiPolygons): Sentinel-2 (S2) layers s2_sat_manure.gpkg: polygons interpreted as manure-on-snow from S2 imagery (id, use_date) s2_non_manure.gpkg: negative samples (non-manure) interpreted from S2 imagery(use_date) s2_groundtruth.gpkg: ground-verified labels (type = manure or fp, verif_date, id, use_date) Landsat 8/9 (L89) layers l89_sat_manure.gpkg: polygons interpreted as manure-on-snow from Landsat 8/9 imagery (id, use_date) l89_non_manure.gpkg: negative samples (non-manure) interpreted from Landsat 8/9 imagery (use_date) l89_groundtruth.gpkg: ground-verified labels (type = manure or fp, verif_date, id, use_date) Provenance and labeling: Polygons were delineated and curated by the dataset creator via manual interpretation. A subset of candidate detections were ground-verified by a volunteer team coordinated by the creator and are provided in the *_groundtruth.gpkg layers; fp denotes a false positive (a candidate verified as not manure). Spatial/temporal coverage: Wisconsin (USA) (approx. lon −92.79 to −87.26, lat 42.61 to 46.14). Imagery “use” dates span January–March 2025 (use_date); verification dates span January–March 2025 (verif_date). How to use: See the accompanying notebook manure_spotter_public.ipynb for end-to-end details on reading these GeoPackages and using them as training/validation inputs for the manure_spotter model workflow.