American Solar Farms (tech.marksblogg.com)

Michigan State researcher Jake Stid and a team of researchers (including alumni from NOAA, NASA and USGS) published GM‑SEUS: a ground‑mounted solar inventory covering 49 US states that catalogs ~15,017 solar arrays and roughly 2.9 million individual panels. The release includes two linked products (arrays and panels) with per‑asset attributes such as capacity (capMW), tilt, avgAzimuth, module type, mount, installation year (1985–2024) and geometry. For AI/ML and remote‑sensing teams this is a high‑value labeled dataset for tasks like object detection/segmentation, panel‑level capacity estimation, siting and degradation studies, and integrating with satellite time series for yield forecasting or grid‑planning research. The authors provide analysis‑ready Parquet versions derived from original GeoPackage files, carefully reprojected (Albers equal‑area projection -> EPSG:4326), spatially sorted with Hilbert encoding, and encoded as ZSTD‑compressed Parquet (row group 15k) with geometries stored as WKB and bbox fields for efficient cloud use. The panels GPKG shrank from 1.1 GB to a 334 MB Parquet and arrays from 108 MB to 37 MB after cleaning (-9999 → NULL) and dropping unused Z. Tools and workflows used include GDAL, QGIS, DuckDB (with h3, spatial and parquet extensions), H3 tiling for heatmaps and common GIS stacks; note some tooling quirks (ArcGIS Pro Parquet issues). Overall, GM‑SEUS provides a compact, well‑structured resource that lowers the barrier for scalable spatial ML and energy systems research.