3.4 Million Solar Panels

The 3.4 GB ZIP of the US solar spatial dataset GM-SEUS v2 was downloaded and converted from GeoPackage → Parquet for analysis. Coordinates were normalized to EPSG:4326 for Rooftop arrays 5,822, arrays 18,980, and panels 3,429,157, with bbox and WKB geometry stored together. The Rooftop arrays conversion succeeded only on DuckDB v1.4.4 and threw an exception on v1.5.1. Storage settings were ZSTD, COMPRESSION_LEVEL 22, ROW_GROUP_SIZE 15000.

• Rooftop arrays: Distribution was compared using an H3 level 4 heatmap and a representative source map. Coverage and boundary shapes differed by source, producing different results even in the same region.

• modType was concentrated almost entirely on c-si, and installation years ranged from 2003 to 2025. 2018 had the most with 225 records.

• The top sources were OSM 2,175, CECSFC 1,835, TZSAM 1,024, and USPVDB 485.

• Panels: The arrayID NULL rate was 0.03%, with about 12,653 unique values, pnlSource had 5 values, and Source had 12 values.

• Geometric features such as rowArea, rowAzimuth, rowLength, and rowWidth were also aggregated to check the distribution across the large-scale panel records.

• Visualization revealed an exception where the mirror structures of the Ivanpah Solar Power Facility were captured as panels.

• Arrays: Source-wise distribution was viewed with an H3 level 4 heatmap, and even nearby solar complexes had different detection sources.

• The top sources were OSM 5,222, USPVDB 4,024, TZSAM 3,278, and CECSFC 2,288.

• In the yearly capacity statistics, 2023 had the most with 2,017 records, with an average of AC 34 / DC 44 and a maximum of AC 1,095 / DC 1,423. In 2025, the maximum rose to AC 1,129 / DC 1,468.