Agent Guidance, Natural Earth Countries (1:110m)

World country polygons for joins and rollups, not for legal boundaries or precise areas. The stable key is ADM0_A3. External statistics join on ISO_A3_EH or ISO_A2_EH. The raw ISO_A3 and ISO_A2 columns hold the sentinel -99 for France, Norway, Kosovo, Northern Cyprus, and Somaliland, so a filter like ISO_A3 = 'FRA' matches nothing.

Query the GeoParquet data asset in place with DuckDB spatial, read_parquet('natural-earth-countries.parquet'), or load it with GeoPandas. It streams over HTTP range requests, so query the published URL directly rather than downloading first. For a quick preview use the thumbnail asset.

Quirks

• -99 is the global no-data sentinel, in the ISO columns and in numeric columns like TINY and MAPCOLOR13. Check for it before trusting a value.
• France includes French Guiana as one row, so its bbox spans from South America to Europe and bbox filters on South America catch France.
• South Sudan is SDS here but SSD in the populated places file. An ADM0_A3 join between the two layers drops it unless you remap.
• Antarctica is a country row with population 4,490, and the continent column also holds Seven seas (open ocean) for the French Southern and Antarctic Lands.
• Russia and Fiji cross the antimeridian, so their bboxes span the full longitude range and naive bbox-width logic breaks.

CRS, Degrees In, Degrees Out

EPSG:4326, longitude and latitude in degrees. ST_Area(geom) returns square degrees, which makes Greenland read nearly as large as Brazil. For real areas use the spheroid function, and flip coordinates first because DuckDB's geodesic functions expect latitude first.

Tested Queries

Population by continent.

Which country contains a point.

Related Collections

reference/natural-earth-populated-places joins on ADM0_A3 with the South Sudan remap above. tabular/eurostat-electricity-prices joins on ISO_A2_EH after remapping Eurostat's EL and UK codes, the worked example lives in that Collection's README. Converted from the upstream zipped Shapefile with DuckDB spatial into web-optimized GeoParquet 2.0.