Agent Guidance, Boston Open Space

Boston's official open space inventory, valuable and messy in the way municipal GIS data actually is. No stable key, free-text code columns, and sentinel strings. The quirks below are each verified against the data and each prevents a silently wrong answer.

Query the GeoParquet data asset in place with DuckDB spatial, read_parquet('boston-open-space.parquet'), or load it with GeoPandas. It streams over HTTP range requests, so query the published URL directly rather than downloading first. For rendering use the visual PMTiles asset with its MapLibre styles.

Quirks

• PROTECTION is a slash-separated free-text list with typos, Ch91 and Ch 91, CATMit and CAT Mit. Match with LIKE '%A97%', never equality. 453 sites match A97, Article 97 constitutional protection.
• Null OS_Mngmnt means the steward in OS_Own_Jur manages the site itself. It is a meaningful value, not missing data.
• Sentinel strings pollute several columns, None, NULL, n/a, and angle bracket Null variants in OS_Own_Jur and AgncyJuris, plus one lowercase South end in DISTRICT.
• TypeLong has spelling variants, an and for ampersand form and a trailing-space form. Normalize with trim and replace before GROUP BY, and expect TYPECODE to contradict it on about 13 rows.
• Nine real sites carry ACRES = 0, so WHERE ACRES > 0 drops them. Fall back to geometry area for those.
• DISTRICT is the open space plan's districting, not official Boston neighborhoods, and four large linear parks are assigned the literal value Multi-District rather than a place.

Coordinate Reference System

EPSG:4326, WGS 84, a geographic coordinate reference system whose coordinates are in degrees. Planar distance and area functions return degrees and square degrees, which are not ground units and vary with latitude. For real distances and areas use a sphere or spheroid function, or transform to a projected CRS first. The data asset carries the same code as proj:code.

Area, Use the ACRES Column

The official ACRES column matches geodesic geometry area to a median of 0.003 acres, use it. If you must compute, set geometry_always_xy first, because DuckDB reads the first coordinate as latitude by default and that shrinks Boston by more than half.

Eighteen sites disagree with their geometry by more than 10 percent, documented cases like Condor Street Overlook where ACRES excludes mudflats the polygon includes.

Tested Queries

Acreage by normalized type.

Which open space contains a point. Boston Common from its center.

Protected share by district.

Related Collections

boundaries/us-counties contains Boston at a far coarser scale, and a spatial join is the only bridge. The upstream layer on Analyze Boston is continuously maintained and now carries columns this July 2026 snapshot predates, OS_ID, ZipCode, ParcelNumber, YearAcquired. Converted from the city's Shapefile export with DuckDB spatial into web-optimized GeoParquet 2.0.