TriMet Geospatial Data — agent guide

A cloud-native mirror of the eight geospatial layers TriMet publishes at developer.trimet.org/gis. Every collection is a single GeoParquet file plus PMTiles, in WGS84. Total 15,232 features; the whole catalog is under 5 MB, so nothing here needs partitioning or a query engine beyond DuckDB.

The collections, and how they connect

The joins that work:

• stops ↔ route-stops on stop_id. Exact: all 6,316 stops appear in both, and stop_name never disagrees. This is the join that answers "which routes serve this stop".
• route-stops ↔ routes on (rte, dir). Use it to get public_rte, the rider-facing route name, which route-stops lacks.
• rail-lines ↔ rail-stops on line, a shared code space (not a key — it is many-to-many, and it describes which services run there).

The join that does not exist: rail-stops carries no stop_id, so it cannot be joined to stops or route-stops by key. Match spatially and check the distance; the rail layers are generalized.

There is a data browser at https://cholmes.github.io/trimet-data-browser — map preview, schema, license and direct downloads — if you need to show a human what a query returned, or point them at a file.

Access

Every file streams over HTTP range requests. All are Hilbert-ordered with a GeoParquet 1.1 bbox covering column, so filtering on bbox.* prunes row groups before any geometry is read. S3 URIs are in each collection.json under assets.data.alternate.s3.

Coordinate system, and what follows from it

The GeoParquet is in EPSG:2913 — NAD83(HARN) / Oregon North, international feet — which is what TriMet surveys and publishes in. It is deliberately not reprojected. Only the PMTiles are in WGS84, because vector tiles are Web Mercator by definition.

The practical consequence is a good one: measurements just work, in feet.

Useful conversions: × 0.3048 for metres, ÷ 5280 for miles, ÷ 43560 for acres, ÷ 27878400 for square miles. 400 m is 1312.34 ft, a quarter mile is 1320 ft.

Two things to watch:

• Coordinates are not longitude and latitude. An x of 7633099 is feet east of the projection's false origin, not a degree. Anything expecting lon/lat — a web map, a geocoder, most GeoJSON consumers — needs a transform first.
• always_xy := true when you do transform. Without it DuckDB honours EPSG:4326's authority-declared latitude-first axis order and every result comes back Infinity:

The bbox covering column is in the same feet, so a spatial filter written against it uses projected coordinates, not degrees. The collection's STAC extent stays in WGS84, because STAC requires that regardless of the data's own CRS.

Nothing here carries a pre-computed length or area except the district boundary, which has TriMet's own area_sq_mi and acres.

Catalog-wide caveats

These are alignments and locations, not schedules

Nothing in this catalog is time-aware. There are no headways, no trip times, no service calendars. The frequent flag is the only service-level hint and it is a boolean-as-string, not a frequency. For anything temporal use TriMet's GTFS feed, which is also where the route colors used in these styles come from.

The rail layers are drawn, not measured

rail-lines and rail-stops are, in TriMet's words, "generalized to improve cartographic display at smaller scales". Do not compute track length or platform position from them. routes carries the operational alignment.

Vintages differ by more than a decade

district-boundary was last updated 2013-01-09; stops and route-stops 2026-07-30. Each collection's collection.json carries its own source date in extent.temporal. Do not present the catalog as a single-date snapshot.

Documented code values that do not appear, and undocumented ones that do

• status documents Planned and UC everywhere, but every row in every collection is currently Existing.
• rail-lines and rail-stops document a code BL/NS; the data contains NS/BL. Filter on what the data has.
• route-stops carries a frequent column that TriMet's metadata page for that layer does not document.
• stops has a type value BSC (shared bus and streetcar) that does not occur in routes or route-stops.

Counts of "stops" depend on which layer you ask

6,316 physical stops (stops), 8,314 stop-route-direction rows (route-stops), 169 rail stations (rail-stops, which collapses directional platforms) against 225 rail-typed rows in stops. All are correct answers to different questions. State which you used.

Visualization

Each collection has a styles/ directory with three to five MapLibre GL styles; default.json is the one to use unless you have a reason otherwise. Discover them from collection.json by filtering assets on roles containing style.

The rail and route styles reproduce TriMet's own published cartography rather than inventing a palette — the source SLD and MapLibre style are mirrored into the collections as assets with roles ["style", "source"], so a style can be diffed against its origin.

Provenance and contributing

TriMet is the producer; this catalog is a mirror and carries via links to TriMet's metadata page on every collection. TriMet's pages are the authority on what any field means, and every column description here was taken from them.

The catalog is generated from cholmes/portolan-catalog-trimet and the data browser from cholmes/trimet-data-browser. If you find an error in this guide — a join that does not hold, a recipe that fails, a caveat that is missing — open an issue or send a pull request against the generator, not the published file. Data errors go to TriMet at gis@trimet.org.

License, and a warning

TriMet distributes these datasets free of charge, but the terms of use do not grant redistribution rights. Section 4 of TriMet's Terms of Use reserves reproduction, modification, distribution and republication of site Content without written consent; the redistribution license in section 5 covers the Web Services API, not these GIS downloads. The collections therefore declare "license": "other" with a link to those terms, rather than claiming an open license the source does not offer.

Practically: use the data, and contact gis@trimet.org before redistributing it or building a product on it. If you need transit data under clear open terms, TriMet's GTFS feed is the better starting point.