Canton boundaries

File: data/canton-boundaries.json. The boundary of each of the 162 judicial cantons as GeoJSON geometry, as the National Geographic Institute (NGI) digitised it from the annex to the Judicial Code.

Why it's useful

• The canton of an address in a split city. 55 cantons cover only part of a municipality (Antwerpen, Gent, Brussel, Luik, Charleroi, …), and getCantonsForMunicipality can only list the candidates. With the address geocoded, findCantonsAt(lon, lat) gives the one canton, and so the justice of the peace.
• Maps. The geometry is standard GeoJSON in WGS 84: draw the cantons on a map, join them to other geodata, or load them into PostGIS, QGIS or a spreadsheet with a map.
• Checking the legal text. The NGI's polygons are a second reading of the same annex, next to the territory text of each canton.

The legal text wins

The boundary is the NGI's interpretation of the annex to the Judicial Code, digitised on cadastral municipality boundaries, watercourses and roads. The annex itself, quoted verbatim in each canton's territory, is the law. For a point close to a boundary line, read the text; it names the streets and the house numbers.

Source and updates

Source	NGI, Judicial cantons and arrondissements (Territorial Divisions) on geo.be: the GeoPackage in WGS 84 (EPSG:4326), feature type judicialcanton, revised 16 July 2026
Built by	scripts/parse/boundaries.ts from the weekly snapshot of the GeoPackage
Changes when	The NGI publishes a new version: every year, when the FPS Finance's municipal boundaries are updated, or when a law changes the annex
Covers	21 October 2026 onwards, the day the file was first read: lookups for an earlier day throw a RangeError (time)
Checked	Every week, by the sources workflow: the build fails when a canton has no boundary or a boundary matches no canton
Licence	© NGI, CC BY 4.0

How the NGI drew the boundaries, in its own words: where the annex refers to municipal boundaries, it used the cadastral boundaries of the FPS Finance; where it refers to watercourses or roads, the NGI's own topographic data; where it refers to former municipalities, the NGI's boundaries of those. The scale is 1:10 000.

Fields

Field	Type	Description
id	CantonId	The canton's id: a canton has one boundary, which takes its key
validFrom, validTo	string?	The days the record applies; none: since the coverage starts, or still applies. See time
canton	CantonId	The canton the boundary belongs to
ngiId	string	The NGI's identifier of the object (its tgid), a GUID
cantonCode	{ nl, fr }	The canton code in the NGI's data: six digits (248112 for Gent 1), with an N or F prefix per language in Brussels-Capital and in Sint-Genesius-Rode (N238104)
geometry	MultiPolygon	GeoJSON (RFC 7946): WGS 84, longitude first, exterior rings counter-clockwise, holes clockwise
sources	{ url, retrieved }[]	The NGI file the record was read from

interface CantonBoundary {
    id: CantonId // 'gent-1'
    canton: CantonId // 'gent-1'
    ngiId: string // '{3905ED53-F233-4C4E-8225-4E43FEC7C541}'
    cantonCode: Bilingual // { nl: '248114', fr: '248114' }
    geometry: MultiPolygon // { type: 'MultiPolygon', coordinates: [[[[lon, lat], …]]] }
    sources: SourceRef[]
}

The geometry is the NGI's, vertex for vertex: nothing is simplified. Coordinates are rounded to 7 decimals, about a centimetre, below the source's own precision. Every boundary is one MultiPolygon, even when it has one polygon, so every record has the same shape.

How cantons are matched to the NGI's features

The NGI names each feature in Dutch and in French (Eerste gerechtelijk kanton Gent, Premier canton judiciaire de Gand). A feature is the boundary of a canton when both names, turned into an id the usual way, give that canton and agree on its number. Seven features are named by their local name in both languages where the annex uses an exonym (Canton judiciaire de Halle, where the annex says Hal; Gerechtelijk kanton Huy, where it says Hoei); those are mapped by their canton code in CANTON_ALIASES in the parser. Anything else stops the build.

Size

The boundaries are by far the largest dataset: about a million positions, 23 MB of JSON for the cantons and 8 MB for the arrondissements. They have their own entry points, belgica/canton-boundaries and belgica/arrondissement-boundaries, which read the geometry from data/ with a JSON import, and the package root doesn't re-export them, so nothing loads them unless it asks. In a browser, load the entry point when the map opens, with import(), rather than in the main bundle.

The raw file, data/canton-boundaries.json, is in the npm package and in every release, so Python, PostGIS and GIS tools can read it without the JavaScript.

Finding the canton of a point

import { findCantonsAt, getCantonBoundary } from 'belgica/canton-boundaries'
import { getJusticeOfThePeace } from 'belgica/courts'

// Longitude first, then latitude, as GeoJSON orders them
const [canton] = findCantonsAt(3.7268, 51.053) // Sint-Baafskathedraal, Gent → gent-1
getJusticeOfThePeace(canton.id).address // where to file

findCantonsAt(2.3522, 48.8566) // [] : Paris, outside Belgium
getCantonBoundary('gent-1').geometry // the MultiPolygon

findCantonsAt returns one canton for a point in Belgium, none outside it, and both cantons for a point exactly on the line between two. It needs no dependency: a plain point-in-polygon test over the boundaries, about a millisecond per lookup. Geocoding the address (street and number to coordinates) is up to you: the regions' address registers (Flanders' Adressenregister, Brussels' UrbIS, Wallonia's ICAR) and the federal BeSt address data on geo.be do that.

Arrondissement boundaries

File: data/arrondissement-boundaries.json, entry point belgica/arrondissement-boundaries. The boundary of each of the 12 judicial arrondissements, from the same NGI file (feature type judicialarrondissement), in the same shape: id and arrondissement (the ArrondissementId), ngiId, geometry (a MultiPolygon), sources, and the validity fields.

import { findArrondissementsAt, getArrondissementBoundary } from 'belgica/arrondissement-boundaries'

findArrondissementsAt(4.2372, 50.7339) // Halle → [brussel]
getArrondissementBoundary('leuven').geometry // the MultiPolygon

An arrondissement is the union of its cantons, so the same answer follows from findCantonsAt(lon, lat)[0].arrondissement. The arrondissement boundaries are there for maps, and for drawing an arrondissement without merging 20 cantons.

The NGI names each feature once, in one language (Oost Vlaanderen, Liège, Bruxelles - Brussel). A name matches an arrondissement when it gives its id or one of its names; two are mapped by hand in ARRONDISSEMENT_ALIASES. One of those is a trap: the NGI calls the arrondissement of Leuven Vlaams Brabant, after its province, though the arrondissement leaves out Halle-Vilvoorde, which belongs to the arrondissement of Brussels. So the build checks every arrondissement against its cantons: its area must be the sum of theirs, within 0.1 %, or the build fails.

From an address to the court

A prototype: type an address, and the region's own address register geocodes it (the Adressenregister in Flanders, UrbIS in Brussels, ICAR in Wallonia). The point goes to findCantonsAt, and the canton to getCourtsForCanton. For an address the register knows down to the house number, the point is on the building, so the even and odd sides of a boundary street fall in the right canton. When it knows only the street, the point is somewhere on the street, and the lookup says so. Within 25 m of another canton the lookup warns and shows the law's text for both.

Address

Try: Kerkstraat 10, 2060 Antwerpen · Rue de la Loi 16, 1000 Bruxelles · Place Saint-Lambert 18, 4000 Liège

In other languages

The file is plain GeoJSON geometry per record, so any GIS reads it. In Python with Shapely:

import json
from shapely.geometry import Point, shape

boundaries = json.load(open("data/canton-boundaries.json"))
polygons = {b["canton"]: shape(b["geometry"]) for b in boundaries}

def cantons_at(lon: float, lat: float) -> list[str]:
    point = Point(lon, lat)
    return [canton for canton, polygon in polygons.items() if polygon.covers(point)]

cantons_at(3.7268, 51.053)  # ['gent-1']