GeoSPARQL (geometry + time)

rete stores geometry as standard GeoSPARQL geo:wktLiteral values and answers a focused set of GeoSPARQL filter functions — so you can ask spatial questions (point-in-polygon, intersection, distance) and combine them with ordinary SPARQL, including temporal filters. The whole thing runs in the browser/WASM engine too: the geometry code is pure Rust (core + f64, no GIS C libraries), so a .rete file of polygons-with-time is queryable in place, offline.

Try it in the playground — pick the history.rete dataset and the Geo examples.

Data model

Geometry follows the GeoSPARQL convention — a feature links to a geometry node that carries a WKT literal:

<http://ex/terr/France_1914>
  rdfs:label "France" ;
  ex:year 1914 ;                                   # plain xsd:integer — filter on it
  geo:hasGeometry [ geo:asWKT
    "POLYGON((-4.79 48.32, 2.55 51.09, 7.59 47.59, ...))"^^geo:wktLiteral ] .

• geo: = http://www.opengis.net/ont/geosparql#
• The functions operate on a geo:wktLiteral term, normally reached with the property path ?f geo:hasGeometry/geo:asWKT ?wkt. A plain or xsd:string literal whose value happens to be WKT is also accepted.
• Axis order is CRS84 lon/lat, always. A CRS-URI prefix on a literal (<http://…/CRS84> POINT(…)) is accepted but ignored — rete never swaps axes, so do not feed it lat/lon EPSG:4326 data.
• Time is just an ordinary literal (ex:year 1914, or ex:startYear/ ex:endYear), so a temporal slice is a plain FILTER(?year = 1914).

Functions

geof: = http://www.opengis.net/def/function/geosparql/, uom: = http://www.opengis.net/def/uom/OGC/1.0/.

WKT parsing covers POINT, MULTIPOINT, LINESTRING, POLYGON (with holes), MULTIPOLYGON, GEOMETRYCOLLECTION, the EMPTY keyword, and Z/M ordinates (dropped). A malformed or non-geometry argument is a type error: the row drops out of a FILTER and the variable is left unbound in a BIND — never an engine error.

Example — "which territory contained this point in year Y?"

The headline query combines a temporal filter and a spatial predicate in one FILTER:

PREFIX geo:  <http://www.opengis.net/ont/geosparql#>
PREFIX geof: <http://www.opengis.net/def/function/geosparql/>
PREFIX ex:   <http://ex/>
PREFIX rdfs: <http://www.w3.org/2000/01/rdf-schema#>

SELECT ?territory WHERE {
  ?t ex:year 1914 ; rdfs:label ?territory ;
     geo:hasGeometry/geo:asWKT ?w .
  FILTER(geof:sfContains(?w, "POINT(2.35 48.85)"^^geo:wktLiteral))
}
# → France   (the same point resolves to "Kingdom of France" in the year-1000 snapshot,
#    and "Liao" / "Qing" / … for Beijing across the era snapshots)

Distance ranking (metres → km), bounding boxes, and bbox intersection work the same way:

# nearest territories to a point over London, 1914
SELECT ?territory ?km WHERE {
  ?t ex:year 1914 ; rdfs:label ?territory ; geo:hasGeometry/geo:asWKT ?w .
  BIND(geof:distance(?w, "POINT(0 51)"^^geo:wktLiteral,
       <http://www.opengis.net/def/uom/OGC/1.0/metre>) / 1000 AS ?km)
} ORDER BY ?km LIMIT 8

The example dataset

history.rete (embedded in the playground) is built from aourednik/historical-basemaps (GPL-3.0) — world territorial borders at snapshots from 323 BCE to 1994 CE. Each border polygon becomes a geo:wktLiteral, the snapshot year an xsd:integer, geometry simplified to ~1 km for an in-browser-sized file. Build it with:

python3 scripts/geo_to_rete.py basemaps \
  --years bc323,1000,1492,1815,1914,1945,1994 --prec 2 --min-bbox 0.3 \
  --max-per-year 90 -o history.nt
rete build history.nt -o web/history.rete

The same script also converts an OpenHistoricalMap extract — real dated administrative boundaries pulled from its public QLever endpoint (geo:hasGeometry/geo:asWKT

• start_date/end_date) — via the ohm subcommand.

Scope & limitations

v1 is deliberately tight and exact for the common cases rather than broad:

• Planar Cartesian computation on lon/lat (haversine only for the metre distance). No CRS reprojection — geof:transform and non-CRS84 axis handling are out of scope.
• The five topological relations are computed directly (point-in-polygon + segment crossing), not via a full DE-9IM matrix. Exact for the dominant polygon ⊇ point / nested-territory cases; sfTouches/sfCrosses/sfOverlaps, geof:relate, constructive geometry (buffer/union/boundary/…), geof:area/length, and geof:getSRID are not implemented — an unsupported geof: IRI is rejected at parse time.
• No spatial index: filters scan the candidate rows (fine for the curated demo data and for queries already narrowed by a non-spatial pattern).

Performance of the non-spatial engine is covered in Benchmarks; overall SPARQL coverage in the conformance scorecard.