Geospatial data

Bijection stores points and shapes as ordinary validated values, answers spatial questions with exact reads, commits geofence transitions with the observations that caused them, and draws your data on the workspace map under your read rules. Bulk and specialist geospatial computation runs in your own infrastructure and comes back through customer compute.

Geometry values

Declare a geometry like any other field:

bijection/schema.ts

import { defineSchema, defineTable } from "bijection/server";
import { geoPoint, geoShape } from "@bijection/datasets/geometry";
import { v } from "bijection/values";

export default defineSchema({
  depots: defineTable({ name: v.string(), location: geoPoint() }),
  zones: defineTable({
    name: v.string(),
    area: geoShape({ types: ["Polygon", "MultiPolygon"], size: "large" }),
  }),
});

Values are GeoJSON in WGS84, longitude first. Every write, function argument and synced record is checked: an empty, unclosed or self-intersecting shape, or one in another coordinate system, is refused rather than repaired. The standard size holds up to 512 positions; size: "large" holds up to 40,960. geoShape({ profile: "wgs84-3d-hae-linear-sphere-v1" }) adds heights above the WGS84 ellipsoid.

What runs where

Bijection	Your compute
Validating every geometry, and converting other coordinate systems at the boundary	Converting Shapefile, FileGDB, GeoPackage, GeoTIFF and other formats (GeoJSON, WKT, WKB, KML, KMZ and Geobuf are read natively)
Region, nearest-object and cell queries, region totals and drawn-region filters over spatial indexes	Spatial joins, nearest-neighbour and distance joins, and aggregation over large inputs
Small per-row geometry steps in pipelines	Raster mosaics, reprojection, tiling and analytics
Tracks, current positions and geofences, committed with their observations	Map matching, road networks, routing engines and route optimization
The workspace map: layers, styling, timeline, drawing, saved and shared maps	Vector tiles for layers too large to draw from your records
Publishing geometry to your lake as GeoParquet at exact snapshots	Models, forecasts and simulations

Your compute reads geometry from a dataset as ISO WKB in OGC:CRS84. Distances on the map are great-circle distances; edges are straight lines in longitude and latitude, as GeoJSON defines them.

Bring computed shapes onto the map

A job writes its result to an external dataset and submits it. A collection sync then keeps one of your collections equal to the latest accepted version: each row’s geometry is admitted again and given its spatial index cell, so the map can filter it by region.

bijection/areas.ts

import { defineCollectionSync } from "@bijection/datasets/collectionSync";
import { shapeCell } from "@bijection/datasets/geoIndex";
import type { Area } from "@bijection/datasets/geometry";
import { internal } from "./_generated/api";

export const areaSync = defineCollectionSync({
  collection: "ServiceArea",
  source: "areaRows",
  generations: "areaGenerations",
  retirements: "areaRetirements",
  byKey: "generation_feature",
  key: "feature",
  retire: internal.areas.retire,
  row: (values) => ({
    ...values,
    cell: shapeCell(values.area as Area),
  }),
});
export const retire = areaSync.retireMutation;

Each accepted version replaces the previous one in a single commit; no reader sees a mix of two versions.

Tiles from your compute

Layers too large to draw from records, and imagery, are tiled in your compute as PMTiles archives: vector tiles with one layer named features (tippecanoe -l features), or PNG, JPEG or WebP images; uncompressed or gzip. Your app declares the archive with defineMapArchives from @bijection/datasets/map-archive and adds its upload and record functions to the producer route. Your job records each archive as the archive’s next version with the same producer grant, naming the dataset versions it was built from, and the workspace map draws it with the dataset’s read rule checked on every tile request. A tile URL from your own server can also be added as a map layer, but it is fetched directly by the browser and carries none of your app’s access rules, so use it only for public reference layers.

Get Started

Use-case atlas

Platform

Data and Operations

Guides

Client Libraries

Tools

API Reference

Geometry values

What runs where

Bring computed shapes onto the map

Tiles from your compute

​Geometry values

​What runs where

​Bring computed shapes onto the map

​Tiles from your compute

Geometry values

What runs where

Bring computed shapes onto the map

Tiles from your compute