Random GPS Coordinate Generator

Bounding-box coordinate generator and geospatial workflows

This variant of the random coordinate generator lets you pin the output to an arbitrary bounding box — four numbers (latMin, latMax, lngMin, lngMax) that describe a rectangle on the globe. The default covers the metropolitan region of São Paulo. Bounding boxes are the cheapest filter in any geospatial pipeline: they are O(1) to test and they pre-filter a large set before more expensive polygon-in-polygon checks.

Geocoding and reverse geocoding

Random coordinates pair naturally with two everyday GIS operations:

• Geocoding — turn an address ("Av. Paulista, 1000") into a lat/lng. Common providers: Nominatim (OpenStreetMap, free), Google Maps Geocoding API, Mapbox, and Brazil-specific stacks like ViaCEP+georeferenciamento.
• Reverse geocoding — turn a lat/lng back into an address or administrative unit. Useful for enriching random points with mock street names during test seeding.

From bounding box to polygon: ray casting

A bounding box always contains more area than the polygon it surrounds — Brazil's bounding box covers parts of Peru, Colombia and the Atlantic. To restrict points to the real outline, use ray casting: generate a candidate point in the bbox, fire a horizontal ray, count polygon edges crossed; an odd count means inside, even means outside. Reject candidates outside the polygon and resample. The polygons themselves come from open datasets:

• Natural Earth — country and admin-1 polygons at 1:10m, 1:50m, 1:110m.
• IBGE Malhas — Brazilian states and municipalities as shapefiles or GeoJSON.
• OpenStreetMap via Overpass API — arbitrary tagged regions (parks, neighbourhoods, water).

Spatial indexes: geohash, H3, S2

For large datasets you want to index points so neighbour queries are fast. Three popular schemes:

• Geohash — encodes lat+lng into a base32 string; longer strings = finer cells. 5 chars ≈ 5 km, 6 chars ≈ 1 km, 7 chars ≈ 150 m, 8 chars ≈ 19 m. Cell sizes shrink toward the poles.
• H3 (Uber) — a hierarchical hexagonal grid. Hexagons have uniform neighbour symmetry (every neighbour is exactly one edge away), which beats geohash for radius queries and routing.
• S2 (Google) — quad-tree on a sphere. Used by Google Maps, Foursquare and InfluxDB geo functions.

Uniform random is not realistic — populations are not uniform

A uniform random point in Brazil's bounding box has equal odds of landing in the Amazon rainforest as in São Paulo's downtown — but the city houses millions and the rainforest pixel houses no one. For realistic mock data weight the draw by population density. Free datasets:

• WorldPop — 100 m population rasters by country and year.
• GHSL (Global Human Settlement Layer) — JRC's settlement and built-up rasters.
• Facebook High Resolution Population Density — 30 m grids derived from satellite imagery.

To draw a population-weighted point, treat the raster as a 2D probability distribution: sample a pixel proportionally to its value, then draw a uniform point inside that pixel.

GPS precision in the real world

A smartphone GPS in clear sky delivers ±3-10 m. Indoors or in dense urban canyons, error climbs to ±50 m or worse — Wi-Fi positioning and cell-tower triangulation start to beat GPS. Survey-grade RTK GNSS reaches 1-2 cm after carrier-phase correction. The output of this generator uses 4-6 decimals; that resolution is finer than what any consumer GPS can guarantee, so do not interpret the trailing digits as "millimetres of truth".

FAQ

Can I restrict points to a specific country? The bounding box gets you most of the way — use a tight bbox around the territory. For points only inside the polygon (no spillover into neighbouring countries), pair the bbox with ray casting against a country polygon from Natural Earth or IBGE.

Will the points avoid the ocean? Not by default — a bbox over Brazil will produce some Atlantic points. You need a land/water mask raster (e.g. Natural Earth ne_10m_land) and a rejection step to filter sea hits.

Can I generate points that mimic real traffic? Uniform output does not. Mix in population density (WorldPop) for realism, or seed from real OSM Points-of-Interest if you want plausible business locations.

How does this differ from the worldwide generator? The other variant exposes preset regions (world / Brazil / Portugal / US); this one lets you specify any rectangle. Use this when your target is a city, district or arbitrary research area; use the preset variant for cross-country sampling.