geoscena 0.1.0

AOI public-geodata acquisition + SceneBundle fusion and meshing core for 3D area reconstruction

geoscena

AOI public-geodata acquisition + SceneBundle fusion and meshing core for 3D area reconstruction. geoscena turns a geographic Area of Interest into a projected, fused, mesh-ready SceneBundle (Draco/meshopt-ready glTF layers + a provenance manifest) assembled from open public datasets. It is the reusable, product-agnostic core behind Maqueta; nothing enters a bundle without recorded provenance (source, URL, license, fetch date).

What it does

Given an AOI (a centre + half-size, or a bbox), geoscena:

1. Fetches each available open layer, AOI-scoped, from its authoritative source:
   • Buildings + roads from Overture Maps GeoParquet on S3 (DuckDB spatial pre-filter).
   • Terrain from Copernicus GLO-30 DSM (AWS Open Data, keyless COG windowed reads).
   • Land cover from ESA WorldCover 10 m (AWS Open Data).
   • Water / green / rail context from OpenStreetMap (OSMnx / Overpass).
   • (extendable: Google Open Buildings 2.5D height rasters, GHS-POP, USGS 3DEP lidar.)
2. Fuses them: the height-provenance ladder assigns every building a height from the best available source (measured -> floors x floor-height -> height raster -> prior) and records which source was used per building, so inference is reported, never hidden.
3. Meshes in the AOI local metric frame (metres, Y-up on export): adaptive terrain TIN (Delatin-style greedy refinement), extruded building prisms, buffered road/rail ribbons draped on the terrain, flat water/green surfaces.
4. Writes a SceneBundle: one .glb per layer + manifest.json (the processing->web contract), with per-layer stats, height-provenance mix, and a data-credits list.

Install

pip install geoscena                 # core (numpy/scipy/shapely/geopandas/rasterio/trimesh)
pip install 'geoscena[overture]'     # + DuckDB (buildings/roads)
pip install 'geoscena[osm]'          # + OSMnx (water/green/rail)
pip install 'geoscena[all]'          # everything (incl. lidar, gltf compression helpers)

Python >= 3.11. No native compiler required (the terrain TIN is pure-Python over SciPy).

Use

from geoscena.aoi import AOI
from geoscena.build import BuildConfig, build_scene

aoi = AOI.from_center("Berlin Mitte", lon=13.405, lat=52.517, half_size_m=1200)
bundle = build_scene(aoi, BuildConfig(fetched="2026-07-12"))
bundle.write("out/berlin_mitte")   # -> terrain.glb, buildings.glb, roads.glb, manifest.json

or from the CLI:

geoscena build --name "Berlin Mitte" --lon 13.405 --lat 52.517 --half 1200 \
    --fetched 2026-07-12 --out out/berlin_mitte
geoscena info    # list known sources + licenses

Provenance and licensing

Every layer carries a LayerProvenance (source, URL, license key, fetch date, method). bundle.any_noncommercial() flags AOIs that pulled a non-commercial source. Default sources are all commercial-OK: Overture/OSM (ODbL), GLO-30 (Copernicus free), WorldCover (CC-BY-4.0). See docs/ for the data contracts and per-source cards.

Status

0.x while the SceneBundle contract stabilizes. Consumed by Maqueta via git tag ref until the first PyPI release (ADR-0061 trusted publishing).

Owner: Felipe Santibanez-Leal · [e-mail] · @fsantibanezleal