Open geospatial field test · Rome · 41.90°N / 12.49°E

From source pixels to OGC API, without hiding the stack.

A small, inspectable on-prem experiment combining 5,603 OpenStreetMap road and rail ways with ESA WorldCover 2021. Run it locally, inspect every assumption, and tell us where the evidence fails.

Download the test kit Read the comparison
Central Rome analysis showing selected OpenStreetMap roads over ESA WorldCover land-cover classes
Map data © OpenStreetMap contributors, ODbL 1.0. Land cover © ESA WorldCover project 2021 / Contains modified Copernicus Sentinel data (2021) processed by ESA WorldCover consortium, CC BY 4.0.
5,603OSM road and rail ways
155,417WorldCover pixels touched
763.7 kmApproximate linework
100%Local default run

Two workloads, two honest boundaries

Do not turn one diagram into one oversized claim.

The raster result is computed with Python and Rasterio. Tellurion serves the vector data through OGC API interfaces. Use the selector to see what was actually measured in each lane.

Python/Rasterio ↔ QGIS Processing

80.10%Pixels classified built-up inside the selected line mask
19.10%Pixels in vegetated WorldCover classes
identical1,200×900 output masks, byte for byte
Boundary: “Road corridor” means raster pixels touched by linework at this analysis grid. It is not a legal buffer, accessibility measure or causal result.

Critical comparison

Different tools have different jobs.

A fair comparison starts by naming the workload, runtime and missing evidence. Esri remains deliberately unmeasured because no licensed ArcGIS runtime was available.

ToolRole in this testMeasured resultWhat the result cannot prove
TellurionEmbedded-first vector service2,714.74 req/s; 34.18 ms p95Production, distributed or Italy-scale capacity
GeoServerMature Java service platform856.35 req/s; 107.19 ms p95A broad product verdict from one endpoint
QGISDesktop and Processing reference4.38 s median complete CLI workflowQGIS Server performance; startup dominates this small run
RasterioLocal analysis implementation0.59 s median; byte-identical maskEquivalent user experience or full GIS capability
EsriCommercial comparison boundaryNot timedAny performance claim without a licensed runtime
Open comparison graphic Inspect metrics JSON

Tellurion 0.3.0 · public binaries

Download once. Run on your own machine.

Large binaries stay out of Git history and are published as versioned GitHub Release assets. Every archive includes the BUSL 1.1 licence, commercial-licensing notice, build metadata and example configuration.

macOSApple Silicon · aarch64Download
Linuxx86_64 · static muslDownload
Windowsx86_64 · GNUDownload
Open the STAC-to-map demo Explore all live demos Release notes and source Verify SHA-256 Request a platform or report a problem

Licence: evaluation, development, testing and training are free. Personal non-commercial production use is allowed. Organisational production use requires a commercial licence before 2030-07-23. The benchmark numbers on this page remain the recorded Tellurion 0.2.0 run; publish new 0.3.0 measurements separately.

Reproduce before believing

Two commands. Local by default.

  1. Download and unzip the versioned kit.
  2. Run the bundled snapshot without network access.
  3. Optionally refresh OSM and the WorldCover COG window.
  4. Share the exact command, environment and metrics when disagreeing.

SHA-256:
28d84f5d9d1ebbc6333f8054f8ec37d15e061ff2b3eeb79f1d7d70d5b952d98c

shell · local snapshot
unzip tellurion-italy-demo-v0.2.0.zip
cd tellurion-italy-demo
chmod +x reproduce.sh
./reproduce.sh

# Refresh both internet sources
./reproduce.sh --refresh

# Install the current public macOS/Linux binary
./reproduce.sh --install
./reproduce.sh --load-rome
./reproduce.sh --serve-rome

Availability: Tellurion 0.3.0 binaries are public for macOS Apple Silicon, Linux x86_64 and Windows x86_64. The shell installer supports macOS/Linux directly and verifies the selected archive against the release SHA-256 manifest.

Ten field notes

The evidence, one claim at a time.

Each article exposes one boundary, one measurement or one reproducibility decision. The sequence is an invitation to challenge the work—not a launch campaign built around invented certainty.

Sources, licences and attribution

Open data is not attribution-free data.

The repository code is AGPL-3.0. OpenStreetMap data is ODbL 1.0 and must credit OpenStreetMap contributors. ESA WorldCover 2021 v200 is CC BY 4.0 and requires the WorldCover/Copernicus acknowledgement shown on this site and its map images. See NOTICE.md for the full record.

OpenStreetMap copyright & ODbL Overpass API ESA WorldCover data, licence & citation WorldCover 2021 v200 DOI CC BY 4.0 Project AGPL-3.0 licence

Community review

Find the first weak assumption.

Run the kit, attach your metrics and describe the environment. A reproducible disagreement is more useful than agreement with the headline.

Open a test report