Open mobility data is everywhere, but different sources answer different questions -- and the differences are easy to miss until results conflict. This talk presents a reproducible, open-source workflow to compare three widely accessible mobility proxies on a common spatial reference.

Using Copernicus Urban Atlas polygons and the DEGURBA classification (derived from 1 km2 population grid cells), we convert each source into comparable density-normalised temporal indicators (day/night ratios, intraday profiles, weekday/weekend patterns). The three sources:

• Eurostat experimental MNO statistics (aggregated, anonymised mobile network operator statistics published by national statistical offices)
• OpenStreetMap public GPS traces (community-contributed trace archive; participation bias applies)
• Census commuting flows from Eurostat (static origin-destination baseline)

We apply a simple clustering step (HDBSCAN) to group similar temporal profiles into functional signatures (residential, office, late-evening activity, mixed-use), and use UMAP only as a visual explanation aid. Instead of selling one "best" dataset, we provide a practical decision guide: which source is best for presence vs flows, what biases to expect, and how to combine sources when a single source falls short.

Early findings: MNO statistics capture temporal presence well but availability and comparability vary by country; OSM GPS traces reflect contributor behaviour more than population-level patterns; census flows miss intraday dynamics but anchor the OD baseline. Where sources agree, classification is robust; where they diverge, the divergence reveals structural data limitations worth knowing.

What you take away:

• A decision matrix with concrete rules of thumb (e.g., intraday presence -- start with MNO; commuting structure -- census OD; fine-grain routes/activities -- OSM, with known participation bias)
• Typical biases and coverage gaps of each source across different European urban contexts
• A reproducible pipeline: Python + PostGIS + OSRM + QGIS-ready layers and scripts
• We will publish the full pipeline as an open repository (Docker/conda, pinned environments, end-to-end scripts to regenerate figures and maps)

This talk is for anyone choosing open mobility data for urban analysis who wants to stop guessing and start comparing.