OpenPOIs

An open-source toolkit for comparing and conflating Points of Interest (POIs) across major geospatial datasets.

What is this?

OpenPOIs downloads current US-wide POI snapshots from multiple publicly available sources — currently OpenStreetMap and Overture Maps — and conflates them into a single unified dataset. The web map lets you explore each source side by side. Each POI in the conflated dataset is given a confidence score, which is the probability that the POI currently exists based on available data from both sources.

Data Sources

All snapshots are US-wide and are periodically refreshed. The release date for each snapshot is tracked in the project configuration.

The Conflation Process

Conflation matches POIs across both sources into a single unified record. The process works as follows:

Taxonomy normalization: Each source uses its own category system. OpenPOIs maps both to a shared set of ~95 labels (e.g., Cafe, Bank, Pharmacy) using manually curated crosswalks. This shared taxonomy is used both for filtering in the web map and to set per-category match distances. See the full taxonomy reference for a complete mapping.

Spatial matching: Within each shared-label group, nearby POIs from different sources are candidate matches. Match radii vary by category — a small neighborhood coffee shop uses a tighter radius than a large hospital campus. Conflated POIs found in both OpenStreetMap and Overture Maps are assigned a match score (separate from the confidence score below), indicating the probability of a true cross-source match.

Confidence scoring: Each conflated POI receives a confidence score from 0 to 1, representing the estimated probability that the POI currently exists.

The confidence score is visualized on the map with a color ramp:

Data Access & Licensing

The raw snapshot and conflated Parquet files are publicly hosted on Source Cooperative and can be queried directly with pyarrow, DuckDB, or any tool that reads HTTPS Parquet.

The OpenPOIs dataset is released under the Open Database License (ODbL). Any public use must credit OpenStreetMap contributors, the Overture Maps Foundation, and OpenPOIs. Any derivative database must be shared under the same license.

See the OpenStreetMap copyright and Overture Maps attribution pages for more details about upstream attribution requirements.

Python API Documentation

The openpois Python package powers this project. It provides modules for downloading POI snapshots, conflating across sources, and modeling historical POI change rates.

We fit an empirical Bayes Poisson model to estimate per-category change rates from historical OpenStreetMap data, giving a sense of how likely a given POI is to have changed since it was last observed. A full methods writeup is coming soon.