The spatial join?

Vikram Gundeti – CTO¹ of Foursquare² – on LinkedIn:

Data scientists love data except when it’s geospatial data. The reason? Traditional GIS infrastructure is notoriously tricky for data scientists. (…)

We believe that geospatial intelligence should be optimized for modern machine learning workflows, not cartographic display. Bootstrapped with over 20 raster and vector datasets pre-indexed to H3³ cells (…), Spatial H3 Hub⁴ allows data scientists to enrich their ML⁵ models with demographic, economic, and environmental features using simple joins—no coordinate systems, no projections, no specialized tools.

See also Google’s recently announced AlphaEarth Foundations, an AI model that “integrates petabytes of Earth observation data”.

Is the future of geospatial mapless and agent-ready, as Vikram puts it? A future sans spatial join? Between these developments and others, such as GERS⁶, I’m inclined to believe it (to a degree).

Footnotes

1. Chief Technology Officer

2. A location technology company (Wikipedia, Crunchbase). You may remember the Foursquare app that let you check-in to places and enabled social location sharing. You may also have seen the recent announcements of a new product of theirs, Foursquare Spatial Desktop or “FSQ Spatial Desktop”, built on top of DuckDB and SQLRooms.

3. H3 is a hexagonal (mostly, except for 12 pentagons – think football (or soccer)) discrete global grid system (DGGS) (Apache 2.0-licensed) created by Uber

4. A product of Foursquare, an H3-based collection of geospatial datasets.

5. Machine learning

6. The Global Entity Reference System by the Overture Map Foundation (OMF)