Covering posts from 0800 ET October 5 to 0800 ET October 6. Sources: 168 geospatial feeds.
1. Making open data work for AI without burning the commons
Bill Dollins revisits his open data and AI thread after using Overspan, a full-planet Overpass service with an MCP interface and paid plans that cap requests and query resources. Spatial Reserves profiles Cobb County, Georgia, whose GIS analysts built a retrieval-augmented chatbot over their ArcGIS Hub data library. David G. Smith adds a useful caution, relaying Thore Graepel's argument that LLMs predict tokens rather than reason.
Why this matters: The open data debate is shifting from whether to share to who pays for machine-scale demand. Dollins' metered MCP service and Cobb County's retrieval layer are two practical answers, and Graepel's warning is why curated context beats raw model confidence.
2. Sovereign and commercial orbit keep filling in
Meridian Space and Kongsberg NanoAvionics confirmed their first reflectarray communications payload is active in orbit, pitched at governments that want to own and control sovereign LEO constellations. SFL Missions confirmed two more GHGSat greenhouse gas microsatellites, C18 and C19, launched October 1 and bringing its GHGSat builds to 13.
Why this matters: Sovereignty talk has moved from op-eds to hardware and procurement. Meridian sells the ownership argument directly to governments. GHGSat is a rarer proof point, a commercial emissions-monitoring constellation that keeps adding satellites, which tells you somebody is buying the data.
3. Risk mapping that names the place and the problem
Spatial Source reports that eight in 10 cities in Great Britain are already highly vulnerable to urban heat islands, with the mapping projected to the end of the century. It also covers drones tracking individual wilding pine trees in New Zealand. EarthDaily argues you can't cut Canada's wildfire risk without knowing where it's building.
Why this matters: These are specific hazards in specific places with an obvious buyer. That's the demand-side texture the feeds usually lack. Conservation and vegetation work is a chronic content gap, and the wilding pine story is a rare ground-level example.
1. Revisiting Open Data and AI — geoMusings by Bill Dollins Dollins ties his earlier prototype, an OSM extract in PostGIS exposed over REST and MCP, to Overspan's hosted Overpass service. It's a concrete look at who absorbs the load when AI agents query open data, written by someone running both approaches. → Revisiting Open Data and AI
2. A Chatbot for a GIS Hub’s Data Library — Spatial Reserves Walks through how Cobb County GIS analysts used retrieval-augmented generation so a chatbot answers public and internal questions from their own datasets. It's applied AI at a local government with a named method and a clear audience, which is rarer than the announcements. → A Chatbot for a GIS Hub’s Data Library
3. MeridianWay: Walking in the sun or in the shade — Spatialists – geospatial news An open-source web app that computes a sunny and a shady walking route for any date, time and place. It uses swissBUILDINGS3D from swisstopo in Switzerland and OpenStreetMap data elsewhere, so the data choices are as interesting as the idea. → MeridianWay: Walking in the sun or in the shade
4. Fixing the gap between your water audit and field data — Fulcrum Argues that a utility's water audit is only as good as the leak survey, valve and meter records behind it, down to one boundary valve left open skewing overnight flow readings. It's vendor content, but it speaks to a commercial vertical the feeds rarely cover from the customer's side. → Fixing the gap between your water audit and field data
5. Aerial imagery deployed to spot invasive trees — Spatial Source Drones are being used in New Zealand to find individual wilding pine pest trees. It's a short piece, but it's concrete applied remote sensing for land management, a topic that doesn't get much airtime here. → Aerial imagery deployed to spot invasive trees
Powered by Neptune