2026-11-04 –, Carr
Distributed wildfire modeling generates thousands of COGs per run. We needed them discoverable, viewable, and traceable with minimal custom tooling. This talk covers the STAC catalog we assembled from pgSTAC, stac-fastapi-pgstac, and titiler-pgstac, and what we learned.
At Timmons Group, we build wildfire modeling tools for state and local agencies. Fire behavior models have traditionally run on a scientist’s workstation, one landscape at a time. To reach regional and eventually CONUS scale, we built a cloud pipeline that tiles the landscape, runs a 216-member weather ensemble on every tile in parallel, and mosaics the results into Cloud-Optimized GeoTIFFs (COGs).
That solved our compute problem and set up a new one: a single statewide run produces hundreds of thousands of rasters, and the scientists behind each run need to find and trust their outputs.
This was our first time reaching for STAC. Instead of writing a catalog, we assembled the eoAPI stack (pgSTAC, stac-fastapi-pgstac, and titiler-pgstac), fed it with event-driven ingestion the pipeline never sees, and modeled our items so every raster points back to the run that made it.
I’ll share what worked, what bit us, and what we’re still working out, with a short demo from search to map tiles. I’ll close with an invitation to help shape shared STAC conventions for wildfire data.
Kirk Davis is a Senior Data Engineer at Timmons Group with over 13 years of GIS experience, specializing in leveraging large geospatial datasets through APIs, automation, and analytics. He focuses on solutions at the intersection of geospatial software, wildfire, and forestry.