FireSpec
Hyperspectral burn-severity and fuel-condition mapping of the January 2025 Los Angeles fires, from Planet Tanager-1 imagery.
Product and engineering for geospatial data.
I manage geospatial and Earth observation data products and build much of them myself: the discovery with the people who use them, the requirements, and the pipeline. The work so far covers satellite object detection, hyperspectral wildfire mapping, real-estate site selection and utility risk. Mostly Python, and open source where the work allows.
The workWork
detr-geo
Runs RF-DETR over a GeoTIFF and hands the detections back as GeoJSON, GeoPackage or Shapefile, in the image's own coordinate system.
FirstLight
Turns a plain-language description of an event into a structured query, finds the satellite imagery, picks a detection algorithm to suit what it found, and writes the report.
FireSpec: Planet Tanager-1 shortwave-infrared bands, 23 January 2025, what the sensor recorded and not a FireSpec result. Tanager STAC Data © 2025 Planet Labs PBC. All Rights Reserved. detr-geo: the model's raw detections on NAIP imagery (USDA, public domain), from weights fine-tuned on xView (CC BY-NC-SA 4.0). FirstLight: Sentinel-2 false colour, 9 May 2026, surface reflectance rather than an analysis product. Contains modified Copernicus Sentinel data 2026.
Writing
What 426 bands found in the Palisades burn scar
Spectral unmixing of the January 2025 Palisades Fire with Planet's Tanager-1, checked against a reference library that never saw the image.
Pushing to publish
Notes on wiring a git push straight through to a live page.
First post
A first post to prove the blog pipeline end to end.