firestorm-lightning-data

Real-time lightning feed for FIRESTORM, mirroring GOES-R GLM Level-2 LCFA flashes from NOAA's Open Data S3 buckets to a slim JSON the FIRESTORM frontend reads via raw.githubusercontent.com.

Same architectural shape as firestorm-aircraft-data, firestorm-wind-data, firestorm-news-data: GitHub Actions cron poll → public S3 source → slim JSON → frontend fetch().

What this replaces

Until v2_141, FIRESTORM's lightning layer rendered 40 randomly-generated points around hand-picked region centroids (generateDemoLightning(40)). The visual presentation gave no indication the data was synthetic. This pipeline replaces that demo path with real strikes.

Source

GOES-R Geostationary Lightning Mapper (GLM) Level-2 Lightning Cluster Filter Algorithm (LCFA) product:

• GOES-East (G19): s3://noaa-goes19/GLM-L2-LCFA/<YYYY>/<DDD>/<HH>/ (replaced G16 in April 2025)
• GOES-West (G18): s3://noaa-goes18/GLM-L2-LCFA/<YYYY>/<DDD>/<HH>/ (replaced G17 in 2023)

Public, anonymous, no auth, no egress charge. NOAA Open Data on AWS.

Native cadence: ~20 seconds per satellite. Each file is netCDF-4, ~50–500 KB.

Output

data/lightning.json — flat JSON consumed by the FIRESTORM frontend.

{
  "generated_at": "2026-05-20T22:00:00+00:00",
  "window_minutes": 15,
  "counts": { "total": 1234, "g19": 800, "g18": 434 },
  "source": "NOAA GOES-R GLM L2 LCFA via NOAA Open Data on AWS S3 (public, no auth)",
  "flashes": [
    { "lat": 38.4, "lng": -98.6, "energy_fJ": 142.7, "age_sec": 47, "type": "flash", "sat": "g19" },
    ...
  ]
}

Cap of 5000 flashes per cycle, sorted by energy descending. The frontend only renders ~100–500 markers usefully, so the cap biases toward the strongest / most operationally meaningful strikes.

Caveats (read these once)

• GLM cannot distinguish CG from IC by itself. It's an optical sensor measuring the flash bloom in the 777.4 nm oxygen line; ground networks measure the EM pulse. Every flash is labeled type: "flash". CG-only classification requires NLDN cross-correlation — out of scope here.
• Detection efficiency: ~70% for CG strikes, much higher for total flash count. Fine for situational awareness; not a substitute for NLDN if sub-second latency or guaranteed CG detection matters.
• End-to-end latency: strike → JSON ≈ 2–7 minutes. (Strike → GLM downlink ~30s → ground processing ~30s → S3 publish ~30s → cron poll up to 5 min → push.)
• Coverage gaps: GOES is geostationary at ~35,786 km altitude. Polar regions above ~55° latitude have degraded view geometry. CONUS, Caribbean, Mexico, most of S. America (G16) + Pacific, Alaska, western CONUS (G18) are well covered.

Running locally

pip install boto3 botocore netCDF4 numpy
python fetch_glm.py
# → writes data/lightning.json

No credentials needed. The script uses boto3 with UNSIGNED config against NOAA's public buckets.

Cost

$0. NOAA pays for the bucket; GHA cron is free for public repos; no AWS account required to read the data.

License

Pipeline code is not licensed for redistribution outside FIRESTORM. The underlying GLM data is U.S. federal government produced and is in the public domain.