FireEye Platform

FireEye is a public-facing geospatial platform for exploring wildfire risk, active-fire information, forecasts, and related spatial data across the United States.

Current Platform

The platform combines:

• a SvelteKit and MapLibre web map;
• PostgreSQL/PostGIS for spatial data and area queries;
• Martin for raster and vector tiles;
• Nginx for the public HTTPS entry point; and
• scheduled data pipelines for source data, model outputs, and downloadable products.

FireWx-FM

FireWx-FM predicts 12-hour active-fire occupancy probability on a 5 km EPSG:5070 Lower-48 grid. Its standard input is a 16-channel tensor containing NOAA HRRR weather fields, validity masks, LANDFIRE fuel and canopy layers, housing density, and population. NASA FIRMS detections define the occupancy target.

The model probability map is converted into a web-serving layer and can also be stored as a queryable spatial product. It is a research and decision-support product and should not be treated as an emergency-response alert without local validation and expert review.

Multi-Hazard Extension

The next practical extension is to evaluate flood or drainage-model outputs as separate, documented layers. A pilot should specify:

1. variables, units, spatial resolution, projection, and forecast horizon;
2. conversion to a raster or vector product with time metadata;
3. spatial and temporal alignment with the platform;
4. a map layer, legend, and area-summary interface; and
5. numerical checks against the source model output.

Wildfire and flood outputs should remain separate until their meanings, calibration, and update schedules have been validated. A combined risk score should only be defined after that check.

Links

• Platform: https://rai-fire.com/
• Full public code and paper artifacts: https://huggingface.co/RAI-Lab/Wildfire-FM
• Full platform overview: https://huggingface.co/RAI-Lab/Wildfire-FM/blob/main/docs/FIRE_EYE_PLATFORM_OVERVIEW.md