CIVTAK26 · Nov 18–20, 2026 · Arlington, VARegistration open →
← All plugins

FireCast v0.5

Satellite heat detections, and a fire spread model on the device with the buildings and people it reaches.

Wildfire

01 What it does

Satellite heat detections on the ATAK map, drawn at the ground each one actually covers.

Every public feed draws a fire detection as a pin, and almost none of them mean what a pin means. A VIIRS detection means "something inside this 375 m square was hot at 23:38 Zulu". A GOES detection seen at 58 degrees off nadir means "something inside this 4 km by 2 km patch was hot", and that is the same dot on every other map. An operator deciding whether to drive to it needs the difference, so FireCast draws the footprint the record itself describes and puts the rest in words on a detail panel.

Technical details features in full, requirements, network

Features in full

Feeds, all public NIFC services and none of them needing a key or a sign-in:

  • VIIRS: Suomi-NPP, NOAA-20 and NOAA-21, a few passes a day. The footprint is the sensor's own Scan and Track for that pixel, near 375 m across the swath.
  • MODIS: Terra and Aqua. Same fields, but a MODIS pixel is 1 km at nadir and grows past 4 km at the edge of the swath, so the drawn box grows with it.
  • GOES: GOES-18 and GOES-19, a scan every few minutes. The feed carries no footprint, so it is computed from the record's satellite view angle and the geostationary geometry; the long axis points at the satellite, because that is the direction the tilt stretches the pixel.
  • RADR: burn areas mapped from Landsat and Sentinel passes over the last fourteen days, as real polygons - where fire has been, not where it is.
  • NASA Fire Event Perimeters: NASA's Fire Event Data Suite (FEDS) draws an outline around every fire's VIIRS detections in the lower 48 every 12 hours, so a fire nobody has mapped has a perimeter. Open data on NASA VEDA, no key. An algorithm's 375 m outline, never presented as the incident's own.

On top of that: NIFC's static industrial heat source mask as a filter, with the masked ground drawable so you can see what is being flagged; age as color, red within the hour through amber to gray past a day, and fading with it; a time window from three hours to three days; a confidence filter; and a list of what is in the map view, newest first, that opens the same detail panel as tapping the map. Where is it burning? lists the GOES fires and NASA perimeters in one list, each with its county and state, in the order the operator picks.

Beside the detections, two layers of other people's fire spread runs, each a switch in the pane with a list behind a long press and a time slider on the run you pick:

  • FIRIS Model Runs: the spread runs California's fire analysts publish for ATAK.
  • PyreCast Forecasts: the ELMFIRE forecasts PyreCast (CloudFire, Inc.) publishes for large fires across the country, shown with their permission, with their logo and a link to pyrecast.com beside them. PyreCast ranks its runs by the area burned at the end of the forecast; the plugin shows the middle run unless you pick the smaller or larger end. Forecasts are kept on the device and come back with no signal.

The fire model runs on the device. Start from a point, a perimeter you draw, any shape already on the map (a FIRIS perimeter in Feature Layer, a shape a contact sent), or any NASA perimeter, every piece of it. Set when it ignited if you know, and use your own weather observation or a Kestrel Wx station's if you have one. Runs stay on the device and come back after a restart with no signal. Nothing about a run is sent anywhere unless the operator sends it.

What goes into the model, all fetched for the run from public sources:

  • Ground: LANDFIRE (USGS and the Forest Service) in 30 m cells, for a square 4 km each way around the start (10 km for a three-hour run, up to 20 km each way for a large perimeter): the fuel model (one of the 40 Scott and Burgan models), slope and aspect, the tree canopy (cover, height, base height, bulk density), and the vegetation type, which tells houses, roads and parks apart inside LANDFIRE's single urban class. The fire moves in 15 m steps (30 m for three hours, 60 m for the largest squares).
  • Wind: the nearest weather station that reported in the last three hours, picked by distance and height (RAWS, National Weather Service, and in California the SCE, SDG&E, PG&E and HPWREN utility stations); the National Weather Service hourly forecast for the hours ahead; and, where available, NOAA's HRRR 3 km wind forecast, which varies across the ground and hour by hour. Or the operator's own observation or a Kestrel station.
  • Fuel moisture: dead fuel (1 h, 10 h, 100 h) from the Forest Service's fire danger figures for the nearest station, else the WFAS national grid; live fuel from the nearest National Fuel Moisture Database sample of the last 45 days, else the seasonal figure.
  • Behavior: the Forest Service Fire Lab's public domain Behave library (the engine in BehavePlus), compiled into the plugin through the NDK: Rothermel surface spread, Scott and Reinhardt crown fire, Albini ember distances.
  • Spread: cell by cell over the real ground, each cell at its own fuel, slope, canopy and wind. Embers carry it across a road or block narrower than they fly; roads slow it; commercial blocks, water, bare and farmed ground stop it. Houses burn house to house at the rate measured in nine North American conflagrations (Moran, Seielstad and Pietruszka, 2025).
  • From a perimeter with a known ignition time, the model first replays the fire from ignition to when the perimeter was mapped, compares that with the area it really covered, and scales its rates to match, at most three times faster or slower. The page says when it did, and when it did not.
  • Predicted Fire Size: each run is repeated 30 times with the wind speed, wind direction and fuel moisture nudged the way forecasts miss, ranked by area burned. The operator picks Smallest to Largest; the rings, slider, acres, buildings and people all follow that pick.
  • Not in it: firefighting, wind the fire makes itself, wind bent by terrain finer than the forecast grid, fuel breaks narrower than 15 m, and fuel that has changed since LANDFIRE mapped it.

Buildings and people reached, from FEMA's national building inventory (USA Structures, built by Oak Ridge National Laboratory for FEMA), which lists every structure with its location, occupancy class and a population estimate:

  • Buildings: every structure in the inventory whose location the fire reaches by that time. A building's own cell does not burn, so a building counts when the fire reaches any model cell touching it.
  • Homes: the buildings among them whose FEMA occupancy class is Residential.
  • People, shown as "up to": the sum of FEMA's upper estimate (the top of its 95% range) for each building reached, its median where there is no upper. The upper figure is used because the median is zero for most second homes and cabins, which read as nobody there beside a count of homes. It is how many people could be inside, not a count of who is home at that hour.
  • Named, not just counted: schools, hospitals, nursing and care homes, fire and police stations, hotels, churches, community centers and utility sites the fire reaches, with the time it reaches each.
  • If the inventory cannot be reached, the page says the buildings were not counted; it never shows zero for a count it does not have. Past 8,000 buildings the page says the count is a floor.
  • FIRIS model runs are counted against the same inventory and in the same words; there a building counts when it is inside that hour's FIRIS perimeter.

The detail panel is the point of the plugin. It says how big the footprint is and how that was arrived at, when the satellite saw it and how long it has been burning where the feed tracks that, the radiative power and brightness temperature, what the feed's own confidence and classification mean - and a section headed "What this does not mean", listing the things a heat detection does not say: that the fire is at the center of the box, that an empty map is empty ground, that a hot pixel is a wildfire rather than a flare or a solar farm.

Equipment required

An Android device running ATAK-CIV 5.6, 5.7 or 5.8, with a network connection to fetch detections and run the model. The last set fetched is kept on the device and redraws without one, with its age on the pane.

Works with

Any device ATAK-CIV supports.

Network and security

Outbound HTTPS (TCP 443) only; nothing inbound.

  • services3.arcgis.com, services2.arcgis.com - NIFC's and FEMA's public ArcGIS organizations: the heat detection feeds, weather stations and the building inventory.
  • fire.data.nesdis.noaa.gov - NOAA's GOES fire detections.
  • edcintl.cr.usgs.gov - LANDFIRE fuel, slope and canopy.
  • api.weather.gov - National Weather Service observations and gridded forecast.
  • nomads.ncep.noaa.gov - NOAA's HRRR wind forecast.
  • fems.fs2c.usda.gov, www.wfas.net - Forest Service fuel moisture.
  • data.pyrecast.org - PyreCast forecasts (www.pyrecast.com for the link in the pane).
  • atak.s3.us-west-1.amazonaws.com - FIRIS model runs.
  • openveda.cloud - NASA's satellite fire perimeters (FEDS), for a model run's start.
  • geocoding.geo.census.gov - county and state of each fire in that list (Census Bureau).

No sign-in, no token, no credential of any kind is sent or stored. The device's position is sent only as part of a request to the services above.

More on GitHub →

02 User manual

Download the PDF · The same manual ships inside the plugin, under ATAK’s Tool Preferences.