Live · NASA FIRMS
Fire starts small.
We see it first.
Two decades of NASA satellites, finally speaking one language. Open the map, search your city and know when something burns nearby.
- Detections, last 24 h
- —
- Active satellites
- 3
- Updated
- —
01 · The problem
Two eyes, two measures.
One history split in half.
Since 2000, MODIS has logged the planet's thermal anomalies. Since 2012, VIIRS does it in finer detail. Both count fire, but not the same way.
MODIS sees the ground in squares of about 1 km. VIIRS uses 375 m squares and catches small fires the other misses. Adding the two series naively creates a false jump in history: fire seems to rise when only the instrument changed.
And MODIS is near the end of its life: its satellites are running low on fuel and drifting out of orbit. Anyone monitoring fire needs a series that crosses that handover without breaking.
02 · How it works
From satellite to your phone,
no noise in between.
-
1
Capture
On every satellite pass, NASA FIRMS publishes where the surface is abnormally hot. Fieremap reads MODIS, VIIRS NOAA-20 and NOAA-21 and drops what does not look like fire: low confidence, industry, volcanoes.
-
2
Harmonize
Where both sensors watched the same ground, we measure how much each over- or under-counts, region by region and month by month. That ruler stitches MODIS's past to VIIRS's present.
-
3
Alert
Pick a city and a radius. When a new detection appears inside it, or the month strays far from the region's normal, a notification arrives with the place and the time of detection.
03 · The harmonized series
Twenty-five years of fire,
at a glance.
Pilot region: Central Brazil (Cerrado), real NASA data. Left alone, MODIS and VIIRS tell two different stories. Calibrated on the years they overlap, they tell one.
- Hold-out R² (2020+)
- —
- Hold-out error
- —
- Overlap months
- —
Same filters
Only presumed vegetation fire. MODIS with confidence of 30% or more, VIIRS without the low class. Volcanoes and doubtful pixels stay out before any comparison.
Calibrated on the overlap
From 2012 both sensors watch the same ground. We train on 2012–2019 and test on 2020 onward, data the model never saw. The score above is that test.
Honest about limits
It says how much fire a month had compared with normal, not where each fire is. Months before 2012 are estimates and appear faded. One region calibrated so far.
04 · What it is for
One fire history,
four decisions.
A harmonized series is only useful if someone can act on it. Fieremap points it at the people who live, move and produce next to fire.
- Safety
Know what burns near you
Pick a place and a radius. See the nearest detection, how strong it is and when the satellite saw it, then get warned when something new appears.
Live - Transport
Roads and rails at risk
Detections near highways and railways, to flag stretches where smoke or flames can close the way.
Next - Production
Fire over crops and pasture
Detections over farmland and grazing, with the month compared to normal, so a producer sees pressure before the loss.
Next - Forecast
What next month looks like
From the harmonized series: — detections expected in — (likely range —), — of a normal month. Over past months this beats using the normal alone by —: a modest, honest gain.
Live
05 · Alerts
Your city, your radius,
your warning.
No tracking: only the city you choose and the distance that makes sense to you. Alerts on the phone app and on Telegram, with simple sign-up and your data always under your control.
- No password to keep: sign in by email link
- Your location never goes beyond what you typed
- Delete your account and data in one tap
Roadmap: the map and the harmonized series are live; accounts and push alerts are being built.