Emberline

NASA FIRMS · 2003–2025

23 years of fire.One steady line

In 2012 the fire satellites changed and the counts jumped overnight. The fires did not. Emberline repairs the break, country by country, so you can tell whether a month is really unusual.

The record · fire detections per year

Australia · 2003–2025 · 2003–11 average = 1

2.95.712012 · VIIRS200320052012201520202025

Joined naively (MODIS to 2011, VIIRS from 2012), Australia's fire count jumps ×3.69 in 2012. The sensor changed, not the fire.

The story ↓The proof ↓Your area ↓The tools ↓Live globe →

At a glance, read from the data files

One day of fire, and the long record it joins.

The live count is today's raw satellite view. Emberline's job is the long record: every year from 2003 to 2025 converted to one unit (Aqua MODIS-equivalent), so a fire season in 2005 and one in 2025 can be compared on the same footing. 2001–2002 are not complete Aqua reference years (Aqua launched in 2002).

Near-real-time, raw detections, last 24 hours

loading

active-fire detections Each one a satellite pixel flagged as burning, seen from orbit in the 24 hours before the snapshot.

As of not available

Source: NASA FIRMS near-real-time global 24 h feeds — VIIRS on NOAA-20 and NOAA-21, MODIS C6.1 on Terra and Aqua. Raw and uncorrected; not the Suomi-NPP record Emberline harmonizes, whose live feed is empty. Includes volcanoes and gas flares.

Time now, UTC

loading UTC

Satellite fire records are stamped in UTC. Emberline converts each detection to local solar time (UTC + longitude ÷ 15) before any time-of-day analysis.

The record

loadingdetections

Daytime vegetation-fire detections, raw, loading, in loading FIRMS countries and territories; loading of them have enough data to fit. Exact total —.

Source: NASA FIRMS archive (MODIS C6.1, VIIRS Suomi-NPP) · countries.json

Fit on test years, median country R²

loadingR²

The web app · who it is for

Pick an area. Ask a question. Get an answer you can take away.

Choose a country, draw your area of interest on the map, or type latitude and longitude, then a date range. Every view downloads: the calendar and the hotspots as CSV, charts, and a brief that opens with the findings.

The story

How joining two satellite records creates a jump.

Four short steps, then the jump itself, then the repair. Scroll through, or skip to the proof.

01, From orbit

Five satellites circle Earth fourteen times a day.

Terra and Aqua carry MODIS; Suomi-NPP, NOAA-20 and NOAA-21 carry VIIRS. Each crosses a given place at about the same local time every day.

02, Heat

They don't see flames. They see heat.

A fire's radiated power, in megawatts, shows up in the infrared. Big fires shine; small ones barely register.

03, 2000

MODIS sees the world in 1 km squares.

A small crop burn inside a square is too faint to cross the threshold. For twelve years, it simply isn't in the record.

04, 2012

VIIRS sees in 375 m squares. Seven times sharper.

Suddenly the small fires appear. Join the two records and the count jumps — not because there is more fire, but because there is more seeing.

05, The jump, Canada, detections per year

A naive join says fire exploded.

2012 · VIIRS record begins

—

Yearly detections, 2012–14 against 2009–11, with MODIS and VIIRS naively joined.

—

Harmonized. Aqua MODIS alone, an unchanged sensor, says —. Much of the jump was the sensor.

Much of it was detection, not fire. Keep scrolling.

06, Still shifting

And the old satellites are drifting.

A satellite in a sun-synchronous orbit passes over a place at the same local time every day. Terra and Aqua no longer do. Each clock below is measured from the fire detections themselves, year by year.

What the hand shows. When each satellite passes over Australia: the average local solar time of its daytime fire detections that year (UTC + longitude ÷ 15, per detection). Noon is at the top.

Why it matters. Fires burn hardest in the early afternoon (shaded). Since 2022 Terra and Aqua have left their fixed orbits: Terra now passes earlier, Aqua later. Each sees a different slice of the day's fire — an orbital artifact, not a change in fire.

Why Emberline is not exposed. From 2012 the harmonized record is built from Suomi-NPP, which has held its overpass, and the model is fitted on 2012–2021 only, before the drift began.

—

frozen orbits (2012–2021)

Terra MODIS

—

2012–21
—
2025
—
Shift
—
Nominal
10:30

Aqua MODIS

—

2012–21
—
2025
—
Shift
—
Nominal
13:30

Suomi-NPP VIIRS

—

2012–21
—
2025
—
Shift
—
Nominal
13:30
Shaded: typical daily fire peak, around 13:00 local solar time (published diurnal cycles, Andela et al. 2015)Times: measured over Australia from the detections · 2012–21 = mean of the frozen-orbit years · nominal = design overpass, not measured

Every satellite, and the drift year by year →

03, Proof

Three checks the repair could have failed.

None of them uses the fitting target. Each asks one thing: after the repair, does 2012 still look special?

Step collapse

not recorded

Loading proof.json

False-Record Counter

not recorded

Loading proof.json

Find the Seam

not recorded

Loading proof.json

Four corrections, the same held-out test

The sensor jump collapses onto the unchanged sensor

Each country's change in yearly fire counts from 2009–11 to 2012–14. Across: Aqua MODIS alone, a sensor that did not change. Up: the same step in our harmonized record (ember) and in the naive splice of MODIS onto VIIRS (blue). On the diagonal the record moves exactly as the unchanged sensor does. Hover or tap a dot.

Scores on years kept back for testing

The four countries the method was built on.

Worked example, for a land manager or emergency planner

How unusual was that month?

Five steps with real numbers. This is historical context: it describes the record, it does not predict fires.

04, Your area

Is this fire season really unusual where you are?

Pick a country. The calendar is in one unit for every year, so the latest season can be set fairly against the ones before it.

Canada, harmonized detections per month, 2014–2025

Is the fire season starting earlier? Less than the raw data says

Chile, season onset trend, 2003–2025

Raw recordnot recorded
Harmonizednot recorded

Negative = earlier. Descriptive least-squares trends. Every country →

Everything inside

One record. Ten ways in.

Pick a place. See its fires.

Draw any area on the map, choose the years, and download the harmonized calendar as CSV. No account, no API key.