The Long Swell
Fires NASA FIRMS SNPP, wk Sep 19-25

Uganda is burning at 3.9x its average for this week of the year

Where the detections fell. 5.0% of this week’s detections fell on cropland against the 20.0% of this country that IS cropland, so they sit on farmland LESS often than chance. Whatever is burning here, it is mostly not farmland. The instrument sees a thermal anomaly, a time and a radiative power. It does not see what is alight, so this is an inference from where the heat sits against a land-cover map, not a measurement of what burned.

Every figure below compares a country only with itself, against the same calendar week in each of 14 earlier years, 2012 to 2025. That is what separates an unusual week from a merely busy one.

Against their own record

3.9x Uganda Second-heaviest of 15 seasonson farmland 0.25× less often than chancebehaves like fire: 0.0% of detections recur in the same cell 3.8x Papua New Guinea Second-heaviest of 15 seasonscropland: not assessedbehaves like fire: 2.4% of detections recur in the same cell 1.9x Republic of the Congo Just past its 2015 record weekcropland: not assessedbehaves like fire: 0.2% of detections recur in the same cell 1.7x Mexico Second-heaviest of 15 seasonson farmland about as often as chancebehaves like fire: 33.3% of detections recur in the same cell 1.4x Colombia Just past its 2012 record weekcropland: not assessedbehaves like fire: 0.0% of detections recur in the same cell

High against their own average, and nothing else agrees

2.8x Indonesia Second-heaviest of 15 seasonson farmland 0.46× less often than chancebehaves like fire: 16.8% of detections recur in the same cell ENSO-loaded window 2.2x Republic of Serbia Third-heaviest of 15 seasonssmall baseline; rank is the sturdier readingon farmland 3.10× more often than chancebehaves like fire: 2.6% of detections recur in the same cell

We take three readings of each country: how it compares with its own average for this week of the year, whether it is setting a record, and how far it stands out from its own year-to-year spread. For the countries above, at least two agree. For these, only the first does. That does not make them quiet, and several are having one of their heaviest weeks on this record; it means the evidence is thinner, so they are listed apart rather than ranked among the others.

One email a week

We find climate signals in the data, and send you the ones that matter.

Confirmation email required. No spam, and every figure here stays free and public whether you subscribe or not. What you get, and how often →

Large, and within their own normal range

1.0x Democratic Republic of the Congo Fire week at 1.0x the seasonal normwithin its historical range; shown for scaleon farmland about as often as chancebehaves like fire: 0.2% of detections recur in the same cell not ENSO-linked

These countries are burning at or above their own average for this week and did not clear the anomaly gate. They are here because a fire map that omitted them would misrepresent where the world is burning, and they are listed separately because presenting them as anomalies would misrepresent whether it is unusual.

Shown every week, whatever they are doing

1.4x Spain Fourth-heaviest of 15 seasonsshown every week so this country can be checked, whatever it is doingon farmland about as often as chancebehaves like fire: 21.1% of detections recur in the same cell not ENSO-linked 1.2x France Fourth-heaviest of 15 seasonsshown every week so this country can be checked, whatever it is doingon farmland 0.28× less often than chancebehaves like fire: 23.2% of detections recur in the same cell not ENSO-linked 0.6x United Kingdom Fire week at 0.6x the seasonal normshown every week so this country can be checked, whatever it is doingonly 16 detections on farmland, too few to compare against chancebehaves like a fixed source: 19.3% recur in the same cell not ENSO-linked 0.5x United States of America Fire week at 0.5x the seasonal normshown every week so this country can be checked, whatever it is doingon farmland about as often as chancebehaves like fire: 21.2% of detections recur in the same cell not ENSO-linked 0.1x Canada Fire week at 0.1x the seasonal normshown every week so this country can be checked, whatever it is doingon farmland 1.76× more often than chancebehaves like fire: 11.8% of detections recur in the same cell not ENSO-linked

These countries are on the page every week so that a reader can check them rather than take our word for which places matter. Their presence here is not a signal in either direction; read the week each one is having off its own row.

How big the season is, for the European Union

A single week can add a third of a season. The largest weekly increment has been a median 25% of the whole season across prior years, and in 2025 weeks 32 and 33 together were 57% of the record. A point estimate would be false precision over exactly that.

2025 record 1035krec 1035kmedian 839kmed 839khighest analog 1131khigh 1131klowest 665klow 665k2026, 623k so farAprJunAugOct

5 of the 20 prior seasons would have finished above the 2025 record from here, and the median lands below it. This season could exceed the record; it is not on course to.

European Union only. United Kingdom (EFFIS reports it under Non_EU, not EU). It is currently the most extreme country in this channel's detection record, so a total that omits it must say so.

Basis. EFFIS's published burnt-area product, aggregated and projected here. The detection charts beside this are Measured. D-033. EFFIS mapped burnt-area perimeters, weekly.

Method. Every complete prior season's multiplier from this same week to its own year end, applied to where this season stands. No fitting and no parameters. The same method analog.py uses for the ONI, on a different quantity. An envelope of what prior seasons did, not a probability distribution and not a forecast. Twenty seasons is twenty samples and the fire regime is not stationary. The median is what a typical season did from here, never what is expected.

Stability. Leave-one-out over the 20 seasons moves the minimum by 2%, the median by 1% and the maximum by 4%. No band edge depends on a single season, including the maximum.

Known biases, both running toward the more alarming answer.

Ranked by how far above normal, not by size. Angola, DR Congo and Zambia each recorded more detections this week than anywhere on this list, and none of them appear on it: the southern African savanna burns on that scale every dry season, so all three sit at or near their own normal. Volume is not news; departure is.

How these figures are built, including the thresholds this page applies and what the instrument cannot see. Every country we have assessed, including the ones not currently qualifying. A country dropping out of a week's findings is not the same as our never having looked at it.

What a detection is. One satellite pixel, about 375 m across, seen radiating enough heat to be flagged as actively burning on a single overpass. A large fire front registers as many pixels at once, and a fire that keeps burning is counted again on every pass it stays hot. Detections are not burned area.

Method. NASA FIRMS, VIIRS on Suomi-NPP only so that every year is measured by the same sensor, low-confidence detections excluded, detections assigned to countries by national boundary polygons applied identically in all years. The window is seven whole UTC days, refreshed once daily, so no figure here is a partial day.

Attribution. Southern and western European fires are driven by heat, drought and wind, and people supply nearly all ignitions. Canadian boreal fire has its own drivers. None of this week's items is presented as an El Nino signal; each carries its own tag.