Skip to content
worldfires
MG · Madagascar · –

Amparafaravola fire, October 2026 Amparafaravola

Dying down · last detection 15 Nov 2026, 10:21 UTC

The fire started on 19 October 2026 in Amparafaravola, Madagascar, and satellites saw it burning for 27 days. Copernicus’s global GWIS system rebuilt a burnt perimeter of 2,078 hectares from heat detections, spread across a single municipality.

Number 115 in Madagascar in 2026, of 131.

GWIS near-real-time perimeter, from the layer that changed in August 2025: this fire is not compared with fires from earlier years, because since then the GWIS layer gives more and larger perimeters.

Compare with…

Is it still active?

No detections since 15 Nov 2026, 10:21 UTC. That usually means no flame is visible from space any more, although hotspots may remain. Satellites pass a few times a day and cloud or smoke can hide the fire: a lack of detections is not an official extinction. Each country’s emergency service declares a fire contained or out.

Burnt area
2,078ha
GWIS perimeter (Copernicus)
Duration
27days
19 Oct 2026 – 14 Nov 2026
Municipalities hit
1
districts
Started on
19 Oct
2026
Heat detections
60
in a single hotspot: 34 MW
Danger at start
31
High · FWI index
Average daily power
32MW
detected power per day, on average

Where it burned

HOTSPOTS BY DAY 19 Oct14 Nov
15 Nov 2026

How it evolved, day by day

19 Oct 2026 · 2 MW · 1 20 Oct 2026 · 0 MW · 0 21 Oct 2026 · 13 MW · 2 22 Oct 2026 · 7 MW · 1 23 Oct 2026 · 45 MW · 6 24 Oct 2026 · 116 MW · 7 27 Oct 2026 · 2 MW · 1 28 Oct 2026 · 87 MW · 9 29 Oct 2026 · 62 MW · 9 31 Oct 2026 · 53 MW · 6 1 Nov 2026 · 15 MW · 4 2 Nov 2026 · 23 MW · 4 3 Nov 2026 · 3 MW · 1 4 Nov 2026 · 0 MW · 0 5 Nov 2026 · 0 MW · 0 6 Nov 2026 · 34 MW · 1 7 Nov 2026 · 58 MW · 5 10 Nov 2026 · 0 MW · 0 12 Nov 2026 · 9 MW · 1 13 Nov 2026 · 0 MW · 0 14 Nov 2026 · 5 MW · 1 15 Nov 2026 · 11 MW · 1
19 Oct 24 Oct 1 Nov 6 Nov 14 Nov

Fire radiative power detected each day, in megawatts, and number of hotspots. The magenta bar marks the peak day.

Fire front film How it spread from one satellite pass to the next (41 passes) and with what wind. Watch the film

The weather it burned in

33°high
52km/h · peak gust
23%lowest humidity

It had gone 2 days without rain (1 mm or more) when it started. Rain came after the fire: 2.6 mm on 15 Nov.

This area had burned before

No earlier perimeters overlap this fire since 2016, when the satellite record begins.

How big is it?

10 km
It burned the equivalent of 1.3% of the area of London (21 km² against 1,583 km²).

London boundary: geoBoundaries · Office for National Statistics (Open Government Licence v3.0).

This compares size, not location: both shapes are at the same scale, north up, centred on each other.

What happened to each town

  1. Amparafaravola where it started
    2,092 ha reached on 19 Oct
    0.4 %

Burnt area inside each municipality, from crossing the GWIS perimeter with administrative boundaries. The distance runs from the point marking the centre of each area to the nearest fire: staying a kilometre away does not mean it reached the houses.

People and buildings inside the perimeter

GHSL 2025
Built-up area< 0.1 ha
Residents< 20

Inside the perimeter there was no noticeable built-up land and almost nobody lived there (GHSL 2025, satellite-based estimate).

This is an estimate on 100 m cells, made from satellite imagery and census data: it does not mean that homes burned or that anyone was harmed. The perimeter also contains land that never burned. The GHSL 2025 epoch is a projection made from 2020 and earlier data.

Source: Global Human Settlement Layer (GHS-BUILT-S and GHS-POP R2023A), European Commission Joint Research Centre (JRC), © European Union, CC BY 4.0. Sum inside the perimeter computed by WorldFires. human-settlement.emergency.copernicus.eu

The weather day by day

ECMWF ERA5 reanalysis at the fire location, the standard reference for past weather, and the Canadian Fire Weather Index (the same system Copernicus uses), computed by WorldFires from that data. Days after the fire run until the first rain.

Open-Meteo.com (CC BY 4.0) · forecast models, provisional until the ERA5 reanalysis arrives

Day High Min. humidity Peak gust Rain Danger (FWI)
19 Oct 28.4 °C 33 % 32 km/h 0.0 mm High 31
20 Oct 28.6 °C 41 % 41 km/h 0.0 mm High 23
21 Oct 31.8 °C 35 % 30 km/h 0.7 mm High 27
22 Oct 31.7 °C 41 % 35 km/h 0.9 mm High 28
23 Oct 31.2 °C 34 % 31 km/h 0.1 mm High 31
24 Oct 31.4 °C 48 % 42 km/h 3.3 mm High 25
25 Oct 28.6 °C 49 % 42 km/h 2.6 mm Moderate 16
26 Oct 29.3 °C 45 % 38 km/h 5.6 mm Moderate 12
27 Oct 30.3 °C 35 % 31 km/h 0.2 mm Low 10
28 Oct 32.3 °C 23 % 38 km/h 0.2 mm High 23
29 Oct 32.8 °C 25 % 40 km/h 1.5 mm High 37
30 Oct 28.8 °C 42 % 52 km/h 11.6 mm Low 8
31 Oct 30.3 °C 39 % 35 km/h 5.7 mm Very low 1
1 Nov 30.0 °C 42 % 38 km/h 0.6 mm Low 9
2 Nov 30.9 °C 39 % 37 km/h 1.4 mm Moderate 11
3 Nov 30.6 °C 43 % 28 km/h 1.5 mm Low 9
4 Nov 30.7 °C 39 % 37 km/h 1.0 mm Moderate 13
5 Nov 29.7 °C 37 % 36 km/h 0.0 mm Moderate 21
6 Nov 29.0 °C 41 % 35 km/h 0.1 mm High 24
7 Nov 30.2 °C 46 % 35 km/h 0.5 mm Moderate 19
8 Nov 30.1 °C 47 % 41 km/h 4.8 mm Moderate 14
9 Nov 30.1 °C 49 % 34 km/h 2.0 mm Low 6
10 Nov 28.5 °C 44 % 32 km/h 0.3 mm Low 9
11 Nov 29.3 °C 42 % 31 km/h 2.9 mm Moderate 19
12 Nov 29.4 °C 46 % 31 km/h 0.7 mm Low 6
13 Nov 31.6 °C 39 % 35 km/h 0.6 mm Moderate 17
14 Nov 32.0 °C 32 % 32 km/h 0.4 mm Moderate 16
15 Nov after 30.9 °C 40 % 25 km/h 2.6 mm Moderate 16

Before and after, from space

Drag the line to compare the same ground before the fire and after it.

Before · 12 Oct 2026
After · 15 Nov 2026

If the day is cloudy, change the date: each satellite passes once a day. In false colour the burnt ground looks brown or red.

The smoke

CAMS · Copernicus

On 25 October 2026 the air beside the fire reached 30 µg/m³ of PM2.5. That day averaged 18: 1.2 times the WHO daily guideline (15).

PM2.5 peak30 µg/m³
Days over the guideline3 days
PM1031 µg/m³
Optical depth0.65
17 Oct22 Oct27 Oct1 Nov6 Nov
Good Fair Moderate Poor Very poor Extremely poor

Particles from the Copernicus CAMS model at the centre of the perimeter; these are not measurements in each town. The pale bar is the worst hour, the solid one the daily average.

Open-Meteo.com (CC BY 4.0) · Copernicus Atmosphere Monitoring Service (CAMS)

Detections table · 60Every satellite heat detection, one by one, with time, satellite, power and coordinates.
⬇ Download CSV Times in UTC. Each detection is a 375 m pixel: a large fire yields hundreds per pass.
Date and timeSatelliteFRP (MW)LatLon

Frequently asked questions

How long did the Amparafaravola fire last?
According to the heat detections seen by satellite inside the perimeter, it burned from 19 October 2026 to 14 November 2026 (27 days). The perimeter dates it from 21 October 2026 to 7 November 2026.
How much area did it burn?
The perimeter GWIS (Copernicus) rebuilt from heat detections is 2,078 hectares (20.8 km²). It is an estimate: official national figures may differ.
Which municipalities burned?
The fire only entered Amparafaravola (2,092 ha, 0.4 %).
Is the Amparafaravola fire still active?
No fire detections since 15 Nov 2026, 10:21 UTC. A lack of detections is not an official extinction.
What was the weather like?
During the fire the temperature reached 32.8 °C, humidity dropped to 23 % and gusts hit 52 km/h. The fire danger index (FWI) reached 37, class High.
Where does this data come from?
From the GWIS (Copernicus) burnt perimeter, NASA FIRMS VIIRS hotspots and the administrative boundaries of Madagascar (geoBoundaries). It is not an official report.

Where to verify it

A satellite detection is not an official report. Final figures and the fire’s status are published by each country’s emergency service.

Satellite cross-check

History of this page · 1 Changes in area, dates, name, official record or merged pages, compared every night since 8 November 2026.
  1. Page published (2,078 ha).

RSS Changes to this page With cross-checks and a citable version

Does something look wrong on this page?

This page is built from satellite data and official records, and it sometimes gets things wrong: the wrong dates, the wrong town, two fires merged into one, or something that was not a wildfire at all. If you know this fire first-hand (you live nearby, saw it, fought it or reported on it), tell us what really happened and we will check it.

Report an error
What do you think is wrong?

We will only use your email to reply about this report. Privacy.

How we know this
  • Perimeter: GWIS (Copernicus), rebuilt from heat detections rather than mapped on imagery.
  • Dates adjusted with the VIIRS detections inside the perimeter, which said 21 October 2026 to 7 November 2026. 60 Suomi-NPP VIIRS detections on those dates.
  • We have not matched this page with any official record.
  • Weather: ECMWF IFS model, provisional until the ERA5 reanalysis arrives.
  • Built-up area and population come from the JRC Global Human Settlement Layer (epoch 2025, 100 m cells), summed inside the perimeter: an estimate of what was there, not a damage count.

How we build fire pages: methodology

Where this data comes from

The perimeter comes from the Copernicus Global Wildfire Information System (GWIS), which rebuilds it from MODIS and VIIRS heat detections and publishes it under CC BY 4.0. It is not measured on imagery as in Europe, so it is an estimate. Heat detections come from the VIIRS sensors aboard Suomi-NPP and NOAA-20, distributed by NASA FIRMS at 375 m resolution. Municipal boundaries come from geoBoundaries ADM2 (Creative Commons Attribution 3.0 Intergovernmental Organisations (CC BY 3.0 IGO)). The area per municipality is computed by WorldFires by crossing the perimeter with those boundaries, so it may differ from each country’s official figure.