How we build fire pages

Each fire page is a fire of a certain size since 2016. It starts from a burned-area perimeter published by a public service (Copernicus in Europe and the rest of the world, official agencies in the United States and Canada), adds the heat detections seen by NASA’s VIIRS satellites, overlays it on the country’s municipalities and works out the weather with the ERA5 reanalysis.

No person reviews this page by page: these are automatic rules, the same for every fire, which we correct when we find mistakes. The figures are satellite measurements and may differ from official statistics.

The perimeter is the outline of the burned area, and the area on the fire page is taken from it. It comes from a different source depending on the region:

• Europe, North Africa and the Middle East: EFFIS (European Forest Fire Information System, Copernicus), which maps the burned area on satellite imagery. It covers 43 countries and has nothing before 2016.
• Rest of the world (Latin America, Asia, Africa and Oceania): GWIS (Global Wildfire Information System, Copernicus). Its polygons are rebuilt from MODIS and VIIRS heat detections rather than mapped on imagery, and they arrive about 4 hours late. If a polygon includes pieces thousands of kilometres away, we keep only those within 2 degrees of the largest one.
• United States: official public-domain perimeters. For closed years, MTBS (US Geological Survey and US Forest Service), which runs one or two years behind; for years MTBS does not have yet, NIFC WFIGS, one perimeter per incident.
• Canada: CWFIS NBAC (Natural Resources Canada) for closed years and GWIS for the current season.

Prescribed burns flagged as such by MTBS, WFIGS or NBAC do not get a fire page: they are not wildfires.

Only fires above an area threshold that we set per country (paises.min_ha), and that never goes below 100 ha, get a fire page. Where the perimeter comes from GWIS the threshold is higher, from 200 to 5,000 ha depending on the country: in the Amazon, the Chaco or the African savanna, thousands of farm burns every year would pass the European threshold and fill the map with patches that are not wildfires.

If we raise a country’s threshold, the fire pages that fall below it are withdrawn, except those with a proper name set by hand. Smaller fires do count in each country’s official yearly series, but not in the lists or in the totals of fire pages.

A large fire can arrive split into several polygons, mapped on different days. We merge them in three passes:

• Two perimeters are the same fire if their bounding boxes overlap by at least 50% of the smaller one and their dates overlap with a 4-day margin.
• Next, two groups with dates less than 3 days apart whose geometries touch (within about 100 m) are merged, unless the result would be more than 75 km across: this stops fires that touch each other across a whole district from chaining into a single one.
• Finally, pairs of perimeters that the news or an official record say are the same fire are joined by hand, even if they are apart.

The fire page is named after the municipality where the fire started: the one containing the centre of the detections from the first satellite pass (the first three hours). If there are no detections, it is the municipality EFFIS assigns to the largest perimeter and, failing that, the one with the most hectares burned. A published fire page is only renamed if its municipality is no longer among those affected or if the new one burned earlier. If the news or an official record give it another name (Sierra de la Culebra, Park Fire), that name wins, and the old addresses redirect to the new one.

The starting dates are those of the perimeter, which are sometimes a few days off. We correct them with VIIRS heat detections, but only those inside the perimeter or within 1 km of it:

• We look for detections up to 7 days before the start and 7 days after the end of the perimeter.
• The dates are extended day by day backwards and forwards as long as there are no gaps of more than 2 days without detections, and never more than 7 days beyond those of the perimeter.
• Perimeters without an end date (MTBS and WFIGS) are followed with the detections inside them for up to 150 days; the fire ends at the first gap of more than 10 days without detections.

When detections move the dates, the fire page says so and also gives the perimeter’s dates. A stray detection from another fire that falls inside the bounding box but outside the perimeter changes nothing.

For the figures on the fire page (number of detections, fire radiative power or FRP) we count only detections from Suomi-NPP, the satellite the archive was built with: since September 2026 we also receive NOAA-20 and NOAA-21 and, added together, they more than doubled the number of detections. From 28 April to 2 June 2026 Suomi-NPP has no data, and that stretch is counted with NOAA-20, which carries the same sensor in the same orbit. All three satellites are used for the dates and to tell whether a fire is still active.

A fire page is active if there are detections in the last 48 hours, and “burning” if the latest is less than 12 hours old. The end date is the last day with detections or that of the perimeter: it is not the official extinction, which we only give when an official record publishes it.

Copernicus takes days to publish the perimeter. Meanwhile, if the detections justify it, we create a provisional fire page:

• Detections from the last few days are grouped into cells of about 5 km joined to their neighbours, and a cluster is cut if more than 2 days pass without detections.
• A cluster only gets a fire page with 8 detections and 400 MW of accumulated power, or with 1.5 km² on fire (distinct 375 m pixels, about 14 ha each).
• Outside Europe, the estimated area must also pass the country’s threshold.
• A cluster more than half a degree across is split again by detection density (thousands of scattered detections burn at once in the savanna), and one that crosses a border is split by country.

The area of a provisional fire page is an estimate (pixels on fire times 14 ha) and is marked with “≈”. These pages are not counted in the totals. When the perimeter arrives, the same fire page receives it and keeps its address.

Gas flares, volcanoes, steelworks and mines produce heat detections all year round, and GWIS sometimes maps them as fires lasting hundreds of days. We set them aside with two tools:

• Fixed cells: a 0.02-degree cell (about 2 km) is a fixed source if, in a single year, it had fire in 6 different months and on 15 days or more, or if FIRMS (NASA) flagged its detections as volcano, static land source or offshore for 10 days or more.
• GWIS perimeters: one is a fixed source if half or more of its VIIRS detections are of those types or fall in fixed cells, unless it has 500 or more vegetation detections far from any fixed cell (a real fire next to a flare). Without detections, the centre cell and its neighbours decide if the perimeter lasts 30 days or more.

Fixed-source perimeters do not get a fire page, and no provisional pages are created on top of them. Fire pages withdrawn for this reason answer “410 Gone”: that way it is clear they existed and why they are gone.

Some fire pages are not wildfires but vegetation burns that recur every year. We mark them with a class; they appear in the lists with their label and are not counted in the totals, the rankings or the lists of largest fires:

• Reed bed: reed burns in the Danube Delta (Romania and Kiliia raion, in Ukraine), with at least 80% of the fire page in the delta’s municipalities.
• Pastoral: pastoral burns (écobuage) in the French Pyrenees, from November to April and with no record in the BDIFF, France’s official forest fire database.
• Queimada: winter pasture burns in northern Portugal, from November to April and with no record in the ICNF.
• Heather: heather and upland pasture burns in the United Kingdom and Ireland during the burning season, with less than 25% forest, under 1,500 ha and no name in the news.
• Stubble: stubble or pasture burns in Ukraine, only before the invasion of 24 February 2022, in February–April or September–November, with less than 25% forest and outside the Chernobyl exclusion zone.
• Prescribed: a prescribed burn in the United States or Canada, when the source itself says so.

Heather, stubble and queimada are our own assumptions based on the season, the terrain and the lack of an official record, and the fire page says so. Below each total, a note says how many fire pages are left out.

Each perimeter is overlaid on the finest administrative divisions we have for the country (municipalities, freguesias, counties, districts). For each one we calculate the hectares burned inside it, what share of its area that represents (at most 100%), the first day the fire reached it and how many detections fell in it. We also measure the distance from the town centre (its point in Wikidata) to the perimeter.

In provisional fire pages, which have no perimeter, the split is made with the pixels on fire in each municipality.

The boundaries come from geoBoundaries, OCHA COD-AB, Portugal’s CAOP or OpenStreetMap, depending on the country, and are used in simplified form. That is why the municipalities may not add up exactly to the perimeter: on the coast, at borders or where the simplified boundary clips a piece.

For every fire with a perimeter we add up the built-up area and the resident population inside it, using the Global Human Settlement Layer (GHSL) of the European Commission Joint Research Centre: GHS-BUILT-S (square metres built in each cell) and GHS-POP (residents in each cell), release R2023A, on 100 m cells.

• We use the GHSL epoch closest to the year of the fire: 2015 for 2016 and 2017, 2020 for 2018 to 2022 and 2025 from 2023 on. The 2025 epoch is a JRC projection made from 2020 and earlier data.
• The perimeters of a fire are merged before counting, so overlaps are not counted twice, and each cell is weighted by the share that falls inside.
• Figures are rounded, and below 20 people we say that almost nobody lived there.

It is an estimate of what was inside the perimeter, not of what burned: the perimeter includes unburned land, GHSL spreads census population over the buildings the satellite sees, and it cannot tell which houses survived. That is why we do not rank fires by burned houses.

Where an agency publishes its fires one by one, we match each fire page with those records:

• A record is a candidate if it started between 5 days before the start of the fire page and its last day, and its point of origin falls inside the perimeter or within about 2 km of it. For sources without coordinates, its municipality has to be one of those the fire page touched.
• Records under 5 ha do not count, and each record goes to a single fire page: the closest in distance, date and area.
• If several records fall in the same perimeter (a fire complex), they are added together.

We then compare the official area with that of the perimeter:

• Matching: the larger of the two is no more than 1.25 times the smaller. It is published on the fire page, in its FAQ and in its structured data.
• Close: up to 3 times. It is published as “close records”, with caveats and outside the FAQ.
• Discarded: more than 3 times. It is not published; we use it to review quality.

When the record includes the detection time or the extinction date, the fire page gives them and cites the source. The last day with detections is not replaced by the official extinction: they do not mean the same thing.

The weather for each fire comes from the Copernicus ERA5 reanalysis (0.25-degree cells, about 25 km), which covers the whole world, at the location of the fire. We read it from our own copy of the Open-Meteo archive. ERA5 is published about 5 days late: for the most recent fires we use the ECMWF IFS model, and the fire page marks it as provisional until the reanalysis arrives and is redone.

We request data from 60 days before the start to 10 days after the end, and keep the daily table up to the first rain of 1 mm or more after the fire. With daily values, the midday values the FWI needs are approximated: maximum temperature minus 2.8 °C, minimum humidity plus 10 points and 0.73 times the maximum wind.

The FWI is the Canadian Fire Weather Index (Van Wagner, 1987), the same one EFFIS uses, with its classes. The day-length factors go by latitude band, so that seasons fall in the right place in the southern hemisphere. Its moisture codes carry weeks of memory: if there are fewer than 30 days of data before the start, the FWI is not published.

“Days without rain” are the consecutive days with less than 1 mm before the start.

The largest fires are checked against news reports and official sources, starting with those with the most area. This is done by artificial intelligence assistants in two steps: one searches for and summarises the sources, and another, separately, reopens each one to try to disprove it.

• A fire page is only published with at least two sources confirmed by the second step, and only what that step confirmed.
• The area given by the news or an agency is only shown if it is within 25% of the satellite’s: if it differs more, it is usually only part of the fire.
• Other names for the fire are only given if the news actually uses them.

The sources are linked on the fire page with their date, so that anyone can check them.

Every night, a process checks the consistency of all fire pages without human intervention: that the municipalities add up to the perimeter, that dates and days match, that the bounding box is not absurd, that the municipality of origin belongs to the country and is among those affected, or that the centre does not fall on a fixed source.

Anything that makes a fire page misleading (wrong place, dates or source) is an error: the page stays online with a warning, but is removed from search engines, rankings, the RSS feed and the lists of largest fires. A data gap (no detections, incomplete municipal overlay) remains a warning.

If you see something that does not add up, every fire page has a “Something wrong?” form, and you can also write to us.

• This is not an emergency service. A fire that is not here may exist, and one shown as active may already be out. If you are in danger, call your local emergency number.
• The thresholds leave out small fires, and outside Europe the threshold is high.
• GWIS perimeters are rebuilt from detections: they can merge neighbouring fires or exaggerate the area, and during the current season they keep accumulating.
• VIIRS has 375 m pixels and passes about six times a day across the three satellites: clouds and smoke hide the fire, and a short fire may leave no detection at all.
• A heat detection is not a confirmed fire: farm burning and industry also produce them.
• The burn classes are reasonable assumptions, not official data.
• Matching with an official record can go wrong in fire complexes or where the records have no coordinates.
• The weather is that of a cell of about 25 km, with approximate midday values: it is not the wind that blew at the fire front.
• There are no EFFIS perimeters before 2016, and in some countries the archive is not fully loaded yet.

The sources and their licences are listed in the terms of use; the data we produce can be reused through the API.