Each bar is a week's count of fires that NASA satellites detected in eastern and southern Ukraine in places that, before the full-scale invasion, almost never burned at that time of year. Most are likely the war: shelling, strikes and burning positions. In peacetime 2021 the same count averaged under five a week.
Unusual fires per weekAutumn: crop burning, not comparablePeacetime level
For every 10 km square of Ukraine, ten years of detections before the invasion (2012–2021, industrial sites included) show when it normally burns. A fire is counted when its square had no fire within about a month of that date in any of those years, and another such fire appears in the same square within three days; isolated single fires are mostly false alarms. Only the east and south are counted (east of 34.5°E, and south of 48°N east of 31.5°E), because in the north drought wildfires in forests and the Chornobyl zone look the same as shelling.
What it can't see. Clouds hide fires, a satellite passes a few times a day, and small fires escape it. From September to November farmers burn crop stubble across the same fields, and the war's signal is lost in it: peacetime 2021 had more "unusual" autumn fires than autumn 2022. The weekly live view starts in December, when that season is over.
Data: NASA FIRMS, VIIRS on Suomi NPP, through 2025. A count of fires, not of attacks. The approach follows The Economist's use of satellite fire data to track the war.
Each circle is a cluster of places where news outlets reported armed violence — fighting, strikes, bombings, assassinations — in the selected window. The data is the GDELT 2.0 Event Database, which reads world news and updates every 15 minutes.
Size is the number of distinct reports. Colour is the trend: the last three days against the four before. New means there was little or no reporting from that place before the last three days.
This measures reporting, not incidents. A heavily covered place looks bigger than a place with little press, and a single event can be reported hundreds of times. Use it to see where attention and fighting are concentrated, then follow the sources.
Each report is screened by an AI model (Claude, by Anthropic) from its headline and web address. Reports that turn out to be about crime, politics and diplomacy, or something else entirely are set aside. When the model cannot tell, the report stays — the check removes only what it can identify.
Events placed inside the United States are excluded. In this data, US violence codes are dominated by crime and mass shootings and by Washington-datelined policy stories, not armed conflict. US forces operating abroad still appear, placed where they acted.
Headlines are rebuilt from each article's web address. Follow the link for the original.
From U.S. officials lists statements from the Pentagon, the State Department and the regional military commands (via DVIDS) from the last 30 days, matched to a hotspot when their title names its place or country. They are what the U.S. government said, shown beside what the press reported, not a confirmation of either.
The flights layer shows one full day of U.S. military airlift, tanker, surveillance and bomber flights, published the day after they flew. Positions come from adsb.lol's public archive of aircraft transponder broadcasts, picked up by volunteer receivers, under the Open Database License.
It shows only what aircraft broadcast. Many military flights switch their transponders off or fly where no receiver hears them, especially over oceans and conflict zones. Training flights and helicopters are left out to keep the map readable. Executive transport and aircraft whose owners opted out of tracking are never shown. Tracks are thinned and rounded to about a kilometre.
Aircraft broadcast how accurate they believe their own position is. The interference layer shades each one-degree square where more than one in ten aircraft reported degraded satellite positioning over a full day, counting only aircraft that navigated normally elsewhere that day. It uses the same adsb.lol archive, a day behind, from half of all aircraft, and follows the method of gpsjam.org.
Airspace warnings are the European Union Aviation Safety Agency's Conflict Zone Information Bulletins: where it tells airlines not to fly, or to take particular care. Click an area for the bulletin's own wording, altitudes and expiry. Shapes follow countries, provinces and seas, so they are approximate; where a bulletin covers only part of a country's airspace, the map says so. Source: EASA.
The cause is not in the data. Most of it is jamming, but tests and faults look the same. Spoofing, where receivers are fed a false position they trust, is not detected. Areas with no flights or no receivers are blank, not clear.