How Long the News Cares

What it measures

For each major disaster, it counts the news articles about it every day. It finds the day with the most coverage, then counts the days until coverage fell to half of that peak. That number is the attention half-life. A short half-life means the world looked away quickly.

For floods, storms, fires and droughts, it also shows how long the disaster itself lasted, so you can see whether the news left before the disaster did.

Step 1. The list of disasters

The disasters come from GDACS, the Global Disaster Alert and Coordination System run by the United Nations and the European Commission. GDACS gives every earthquake, tropical cyclone, flood, volcanic eruption, drought and forest fire an alert level of Green, Orange or Red, based on its expected impact on people.

The page uses every Orange and Red alert from the last 365 days. Green alerts are left out: they are small events that the news is not expected to cover. Each disaster keeps the name, dates, countries and severity that GDACS gives it.

GDACS sometimes splits one long flood or fire into several events. When two events of the same type in the same country overlap or are less than four days apart, they are joined into one disaster. Events spread over more than 3 countries are left out, because news about a drought across half a continent cannot be pinned to one event.

Step 2. The news

The news comes from GDELT, a project that reads news websites in most countries of the world and records what each article is about. This page uses GDELT's daily file of English-language news, which covers tens of thousands of articles a day. For each article, GDELT lists the places it names and the themes it covers, such as NATURAL_DISASTER_FLOOD.

An article counts for a disaster when it has a theme for that type of disaster and it names a place in the affected country. A flood in Nepal is matched with articles that have a flood theme and name a place in Nepal.

The news on file starts on 9 Sep 2025 and runs to 6 Oct 2026.

Step 3. The normal level

Some countries are in the news for floods or fires almost every day. To measure the coverage caused by one disaster, the page first works out what normal looks like. It takes the daily counts from 21 to 8 days before the disaster started, and uses the lower quarter of those days. The lower quarter, rather than the average, stops leftover coverage from an earlier event from setting the bar too high.

Everything above that normal level is counted as extra coverage caused by the disaster.

Step 4. The peak and the half-life

The daily extra coverage is smoothed over three days, so one odd day does not decide the result. The peak is the highest smoothed day between three days before the start (storms are covered before they land) and 30 days after it.

The half-life is the number of days from the peak until the smoothed coverage first drops to half of the peak or lower. The page also records when it drops to a tenth of the peak, as the point where the news has gone quiet.

If the peak is fewer than 15 extra articles a day, the disaster is marked as barely covered and no half-life is given. A half-life of something that never rose is not a meaningful number. Being barely covered is a finding in itself, so these disasters stay on the page.

What is left out of the totals

• Disasters that started less than three days ago. There is not enough news yet to measure.
• Disasters that started soon after a similar one in the same country. Their coverage runs together, so the normal level and the peak cannot be trusted. They are shown, marked as not measured.
• Disasters with less than a week of news on file before they started.

What the numbers cannot tell you

• English only. The news is in English. Disasters in places where English is widely used, or that matter to English-speaking audiences, will get more coverage. The gap between regions is partly about language, not only attention.
• How long a disaster lasts is not the same as how long people need help. GDACS records how long the hazard lasted: the flood, the storm, the fire. Rebuilding takes months or years after that. So the comparison here is generous to the news. The real gap is larger.
• Matching is by country and theme, not by event. An article about a different flood in the same country, in the same week, is counted too. Joining overlapping events and leaving out disasters that follow each other closely keeps this small, but it does not remove it.
• Natural hazards only. Wars, famines and epidemics are not in GDACS, so they are not here. They could be added with data from ReliefWeb, which needs an approved access key.
• GDELT is not all of the news. It reads a very large set of websites, but not every outlet, and not print or broadcast. Its daily volume also changes over time.

How it updates

Every morning a job on GitHub Actions downloads the newest GDELT daily file and the latest GDACS alerts, recomputes every number on this page, and publishes the site again. Nobody edits the results by hand.

Get the data

Download everything on the findings page as JSON, including the daily counts behind every chart. The settings used are in the file too.