What Counts as a Disaster — and Why Some Never Enter the Record

WND Editorial Team Published: Aug 23, 2026 Updated: Aug 23, 2026 8 min read

Direct answer: the number of disasters recorded in a year is not the number of disasters that happened. Every global database applies entry criteria, and an event that stays below them never appears — no matter how bad it was for the people in it. The most widely used record, EM-DAT, maintained by CRED at UCLouvain, admits an event if at least one of four conditions is met: ten or more deaths, a hundred or more people affected, a declared state of emergency, or a call for international assistance. Miss all four and the event is real, documented locally, remembered for a generation, and statistically invisible.

Every disaster database has a doorway

This surprises people, because a disaster feels like a category of event rather than a category of record. It is not. A database is a set of rows, and something has to decide which events become rows. That decision cannot be made by judgement case by case across a century and two hundred countries, so it is made by rule.

The rule has to be reachable from the outside, too. Nobody at a global database observes the event; they read what was published about it. So the criteria are built out of things that leave a paper trail — a death toll, an affected estimate, an official declaration, a request for help. Each of those is a signal that somebody with authority wrote something down.

Which means the doorway is really two doorways in series. An event must be large enough to qualify, and it must have been reported by someone whose report the database can find.

The events just below the line

Picture a flood that takes out forty houses in one valley. Three people die. A few hundred are wet, displaced for a fortnight, and back in damaged homes by the end of the month. The district handles it. No state of emergency is declared, because the district did not need one. No international appeal goes out, because nobody was going to appeal for forty houses.

That event crossed none of the four criteria on deaths or declarations, and whether it crossed the “affected” one depends entirely on who counted and how — which is its own problem. It may well not enter the global record at all. For the families in the valley it was the worst week of their lives. In the annual total it is nothing.

Now notice that this is not a rare shape. It is the most common shape a disaster takes.

Small and repeated is where most of the harm lives

Attention, funding and record-keeping all sort by size. Harm does not. A hazard that arrives every single year and takes a little each time — seasonal flooding in the same low ground, a landslide on the same slope, heat that kills quietly among people who were already ill — can remove more from a community over a decade than one catastrophic event, and it does it without ever producing a headline, a declaration or a database row.

This is sometimes called extensive risk, as opposed to the intensive risk of the rare large event, and the distinction matters because the two are managed by completely different means. Intensive risk is managed by warning and evacuation. Extensive risk is managed by drainage, building standards, land use and where people are allowed to live — slow, unglamorous work that never has a moment of urgency attached to it.

A record that under-counts the small and repeated will, over time, make the case for the slow work look weaker than it is.

The criteria are not the mistake

It would be easy to read all of this as an argument that disaster statistics are unreliable and should be discounted. That conclusion is wrong, and on a site whose own history pages are built from exactly such a record it would also be dishonest.

Criteria are what make a record usable. Without a threshold there is no consistent unit, and without a consistent unit nothing can be compared to anything — not one year to another, not one region to another, not one decade to the next. A database that admitted every event anyone described as a disaster would not be more complete; it would be a pile of incommensurable reports whose totals moved with reporting effort alone.

So the criteria are doing their job. The error happens one step later, in the reading — when a count of recorded events above a threshold is quoted as a count of disasters, and the gap between those two sentences goes unmentioned. That is a sentence problem, not a data problem, and it is entirely fixable by saying what the number is.

The related trap is a different one and worth keeping separate: older records look calmer than recent ones because observation improved, not because the world was quieter. That is about the record changing over time. This is about the record having an edge at any given time. Both are true at once.

Two databases, two totals, both correct

Because criteria differ, totals differ. A record built for insurance loss will admit an event by monetary damage and ignore one that killed people and destroyed nothing insured. A record built for humanitarian response will admit an event that triggered an appeal. A hazard-specific network will admit anything its instruments detected, which for earthquakes means thousands of events a year that nobody felt.

Two credible sources can therefore report a different number of disasters for the same year and both be exactly right, because they are counting different things and saying so in their documentation. This is the same shape as two agencies reporting different figures for one event, moved up a level: not a disagreement about an event, but a disagreement about what an event is.

The practical consequence is that a disaster count is only meaningful next to its source. A total with no source attached cannot be interpreted at all, because the reader has no way to know which doorway it was measured through.

What our own count is, and what it is not

We should apply this to ourselves, because it applies to us twice.

We inherit every threshold above us. This site does not observe disasters. It reads what monitoring agencies publish, so anything they did not publish does not exist here either, and every criterion they applied is applied to our pages by inheritance.
We add a filter of our own, on purpose. Some feeds publish at a granularity that would bury everything else — automated fire detections and zone-level weather warnings can run to hundreds an hour. A single policy file decides what is kept, and it excludes by source rather than by hazard type, because filtering type by type cannot be won. That is a deliberate editorial choice and it means our live count is a count of what selected agencies published and we chose to keep.
Our historical pages carry the doorway described above, because they are built from a threshold-based record and cannot contain what that record never admitted.

None of that makes the pages wrong. It makes them specific. What we can promise is the thing that lets you check us: every record names the agency it came from and links the original, so any figure on this site can be traced back to the body that published it.

How to read a disaster count

Ask which record it came from. A total without a named source is uninterpretable, not merely unsourced.
Ask what the doorway was. Deaths, affected, damage, declaration, detection — each produces a different total from the same year.
Do not read a low count as a quiet period. It may be a period nobody was recording carefully, or one in which the harm arrived in pieces too small to qualify.
Do not read a rising count as a rising world on its own. Reporting improves, criteria get applied more consistently, and more of what always happened becomes visible. Physical measurement series and attribution studies are the instruments for that question; an event count is not.
Treat “no recorded disasters” as a statement about the record. It is never a statement about a place.

The honest summary

A disaster database is a filing system with an entrance policy, and the entrance policy is the most important thing about it that almost nobody reads. Knowing where the doorway sits changes how you read every total you will ever be shown — and it makes one particular sentence, the one that begins “there were no recorded disasters in...”, say what it actually says, which is that nothing crossed a line that somebody drew.

This article describes how disaster records are compiled in general. It is not a criticism of any database, and the entry criteria described are published openly by the organisations that apply them.

Frequently asked questions

What are the criteria for an event to enter EM-DAT?

At least one of four: ten or more deaths, a hundred or more people affected, a declared state of emergency, or a call for international assistance. Meeting any one is enough. An event that meets none of them does not enter, however serious it was locally.

So are disaster statistics unreliable?

No. They are specific. Criteria are what make a record comparable across decades and countries — without a threshold there is no consistent unit and nothing can be compared to anything. The mistake is in the reading, when a count of recorded events above a threshold is quoted as a count of disasters.

Why do two sources give different numbers of disasters for the same year?

Because they admit different events. One record may enter events by insured loss, another by humanitarian appeal, another by instrument detection. Both totals can be exactly right and neither is comparable to the other. A disaster count is only meaningful next to its source.

What is extensive risk?

Frequent, small-scale, often recurring events — the same low ground flooding every year, the same slope slipping — as opposed to the rare large event. Over a decade extensive risk can remove more from a community, and because each occurrence is small it is systematically under-recorded and under-funded.

Does a country with no recorded disasters have no disasters?

It has no records. That can mean genuinely low hazard, or it can mean events that stayed below the threshold, or that no authority published figures a global database could find. The sentence is always a statement about the record, never about the place.

Is the count on this site a count of disasters?

No, and in two ways. We do not observe events, we read what monitoring agencies publish, so we inherit every criterion they applied. We then filter further — some feeds publish hundreds of automated detections an hour and would bury everything else — so the live figure is a count of what selected agencies published and we chose to keep. Every record names its agency and links the original.

If small events are under-recorded, how should I judge risk where I live?

Local memory and local records are better instruments than a global total for this specific question. Municipal drainage histories, insurance and building records, and what long-term residents remember about which streets flood will describe recurring local hazard far better than any global database, which was never built to see events at that scale.

Sources

About the author

WND Editorial Team — The World Natural Disasters editorial team curates disaster information from official scientific and humanitarian sources, reviews safety guidance against public-agency recommendations, and maintains the historical database.

Written and reviewed under our editorial policy. Factual claims cite the sources listed above.

Found an error? See our corrections policy.