Why the Past Looks Calmer Than It Was — and Why You Cannot Count Disasters to Find a Trend

Direct answer: older decades hold fewer disaster records largely because far less was detected, reported and archived, not because far less happened. Seismic networks, weather satellites, global news and humanitarian reporting systems all arrived and then improved, and each improvement added events to the record that would previously have gone unwritten. Counting entries per decade therefore measures the growth of observation at least as much as anything about the world. It is a real question with real answers — but the event count is the wrong instrument for it, and better ones exist.

What the record actually is

A historical disaster database is not a list of disasters. It is a list of disasters somebody documented, in a form somebody later collected. Every entry had to clear four separate hurdles: it had to be detected, judged significant enough to write down, published somewhere durable, and then found by whoever built the database.

Each of those hurdles has moved. A seismograph network that did not exist could not record a moderate earthquake in a remote range. A country with no meteorological service produced no cyclone report. An event in a region with no press and no international presence left no trace that a compiler could later find. None of that means the ground was still.

The small events vanish first, and they are most of the list

Detection thresholds bite hardest at the bottom of the scale, and the bottom of the scale is where nearly all the events are. A very large earthquake or a major cyclone has always been hard to miss — it destroys things people write about. Modest events are the ones that did not survive into the record, and they vastly outnumber the severe ones.

So the shape of an old decade is distorted twice over: it holds fewer entries, and the entries it does hold skew large. That makes any average computed across it wrong in a specific and misleading direction — the past looks quieter, and simultaneously looks like it only produced catastrophes.

Where the event happened mattered more than it does now

Historical records are denser near cities, near coasts with shipping, near colonial administrations that kept archives, and near anywhere with a newspaper. They are thinner in mountains, in deserts, in the interior of large land masses, and at sea. That is a map of who was writing things down, and it is the same bias described for the present day in what “no alerts” actually means — only much stronger, because the reporting infrastructure was much thinner.

Impact figures are worse than event counts, not better

It is tempting to sidestep the counting problem by looking at deaths or losses instead. Both carry their own drift.

• Deaths depend on how attribution was done, which has changed substantially — a subject in its own right, covered in how death tolls are counted. Death tolls have also genuinely fallen for many hazards because warning and response improved, so a downward line here is partly a real success and partly a change in counting.
• Economic losses rise almost mechanically over time, because there is more built in harm's way every decade and because money is not comparable across years without adjustment. A loss series that has not been corrected for exposure and inflation will trend upwards whatever the weather does.

So what is the right instrument?

This is the part that matters, and it is why the section exists.

That the event count cannot answer the question does not mean the question has no answer, and it certainly does not mean nothing is changing. It means the record of who wrote things down is the wrong tool. The right ones are measurements and studies built to avoid exactly this problem:

• Physical measurement series — temperature, sea level, ocean heat, sea ice — are far better constrained than event counts, because an instrument reading does not depend on somebody deciding an event was newsworthy.
• Attribution studies ask a narrower and more tractable question about a single event or a single class of events, using physics and modelling rather than counting entries in a catalogue.
• Homogenised datasets are built specifically to correct for changing observation practice, and they say so in their documentation. If a series has not been corrected, it is not evidence about the world.

The distinction to hold on to: an unreliable count of events is not evidence of stability. Absence of a trend in a biased dataset is not a finding.

What this means for our archive, plainly

Our historical section holds a limited number of documented major disasters. It is a reference for looking up individual events, and it is honestly not a dataset:

• Do not compute frequencies from it. The number of entries in a decade reflects what we have collected, not what occurred. Our methodology says the same thing in the site's own terms, and llms.txt says it to AI systems for the same reason.
• Do not compare regions using it. Coverage differs by region far more than hazard does.
• Do use it to read about a specific event, with its sources named, which is what it is for.
• Live data is a different thing entirely. The current feeds are dense and consistent enough to tell you what is happening now; they are not long enough to tell you what is changing, and we do not present them as though they were.

The honest summary

The historical disaster record is a record of attention. It thickens as instruments, agencies and journalism spread, and that thickening is easily mistaken for the world getting louder. Read old decades as under-observed rather than as calm, treat any per-decade count as a statement about the archive, and take questions about change to the measurements built to answer them.

This article is about the limits of historical disaster data. It takes no position on any specific attribution finding; for those, use the peer-reviewed literature and the physical measurement series rather than any aggregator, including this one.

Frequently asked questions

Why do disaster databases show fewer events in older decades?

Mainly because far less was detected, reported and archived. Seismic networks, weather satellites, national meteorological services, global journalism and humanitarian reporting systems all arrived and then improved, and each improvement added events to the record that would previously have gone unwritten. The thinner past is largely a record of when we learned to observe.

Can I count entries per decade to see whether disasters are increasing?

No. That measures the growth of observation at least as much as anything about the world, and the bias runs in one direction — upwards over time. It is a real question with real answers, but the event count is the wrong instrument. Physical measurement series and attribution studies are built specifically to avoid this problem.

Does this mean nothing is actually changing?

No, and that conclusion does not follow. An unreliable count is not evidence of stability — absence of a trend in a biased dataset is not a finding. It means the archive of who wrote things down cannot answer the question. Temperature, sea level and ocean heat series are far better constrained, because an instrument reading does not depend on somebody deciding an event was newsworthy.

Why are economic loss figures unreliable for trends?

Because they rise almost mechanically. There is more built in harm's way every decade, so the same physical event destroys more, and money is not comparable across years without inflation adjustment. A loss series that has not been corrected for exposure and inflation will trend upwards regardless of what the weather is doing.

Why do death tolls sometimes fall over time?

Partly because warning systems, building standards and emergency response have genuinely improved, which is a real and welcome result. Partly because how a death is attributed to a disaster has changed. Both effects are present in the same series, which is why a downward line cannot be read as purely one or the other.

Are old records biased towards certain places?

Strongly. Historical records are denser near cities, near shipping routes, near administrations that kept archives, and anywhere with a newspaper. They are thinner in mountains, deserts, continental interiors and at sea. That is a map of who was documenting events, not of where hazards occurred.

What is the historical section on this site actually for?

Looking up individual documented events with their sources named. It holds a limited number of major disasters and is deliberately not presented as a dataset. Do not compute frequencies from it and do not compare regions with it — the entry count reflects what has been collected rather than what occurred.

Sources

• CRED / EM-DAT — Database documentation on data quality and coverage over time
• WMO — Atlas of Mortality and Economic Losses from Weather, Climate and Water Extremes
• USGS — Earthquake catalogs, network history and catalogue completeness
• NOAA NCEI — Historical weather and climate records and homogenisation
• IPCC — Assessment reports on detection, attribution and observational uncertainty
• UNDRR — Sendai Framework monitoring and disaster loss data quality
• IDMC — Internal Displacement Monitoring Centre, on the limits of historical displacement data

Found an error? See our corrections policy.