The reported count jumped. The historical record changed underneath it.
Are reported disasters changing?
Basis: EM-DAT annual counts, 1970–2025 (supplied OWID extract)
The thin line is the noisy year-to-year count; the bold step line bins it by decade to show the underlying shift. Shaded region = EM-DAT's own "Historic" (pre-2000) caution period. See ① – ④ in Fig. 1 above for the milestones driving each step.
The raw numbers are real: the 1970s averaged 89 recorded disasters a year; by the 2000s that had risen to 394 a year — a 4.4× increase. Much of that climb overlaps with CRED's founding in 1973, EM-DAT's establishment in 1988 and rapid improvements in global communications.
That build-out overlapped with the spread of satellites, the internet, and personal computers — the infrastructure that lets a small flood in a small country get recorded at all. EM-DAT has flagged all pre-2000 data as "Historic" since 2023 for this reason (annotation ③).
After 2000, the record settles into a much narrower band rather than continuing the old climb. The supplied extract peaks at 450 reports in 2023, but six years do not establish a new long-run trend. Use the post-2000 series to ask focused questions — not the pre-2000 jump to claim a fivefold rise in real-world disasters.
Decade breakdown
| Decade | Avg / yr | Index |
|---|
Index = decade average ÷ 1970s average. The climb from 1.0× to 4.4× happens entirely before 2000; the 2020s figure (4.8×) is a separate, later high.
Floods and storms dominate the record — and the reporting problem.
Which types are driving the trend?
Basis: EM-DAT type breakdown, 1970–2025 (supplied OWID extract)
Two hazard families do almost all of the work. Since 1970, EM-DAT has logged 5,977 floods and 4,545 storms — together 72% of every disaster recorded in the 2020s — versus 1,239 earthquakes and 1,208 droughts.
The growth rates tell the sharper story. Comparing 1970s averages to 2015–2024 averages: floods grew 6.5× (26 to 171 a year) and storms grew 4.1× (29 to 118 a year), while earthquakes grew only 2.3× and droughts just 2.1×.
That gap is consistent with unequal historical coverage: large earthquakes were harder to miss, while smaller floods and storms were more likely to enter the record as satellites, communications and local reporting improved. It is also consistent with genuine changes in some weather hazards. This chart alone cannot apportion the difference between those mechanisms.
Total events recorded by type, 1970–2025 (EM-DAT).
Annual average, 1970s vs. 2015–2024 — the growth-multiple bar reveals which types actually drove the rise.
One earthquake or one heatwave can outweigh a decade of ordinary years.
How many people die from disasters?
Basis: EM-DAT global deaths by cause, 2000–2025 (supplied OWID extract)
Global disaster deaths per year, 2000–2025, log scale (EM-DAT). Callouts show which single hazard drove each spike.
Deaths from extreme temperature events specifically — three of the four worst years on record happened in 2022–2024.
Global deaths from disasters bounce enormously year to year, and almost never for a gradual reason. 2004 (242,175 deaths) was the Indian Ocean earthquake and tsunami. 2008 (235,355) was dominated by Cyclone Nargis alone — 140,983 storm deaths in a single year. 2010 (307,988, the worst year in this series) combines the Haiti earthquake with a deadly Russian heatwave. 2023 (135,792) combines the Turkey–Syria earthquake with a brutal year for heat.
Add up 2000–2024 by cause and earthquakes are the single largest killer (790,018 people) — more than five times as many floods happen every year, yet floods killed far fewer (136,843) over the same 25 years, largely thanks to warning systems and evacuation.
The category to watch is extreme temperature: 352,756 deaths since 2000 in the supplied extract, heavily concentrated in a handful of years. The recent clustering is concerning, but heat mortality is difficult to count consistently and EM-DAT captures only a subset of excess deaths. Read it alongside dedicated mortality research, not as a complete trend by itself.
Pre-1970 EM-DAT records include mega-catastrophes rarely discussed today — the 1931 China floods (an estimated 3.7 million deaths) and 1928 (3.0 million) — a reminder that the historic record is dominated by a handful of enormous, imprecisely verified events, which is part of why EM-DAT urges caution before 2000.
The nominal bill rises. The cause does not travel alone.
What's the economic cost?
Basis: EM-DAT global economic damage, 1970–2025 (supplied OWID extract); data completeness from Jones et al. (2022)
Recorded global disaster damage climbs across each decade in this extract: $5.4B → $18.5B → $69.8B → $89.3B → $167.0B → $228.8B in average annual nominal U.S. dollars. The single costliest year is 2011 at $371B, driven by the Tōhoku earthquake and tsunami.
By hazard, storms are the biggest line item globally — $1,865.5B in damage from 2000–2024, well ahead of floods ($799.4B) and earthquakes ($653.8B). Droughts, wildfires, and extreme temperature together add up to a few hundred billion more.
But these totals are not normalized for inflation, population, wealth or the amount of infrastructure exposed to hazards. The IPCC finds that development has placed more assets in harm's way even as climate change contributes to losses in many sectors. The raw dollar curve combines both, plus changes in reporting.
None of this is evenly measured, though: more than 40% of all recorded disasters have no damage estimate at all, and the gap is worst exactly where vulnerability is highest — only 13% of African disasters and 23% of South Asian disasters come with any economic-loss figure attached.
Global economic damage from disasters, 1970–2025, nominal US$ (EM-DAT). 2025 is partial.
Share of global economic damage by hazard, 2000–2024 (EM-DAT).
Share of disasters with any recorded economic-loss estimate, by region, 2000–2019 (Jones et al., 2022 / CRED — not part of the uploaded extract).
The planet just logged its three warmest years ever measured — back to back to back.
How has global temperature changed?
Basis: annual temperature anomalies, 1940–2025 (supplied ERA5/OWID extract), converted to a pre-industrial baseline
Global average surface temperature, 1940–2025, converted to degrees above the 1850–1900 pre-industrial baseline using Copernicus's published +0.88°C offset from its native 1991–2020 baseline. Underlying series: ERA5-based annual anomalies (OWID extract).
This is the one number in this report with no reporting-bias asterisk: reanalysis temperature records are cross-checked by half a dozen independent agencies that all show the same curve.
By this measure, 2024 was the warmest year on record at 1.60°C above pre-industrial — the first calendar year to clear the Paris Agreement's 1.5°C marker. 2023 came in at 1.48°C and 2025 at 1.47°C — meaning the three warmest years ever recorded happened back to back to back, and 2025 was barely 0.01°C behind 2023 despite cooling La Niña conditions for most of the year.
The simple 10-year average (2015–2024) sits at about 1.3°C above pre-industrial. Copernicus estimates the current long-term warming level at around 1.4°C. One calendar year above 1.5°C does not by itself mean the Paris Agreement's long-term threshold has been crossed.
What can — and can't — we conclude about the climate connection?
Put the five readings together, and the honest answer is: it depends which claim you're checking.
Basis: synthesis of Readings 01–05 and IPCC AR6 assessments
Can conclude
- Human-caused warming is virtually certain to be the main driver of the observed rise in global heat extremes — the IPCC's strongest confidence category.
- Specific heatwaves and heavy-rainfall events can be studied with event-attribution methods that estimate how warming changed their probability or intensity.
- Climate change has already caused losses and damages, while exposure and vulnerability shape whether a hazard becomes a disaster.
Can't (yet) conclude
- Raw global disaster counts are not clean climate evidence because the historical rise is heavily affected by improvements in reporting.
- Confidence in a trend varies sharply by hazard: high for heat extremes, low for overall tropical-cyclone frequency, genuinely complex for localized convective storms.
- Global nominal-loss growth is not cleanly attributable to climate change as the primary driver without normalizing for inflation, wealth and exposure.
- Attribution science itself is geographically uneven, with far fewer studies covering the developing world — so an absent signal there isn't proof of an absent effect.
Confidence, claim by claim