Is the number of natural disasters increasing?

If we want to reduce the risks of disasters, we need to track where they’re happening; what types of events they are; their human and economic impacts; and how these trends change over time.

High-quality data helps us see patterns in the data on factors such as increased resilience, climate change, and humanitarian response.

There are now several dedicated research groups that publish in-depth databases of disaster records.

One of the most widely cited is the International Disaster Database (EM-DAT) of the Centre for Research on the Epidemiology of Disasters (CRED). It is open-access and free and lets anyone dig into the specific details of each recorded disaster. At Our World in Data, we rely on EM-DAT as our main data source for disasters. It’s also used by organizations such as the United Nations, World Meteorological Organization, UNFCCC, and many academic researchers.

But no disaster database is perfect. Data is incomplete. Its quality varies over time. And some events are either unreported or hard to quantify.

That’s why it’s important to understand the biases and limitations of data sources so that they can be interpreted usefully.

Many of them are explained by EM-DAT itself in its documentation.

In this article, we explore several of these biases, which can lead to incorrect conclusions when analyzing historical trends.

EM-DAT publishes data on the number of events, deaths, numbers affected, and other metrics since 1900.

If you look at the chart of the number of disasters over time — shown below — you see a very steep rise since the 1980s.

Many organizations, such as the United Nations and World Meteorological Organization, have reported on this as a dramatic rise in actual disaster events. Here are a couple of high-profile examples:

These points are correct regarding the number of reported disasters but are unlikely when it comes to the true number of events.

EM-DAT — the data source that these reports rely on — says as much in its limitations section:

“Time biases result from unequal reporting quality and coverage over time […] Technologies and initiatives can be considered responsible for the dominant trend observed. Therefore, it is challenging to infer insight into the actual drivers of disasters such as climate change, population growth, or disaster risk management.”

As it explains, CRED wasn’t created until 1973, when it started compiling disaster data. It wasn’t until 1988 that it took over the disaster database and established EM-DAT. This period of trying to develop the first database over the 1970s and 1980s coincides with the steep rise in the number of reported disasters.

EM-DAT also notes that communication technologies developed rapidly over this period — most notably satellites, the Internet, and personal computers. This is likely to have caused a step-wise change in the frequency of reporting across the world. Several publications from CRED over the years warn about the overinterpretation of these trends.²

Many smaller events in the past aren’t captured

In its 2004 report, Thirty Years of Natural Disasters 1974-2003: The Numbers, CRED notes that historical data was based on retrospective analysis, resulting in:

“a list that included mainly events of major importance, as neither humanitarian aid nor telecommunications were particularly developed and few organizations were interested in compiling data on natural disasters. When active registering of disasters took on a more important role, both the larger disasters were recorded, together with increasingly more of the smaller ones.”

We’ve replicated this analysis with the latest data from EM-DAT in the chart below. It shows the share of reported disasters of different sizes as a share of the total. The further we go in the past, the more we see large events dominating the records. Small and medium events were missed at the time and are impossible to identify today. Every new decade, as reporting and communication improved, more small events made it into the reports. You can also explore the annual data.

A recent paper published in Environmental Hazards analyzed the trends in the number of reported disasters in the EM-DAT database since 1900.³ It looked at the number of events over time at different thresholds of “recorded deaths”. So, the number of recorded events with deaths below 200, 500, 1,000, 2,000, and 5,000.

It found that the trend for small events – with less than 200 deaths – increased greatly in the 1980s and 1990s. This suggests that small events are missing in earlier periods of the database, and only in the past 30 or 40 years have they been reported more consistently.

We did not use the data or results from this paper but instead replicated this analysis based on the latest data from EM-DAT. The only difference is that we did not include epidemics to focus on meteorological, hydrological, and geophysical events. The results are shown in the chart below. They mirror the findings, with a significant increase in the number of small events recorded.

We should be cautious about reporting increases in the number of disasters using EM-DAT

EM-DAT itself strongly recommends that users exclude or are cautious about interpreting trends using pre-2000 data because of this consistency issue. In fact, since September 2023, it has labeled all pre-2000 data as “Historic” to differentiate it from recent records.

If you look at data from 2000 onwards, there is no clear increase in the number of global disasters.

To be clear: this does not mean that there is no increase in disasters, especially when looking at specific types of events, or specific locations. To establish clear trends on this, people need to look at more focused academic literature. It may also be the case that other databases do find an increase. It also says nothing about the intensity of disasters.

But EM-DAT shouldn’t be used as evidence that there has been a four- or five-fold increase in the actual number of disasters globally. While there may be an increase, at least some of this is down to improvements in reporting.

EM-DAT publishes a range of indicators: the estimated number of deaths, people left homeless, numbers affected, and economic damages.

However, the completeness of this data varies. The metric where data is largely missing is economic damages. In a paper published in Nature Scientific Data, Rebecca Jones et al. (2022) look at the share of disasters between 1990 and 2020 that have missing data.⁴ More than 40% of the events did not have estimated monetary damages. This was even worse for insured damages, which were missing 88% of the time, and 96% had no records of reconstruction costs.

Unsurprisingly, data coverage tended to be poorer in low-income countries where statistical capacity and reporting are more limited. From 2000 to 2019, only 13% of disasters in Africa and 23% in South Asia reported any economic losses.

Data for total deaths appeared to be much more complete. The Jones et al. (2022) paper reported that just 1.3% of events did not have death estimates recorded. We tried to replicate these estimates using the publicly available EM-DAT dataset. A much higher share of events had no data (i.e., blanks) for recorded deaths: 30% at the time of publication. However, since events in EM-DAT with zero deaths are also recorded as blanks, this estimate includes non-fatal events and those with genuine data gaps. It’s therefore difficult to give a concrete estimate for what share of events have real gaps in their death records.

Users of EM-DAT data should be aware of these potential gaps in data coverage. Particularly for lower-income countries and records from previous decades when reporting was much patchier.

One disaster category that is included in EM-DAT but is likely to be very patchy is extreme temperature. There are several reasons why.

First, the inequalities in reporting on heatwaves across regions are stark. Many regions have poor data coverage, and Sub-Saharan Africa is almost completely overlooked.

Over half of heat events in EM-DAT were reported across only 9 countries: Japan, India, Pakistan, the United States, France, Belgium, the United Kingdom, Spain, and Germany. It’s highly unlikely that these are the only countries experiencing extreme heat events. Such events are just not being recorded or estimated in other regions.

Second, proper quantification of the health effects of extreme temperatures is difficult. We often think about acute or very sudden deaths such as heat strokes. But most heat-related deaths come from an increase in the risk of less obvious conditions such as cardiovascular disease.⁵ These seemingly indirect deaths are not noted as being “heat-related” at the time and can only be estimated using various statistical methods later. This is true for both hot- and cold-related deaths.

There is an emerging literature that tries to quantify heat-related mortality — with projections into the future — but heat deaths are not always captured (at least not fully) in disaster databases.⁶

We therefore suggest caution when using heat death figures. Due to reporting biases, they are likely to be underestimates (for both cold and heat-related deaths) with large inequalities.

Disasters often have near-term and acute impacts on human mortality, health, and infrastructure. However, they can also lead to some indirect, medium-to-long-term impacts that are hard to capture.

This is often the case for events such as droughts, where indirect impacts such as malnutrition, food insecurity, and potential impacts of water shortages are harder to quantify — especially in an ongoing or very recent event. This is explained in more detail in the UNDRR’s Special Report on Drought 2021.

Therefore, users should know that disaster databases may fail to capture the full chain of disaster impacts.

Republished under a Creative Commons BY license from Our World In Data. Read the original article.