One of the sustainable development goals (target 11.5) is to significantly reduce, by 2030, the number of deaths and people affected by disasters. Although evidence suggests that disaster-related fatalities have fallen globally since the 1970s1, the 2022 IPCC report2 called for continued efforts to minimize climate risks and strengthen community resilience. Understanding what prompts countries to take measures to mitigate disaster-related fatalities is thus an urgent scientific and practical challenge.

Scientific and public discourses have increasingly focused on conditions for transformative adaptation3, including measures to reduce societal vulnerability. Disruptive disasters are often depicted as seminal events that provide opportunities for ambitious risk reduction actions, yet scholars debate whether disasters actually play this role. While some studies find that fatal disasters, including major floods4,5,6,7, enhance risk mitigation8, especially if exposing flaws in pre-existing policies9 and sparking debates about causes and consequences10,11, other studies8,12,13 found negligible or no effects. Others yet have pointed to the risk that certain disasters may, in reality, increase vulnerability through maladaptive responses14.

These sometimes-contradictory insights are mainly derived from single-case studies of individual countries or events. Except for a few large-n studies, time-series and cross-sectional studies have been rare. Thus, we lack knowledge of how common it is that major disasters are followed by decreasing mortality in subsequent disasters. One central hypothesis holds that adaptation is generally most likely after large-scale catastrophic disasters. Compared to lower magnitude events, these ‘mega-disasters’ are likely to challenge dominant ideas and policies15. The scale of these events often comes as a surprise, introducing a sense of novelty as previously unimaginable scenarios materialize and expose neglected risks and unanticipated capacity needs. Meanwhile, past studies have only partially controlled for event magnitude or relied on various proxy measures of magnitude16. Whether catastrophic disasters enable adaptation actions that reduce losses from consecutive disasters remains to be explored.

In this study, we assess whether milestone flood disasters – events whose death toll exceeds historical country averages – incentivize countries to take actions that reduce future flood fatalities. The study includes riverine, coastal, ice jam, and flash floods. Floods can enable adaptation by a range of measures that reduce fatalities, typically by targeting human settlement, public awareness, flood defense, and early warning systems17. Decisions about these measures are often taken at the national level18, and much previous work on flood losses consists of country-level comparisons8,17,19. Our study spans the 23 countries worldwide with the greatest number of flood disasters in the period from 1970 to 2021 (Fig. 1a). Some flood-prone countries, for instance, the United Kingdom and the Netherlands, were excluded as the number of floods reported for these countries was below the cutoff point applied in this study. Supplementary Table 1 and Supplementary Table 3 provide details on milestone flood events included in the study.

Fig. 1: Summary of study results.
figure 1

Map (a) shows countries (n = 23) in the study. b shows that most countries are close to the no-change line (dotted) with close to flat trends. Some error bars in (b) extend beyond the plotted interval. Only Indonesia (c) shows a statistically significant difference in mortality rate, with a decreasing trend after the milestone event. All other countries show no significant difference, as illustrated here by Viet Nam (d). Data span all flood types.

Full size image

This sample reduces the confounding effects of event frequency since all countries recorded a large number of flood events in the study period (ranging from 47 events in Algeria to 279 events in India). Most countries here are classified by the World Bank as lower- or upper-middle-income countries, except France and Australia (high-income) and Afghanistan and Ethiopia (low-income). Our study hereby reduces, albeit not eliminates, the influence of development factors, including, for example, income inequality and quality of institutions, which have been shown to reduce flood disaster fatalities17. However, since the sample is too small to control for income level fully, this remains a potential confounder.

By examining the deadliest floods affecting a country over time relative to its historical average, we are able to utilize disaster mortality trend lines as a proxy measure for adaptation action outcomes. This approach enables an assessment of adaptive potential as a reduction in fatalities from subsequent disasters can be seen as evidence of adaptation measures initiated after previous disasters8. Mortality rates were calculated for each milestone event, which normalized fatalities by the number of affected people (Methods). Eight countries (Brazil, Sri Lanka, Argentina, Haiti, Nigeria, Nepal, Bolivia, and Algeria) recorded one milestone flood, while the remaining 15 countries recorded two or more flood disasters during the milestone event year (Supplementary Table 1). There is also a temporal spread of milestone event years in the sample, with the first occurring in 1978 (India) and the most recent in 2013 (Argentina). As a result, the length of the period for calculating mortality rate trends varies across countries.

We estimated bivariate linear regression models of the difference in mortality rate before and after the milestone event year in each country, respectively (Supplementary Table 2). The models confirm that Indonesia (Fig. 1c) is the only country with a statistically significant difference in mortality rate trends before and after the milestone event year (βcoeff. = 0.12, p < 0.05), indicating a small decrease in fatalities in the period after the milestone event year. For Indonesia, this result includes the 2001 floods affecting the island of Nias 1400 km northwest of Jakarta, for which EM-DAT reports 257 casualties and over 3500 affected people. The fact that none of the other countries displayed a significant downward trend indicates that decreased flood mortality after milestone events is rare overall. Figure 2 shows mortality rate trends by country before and after milestone event years.

Fig. 2: Flood mortality trends by country.
figure 2

Mortality rate trends (log10) by country for all floods (riverine, coastal, ice jam, and flash floods) before and after milestone events (vertical dashed line), 1970−2021.

Full size image

The results remain similar when controlling for flood frequency, intensity, and subtypes, suggesting that the trend is robust. Since we focus on the most flood-prone countries globally, the results suggest that mortality trends are unassociated with the frequency of past flood disasters. We performed separate analyses to investigate whether the results hold for riverine (Supplementary Fig. 1) and flash floods (Supplementary Fig. 2), respectively. Results for these analyses, including separate bivariate regression models by country, confirm that milestone events are not generally followed by reduced fatality trends for either one flood subtype. The lack of any statistically significant effects in these models suggests that milestone flood disasters, regardless of flood subtype, are not systematically followed by measures that reduce fatalities from subsequent disasters.

This exploratory study suggests that adaptation policy punctuations triggered by milestone flood disasters are uncommon. When mortality rates are normalized by the number of affected people, most countries are located near the no-change line with close to flat fatality trends (Fig. 1b). This pattern indicates that historically in the most flood-affected countries around the world, milestone flood disasters have generally been unassociated with a reduction in fatalities from subsequent floods. This finding is inconsistent with the assumption that major catastrophic disasters increase the likelihood of adaptation measures that reduce disaster risk20.

The study corroborates results reported elsewhere that higher climate risks do not necessarily lead to fewer flood fatalities8. In addition, this study explores mortality rate trends after the most fatal flood disasters, which differs from previous studies assessing impacts of flood magnitude measured by, for example, precipitation anomalies and annual rainfall variability. The results of this study suggest that even the most devastating flood disasters, alongside precipitation anomalies, may not provide momentum for measures that reduce future flood fatalities. One implication is that flood disaster magnitude, in terms of high-loss/low-probability events, does not necessarily increase the probability of effective adaptation.

The finding reported elsewhere that more recent disasters (happening within roughly a decade) lead to adaptation measures that reduce fatalities could not be supported here. Studies of hurricanes21 and floods8 have found that more recent events matter more for reducing fatalities in subsequent events, suggesting that memories and knowledge from past shocks fade with time. In contrast to these findings, we found no difference between mortality rate trends after recent and more distant milestone events (Fig. 2).

Several caveats apply to the study and suggest directions for future research. Most countries in the study classify as low- and upper-middle-income countries, which generally have low institutional capacity to achieve effective adaptation in the wake of disasters16. Recurring disasters have been hypothesized to reinforce this effect as short-term reconstruction needs can override long-term adaptation measures12,13. Although this study finds no association between milestone flood events and mortality across income levels, this remains to be tested with a more diverse sample of countries across income levels. Other studies17,22 have demonstrated that flood fatalities decrease with increasing income, but, again, these do not investigate the effects of milestone events specifically.

It is also possible that countries have initiated, yet not effectively implemented, adaptation measures after milestone events, or, alternatively, invested in measures with limited risk reduction potential23. Implementation gaps and maladaptation after milestone events could be studied more closely by estimating the effects of different types of adaptation measures. Studies24 point to certain cost-effective measures that contribute to mitigating disaster fatalities (e.g., medical facilities, road infrastructure, electrification, and financial accessibility), but these should be investigated further8. The trend analysis here may also not capture effective risk reduction measures that take several decades to implement. For example, in the United Kingdom and Netherlands, the 1953 coastal floods led to major flood protection measures, including the Thames Barrier and the Delta Works programme, which were completed in 1982 and the 1990s, respectively25. Such long-term measures may not have been finalized yet in countries with more recent milestone events, such as Argentina and Nigeria.

To reduce disaster fatalities, disaster risk reduction discourse urges communities to “build back better” after disasters by adopting adaptation measures to reduce vulnerability. Our study suggests that this is generally uncommon, which is inconsistent with success stories reported in some case studies of flood adaptation26 and comparisons of mortality after consecutive floods affecting the same region27. Investigating circumstances that constrain adaptation after disasters, such as blame-game politics, inertia, or exposure to repeated disasters12, is an essential next step in this research. More robust analyses are also required to uncover milestone events that deviate from this general pattern, as indicated by trends of decreased mortality. This can be achieved by studies controlling for different types of adaptation actions and their potential effects on mortality and taking into account that implementation of risk reduction measures may take decades to complete.

Methods

The flood mortality rates utilized for this study normalized fatalities by the total number of affected people, which indirectly accounts for flood disaster intensity. As a result, the mortality rates can capture increased flood intensity, as indicated by an increase in affected populations.

Data were retrieved from the International Disasters Database28 (EM-DAT), which includes events with ten or more casualties, 100 or more affected people, a state of emergency declaration, and/or an appeal for international assistance. Coastal, riverine, flash floods, and ice jam floods, 1970−2021, were included. Events before 1970 were excluded from the study due to underreporting19. We extracted each event’s ID, country, region, continent, starting year, fatalities, affected individuals (injured, homeless, affected), and economic damage (in thousands of US$). Most countries in the study were classified (by the World Bank, as of 2021) as lower- or upper-middle-income countries, but in some cases, the classification has changed over time.

Countries were ranked in descending order according to the number of flood events. The top 30 countries (Supplementary Table 1, Fig. 1a) were included to ensure sufficient observations for the trend analysis. Seven countries (Bangladesh, China, Kenya, Malaysia, Philippines, Romania, USA) were excluded due to too few events (n < 5 before and/or after the milestone event) or missing data.

To normalize the different levels of flood severity over time, mortality rates were computed by dividing fatalities by the total number of affected people. Moreover, since total fatalities are not included in the total affected in EM-DAT, we calculated the mortality rate (R) for each event as follows:

$$R=\frac{{total}\,{fatalities}}{{total}\,{fatalities}+{total}\,{affected}}$$
(1)

Mortality rates were log10-transformed for visualization and plotted over time (Fig. 2, Figs. S1, S2). Milestone events were defined as the first in chronological order between first, the first event belonging to the 99th percentile in terms of fatalities, and, second, the event with the greatest number of fatalities. In cases where countries recorded several events during a milestone event year (Supplementary Table 1), only fatalities from the deadliest event were included.

The study fitted linear trends for all floods before and after the milestone event. We then compared the angles between the first and the second linear trend and the y-axis, referred to as α and β, respectively. Instances when α is greater than β provide evidence of decreased mortality after milestone events (Fig. 1b). Results were corroborated by using linear regression to compare differences in mortality rates before and after milestone events. We estimated models for all floods and separate models for riverine and flash floods, respectively. No models were estimated for coastal and ice jam floods due to too few data points. None of these models returned p-values below 0.05.

A challenge of utilizing this approach is whether the number of included events is sufficient to derive trend lines. Nonetheless, our study was able to capture strong trends for analysis. Future work can enhance the number of events by adding additional hazard types and countries.

It is important to acknowledge that several caveats apply to our study. Despite being one of the most reliable and comprehensive global sources of disaster information, EM-DAT’s data are heterogeneous and sometimes subject to uncertainty due to missing information. Another issue is that it can be difficult in larger countries to trace the effects of measures to reduce disaster risk in one specific flood-affected region. Although we found consistent results independent of country size (Fig. 2), determining the scale of adaptation measures (local, regional, national) is potentially arbitrary. Our results were also robust for sub-samples of different flood types (Supplementary Figs. 1, 2).