Human Survivability Under the Global Polycrisis

The record kept by the EM-DAT international disaster database stretches back to 1900. Its early decades are dominated by catastrophes of a scale that has not been repeated: the Chinese floods of 1931, which EM-DAT records as killing about 3.7 million people, recurring droughts and famines, and earthquakes that levelled cities. Recorded deaths averaged about 520,000 a year in the 1920s and 330,000 in the 1960s. By the 2010s the average had fallen to about 45,000 a year, and in 2020–24 it was about 60,000, inflated by the 2023 earthquakes in Türkiye and Syria.

Population quadrupled over the same century, so the decline per person is steeper still: from about 99 deaths per million people a year in the 1960s to 25 in the 1970s, 8 in the 1990s and 6 in the 2010s. Meanwhile, the number of people recorded as affected by disasters rose from a few million a year in the 1950s to around 180–230 million a year in the 1990s–2010s. Part of that rise reflects better reporting, but part is real: more people live in floodplains, coastal cities and drylands. More people are hit; far fewer die.

The record should be read with care. Reporting improved enormously, which makes early totals uncertain in both directions; heat deaths are badly under-recorded outside Europe; and a single event can dominate a decade. But the direction is not in doubt, and it is the starting point for any discussion of survivability: the gap between exposure and death is where capacity lives. The masthead trace shows the record year by year, with this report's projections to 2100 for four scenarios.

Capacity is hard to measure directly, but education is a strong proxy. Educated populations understand and act on warnings, build and maintain safer homes, earn the incomes that buy protection, use health services and demand competent governments. Lutz, Muttarak and Striessnig argued in 2014 that universal education is key to climate adaptation, based on evidence that education reduces disaster vulnerability. This report tests the relationship with current data, using the Wittgenstein Centre's estimates of mean years of schooling among adults aged 25 and over, weighted by population across age groups.

The measure of vulnerability is deaths per 1,000 people affected by floods, storms, earthquakes, landslides, heat, wildfires and volcanoes — excluding droughts, for which deaths are rarely recorded alongside the very large numbers affected. Across 147 countries and six decades since the 1970s, the median country's fatality ratio fell from about 0.7 per 1,000 affected in the 1970s–1990s to 0.3 in the 2010s, while its adults' schooling rose from about 3 years to more than 8.

Comparing each country with itself, and removing changes common to all countries in each decade, one more year of adult schooling is associated with about 32% fewer deaths per 1,000 affected (t = −2.4). The estimate barely changes when income is added, and income itself adds nothing once schooling is included. The uncertainty is wide — the 95% range runs from about 7% to 50% per year — and schooling stands in for the broader capacity it helps create. But it is the most robust lever in the data, and it can be projected, because the Wittgenstein Centre projects schooling for every country under each scenario.

The IPCC's Sixth Assessment Report quantified how the frequency of extremes rises with global warming. A hot extreme over land that occurred once a decade in the late nineteenth century is now about 2.8 times as frequent at roughly 1 °C of warming; it becomes 4.1 times as frequent at 1.5 °C, 5.6 times at 2 °C and 9.4 times at 4 °C. A once-a-decade heavy daily rainfall becomes 1.5 times as frequent at 1.5 °C and 2.7 times at 4 °C, and a once-a-decade agricultural drought in drying regions 2.0 and 4.1 times. These are median estimates, with uncertainty ranges shown in the IPCC figure.

The warming those multipliers apply to depends on the path. Using the IPCC's assessed central projections, warming averages about 1.8 °C in 2081–2100 under SSP1-2.6, 2.7 °C under SSP2-4.5, 3.6 °C under SSP3-7.0 and 4.4 °C under SSP5-8.5, and keeps rising to 2100 in all but the first. Each country's hazard in the projection is weighted by its own mix of disaster deaths since 2000: heat-dominated tolls, as in Europe, rise fastest; flood and drought tolls rise more slowly; and deaths from earthquakes, volcanoes and storms are held constant, since warming does not change earthquake frequency and the IPCC projects more intense rather than more frequent tropical cyclones.

Exposure is the third force. The Wittgenstein Centre's projections show world population peaking within the century in most scenarios, but growing strongly in the poorest and most hazard-exposed countries under SSP3, where fertility stays high and education spreads slowly. In that world, the population of sub-Saharan Africa in particular grows several-fold in exposure-weighted terms — more people, in the places where disasters are deadliest, at the time hazards are most frequent.

The projection multiplies each country's average yearly disaster deaths in 2000–24 by three factors: its population relative to 2020 under each scenario, its hazard frequency relative to 2020 at each year's warming, and the protective effect of its projected gain in adult schooling. In the sustainability scenario SSP1, schooling rises from a population-weighted 8.1 years to 13.6 by 2100, warming stays below 2 °C and population shrinks slightly; expected deaths fall by more than 80%. In the middle-of-the-road SSP2 they fall by about 69%, and in the fossil-fuelled SSP5 — fast growth and education, but 4.7 °C of warming by 2100 — by about two-thirds, because capacity rises so fast.

The fragmented SSP3 world, with high warming, slow development and rapid population growth in the poorest countries, is different. Schooling rises only from 8.1 to 9.8 years, population exposure grows by about half and hazard frequency by almost as much. Expected deaths stay near today's level worldwide but rise in 107 of 157 countries, and the race is lost decisively in sub-Saharan Africa. To hold world disaster deaths steady in SSP3 would take about 2.1 additional years of adult schooling — more than that world delivers. Without any capacity gains, deaths would rise by half in SSP2 and SSP5 and more than double in SSP3.

Everything hinges on the capacity effect, so the slider lets you change it. At the weakest effect consistent with the data, about 7% per year of schooling, expected deaths fall only modestly in SSP1, rise by about a tenth in SSP2 and SSP5 and nearly double in SSP3. At the central estimate the picture is optimistic for all but SSP3. Either way, the ranking of futures does not change: the worlds that invest in people and limit warming keep survivability rising; the fragmented world does not.

Under SSP3, expected disaster deaths in sub-Saharan Africa rise about fourfold by 2100. Its exposure-weighted population grows roughly fivefold, hazards nearly double, and schooling gains, though real, are too slow to offset them. The Middle East and North Africa and North America also see rising tolls, and Europe and Central Asia — whose recorded toll is dominated by heatwaves — rises by about a third, because heat extremes multiply fastest with warming and its adults are already so well educated that further schooling buys little extra protection.

That European result exposes a limit of the education proxy and a real policy point: in rich, highly educated societies, capacity has to grow through other channels — heat-health warning systems, cooling for the vulnerable, urban design, health care for the old — because the historical gains from education are largely banked. Europe's heat deaths in 2003 and 2022 were the products of societies with very high schooling and income. South and East Asia and Latin America, by contrast, see declines even under SSP3, because large schooling gains outweigh rising hazard.

In the other scenarios, every region's expected toll falls by 2100 except North America, where it rises under SSP2 and SSP5, and Europe and Central Asia, which stays roughly flat under SSP5's 4.7 °C. The regional chart follows the scenario and capacity effect chosen in section 4, and the atlas below shows the same picture country by country, including the schooling shortfall: the years of schooling a country would need to gain to hold its disaster deaths at today's level, minus the years it is projected to gain.

The atlas follows the scenario and capacity effect chosen in section 4. Its default layer shows the projected change in expected disaster deaths by 2100. The shortfall layer shows where projected schooling gains fall short of the gains needed to hold deaths at today's level; under SSP3 the largest shortfalls are in Somalia, Niger, the Democratic Republic of the Congo, the Central African Republic, Burkina Faso and Uganda, along with several heat-exposed, already highly educated countries.

The schooling layer shows adult educational attainment in 2100 under the chosen scenario, and the last layer today's disaster death rate. The country card lists each country's baseline toll and rate, population and hazard change, schooling path, the gain it would need, its projected toll in 2100, and the mix of hazards behind its baseline — heat, floods, drought, or earthquakes and storms, which the projection holds constant.

Country results inherit the noise of the disaster record. A country whose 2000–24 toll was dominated by one event — an earthquake or a cyclone — carries that event into its baseline, and countries with very small tolls can show large percentage changes from small numbers. Countries with fewer than 1 million people, including many small island states with high exposure, are not included.

Disasters are visible and countable, which is why they dominate discussion of survivability. But they are far from the largest threats to life. Air pollution caused about 8.1 million deaths in 2021, according to the State of Global Air report — more than a hundred times the annual toll of recorded disasters. The World Health Organization estimated 14.9 million excess deaths associated with the COVID-19 pandemic in 2020–2021. Heat kills far more than the disaster records capture: the Lancet Countdown reports that heat-related deaths have risen by 63% since the 1990s.

Other threats are systemic in a second sense: they spread across borders and systems and undo capacity itself. The UCDP recorded 65 state-based armed conflicts in 2025, the most since 1946, and organised violence killed about 245,000 people that year. Conflict is also the main driver of famine: the 2026 Global Report on Food Crises mid-year update projects about 327,000 people in catastrophic food insecurity in 2026 and a risk of famine in Somalia, South Sudan, Sudan and Gaza. And debt erodes the fiscal capacity on which protection depends: 3.4 billion people live in countries that spend more on interest than on health or education.

These threats change what survivability means. Schooling, warnings and safe buildings protect people against floods and storms; they do far less against a pandemic, a war or a debt crisis that empties health budgets. Capacity against systemic threats is collective — global health security, conflict prevention, clean air policy, crisis finance — and its absence is what turns a hazard into a catastrophe. The table summarises the scale of each threat and the kind of capacity it requires.

The projection's clearest message is that the scenarios differ less in their hazards than in their capacity. SSP5 warms the world far more than SSP3 by 2100, yet expected disaster deaths fall by two-thirds in SSP5 and stagnate in SSP3, because SSP5 educates its people. Investing in human capacity — above all universal, good-quality secondary education, especially for girls in the countries where populations grow fastest — is therefore one of the most powerful survivability policies available, with returns that compound over decades.

But capacity is not a substitute for limiting warming. Every scenario's toll is higher with more warming than it would otherwise be, heat extremes multiply fastest of all, and the protective returns to education flatten in societies that are already well educated, as Europe's heat record shows. Beyond schooling, the most direct capacity investments are early warning for everyone, health systems that can absorb surges, social protection that scales up in shocks, safe buildings and land use, and heat-health action plans with cooling for the vulnerable.

Finally, survivability depends on capacity that no country can build alone. The systemic threats in section 7 — pandemics, conflict, famine, pollution and debt — require shared institutions: global health security, conflict prevention and humanitarian access, clean-air and climate cooperation, and an international financial architecture that keeps fiscal space for protection in a crisis. A world that fragments, as SSP3 imagines, loses on every count at once.

Disaster deaths and people affected by country, year and hazard come from EM-DAT via Our World in Data (1900–2024); population from the UN World Population Prospects 2024 via Our World in Data. Adult schooling is the population-weighted mean years of schooling of people aged 25 and over from the Wittgenstein Centre Human Capital Data Explorer v3.3, read directly from the centre's published data files, for the 1950–2020 reconstruction and SSP1, SSP2, SSP3 and SSP5 projections. Warming trajectories are the assessed means for SSP1-2.6, SSP2-4.5, SSP3-7.0 and SSP5-8.5 behind IPCC AR6 WGI Figure SPM.8, and frequency multipliers the medians behind Figure SPM.6.

The capacity elasticity comes from a country-decade panel, 1970s–2020s, of log deaths per 1,000 affected (excluding droughts) regressed on adult schooling with country and decade fixed effects and standard errors clustered by country (147 countries, 684 country-decades): −0.38 (s.e. 0.16), or −0.35 with log GDP per person added. Projected expected deaths for each of 157 countries with at least 1 million people equal average yearly deaths in 2000–24 × population ratio × hazard ratio × exp(β × schooling change since 2020). The hazard ratio weights heat (heat and wildfire deaths), heavy-rainfall (floods and landslides) and drought multipliers by each country's 2000–24 death mix, holding earthquake, volcano and storm deaths constant.

Limitations: EM-DAT under-records heat deaths and deaths in data-poor places, and the baseline carries the noise of a few mega-disasters; schooling is a proxy for broader capacity and its elasticity is uncertain; the model omits sea-level rise, cyclone intensification, compound and cascading events, conflict-driven collapses of capacity, and adaptation beyond schooling. Related reports: the geography of human survivability under climate stress, wet-bulb temperatures and unsurvivable heat and cumulative exposure across the polycrisis century.