Global hazard trends

A hydro-meteorological hazard is a hazard caused by atmospheric, hydrological or oceanographic processes — that is, one driven by weather, water or climate.

Two examples are tropical cyclones (storms) and floods (others include droughts and heatwaves).

The overall trend is a decrease: the number of deaths from disasters has fallen over the period shown.

The magnitude of the fall is large — from around 2 million per year in the 1920s to under 100,000 per year in recent decades, a decline of more than 95%.

However, the decline is not steady: there are sharp spikes (anomalies) in individual years — for example single catastrophic events such as a major earthquake or cyclone — that stand well above the falling trend.

The paradox is that two impacts of hazards are moving in opposite directions: the money value of damage is rising while the number of people killed is falling.

Economic losses rise because population growth, urbanisation and economic development have put more, and more valuable, buildings and infrastructure into hazard-prone areas — for example dense, high-value development on coasts and floodplains. When a hazard strikes it therefore destroys more assets worth more money.

Death tolls fall because prediction, warning systems, evacuation, hazard-resistant design and preparedness have improved, so people can be moved or protected before impact.

The two trends fit together because these protective measures save lives but cannot move fixed property out of the way — so the same hazard kills fewer people yet destroys more wealth. The exception is in the poorest, most vulnerable countries, where weak warning and response mean death tolls can still be very high.

The rise in recorded climate-related hazards (floods, storms and droughts) can be explained in two competing ways, and geographers debate how much each contributes.

The case for a real increase. A warming climate is expected to increase the frequency and intensity of hydro-meteorological hazards: higher sea-surface temperatures give tropical cyclones more energy, a warmer atmosphere holds more moisture (heavier rainfall and flooding), and shifting rainfall patterns can lengthen droughts. On this view the rising line reflects a genuine climate-change signal. Importantly, geophysical hazards (earthquakes, volcanoes), which are not linked to climate, have not shown the same rise — which supports the idea that the increase in climate-related hazards is at least partly real.

The case for better reporting. Alternatively, more events may simply be detected and recorded now than in the past. Satellites, weather stations and global databases such as EM-DAT capture events in remote areas that once went unrecorded. Better communications and media mean more events are reported, and because population growth and urbanisation have put more people in hazard-prone areas, more events now affect people and so count as disasters. A wider definition of 'disaster' also inflates the count.

Evaluation. The truth is probably a combination: reporting has certainly improved, which exaggerates the apparent rise, but the fact that climate-related hazards have risen while geophysical hazards have not suggests a real underlying increase driven by climate change as well. Distinguishing the two is difficult because the reporting improvements and the climate signal overlap in time.

Advances in prediction and forecasting technology have reduced hazard impact mainly by giving people time to respond before an event, though the amount of warning possible varies greatly between hazard types.

Volcanoes — monitoring precursors. Rising magma produces measurable warning signs, so volcanoes can often be predicted. Tiltmeters detect ground deformation as the volcano swells, gas sensors detect rising sulphur dioxide, and seismographs record increasing earthquake activity. At Mount Pinatubo (Philippines, 1991), monitoring by the USGS and PHIVOLCS allowed a timely evacuation of tens of thousands of people, saving an estimated 5,000+ lives despite one of the largest eruptions of the century.

Tsunami — ocean sensors and warning centres. Because tsunami travel at a known speed across the ocean, they can be detected in transit. DART buoys (Deep-ocean Assessment and Reporting of Tsunamis) measure changes in water pressure on the sea floor and relay data to warning centres. After the 2004 Indian Ocean tsunami (~230,000 deaths, when no warning system existed) an Indian Ocean Tsunami Warning System was established, so later events could trigger coastal evacuation within minutes.

Tropical cyclones — satellite tracking. Satellites monitor cyclone position, size and intensity, and computer models project a forecasting cone of likely tracks, giving several days' warning of landfall. This underpinned the evacuation of over a million people before cyclones striking Bangladesh and India, cutting death tolls dramatically compared with unwarned events.

The limit — earthquakes. In contrast, earthquakes cannot be reliably predicted because there are no consistent precursors. Technology can only produce long-term probability forecasts and hazard maps showing where large earthquakes are likely, not when they will strike — which is why earthquakes still cause sudden, high death tolls even in monitored regions.

Overall, prediction technology has been highly effective at reducing deaths from predictable hazards (volcanoes, tsunami and cyclones) by enabling warning and evacuation, but it has done far less for unpredictable earthquakes and little to reduce economic losses, since property cannot be moved.

Although hazards occur across the world, death tolls have fallen much faster in developed countries than in the poorest developing countries, because the ability to predict, warn and respond depends heavily on wealth and development.

In developed countries (e.g. Japan, USA), governments can afford prediction technology (satellites, seismic monitoring), hazard-resistant infrastructure (earthquake-proof buildings, sea defences), organised evacuation with transport and shelters, and effective emergency services. Populations are also better educated about hazards, so warnings are understood and acted on. As a result, a strong hazard tends to cause relatively few deaths — for example, well-engineered buildings and drills keep Japanese earthquake death tolls low relative to the shaking involved.

In developing countries (e.g. Haiti, Bangladesh in the past), by contrast, limited money means poorly built housing, weak or absent warning systems, few shelters and evacuation routes, and under-resourced emergency services. High population densities in vulnerable areas (floodplains, unstable slopes, coasts) add to exposure. The 2010 Haiti earthquake killed over 200,000 people largely because of poor building quality and weak preparedness, whereas a similar-magnitude event in a developed country would kill far fewer.

However, some developing countries have made rapid progress: Bangladesh cut cyclone deaths from around 500,000 in 1970 to a few thousand in later cyclones by investing in forecasting, warnings and cyclone shelters, showing that governance and targeted investment — not just national wealth — shape the trend.

Overall, the death-toll trend falls fastest where a country can afford and organise prediction, protection and response, so the gap between rich and poor countries reflects differences in vulnerability and capacity to cope rather than in the hazards themselves.

Reducing the impact of natural hazards depends on two broad sets of measures: prediction and forecasting (knowing that a hazard is coming and where) and responses (warning, evacuation, resistant design, zoning, relief and reconstruction). This essay argues that the two are interdependent — prediction is only useful if it triggers an effective response — but that response measures are ultimately the decisive factor, because prediction saves no one unless people can act on it, and some deadly hazards cannot be predicted at all.

The case that prediction and forecasting have been most important. For predictable hazards, forecasting is what makes life-saving possible. Volcanic eruptions can be forecast from measurable precursors — ground deformation (tiltmeters), gas emissions and seismic activity — and at Mount Pinatubo (1991) monitoring gave enough warning to evacuate tens of thousands and save an estimated 5,000+ lives. Tropical cyclones are now tracked by satellite with a multi-day forecasting cone, so the path and timing of landfall are known in advance; tsunami are detected by DART buoys feeding warning centres set up after 2004. Without this forecasting, no evacuation could be organised in time, so it can be argued that prediction is the essential first step on which everything else depends.

The case that responses have been more important. However, prediction on its own saves no lives — it must be converted into action. Bangladesh shows this clearly: cyclone forecasting existed before 1970, yet the 1970 Bhola cyclone killed around 500,000 people because there was no effective response. After investing in sirens, volunteer networks, evacuation routes and thousands of cyclone shelters, later cyclones of similar strength killed only a few thousand — the improvement came overwhelmingly from response and preparedness, not from better forecasting. Response measures also protect against hazards that cannot be predicted: since earthquakes have no reliable precursors, the fall in earthquake deaths in places like Japan comes almost entirely from hazard-resistant building design, land-use zoning, drills and emergency planning rather than prediction. The contrast between the 2010 Haiti earthquake (>200,000 deaths, poor construction and weak response) and comparable events in Japan demonstrates that how a society is built and prepared matters more than any forecast.

The role of development and vulnerability. Both prediction and response depend on wealth, governance and vulnerability, which cuts across the debate. Rich, well-governed countries can afford both advanced forecasting and organised response, so their death tolls have fallen fastest. Poorer countries may receive the same global forecasts (e.g. cyclone tracks) yet still suffer high death tolls if they lack shelters, transport and education — again pointing to response capacity as the limiting factor. Neither prediction nor response reduces economic losses much, because fixed property cannot be moved, which is why losses keep rising even where deaths fall.

Conclusion. Prediction and response are not alternatives but a chain: forecasting provides the warning, but only an effective response turns that warning into saved lives, and for unpredictable hazards response is the only tool available. The evidence — Bangladesh's transformation through shelters and Japan's low earthquake death tolls from resistant design — shows that improved responses have been the more decisive factor, with prediction essential but insufficient on its own. The most successful reductions in impact occur where accurate prediction and well-resourced response are combined, and where underlying vulnerability is also reduced.

Global economic losses from natural hazards have risen sharply — from tens of billions of US dollars a year in the 1960s to over US$300 billion in recent peak years — even as death tolls have fallen. This essay evaluates whether this rise is inevitable, arguing that the underlying trend is very difficult to reverse because it is driven by development itself, but that losses are not wholly inevitable because good planning, zoning and design can reduce them.

The case that rising losses are inevitable. The rise reflects the same processes that make countries wealthy. Population growth and urbanisation concentrate more people and property in hazard-prone locations, especially coasts and floodplains, because these are attractive sites for cities and trade. Economic development means the assets exposed — housing, factories, technology, transport networks — are worth ever more money, so the cost of the same physical damage rises over time. Because the most productive land (e.g. deltas, coastlines such as Tokyo Bay or the US Gulf Coast) is often also the most hazard-prone, the growth of the economy automatically puts more value at risk. Crucially, the very measures that cut death tolls — warning and evacuation — save lives but cannot move fixed buildings, so a hazard that now kills fewer people still destroys more property. On this reasoning, as long as economies grow and cities expand, total losses will keep climbing.

The case that rising losses are not inevitable. However, losses can be reduced by deliberate planning. Land-use zoning can keep high-value development off the most dangerous floodplains and fault zones, lowering the assets exposed. Hazard-resistant design — flood-proofed buildings, sea defences such as the Thames Barrier, earthquake-resistant construction in Japan — reduces the damage a given event causes, so losses per event can fall even as exposure grows. Improved forecasting can allow moveable assets (vehicles, stock, aircraft) to be protected in advance, and insurance and building regulations create incentives to build more safely. Where such measures are enforced, the loss from a hazard of a given size can be much lower than it would otherwise be, showing that the outcome is a matter of choice and governance, not fate.

Evaluation — losses in money terms versus losses relative to GDP. A key distinction is between absolute and relative losses. In absolute money terms, global losses are very likely to keep rising because of growth and inflation, so that part of the trend is close to inevitable. But losses as a share of the wealth exposed need not rise, and can fall where zoning, design and preparedness are strong — this is why losses relative to GDP have not risen as fast as absolute losses in well-managed, developed countries. In poorer countries, weak planning and rapid, unregulated urban growth can make losses rise faster than wealth, so there the trend is more damaging but still partly avoidable through better governance.

Conclusion. I largely agree that rising absolute economic losses are close to inevitable while populations, wealth and cities keep growing in exposed locations, and while protective measures save lives rather than property. But the claim is too strong if taken to mean losses cannot be influenced: through land-use zoning, hazard-resistant design, forecasting and enforced regulation, the losses from any given hazard — and losses relative to the wealth at risk — can be significantly reduced. The rise in losses is therefore best seen as a strong tendency created by development that can be moderated, but not eliminated, by good planning.

Earthquakes remain among the deadliest natural hazards: single events such as the 2010 Haiti earthquake (>200,000 deaths) and the 2004 Indian Ocean earthquake and tsunami (~230,000 deaths) kill on a scale rarely matched by other hazards. This essay assesses how far this is explained by the inability to predict earthquakes, arguing that unpredictability is an important factor but that vulnerability, building quality and secondary hazards are at least as important.

The case that unpredictability is the key factor. Unlike volcanoes, tsunami or cyclones, earthquakes give no reliable precursors. There is no equivalent of the ground deformation and gas emissions that warn of an eruption, or the satellite track of a cyclone, so scientists can produce only long-term probability forecasts and hazard maps, not a warning of when an earthquake will strike. This means there is usually no time to evacuate: the shaking arrives without notice, so the protective response that has cut deaths from predictable hazards — warning and evacuation — is largely unavailable. Where a hazard can be predicted, as with cyclones in Bangladesh, death tolls have fallen dramatically; the absence of that option for earthquakes helps explain their continuing high death tolls.

The case that other factors matter more. However, the deadliness of earthquakes owes more to vulnerability than to unpredictability, because the main killer is usually building collapse, which is preventable. Japan experiences frequent, powerful earthquakes yet keeps death tolls comparatively low through hazard-resistant design, strict building codes, land-use zoning, drills and rapid emergency response — none of which requires prediction. By contrast, the 2010 Haiti earthquake was so lethal mainly because of poorly built housing, high population density, poverty and a weak state, not simply because it was unforeseen. This contrast shows that how a society builds and prepares can matter more than whether the event was predicted. In addition, modern warning technology partly compensates for unpredictability: earthquake early-warning systems (e.g. in Japan and Mexico) detect the first, faster seismic waves and issue seconds of warning before the destructive shaking — not a true prediction, but enough to stop trains and trigger automatic shut-offs.

Secondary hazards and location. Much earthquake mortality also comes from secondary hazards that unpredictability alone does not capture — tsunami (2004 Indian Ocean), fires, landslides and building collapse. Deaths are heavily concentrated where large populations live in vulnerable, densely built areas near plate boundaries, so exposure and vulnerability shape the toll as much as the lack of warning.

Conclusion. The inability to predict earthquakes is a genuine and important reason why they remain so deadly, because it removes the evacuation option that has tamed predictable hazards. But it is not the main reason: the decisive factor is vulnerability — building quality, preparedness, development level and secondary hazards — as shown by the vast difference in outcomes between Japan and Haiti for earthquakes of comparable strength. Earthquakes are deadliest where unpredictability combines with high vulnerability, so reducing deaths depends less on the (currently impossible) goal of prediction than on resistant construction, preparedness and early-warning technology.