The uneven toll — a climate mortality explorer

Climate disasters don't kill evenly.

1900a century of warming · each stripe is one year2023
~2,000,000
recorded deaths from weather, climate & water hazards, 1970–2021
90%
of those deaths occurred in developing countries
Storms ↓  Heat ↑
deaths from storms have fallen sharply — deaths from heat are climbing

This tool asks two questions: which hazards fall hardest on the already-vulnerable, and where falling death tolls reflect real adaptation rather than better record-keeping.

scroll to begin
02 — the pattern

Regressive, or democratic?

Each line shows how a hazard's death rate changes as you move from the least to the most socially vulnerable places. A steep climb means the poor die far more often — a regressive hazard. A flat line means it kills more evenly.

hover to isolate

Countries split at the median ND-GAIN vulnerability score; each line runs from the mean death rate of the less-vulnerable half to the more-vulnerable half (deaths per 100k, 2000–2020, log scale). Drought is sharply regressive. Heat and wildfire invert the pattern — partly real (wildfire kills in wealthy fire-prone regions) and, for heat, substantially a reporting artifact: high-income countries count heat deaths, most others don't. Means are outlier-sensitive — read as direction, not exact magnitude. Source: EM-DAT / CRED · ND-GAIN.

03 — over time

Which curves are bending?

Death rates per hazard, smoothed across decades. Some have fallen sharply where warning systems and preparedness improved. Others haven't moved. Click a bending curve to see what happened there.

before you explore

How to read this honestly

Rates, not counts

A bigger country isn't a deadlier one. Everything here is deaths per exposed population, so places are comparable.

A falling line has two readings

Fewer deaths can mean real adaptation — or simply fewer, smaller events that decade. We separate the two before claiming progress.

Old data is thin

Early records undercount, especially small events in poorer regions. Trends before the 1990s are read with caution.

Cause is claimed carefully

Where we link a decline to a policy, it's cited evidence framed as a contributing factor — never a proven cause.

05 — three stories

What the curves mean on the ground

06 — explore freely

The map is yours now

Pick a hazard, move through the years, and read the geography of loss directly. You've seen the patterns and the caveats — now dig where you like.

2020s

07 — resilience

Surviving is not the same as recovering

A country can stop its people dying in disasters and still spend decades rebuilding — and rebuilding output is not the same as rebuilding a foundation that holds next time. Three signals, read together, tell those apart.

or try an example
Read the three graphs together. Real resilience means all three point the right way. If the money recovers but fragility stays high, the recovery may be a temporary patch — not durable healing.
1 · Money
Income per person, inflation-adjusted. Up is good.
2 · Lives
Deaths from floods, storms, droughts & wildfires, per 100,000. Down is good.
3 · Fragility
How exposed the country still is, 0–1. Down is good.

Two things stay true everywhere on this page. Falling death tolls are a real achievement — but they count only who survived, not who was ruined: the same storm that now kills ten instead of thousands can still take the homes and harvests behind them. And every falling line here describes a past climate milder than the one ahead. We have become very good at keeping people alive through disasters — not nearly as good at keeping them from losing everything, and the threat is still growing.

what the data leaves us with

Climate change is global.
Its mortality is not.

The heaviest toll falls on those least able to withstand it.

Explore the data yourself →