Skip to content

Attributing an event to climate

What an attribution study actually establishes, why it speaks of probabilities and never of a single cause, and how to read its results without making them say more than they do.

Updated on 15 November 2026 · Natural hazards, Modelling and projections, Heat and heatwaves

After every heatwave, every flood, the same question returns: was this climate change? Put that way, it has no answer. Put differently, it has one, and increasingly quickly.

The question that cannot be asked

A weather event always has an immediate cause: a blocking high, a surge of tropical air, a stalled depression. Climate does not create these configurations; it changes how often they arise and how intense they become when they do.

Asking whether "warming caused" a heatwave is therefore like asking whether a loaded die caused the six just thrown. The six might have come up anyway. What can be established is how heavily the die is loaded.

How it is done

The dominant method compares two simulated worlds. The first is the world as it is, with its observed atmospheric composition. The second is a counterfactual world, identical but with human emissions removed. Hundreds of simulations of each are run, then one counts in how many of them the studied event, or worse, occurs.

The result is expressed two ways. The probability ratio says how much more likely the event has become: "ten times more likely" means it used to occur once a century and now occurs once a decade. The intensity change says how much worse it has become: "2 °C hotter" for a heatwave of the same return period.

Fischer et al., 2015

The two framings are not interchangeable. An event may be barely more likely yet markedly more intense, or the reverse. Coverage generally keeps whichever figure is the more striking.

What attributes well, and what resists

Not all categories of event lend themselves equally to the exercise. Heat attributes best: the signal is strong, the mechanism direct, and models represent it well. Most recent major heatwaves turn out, on analysis, to be near-impossible without warming.

Extreme precipitation comes next, with a solid physical basis — warmer air holds more water vapour — but local variability that demands long records. Droughts are harder, because they combine rainfall, evapotranspiration and human water use. Tropical cyclones resist most: they are rare, intense, and poorly resolved by coarse-grid models.

Bartusek et al., 2022

One case deserves particular attention. Some recent events have exceeded not only observed records but what simulations judged possible — the Pacific Northwest heat dome of June 2021 is the example. Such episodes pose a different question from ordinary attribution: not "how much has the risk risen?" but "do our models bound the possible correctly?"

Fischer et al., 2021

From probability to damage

A second generation of work extends the chain to consequences. If one knows how much a heatwave was worsened, and knows the relationship between temperature and mortality, one can estimate how many deaths are attributable to the human share of the warming.

Mitchell et al., 2016

The same reasoning applies to economic costs. These estimates chain more assumptions than physical attribution alone, and their uncertainty is wider — which the publications state, and which coverage almost always omits.

Newman et al., 2023

Reading an attribution result

Three precautions are enough to avoid the commonest misreadings.

A probability ratio is not a percentage of responsibility. "Ten times more likely" does not mean "90 % caused by climate": the event remains meteorological in form, and internal variability keeps its share in it.

No detected signal is not no effect. For rare or poorly observed hazards, a study may conclude that it cannot decide. That is a statement about data quality, not about climate.

Speed has a price. Analyses published within days of an event use streamlined methods and models already to hand. They serve public debate and are generally confirmed later, but their range is wider than a peer-reviewed study's — and the confidence level they state says so.

Sources

  1. World Weather Attribution — Imperial College London
  2. Attribution of Extreme Weather Events in the Context of Climate Change — National Academies of Sciences, Engineering, and Medicine
  3. IPCC AR6 WGI, Chapter 11: Weather and Climate Extreme Events — IPCC

Studies cited in this brief