“Climate change contributed to more than half of Europe’s heat deaths” is the kind of sentence that invites a fair technical question: how could anyone possibly know that? No death certificate records a cause called climate. The answer is one of the most sophisticated measurement programmes in environmental health — a stack of exposure–response epidemiology, counterfactual climate modelling and demographic data — and understanding it makes the finding more credible, not less.
Step one: the exposure–response function
For each city, epidemiologists model the statistical relationship between daily temperature and daily mortality, using years of records. The curve is typically U-shaped with a city-specific minimum-mortality temperature; above it, deaths rise at a rate that accelerates in extreme heat. The function quantifies risk from cardiovascular, respiratory and renal causes that heat aggravates — which is why heat mortality is estimated rather than counted. Research programmes coordinated by groups such as the Lancet Countdown maintain these baselines across hundreds of locations, updated annually.
Step two: the counterfactual climate
Attribution then asks what temperatures would have been without human influence. Climate models simulate the observed period twice — with actual greenhouse-gas forcing, and with pre-industrial forcing — producing, for each heat event, a probability and intensity shift attributable to emissions. Combining the two steps yields deaths attributable to anthropogenic warming: the mortality expected under observed temperatures minus mortality under the counterfactual. That is the machinery behind the “more than half” estimate for recent European summers, and its uncertainty intervals — driven by both model spread and epidemiological error — are published with it. Methodological standards are documented by bodies such as the European Environment Agency, which integrates attribution outputs into its climate-risk assessments.
Where the uncertainty lives
Honest practitioners flag the soft spots. Mortality displacement (“harvesting”) can inflate short-window estimates if deaths merely move forward by weeks — the better studies model across whole seasons to correct for it. Air-conditioning penetration, demographic change and healthcare improvements shift the exposure–response curve over time, and cities differ in data quality. None of these, reviewers note, plausibly reverses the sign of the result: warming of the observed magnitude mechanically increases exposure to temperatures where the mortality curve is steepest. Our data desk reproduces the sensitivity analyses behind the headline numbers.
From attribution to engineering
Reading the finding like an engineer
Stripped of advocacy, the attribution result is a load calculation. Europe’s population is being exposed to a quantifiably increased frequency of high-mortality thermal conditions; housing, urban form and demography set the vulnerability coefficients; the product is a death toll with a growing anthropogenic share. Engineers respect such numbers because they are actionable at every term of the equation — reduce exposure (urban cooling, building retrofit), reduce vulnerability (health surveillance, social checks), or reduce the forcing itself (emissions). The models also expose where money is best spent: the same analysis that attributes deaths identifies the neighbourhoods and building types where intervention buys the most survival per euro. Attribution began as an argument about the past. Its most important use is as a design tool for the future.
The output of these models increasingly drives design decisions with specifications attached: urban tree-canopy targets modelled in degrees of neighbourhood cooling, reflective-pavement standards, building codes for passive survivability during power cuts, and heat-plan trigger thresholds calibrated city by city. Attribution science began as an academic exercise in assigning responsibility. It is becoming, in Europe’s cities, an engineering input — the dataset that tells a mayor how many lives a given intervention buys. That translation, from probability to policy, is where the models will be judged. More technical climate coverage at our homepage.
