Increasing risk of mass human heat mortality if historical weather patterns recur

nature climate change Article https://doi.org/10.1038/s41558-025-02480-1 Increasing risk of mass human heat mortality if historical weather patterns recur Received: 27 January 2025 Accepted: 8 October 2025 Christopher W. Callahan 1,5 , Jared Trok 1, Andrew J. Wilson Carlos F. Gould 3, Sam Heft-Neal 2, Noah S. Diffenbaugh 1 & Marshall Burke 1,2,4 , 2,6 Published online: xx xx xxxx Check for updates The potential death toll of exceptional extreme heat events is crucial for climate risk analysis and adaptation planning but may not be captured by existing projections. Here we combine machine learning-based projections of five historical European heat waves under present or future global temperatures with empirical exposure–response functions to quantify the potential for extreme heat events to generate mass mortality. For example, if August 2003 meteorological conditions recur at the recent annual global temperature anomaly of 1.5 °C, we project 17,800 excess deaths across Europe in one week, rising to 32,000 at 3 °C. This mortality is comparable to peak COVID-19 mortality in Europe and is not substantially reduced by climate adaptation currently observed across Europe. Our results suggest that while mitigating further global warming can reduce heat mortality, mass mortality events remain plausible at near-future temperatures despite current adaptations to heat. Climate change is increasing the frequency and magnitude of extreme heat events1–4, threatening human health5. Additional warming is projected to generate more intense heat events than even recent record-breaking events6, with the potential for mass mortality events similar to those witnessed in Europe in the summer of 20037, especially during exceptionally hot years such as 20238,9. Projections of increased heat-related mortality from climate change are now numerous10–15. However, these projections generally focus on the long-term population burden of non-optimal temperatures rather than the death toll of individual high-impact events. Exceptional extreme heat events require distinct management strategies compared with typical population burdens, as they can strain health and emergency services beyond what occurs at milder temperatures16. Preparedness for hospital overcrowding and health system surge capacity should therefore be benchmarked to a plausible extreme scenario rather than an average projection17. Quantifying plausible scenarios of extreme events under future climate change requires careful methodological treatment, and there are reasons to believe that existing projections do not capture the most extreme mortality events. In particular, the relatively short records of observations and most global climate model (GCM) simulations make it difficult to assess the probabilities of the most extreme events18. While progress has been made using large initial-condition ensembles to quantify very rare heat mortality19, some of the most extreme events may be poorly captured even by ensembles with many members20. Additionally, GCMs underestimate trends in the frequency and persistence of atmospheric circulation patterns that have contributed to recent rapid warming of heat extremes in populous regions such as Europe21–26. To complement existing work, a promising approach is to develop ‘storylines’ of heat waves that are physically plausible and dynamically consistent. This conditional approach, which emphasizes plausibility rather than probability27, enables exploration of extreme outcomes28,29 and stress tests of adaptation strategies17,30. Plausible storylines must also account for the documented ability of humans to adapt to repeated heat exposure and to change behaviour following past extreme heat episodes31. 1 Doerr School of Sustainability, Stanford University, Stanford, CA, USA. 2Center on Food Security and the Environment, Stanford University, Stanford, CA, USA. 3School of Public Health, University of California San Diego, La Jolla, CA, USA. 4National Bureau of Economic Research, Cambridge, MA, USA. 5 Present address: Paul H. O’Neill School of Public & Environmental Affairs, Indiana University, Bloomington, IN, USA. 6Present address: Frank Batten School of Leadership and Public Policy, University of Virginia, Charlottesville, VA, USA. e-mail: Nature Climate Change Article Major heat-mortality events require several ingredients: large-scale physical drivers of elevated temperatures as well as human health responses to the resulting heat stress. Extreme heat events tend to occur when atmospheric high-pressure systems interact with dry soils to produce land–atmosphere feedbacks that amplify heat accumulation6,21,32,33. In turn, prolonged exposure to high ambient temperatures impairs the ability of the body to dissipate heat, leading to elevated core temperature, increased cardiovascular strain and a heightened risk of heat-related illness and death34. Here we focus on the combination of these geophysical and physiological ingredients in Europe. Hot extremes are increasing more rapidly in Europe than the rest of the hemisphere22,23,26, and tens of thousands of deaths across the continent have been linked to recent summer heat35,36, with climate change causing more than half37. As a result, Europe is a particularly important setting in which to study the risk of mass heat-mortality events. We combine two existing approaches to quantify the risk of mass heat mortality across Europe (Methods). First, we use a recently developed machine learning framework38. In this framework, convolutional neural networks (CNNs) are trained on an ensemble of GCMs from the sixth phase of the Coupled Model Intercomparison Project (CMIP6) to predict daily temperatures in three Intergovernmental Panel on Climate Change (IPCC) regions of Europe from (1) the annual global mean temperature (GMT) in the preceding 12 months, (2) the calendar day and (3) modelled daily meteorological conditions. Then, meteorological conditions from ERA5 reanalysis are used as out-of-sample inputs to the trained neural networks to predict ‘counterfactual’ versions of historical heat waves at varying annual GMT. Our method learns the representation in the GCMs of the meteorological drivers of individual extreme heat events, allowing us to quantify the intensity of surface temperature extremes conditional on historical meteorological patterns, independent of projected changes in the frequency or persistence of those patterns. We predict counterfactual events at varying annual GMT from the preceding 12 months, rather than long-term mean GMT, because individual hot years are plausible before long-term climate targets are reached39, these years pose substantial regional climate risks40 and GMT in the previous 12 months is directly detectable in observations41. For this study, we produce counterfactual estimates of five multiweek periods of extreme heat that occurred in July 1994, August 2003, July 2006, June 2019 and August 2023 (Fig. 1). We choos (...truncated)