Future heat-related mortality in Europe driven by compound day-night heatwaves and demographic shifts

Abstract

Anthropogenic climate change is driving summer heat toward more humid conditions, accompanied by more frequent day-night compound heat extremes (high temperatures during both day and night). As the fast-warming and aging continent, Europe faces escalating heat-related health risks. Here, we projected future heat-related mortality in Europe using a distributed lag nonlinear model that incorporates humid heat and compound heat extremes, strengthened by a health risk-based definition of extreme heat and a scenario matrix integrating time-varying adaptation trajectories. Under 2010–2019 adaptation baselines, future heat-related mortality is projected to increase annually by 103.7-135.1 deaths per million people by 2100 across various population-climate scenarios for every degree of global warming, with Western and Eastern Europe suffering the most. If global warming exceeds 2 °C, climate change will dominate (84.0–96.8%) projected increase in heat-related mortality. Across all socioeconomic pathways, even a 50% reduction in heat-related relative risk through physiological adaptation will be insufficient to offset the climate change-driven escalation of future heat-related mortality.

This study projects heat-related mortality in Europe across various adaptation scenarios by modelling humid and compound day-night heat, using a health-based heat definition. Without heat adaptation, mortality could rise by 103.7–135.1 deaths per million people per 1 °C of global warming.

Introduction

The frequency of extreme heat events has increased rapidly in recent decades, accompanied by notable rises in intensity and duration^1,2. These changes exert profound impacts on human health³. Global climate models project that as the Earth warms, the land areas exposed to humid heat and compound day-night heat extremes will expand^4–7, likely exacerbating heat-related health risks^7,8. Since the 1980s, Europe has warmed at a rate nearly double the global average, making it a critical ‘hotspot’ for heat stress⁹. Simultaneously, demographic shifts—including declining birth rate, increasing life expectancy, and an immigration pattern dominated by net inflows of retirees^10–13—are reshaping European population age structure, featured by a growing proportion of heat-vulnerable older adults. Despite its developed economy which might imply stronger adaptive capacity, nearly 21.4% of European population lived in dwellings that were not comfortably cool during summer by 2020¹⁴. This confluence of amplified warming, population aging, and limited heat adaptability renders the continent highly susceptible to heat stress^13,15. Thus, understanding the spatiotemporal dynamics of heat-related mortality and their underlying drivers is of paramount importance for Europe.

Previous studies have indicated that compound heat extremes (i.e., extreme heat during both day and night) pose significantly higher health risks than daytime-only heat extremes^16–20. However, the discourse on heat-related mortality in Europe has remained largely confined to daily mean temperature and mortality records, with even less attention paid to exploring the additional health impacts of prolonged compound heat extremes. Moreover, most studies defined extreme heat using meteorological criteria or specific temperature percentiles, lacking an explicit health-based rationale for selecting threshold metrics^21–23. Montero et al.²⁴ proposed that extreme heat should be defined based on the cause-effect relationship between heat and population health, yet few studies have evaluated the adverse health effects of such extremes across different age groups using localized, health risk-based thresholds. This gap may lead to an underestimation of societal vulnerability to extreme heat²⁵. Most prior studies estimated heat-attributable excess mortality using a function of ambient dry-bulb temperature^26,27, assuming that heat-related mortality is independent of other factors, such as ambient humidity. However, ambient humidity partially determines the rate of evaporative heat loss from the human body, thereby influencing physiological responses to heat^28,29. Given that the regions affected by high humidity are projected to expand in Europe, accounting for the impact of ambient humidity is crucial for accurately assessing heat-related deaths⁴. Furthermore, most previous studies on heat-related mortality have been geographically restricted, focusing on specific cities, countries, or regions^{26,27,30–33}, with few achieving comprehensive cross-national comparisons covering the entire continent of Europe.

Beyond precise quantification of the heat-mortality relationship, robust projection of heat-related mortality risk requires a more comprehensive understanding of uncertainties embedded in climate change and variability, as well as shifts in population size and age structure. Traditional global climate models participating in the 5th and 6th phases of the Coupled Model Intercomparison Project (CMIP) typically include a limited number of ensemble members. This constraint may lead to underestimation of the uncertainty range for future heat-related deaths, as these models cannot fully sample the internal variability of the climate system^31,34. In contrast, single-model large-ensemble simulations have advantages in sampling internal climate variability and its potential impacts on future changes in heat-related mortality. Furthermore, our ability to account for the evolving vulnerability of individuals to future heat exposure is hampered by the traditional omission of population age structure changes. This gap may result in a conservative projection of heat-related deaths^35–37. Another challenge in projecting heat-related health burden lies in the complexity of adaptive capacity. A growing body of evidence indicates that heat adaptation can mitigate future heat-related mortality^38,39, manifesting through two mechanisms: reduced physiological sensitivity to heat and enhanced heat adaptive capacity driven by socioeconomic developments⁴⁰. However, most prior studies either entirely overlooked heat adaptation^3,41–44 or focused on only one aspect of these adaptive mechanisms⁴⁵ when projecting future heat-related mortality.

Here, we develop a comprehensive, Europe-wide, and high-resolution dataset, that incorporates an integrated metric of temperature and humidity, namely the Humidex, along with three categories of health risk-based heatwaves. We employ a distributed-lag nonlinear regression model with a quasi-Poisson distribution and use data aggregated at hourly to weekly scales to estimate location- and age-specific associations between weekly mortality, Humidex, and extreme heat events (including consecutive and non-consecutive daytime, nighttime and compound heat extremes) across 989 NUTS3-level regions in 34 European countries. After validating the robustness and prediction accuracy of the model, we project per capita heat-related mortality under varying levels of global warming from 2024 to 2100 by combining single-model large-ensemble simulations with projections of population size and age structure. We further apply a decomposition approach to elucidate the roles of climate change, population growth, and aging in driving future changes of heat-related deaths across Europe. Finally, we project heat-related mortality under a range of physiological and socioeconomic adaptation scenarios.

Results

Unprecedented warmth and heat-related mortality in Europe

Europe has experienced a notable increase in mean summer temperature since 1950 (Supplementary Fig. 1a). Four of its five hottest summers on record (2003, 2018, 2019, 2021 and 2022) have occurred in the past decade. The Humidex, which accounts for dry-bulb surface air temperature and relative humidity, has increased at a rate of 0.31 °C per decade, exceeding the warming trend of dry-bulb temperature alone (0.23 °C per decade). Relative to the 1986–2005 climatological mean, the regional mean Humidex anomaly in Europe reached 2.1 °C in 2022. Western and Southern Europe are warming most rapidly, with Humidex anomalies of 2.6 °C and 2.7 °C, respectively, in 2022 (Supplementary Fig. 1b).

Mortality risk increases with the weekly Humidex when it exceeds the minimal mortality Humidex (MMH) threshold. Among the three age groups, children (aged 0–15 years) exhibit greater resilience to heat stress, while the elderly (aged >65 years) face the highest risk from summer heat (Fig. 1a). For the working-age population (aged 16–65 years), the MMH is approximately 12 °C, with a comfort Humidex range (where relative mortality risk increases by less than 1) of 10–16 °C. In contrast, the elderly have a higher MMH (16 °C, p < 0.05) but a narrower comfort Humidex range (11–21 °C). These comparatively higher MMH values for older adults align with previous findings^46,47. Across all age groups, heat-related health effects tend to be immediate, with the strongest impact observed at a time lag of 0 weeks (Fig. 1b). The model incorporating Humidex provides a more robust characterization of the heat exposure-mortality response relationship across Europe compared to the model using dry-bulb surface air temperature alone (Supplementary Note 5). Cross-validation of our estimated heat-related mortality against previous findings^47,48 reveals higher predictive accuracy in the Humidex-based model (R² = 0.79) than in the dry-bulb temperature-based model (R² = 0.72). This underscores the critical role of ambient humidity in modulating heat-related excess mortality.

Fig. 1. Heat–mortality associations and heat-related deaths in Europe.

a Cumulative relative risk–weekly Humidex response curves for three age groups: children (0–14 years), working-age population (15–64 years) and elderly (>65 years). Shown for each age group are the cumulative relative risk of death (lines) with a corresponding 95% confidence interval (shadings). b Heatmap illustrating the lagged effects of heat exposure on mortality. Red indicates an increased mortality risk (RR > 1), while blue indicates a decreased mortality risk (RR < 1). c Bar chart displaying variations in mortality risk with the duration of heat exposure. Each bar represents the absolute mortality risk of compound daytime-nighttime heat extremes (CH) relative to daytime-only heat extremes (DH). Absolute risk refers to the increased mortality risk during hot days compared to normal days. d Estimated total heat-related mortality across 34 European countries during summers (weeks 22–35) of 2010–2022. These nations are geographically divided into four regions (Northern, Southern, Eastern and Western Europe) based on the United Nations Geoscheme. Dark grey bars denote heat-related excess deaths in the elderly group, and light grey bars denote those in the all-age population.

Using a health risk-based approach (Methods), we categorize heat extremes during a week into six distinct types: consecutive daytime-only heat extremes (CDHs), consecutive nighttime-only heat extremes (CNHs), consecutive compound day-night heat extremes (CCHs), non-consecutive daytime-only heat extremes (NDHs), non-consecutive nighttime-only heat extremes (NNHs), and non-consecutive compound day-night heat extremes (NCHs). This classification enables the assessment of how consecutive versus non-consecutive heat extremes impact mortality. We find that heat extremes occurring consecutively (lasting for more than three days within a week) exert a more significant effect on mortality than non-consecutive ones (Fig. 1b). For children, CDHs are associated with a 2.2-fold higher mortality risk compared to CCHs (Fig. 1c). However, for older adults, this effect is reversed: the relative mortality risk of CCHs is over twice that of CDHs of the same duration (Fig. 1c).

Prior to projecting future heat-related excess deaths in Europe, it is critical to evaluate the performance of the heat exposure-mortality response model. We cross-validate the model’s predictive accuracy using a three-step validation approach (Supplementary Note 3), which confirms high predictive accuracy. As shown in Supplementary Fig. 6, there is a high spatial correlation (R² = 0.79) between the heat-related excess deaths predicted by our model and those reported in other studies. Notably, to avoid confounding effects of the COVID-19 pandemic on heat-related mortality assessment, we exclude data from 2020 to 2021 during the training and validation of the heat exposure-mortality response model.

Based on the well-validated heat-mortality association model, we estimate that extreme heat in the summers of 2010–2022 was associated with 368,183 deaths across 34 European countries (95% confidence interval (CI): 261,989–432,305), with the highest mortality recorded in 2022 (Fig. 1d). The elderly are particularly vulnerable to summer heat stress and approximately 89.4% of these deaths occurred among individuals aged 65 years or older. Southern and Eastern Europe experienced the highest absolute number of heat-related deaths. Southern Europe had the highest proportion (91.0%) of elderly deaths, while Eastern Europe had the lowest (84.8%). Italy, Spain, Greece, France, and Germany ranked as the top five countries with the highest heat-related mortality, significantly exceeding other European nations (Fig. 1d and Supplementary Fig. 2). Northern Europe had the lowest total heat-related deaths (14,690, 95% CI: 8636–20,040), yet it had the second-highest proportion of elderly deaths (90.0%), only behind the severely affected Southern Europe.

We calculate cumulative deaths associated with moderate heat as those occurring in summer weeks when Humidex values fall between the MMH and the localized extreme heat threshold (see Supplementary Table 1 for local thresholds at the NUTS-1 level). The cumulative deaths related to extreme heat are calculated for summer weeks when Humidex exceeds the localized extreme heat threshold. The threshold is defined based on the cause-effect relationship between Humidex and population health (Methods). Supplementary Table 6 summarizes the number of deaths among the three age groups attributable to moderate and extreme heat from 2010 to 2022. In the summer of 2022 alone, there were an estimated 46,405 heat-related excess deaths in Europe (95% CI: 39,470–54,125), accounting for 12.7% of all summer heat-related deaths from 2010 to 2022 (95% CI: 10.8–14.8%). Of these deaths, extreme heat was associated with 22,770 (95% CI:19,668–26,016), while moderate heat contributed 23,635 (95% CI: 19,802–28,109). Notably, after 2015, deaths attributed to extreme heat surpassed those from moderate heat (Supplementary Table 6). Our estimate for Spain—where extreme heat-related excess deaths were 0.80 times that of moderate heat—aligns with a recent assessment, which found that extreme heat was linked to 0.79 times as many excess deaths as moderate heat in the summer of 2022 in Spain⁴⁹.

Future burden of heat-related mortality

The aforementioned exposure-response functions for heat-related mortality across the three age groups are used to project future changes in heat-related excess mortality per capita in Europe from 2024 to 2100 (see Methods). We integrate population estimates corresponding to four shared socioeconomic pathways (SSPs) – SSP1, SSP2, SSP3 and SSP5 –with single-model large-ensemble simulations under four future emission scenarios: SSP1-2.6, SSP2-4.5, SSP3-7.0 and SSP5-8.5 (see Methods). Per capita heat-related mortality rates are projected to increase by 217.9% (95% CI: 134.2–327.4%) for SSP1-2.6, 195.9% (95% CI: 140.7–250.4%) under SSP2-4.5, 257.3% (95% CI: 207.2–303.8%) under SSP3-7.0 and 255.9% (95% CI: 213.8–300.1%) under SSP5-8.5 with every degree of global warming (Fig. 2).

Fig. 2. Projected changes in heat-related mortality in Europe.

a Growth rate of heat-related excess mortality per capita under different population-climate scenarios. Squares and circles indicate comparisons for the near-mid-century and mid-end-century periods, respectively. The pre-mid-21^st century growth rate is calculated by comparing mean mortality per capita between 2045–2055 and 2025–2030. The post-mid-21^st century growth rate is calculated by comparing the mean mortality per capita between 2095–2100 and 2045–2055. Black bars represent the 95% confidence intervals of the projected growth rate of heat-related mortality per capita. b Variation in heat-related deaths per million people with global warming. Specific global warming levels are measured using the 20-year moving average of global mean surface temperature anomalies (relative to the 1850–1900 pre-industrial period). The years with the 20-year mean global warming magnitude exceeding the specific level are determined to calculate heat-related deaths per capita. c Regional growth rates of heat-related excess mortality per capita with global warming. Growth rates are represented by the linear slope of heat-related mortality against global warming levels—that is, the percentage change in additional heat-related mortality per capita per degree of global warming. These percentage changes are estimated relative to the 2022 baseline of 53.8 heat-related deaths per million people. Europe (EU) is divided into four regions: Northern (NEU), Western (WEU), Eastern (EEU) and Southern Europe (SEU), with baseline heat-related excess deaths of 5.46, 19.3, 48.4 and 145.0 per million people, respectively. Note that the 2022 baselines reported here differ from those in the previous section due to the exclusion of regions with missing data in projections.

We compare the growth rates of heat-related deaths per capita before and after the middle of the 21st century (Fig. 2a). The difference between 2045–2050 and 2025–2030 (between 2095–2100 and 2045–2050) is used to represent the growth rate before (after) the middle of the 21st century. Under the high-emission scenarios (SSP3-7.0 and SSP5-8.5), the growth rate of heat-related deaths per capita will accelerate significantly in the post-mid-21st century, while under low- (SSP1-2.6) and medium-emission (SSP2-4.5) scenarios, the increase will be more pronounced in the pre-mid-21st century. Heat-related deaths are projected to increase most slowly under SSP2-4.5 (103.7 per million people per degree of global warming) compared to SSP1-2.6 (115.9 per million people), SSP3-7.0 (135.1 per million people) and SSP5-8.5 (133.4 per million people) (Fig. 2b). More than 92.7% of heat-related deaths will be experienced in individuals aged 65 years or older. The growth rates of heat-related mortality with each degree of global warming showed strong spatial diversity in response to climate change and population dynamics (Fig. 2c). The largest increase in heat-related mortality per capita will be found in Western Europe and Eastern Europe, with increase rates of 307.0–548.7% and 325.2–481.4% per degree of global warming, respectively. For an additional 1 °C of global warming, Southern Europe will show the lowest growth rates, at 162.7–212.1%. However, the growth rate in Northern Europe (591.4–600.1%) will exceed that of Western Europe (488.4–548.7%) under high-emission warming scenarios SSP3-7.0 and SSP5-8.5.

To better characterize the spatial distribution of future heat-related mortality burdens, the annual average number of heat-related deaths per million people under global warming levels of 1.5 °C, 2.0 °C, 3.0 °C and 4.0 °C is illustrated in Fig. 3. For each warming level, the annual average deaths are calculated as the 10-year mean value centred on the year when that specific warming level is reached. Across various population-climate scenarios, if global warming exceeds 2.0 °C, the annual average number of summer heat-related deaths per million inhabitants in Europe could reach 160.7–170.2. If global warming continues to rise and surpasses 4 °C, summer heat could cause more than 467.4 deaths per million people annually across Europe, with a distinct north-south increasing trend. By the end of the 21st century, nations with rapidly aging populations—including Germany, Italy, Spain, France and Slovenia—are projected to experience a disproportionate escalation in heat-related health impacts. As global warming level increases from 1.5 °C to 4 °C, new hotspots for high annual average heat-related mortality per million people will emerge, including Hungary (146.4–1105.1), Italy (281.5–1163.0), Greece (283.1–1230.3), Bulgaria (170.0–1314.4), and Romania (152.2–1331.3). The spatial trend of future heat-related mortality burdens aligns with a recent study⁵⁰, though our results indicate a higher heat burden in the coastal region of Western and Southern Europe. Notably, while the prior study assessed the impacts of both cold and hot extremes on human health, the current analysis examines a more comprehensive future heat burden by integrating day-night compound heat extremes and humid heat rather than focusing solely on dry-bulb air temperature.

Fig. 3. Geographical distributions of projected heat-related excess mortality under different global warming levels.

The heat-related excess mortality is the projected annual average heat-related excess deaths per million population at a 1.5 °C, b 2.0 °C, c 3.0 °C, and (d) 4.0 °C. Results are based on the SSP5-8.5 population-climate scenario. Additional projections for alternative scenarios (SSP1-2.6, SSP2-4.5, and SSP3-7.0) are provided in Supplementary Figs. 10–12. Regions with missing data are shaded grey.

Roles of changes in climate, population size and age structure in future heat-related mortality

Our preliminary analysis reveals a nonlinear flattening of the relative risk curve during the adaptation process, which complicates the formulation of optimal assumptions (see Supplementary Note 7). Based on adaptation levels with further improvements observed during 2010–2019, future heat-related mortality burden will be modulated by changes in climate, population size and age structure. To better identify the drivers of future heat-related mortality in Europe, we develop a decomposition method to analyse the individual contributions of climate change, population size change, and aging to heat-related deaths at the country level (see Methods). The annual average heat-related deaths under different global warming levels are calculated as the mean value across all population-climate scenarios for each specific warming level. As shown in Fig. 4, heat-related deaths attributed to climate change are projected to increase by 36.6–401.6 additional deaths per million people per year per degree of global warming. Meanwhile, 13.6–21.1 heat-related deaths per million people per year can be attributed to population aging.

Fig. 4. The relative contributions of climate change, population size changes, and population aging to future heat-related excess deaths per million people in Europe.

a Variations in the individual effects of climate change, population size changes and population aging, along with their total effects (overall impacts) on future heat-related excess deaths in Europe as global warming intensifies. Solid lines represent projected annual average heat-related excess deaths when all factors or a specific factor are accounted for, while dashed lines denote projections with all factors held constant. b Annual average number of heat-related excess deaths attributable to changes in climate, population size and age structure under a 3.0 °C global warming level. Additional projections for alternative warming scenarios (1.5 °C, 2.0 °C, 2.5 °C, and 4.0 °C) are provided in Supplementary Figs. 13–16. c Heatmap displaying the relative contributions of climate change, population size changes, and population aging under global warming levels of 1.5 °C, 2.0 °C, 2.5 °C, 3.0 °C and 4.0 °C. The relative contribution of each factor is determined by comparing the annual average number of heat-related deaths per million people attributable to that specific factor versus the total from all factors.

By comparing the annual average number of heat-related deaths per million people attributable to each factor under 1.5 °C, 2.0 °C, 2.5 °C, 3.0 °C and 4.0 °C warming scenarios (Fig. 4c), we find that climate change and population aging are the primary contributors to the growth of heat-related mortality across all population-climate scenarios, with climate change acting as the dominant driver. When global warming reaches 2 °C, climate change is projected to account for 84.0% of annual heat-related deaths—an impact expected to surpass that of population aging. Population decline will mitigate heat-related deaths to a limited extent (2.5%). The percentage change in future heat-related mortality due to population aging decreases as global warming intensifies, falling from 27.8% under the 1.5 °C warming scenario to 5.1% under the 4.0 °C warming scenario. Furthermore, if global warming exceeds 4 °C, climate change will contribute to 96.8% of the annual increase in heat-related deaths in Europe, substantially exacerbating the heat-related mortality burden.

Implications for integrated European heat adaptation policy

To guide the formulation of integrated heat adaptation policy aimed at reducing future heat-related mortality in Europe, we project heat-related mortality under five physiological-socioeconomic adaptation scenarios (see Methods). The zero-adaptation (0%) scenario assumes a future without physiological heat adaptation. In this case, improvements in heat adaptation capacity are solely driven by socioeconomic development under different shared socioeconomic pathways (SSPs). The remaining four scenarios incorporate incremental levels of physiological adaptation: low (5%), moderate (10%), medium (25%), and high (50%).

As illustrated in Fig. 5, the estimated percentage reductions in future heat-related mortality attributed to socioeconomic adaptations alone are 0.8–14.5% under SSP1-2.6, 3.0–11.9% under SSP2-4.5, 1.4–3.3% under SSP3-7.0, and 2.0–10.8% under SSP5-8.5. Low and moderate levels of physiological adaptation exert only a modest impact on future heat-related mortality. Even a 50% attenuation of the excess relative risk through physiological adaptation is insufficient to offset the mortality effect of increased population exposure to heat in a warming world. Notably, under SSP2-4.5 with 2 °C of warming, high-level adaptation reduces heat-related mortality to 106.5 deaths per million people (95% CI: 68.3–154.2), a level comparable to that observed under SSP1-2.6 with 1.5 °C of warming and no adaptation (112.7 deaths per million, 95% CI: 69.6–169.9). By the end of the 21st century, a 50% reduction in excess relative risk through physiological adaptation could lower heat-related deaths in Europe by approximately 27.3% (95% CI: 26.9-27.8%) under SSP3-7.0 and 32.5% (95% CI: 30.8-34.5%) under SSP5-8.5. Our results also highlight the spatial heterogeneity in the impacts of physiological and socioeconomic adaptations on future heat-related mortality (Supplementary Figs. 24–27). Heat adaptation measures show relatively limited effectiveness in Southern European countries, whereas countries such as Germany, Finland, Sweden, and the United Kingdom exhibit the greatest potential for adaptation to mitigate increases in heat-related mortality.

Fig. 5. Projections of annual heat-related deaths per million people in Europe.

Dashed lines denote projections for the baseline no-adaptation scenario. Solid lines represent projections for five adaptation scenarios: zero-adaptation (0% reduction in excess relative risk), low adaptation (5% reduction), moderate adaptation (10% reduction), medium adaptation (25% reduction), and high adaptation (50% reduction). Note that the baseline no-adaptation and zero-adaptation scenarios represent distinct future states. The zero-adaptation scenario assumes no physiological heat adaptation but incorporates socioeconomic heat adaptation. In contrast, the baseline no-adaptation scenario assumes that future relative risk remains at the current level.

Discussion

We investigate the health impact of compound day-night humid heat extremes on three age groups and project future heat-related mortality in Europe from 2024 to 2100. When exposure lasts for more than three days, compound day-night heat extremes pose a much greater adverse impact on older adults than daytime-only heat extremes, with mortality risk increasing by ~2.0-fold or more. Consistent with previous studies^35,36,51, countries with the highest heat-related mortality include Italy, Spain, and Greece. Under the adaptation level observed during 2010–2019, our projection indicates substantial increases in heat-related deaths during the summer (weeks 22–35). The growth rates per degree of global warming vary by climate-population scenario: 217.9% (95% CI: 134.2–327.4%) for SSP1-2.6; 195.9% (95% CI: 140.7–250.4%) for SSP2-4.5; 257.3% (95% CI: 207.2–303.8%) for SSP3-7.0; and 255.9% (95% CI: 213.8–300.1%) for SSP5-8.5. Notably, Western Europe is projected to experience the largest increase (307.0–548.7%), followed by Eastern Europe (325.2–481.4%). We identify distinct patterns of the two major drivers of future heat-related mortality: climate change and population aging. As global warming intensifies, the contribution of climate change to the heat-related mortality burden is projected to increase from 74.6% to 96.8%. In contrast, the contribution of aging will follow a different trajectory, peaking at ~27.8%.

To the best of our knowledge, few studies to date have used a weather indicator that combines dry-bulb air temperature and humidity to model the responses of Europe-wide mortality to day-night compound, daytime-only, and nighttime-only heat extremes across various age groups. A related study¹⁶, which examined the relationship between hot nights and mortality in four Southern European countries, reported excess mortality risks from compound heat ranging from 12% to 37%, aligning well with our findings (over 30%). Recent research has also highlighted the importance of incorporating humidity when assessing the human health impact of heat stress¹⁸. Our findings indicate that persistent extreme humid heat during both day and night is associated with higher mortality rates among the elderly. Sweating is a crucial thermoregulatory mechanism for the human body⁵². As a result, reduced sweating efficiency due to high ambient humidity may elevate body temperature and increase heat-related mortality risk. Nevertheless, some studies argued that accounting for humidity does not significantly improve the predictive accuracy of heat exposure-mortality response models^29,53,54. We note that most of these studies were conducted in predominantly dry conditions, potentially overlooking a global shift toward more humid heat (Supplementary Fig. 16). In this sense, our results provide a contrasting perspective on the humid heat-related mortality burden. Model validation in our study confirms that incorporating humidity improves the representation of heat-mortality relationship across different time periods. Models including humidity outperform those using dry-bulb air temperature alone in estimating heat-related mortality (R² = 0.79 vs. 0.72). Despite this, the role of humidity in such models may vary with factors like local climate conditions and data characteristics. Our study contributes to the ongoing discourse by emphasizing the necessity of including humidity in future health impact assessments as humid heat becomes more prevalent. It also highlights the importance of addressing the potential health impacts of humidity during extreme heat events. Further research is needed to determine whether the impact of humidity is becoming more pronounced over time and how it interacts with other factors, such as surface air temperature and population vulnerability.

Building upon previously established epidemiological frameworks^43,44,55,56, the baseline no-adaptation projections employ a fixed heat-mortality curve shape and static thermal vulnerability parameters. However, compelling evidence reveals the temporal dynamics of human heat adaptive capacity^40,57,58, suggesting that conventional projections may overestimate mortality risks given growing heat awareness and expanding cooling infrastructure. To address this gap, we conduct multi-dimensional scenario analyses integrating physiological and socioeconomic adaptations. Our findings indicate that even a 50% reduction in heat-related mortality risk through physiological adaptation fails to offset the amplified mortality risk induced by future warming, underscoring the need for extraordinary adaptation measures to mitigate climate change-related health costs. Across all SSP scenarios (SSP1, SSP2, SSP3, SSP5), socioeconomic adaptations show limited efficacy in reducing heat-related deaths (15%), with particular modest effects under SSP3-7.0 (1.3–3.3%). Compared to socioeconomic adaptation—whose impacts can be quantified with measurable metrics—physiological adaptation mechanisms exhibit multidimensional complexity. Analysis of the latest relative risk (RR) curve (2010–2019) reveals a 36.8% reduction in projected heat-related mortality compared to the baseline period (2000–2009) (Supplementary Table 7). This nonlinear progression indicates temporal non-stationarity in adaptation rates (Supplementary Note 7), reflecting the dynamic nature of human heat adaptive capacity, a phenomenon linked to social infrastructure support networks, healthcare accessibility, and health education^58–60. Current modelling frameworks oversimplify adaptation trajectories by neglecting two critical aspects: the upper limits of physiological adaptation under projected climate scenarios and the varying heat adaptation capacities across different population groups and regions. Additionally, nonlinear interactions between physiological and socioeconomic adaptations may influence future heat-related mortality. Therefore, future research should focus on developing methodologies to determine physiological adaptation ceilings, establishing standardized metrics for quantifying differential heat adaptation capacities across population groups and regions, and investigating how nonlinear interactions between human acclimatization and socioeconomic development—particularly educational attainment as a potential modifier of adaptive behaviours—shape heat adaptation outcomes.

Based on adaptation levels with a further improvement observed during 2010–2019, our analysis reveals a striking finding: under the high-emission scenario (SSP3-7.0), heat-related deaths in Northern Europe are projected to increase by 591.4% (95% CI: 390.6–809.9%). This rate is comparable to that of Southern Europe (171.7%, 95% CI: 150.0–192.5%)—an area previously overlooked in many studies^47,61,62. Although current heat-related deaths in Northern Europe are less than one-tenth of those in Southern Europe, the low heat resilience and limited access to air conditioning may leave the region vulnerable to climate change. High-latitude cities in Northern Europe have historically prioritized solar efficiency and minimized ventilation in architectural designs to resist cold environments⁶³. A recent analysis⁶⁴ shows that surface air temperature trends in Lithuania, Finland, and Latvia have outpaced the European average, corroborating our findings and underscoring the need for intensified monitoring of these regions. However, statistics from the World Health Organization (WHO) Regional Office for Europe indicated that, except for Sweden, most Northern European countries still lack comprehensive heat-health action plans and early warning systems⁶⁵. An even more urgent challenge than global warming is Europe’s rapidly ageing population^12,66,67. Our study projects that if global warming reaches 1.5 °C, elderly heat-related deaths in Southern Europe will account for over 93.8% of all heat-related deaths per year in Europe. Given their heightened vulnerability, elderly individuals should be the primary focus of Europe’s heat response strategies^68,69. However, most current heat-health strategies in Europe merely acknowledge vulnerable groups but offer no targeted actions for them. This may also explain the abnormally high heat-related mortality in Italy, a country long at the forefront of summer heat^47,50,70,71. Italy introduced heat health prevention plans and warning systems as early as 2004⁷², making it one of the largest air conditioning markets in Europe. However, both health concerns about the side effects of air conditioning and environmental awareness⁷³ have limited the use of air conditioning⁷⁴. The energy-saving law⁷⁵ and high electricity prices⁷⁶ may also potentially exacerbate regional vulnerability in Italy. These dynamics underscore the need for nuanced investigations into the drivers of vulnerability and adaptation across Europe, with the aim of developing coordinated, context-specific strategies to mitigate heat-related health impacts.

Some limitations should be acknowledged. First, using weekly data may underestimate heat-related deaths due to challenges in resolving the effects of continuous heat extremes. Due to the lack of finer-resolution mortality data, we use weekly Humidex and mortality data to establish heat-mortality model, which categorically aggregate different durations of continuous heat extremes on a weekly basis. Additionally, our analysis does not distinguish indoor and outdoor exposures, potentially overestimating projected mortality rate. Individuals spending more time indoors may benefit from cooling measures and thus have greater resilience to heat stress. Future research should incorporate indoor exposure data to refine heat-related mortality risk estimate and clarify how living environments modulate individual vulnerability to extreme heat. Second, the mortality data used lacked precise cause-of-death information. To mitigate measurement errors from confounding factors (e.g., crime, air pollution), we incorporate region-specific fixed effects and seasonal covariates, assuming these uncertainties follow a zero-mean random distribution. Third, our study use a single set of single-model large-ensemble simulations, namely the new Max Planck Institute (MPI) grand ensemble with CMIP6. While it helped to assess the uncertainty arising from internal climate variability, it may have constrained our understanding of model structure uncertainty. However, projecting heat-related mortality changes under specific global warming levels may mitigate the effect of model structural uncertainty on future projections.

Methods

We use historical weekly all-cause mortality records and climate data at the NUTS-3 level from 2010 to 2019 to establish the association between summer heat and mortality. Leveraging these datasets, we develop a distributed lag nonlinear model with a quasi-Poisson distribution, which accounts for consecutive and non-consecutive daytime-only, nighttime-only and day-night compound heat extremes (Fig. 6). We design a three-step procedure framework to rigorously evaluate the model’s generalization capacity and robustness. We conduct a model-based prediction to estimate future heat-related excess deaths up to 2100 using climate projections under four representative emission scenarios (SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5) and population sizes from the Gridded Population of the World v4 and EUROPOP2019 baseline projections. Subsequently, we employ a decomposition approach to quantify the relative contributions of three factors—climate change, population size change, and population aging—to the projected future burden of heat-related mortality. Finally, we extend the projections to assess future heat-related mortality under diverse physiological-socioeconomic adaptation scenarios.

Fig. 6. Schematic flowchart illustrating the data and models used in this study.

Here, CDH, CNH, CCH, NDH, NNH, and NCH denote consecutive daytime-only heat extreme, consecutive nighttime-only heat extreme, consecutive day-night compound heat extreme, non-consecutive daytime-only heat extreme, non-consecutive nighttime-only heat extreme, and non-consecutive day-night compound heat extreme, respectively. The regression model is trained using Humidex and mortality data during the summers (weeks 22–35) of 2010–2019. To minimize the confounding effects of the COVID-19 pandemic on mortality, data from 2020 to 2021 are excluded from both the model training and validation processes.

Mortality records

We use Eurostat mortality records during the summers (weeks 22–35) from 2010 to 2019 to minimize the confounding effects of the COVID-19 pandemic on mortality. This dataset archives the weekly counts of all-cause deaths at the third level of the Nomenclature of Territorial Units for Statistics (NUTS-3), which are reported by 34 European nations (Albania, Austria, Belgium, Bulgaria, Croatia, Cyprus, the Czech Republic, Denmark, Estonia, Finland, France, Germany, Greece, Hungary, Iceland, Italy, Latvia, Liechtenstein, Lithuania, Luxembourg, Montenegro, Malta, Netherlands, Norway, Poland, Portugal, Romania, Serbia, Slovakia, Slovenia, Spain, Sweden, Switzerland, and the United Kingdom; last accessed January 19, 2025). At the NUTS-3 level, some countries, including Germany, Slovenia, Ireland, Estonia, and Malta, have missing data. To fill in these missing observations, our analysis is adjusted to use the finest available spatial units, such as the NUTS-1 level data for Germany. Population subgroups are categoried by gender and three age groups: children (aged 0–14 years), the working-age population (aged 15–64 years), and the elderly adults (aged 65 years or above).

Climatic data

To calculate the Humidex, we use an improved land component dataset of the fifth-generation global climate reanalysis of the European Centre for Medium-Range Weather Forecasts (ERA5-Land). The population-weighted approach performs better in detecting the effects of climate change, particularly in regions with large climate and population variability, as suggested by Evan et al.⁷⁷. Therefore, climate variables from ERA5-Land are retrieved as hourly population-weighted means at the NUTS-3 level. For historical data, population weighting is determined using gridded population data with a 1-km horizontal resolution (available at https://www.worldpop.org/). The weight of each grid cell is calculated as the ratio of the population residing in it to the total population within the corresponding NUTS-3-level region. Gridded climate variables are simply resampled to 1-km grids to align with the population data resolution. Humidex⁷⁸ is computed based on the 2-m dew-point temperature ($T_d$; K) and the 2-m air temperature ($T_a$; °C), as shown in Eqs. (1) and (2):

$$Humidex=T_a+\frac{5}{9}*\left(e-10\right)$$ 1

$$e=6.11*exp\left(5417.753*\left(\left(1/273.16\right)-\left(1/T_d\right)\right)\right)$$ 2

where e is vapour pressure (hPa) and exp denotes the exponential function in the natural base.

In addition, we incorporate monthly relative humidity and surface air temperature data from the new MPI Grand Ensemble with CMIP6. These grand ensemble simulations were performed under combined natural and anthropogenic forcing, including historical runs up to 2014 and future projections under various SSPs, such as SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5. For each scenario experiment, there are 50 ensemble members for the years from 2015 to 2100, available at horizontal resolution of 1.875° × 1.875°. Each member is subjected to identical external forcing but initialized with slightly different initial conditions. We calculate the 2-m dew-point temperature ($T_d$) based on relative humidity (RH, %) using an algorithm from an earlier study (see Eq. (3))⁷⁹:

$$T_d=T_a-\left(\frac{100-RH}{5}\right)$$ 3

To estimate future summer Humidex, we calculate model-based projected changes of seasonal mean Humidex relative to 2022 and then add these projected changes to the observed hourly Humidex evolution during the summer of 2022. To facilitate more sophisticated projections, we use the model developed by Wu et al.⁸⁰ to generate a high-spatiotemporal-resolution (0.01° and hourly) reconstruction dataset of Humidex. To align the spatiotemporal resolutions of the model simulations with hourly data, the model projections are interpolated to the same spatial resolution using a nearest-neighbour resampling algorithm. The accuracy of the Humidex data is validated against the in-situ observations from the National Centers for Environmental Information⁸¹. More details can be found in Supplementary Note 2.

Population and age structure projection data

Since 2014, Eurostat has provided population statistics by sex, five-year age intervals, and region of residence at the NUTS-3 level (last accessed January 19, 2025). We aggregate these data into three broader age groups: 0–14, 15–64, and 65 or older. Due to limitations in data availability, population age structure data from 2010 to 2013 are estimated by multiplying the total population at the NUTS-2 level for each specific year by the age-group proportions in 2014. Meanwhile, several regions—including Italy, Poland, the Czech Republic, Norway and the United Kingdom—exhibit a ‘b’ (break in time series) in their observation flags or lacked updated data after 2019, We fill in these missing data after 2019 using their latest available population numbers. Future gridded population projections from 2020 to 2100 are derived from Gridded Population of the World v4 (last accessed June 27, 2024), which provides data at a spatial resolution of 1 km under different SSPs⁸². These SSPs depict five alternative trajectories for future world, each grounded in distinct assumptions regarding social, economic, demographic, environmental, and technological developments. We focus on four scenarios: SSP1 (sustainability), SSP2 (middle of the road), SSP3 (regional rivalry), and SSP5 (fossil-fuelled development) scenarios. The population proportion of each age group and the projected total population under corresponding scenarios are multiplied to estimate the population size of each age group in each sub-region. The proportions of the three age groups are derived from the latest Eurostat population projections via EUROPOP2019 (last accessed: January 19, 2025). These projections delineate prospective changes in population size and demographic composition across 30 European countries from 2020 to 2100. This dataset includes three scenarios: baseline, no migration, and no inter-regional migration. We use the baseline projections, which were built on a set of assumptions on age-specific fertility rates, mortality probabilities and migration patterns, comprising both international and internal movements. The selected baseline scenario describes a projected trajectory of the population age structure based on these assumptions. For missing data pertaining to the United Kingdom, we employ national population projections provided by the Office for National Statistics. This dataset provides population sizes for the United Kingdom and its constituent countries over the next 100 years. We tally the projected age structure at the NUTS-1 level and assume that the population age structure at the NUTS-3 level is consistent with its corresponding NUTS-1 level statistical unit.

Social-economic data

To estimate the log-linear relationship between relative risk (RR) and the logarithm of GDP per capita, we use historical gross domestic product (GDP) data, which are sourced from Eurostat (https://ec.europa.eu/eurostat/). The dataset provides annual GDP per inhabitant at the NUTS-3 regional level, spanning from 2000 to 2023, with values reported in millions of euros. Future GDP projections are extracted from the gridded GDP projections compatible with the five SSPs by Murakami et al.⁸³. These projections, reported in millions of United States dollars (USD) at 2005-year rate, cover the period from 1850 to 2100 by 10-year intervals at a 1/12-degree spatial resolution. All data are subsequently converted to uniform units of million USDs in 2005-year rate for further analysis.

Health risk-based definitions of daytime-only, nighttime-only and day-night compound heat extremes

Each day is divided into two subperiods: daytime (from sunrise to sunset), and nighttime (from sunset to sunrise). Precise sunrise and sunset timings for each day and location are determined using the SunCalc package on the Google Earth Engine platform. Previous studies have shown that human thermoregulation varies with environmental conditions⁸⁴, indicating that long-term evolution may lead to adaptive diversification to heat extremes. Following the method proposed by Montero et al.²⁴, we determine localized health risk-based heat thresholds based on the relationship between Humidex and mortality. The procedure involves three steps. First, we remove the seasonality, periodicities, and long-term trends of the mortality time series using the Box-Jenkins stochastic procedures (ARIMA). Then, we establish a correlation framework between mortality residuals and binned Humidex values (grouped in 2 °C increments) across all the NUTS-1 regions. Finally, we identify the daytime/nighttime heat threshold as the mid-point of the Humidex interval where the residual mortality values show a statistically significant increase compared to the population-mean baseline. We calculate the population-mean baseline as the 10-year average of residual weekly mortality data from 2010 to 2019. Statistical significance is assessed using the one-sample Student’s t test, with the significance level set to 0.05. For regions where no statistically significant differences are observed at the 0.05 level, we relax the threshold to 0.1 to accommodate the reduced statistical power due to data sparsity. If statistical significance remains unachieved, we define the extreme daytime/nighttime heat threshold as the 98th percentile of its local maximum/minimum weekly mean Humidex values. All the parameters for the detrending model are optimized using an automated function from the ‘tseries’ package in R 4.1.1. The heat thresholds of each region can be found in Supplementary Note 1. A sub-period (daytime/nighttime) is classified as a heat extreme if its mean Humidex value exceeded the threshold. If the heat extreme occurs solely during the daytime or nighttime, we label it as a daytime-only heat extreme (DH) or a nighttime-only heat extreme (NH). If the heat extreme occurs throughout both the daytime and nighttime, we define it as a day-night compound heat extreme (CH). Additionally, DHs, NHs, and CHs within a week are further categorized into consecutive (CDH, CNH, and CCH) and non-consecutive (NDH, NNH, and NCH) heat extremes.

Humidex-mortality relationship

We train a regression model using Humidex and mortality during the summer (weeks 22–35) from 2010 to 2019. Specifically, a distributed lag nonlinear model²⁷ with a quasi-Poisson distribution is used to quantify the geographically dependent relationship between weekly counts of all-cause mortality and weekly mean Humidex, as well as consecutive and non-consecutive daytime-only, nighttime-only, and day–night compound heat extremes, as formalized in Eqs. (4) and (5):

$${\mathrm{Y}}_{\mathrm{a},\mathrm{t},\mathrm{c}}~\mathrm{quasi}-\mathrm{Poisson}\left(\mathrm{E}\left[{\mathrm{\mu }}_{\mathrm{a},\mathrm{t},\mathrm{c}}\right]\right)$$ 4

$$\log \left(E\left[{\mu }_{a,t,c}\right]\right)=f\left(H_{t,c},\theta \right)+W\left({CDH}_{t,c}\right)+W\left({CNH}_{t,c}\right)+W\left({CCH}_{t,c}\right)+W\left({NDH}_{t,c}\right)+W\left({NNH}_{t,c}\right)+W\left({NCH}_{t,c}\right)+\log \left({pop}_{a,c}\right)+ns\left(year,df=4\right)+ns\left(week,df=3\right)+\alpha +{\tau }_c$$ 5

where f represents a cross-basis function that models the heat exposure-mortality response relationship between the mean Humidex ${\mathrm{H}}_{\mathrm{t},\mathrm{c}}$ and mortality $E\left[{\mu }_{a,t,c}\right]$ for age group a in week t and region c using a B-spline of degree 2 with one knot at the 80th percentile of the weekly Humidex distribution and a time-lag effect function using a natural spline with integer lag values $\theta$ of 1, 2 and 4 weeks chosen by the log-knots function. $W\left({DH}_{t,c}\right)$, $W\left({NH}_{t,c}\right)$, and $W\left({CH}_{t,c}\right)$ are strata functions that model the effects of daytime-only, nighttime-only, and day-night compound heat extremes, respectively. Each stratum function classifies different durations of summer heat (consecutive and non-consecutive daytime-only, nighttime-only or day-night compound heat extremes) within a week into three categories: 1 day, 2 days, and more than 3 days. $\log \left({pop}_{a,c}\right)$ serves as an offset term to adjust for the population size of each age group in each region, considering the possibility that regions with smaller populations might erroneously exhibit higher expected number of deaths than those with larger populations. In addition, we use two natural cubic splines of time with 4 and 3 degrees of freedom (df) to control the long-term trends and seasonality. α is the intercept. ${\tau }_c$ is a factor variable accounting for geographic variability across 989 NUTS-3-level regions. The model is fitted within three age groups: children (0–14 years), working-age population (15–64 years), and elderly adults (65+ years).

The model accounts for both measured (i.e., the Humidex and consecutive and non-consecutive daytime-only, nighttime-only, and day-night compound heat extremes) and unmeasured time-invariant confounders by analysing temporal variations of dependent and independent variables for each location²⁷. A three-step validation procedure is designed to validate the model’s robustness and generalizability. First, the data are divided into training (2010–2014) and test (2015–2019) sets to verify the estimates of the risk contributions at different temperature exposure intensities. Second, we employ a bootstrapping technique to validate the sensitivity of our models to data variability. Third, cross-validation is performed by comparing the model’s prediction of heat-related mortality with findings from prior studies. More details are provided in Supplementary Note 3. Results of the sensitivity analysis of parameter settings are available in Supplementary Note 4. Additionally, a comparison analysis is conducted between the 2-m dry-bulb air temperature and the Humidex to assess their sensitivities in characterizing the heat-mortality response (Supplementary Note 5).

Projection of heat-related mortality burdens

We project the future burden of heat-related deaths in Europe under SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5. The above-validated heat exposure-mortality response model is integrated with the projections of the weekly mean Humidex to compute the number of heat-related deaths in the summer associated with global warming by the end of the twenty-first century. Regional heat-related mortality is calculated using Eqs. (6) and (7), and then summed to obtain the sub-national and European heat-exposure burdens^47,85.

$${\mu }_{a,w,c,x}=O_{a,w,c}\times {RRa}_i\times {pop}_{a,c,x}/{pop}_{a,c}$$ 6

$${RRa}_i=\exp \left({{rt}_i+rdh}_i+{rnh}_i+{rch}_i\right)-1,$$ 7

where ${\mu }_{a,x,w,c}$ represents the heat-related mortality of age group $a$ during week$w$ in sub-region $c$ under population-climate scenario $x$. $O_{a,w,c}$ is the observed baseline mortality of age group $a$ in sub-region $c$ during week $w$ (calculated as the weekly average mortality from 2010 to 2019). ${RRa}_i$ is the increase in relative risk associated with the combined effects of climate change, where $rt$, $rd\mathrm{h}$, $rn\mathrm{h}$, and $rc\mathrm{h}$ are the cumulative heat mortality effects of high Humidex, daytime-only, nighttime-only, and day-night compound heat extremes at the intensity level$i$, respectively. ${pop}_{a,x,c}$ is the projected population size of age group $a$ in sub-region $c$ under scenario $x$ and ${pop}_{a,c}$ is the baseline population size by age structure (calculated as the average population from 2010 to 2019).

In addition to the baseline no-adaptation projections, our study incorporates a methodological framework to assess heat-related mortality under diverse combinations of adaptation scenarios, which features two components: (1) Physiological adaptation scenarios: Five distinct physiological acclimatization trajectories are parameterized with the percentage change in excess relative risk (RR) over time, such as zero-adaptation (0% reduction in excess RR), low adaptation (5% reduction), moderate adaptation (10% reduction), medium adaptation (25% reduction), and high adaptation (50% reduction) scenarios. (2) Socioeconomic adaptation pathways: Based on established health economics literature^40,86, we use the GDP per capita under the respective SSPs to characterize future socioeconomic adaptive capacity. Specifically, leveraging the log-linear association between RR and GDP per capita, the ratio of the logarithm of future GDP per capita and the logarithm of current GDP per capita under each SSP is used to simulate future socioeconomic adaptation.

For a given adaptation scenario, future heat-RR is calculated as Eq. (8):

$${\mathrm{RRa}}_{\mathrm{i}}\prime ={\left[\tau \left(\left({\mathrm{RRa}}_{\mathrm{i}}-1\right)+1\right)\left(\left({\mathrm{RRa}}_{\mathrm{i}}-\mathrm{b}\right)\mathrm{x}+\mathrm{b}\right)\right]}^{1/2}$$ 8

where ${\mathrm{RRa}}_{\mathrm{i}}$ is the heat-RR under the baseline no-adaptation scenario. $\tau$ represents the attenuation rate associated with five physiological adaptation scenarios. x denotes the ratio of the logarithm of future GDP per capita to the logarithm of current GDP per capita. b is the intercept of the relationship between heat-related RR and the logarithm of GDP per capita. The identification of the association between heat-related RR and the logarithm of GDP per capita is detailed in Supplementary Note 6.

For each scenario, we sum the mortality across all weeks to estimate the total number of heat-related deaths. This estimation procedure is repeated for all ensemble members, enabling the calculation of the mean and 95% confidence intervals of total heat-related mortality. The impact of internal climate variability is largely represented by the 95% confidence interval. We calculate the changes in future mortality per capita ($\mathrm{\Delta }{\mu }_{y,c}$) relative to the baseline of 2022 (${\mu }_{2022,c}$) in each region, as in Eq. (9). For regions with missing data in 2022, we use the most recent available historical mortality data as the baseline:

$$\mathrm{\Delta }{\mu }_{y,c}=\left({\mu }_{y,c}-{\mu }_{2022,c}\right)/{\mu }_{2022,c}$$ 9

where ${\mu }_{y,c}$ represents the heat-related excess mortality per capita in year $y$ and region $c$. The annual average number of heat-related deaths under specific warming levels is calculated as the 10-year mean of heat-related deaths, centred on the year when each warming level is reached.

Specific levels of global warming

For each member of the MPI Grand Ensemble, we compute monthly anomalies (relative to the 1986–2005 baseline) of surface air temperature for each grid cell. We then calculate the annual mean global mean surface air temperature anomalies for each ensemble member by weighting the gridded values by the cosine of latitude and averaging them to achieve the ensemble mean magnitudes of global warming. Specific levels of global warming are determined using a 20-year average of the ensemble mean^87–89. To correct model bias in historical warming magnitudes, we calculate the simulated temperature changes relative to 1986–2005 and then added these values to the observed global mean surface air temperature anomalies above pre-industrial levels (1850–1900). As noted in the Sixth Assessment Report of the Intergovernmental Panel on Climate Change, the global mean temperature for 1986–2005 was 0.6 °C warmer than that during the pre-industrial period⁹⁰.

Roles of climate and population changes in future heat-related excess mortality in Europe

The individual contributions of climate change, population size changes, and population aging to future burden of heat-related excess mortality in Europe are assessed using a decomposition method⁹¹. The decomposition method isolates the effect of each factor by adjusting whether the factor changes or remains constant. The weekly mean Humidex, the three types of heat extremes, and their durations serve as indicators of the effects of climate change. Population aging refers to changes in age composition or population structure. We first project the burden of heat-related excess mortality by considering changes in all three factors. Subsequently, we repeat the projections under identical conditions, except that one factor is held constant at its 2022 level. The differences between the two sets of projections represent the respective roles of the factors (climate change, population size change and population aging) in the projected total heat-related excess mortality.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

X.L.W., J.W. and Y.G. conceived and designed the study, interpreted the findings, and wrote the manuscript. X.L.W. built the model, collected data, finalized the analysis and drafted the manuscript. J.W., Z.P.R., D.Z., S.J.L., and J.H.W. commented on and revised drafts of the manuscript. X.L.W., J.W., Y.G, Z.P.R., D.Z., S.J.L., and J.H.W. read and approved the final manuscript.

Peer review

Peer review information

Nature Communications thanks Dominic Royé, and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Data availability

The data used to interpret the findings in this study are available publicly. The all-cause mortality records from Eurostat can be downloaded from https://ec.europa.eu/eurostat/. The ERA5-land data were downloaded from https://cds.climate.copernicus.eu/. The hourly Humidex data generated in this study have been deposited in the Zenodo database (10.5281/zenodo.15747415). The data from the new MPI grand ensemble with CMIP6 were accessed from https://esgf-metagrid.cloud.dkrz.de/search. The future gridded population projections were derived from the Gridded Population of the World (https://sedac.ciesin.columbia.edu/data/set/popdynamics-1-km-downscaled-pop-base-year-projection-ssp-2000-2100-rev01/data-download). The Eurostat (https://ec.europa.eu/eurostat/) and the Office for National Statistics (https://www.ons.gov.uk/) provide associated population projections. The history gross domestic products (GDP) were sourced from Eurostat (https://ec.europa.eu/eurostat/). Future GDP projections were extracted from https://gcp-tsukuba.github.io/SSP-downscale/. The NUTS-level 3 administrative boundaries (version 2021) for Europe were obtained from Eurostat (https://ec.europa.eu/eurostat/web/gisco/geodata/ statistical-units/territorial-units-statistics, last accessed: 2023-04-24), while the global national boundaries were obtained from GADM (https://gadm.org/, last accessed: 2023-08-16).

Code availability

This work was analysed using R (Version 4.1.1) and Python (Version 3.9.4). The scripts used to reproduce the analysis and other supporting information in this paper are available at 10.5281/zenodo.15747164.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-025-62871-y.