Following the Intergovernmental Panel on Climate Change (IPCC) convention, risk in this study is defined as “the potential for adverse consequences”36, mainly in the form of human mortality. IPCC’s definition of exposure is “the presence of people…in places and settings that could be adversely affected”36. However, except where otherwise specified, risks in this study are assessed from an individual’s perspective, for each city’s average inhabitant. The same average inhabitant is always assumed to be exposed, and intra-city risk differentials are indicative of the potential difference in risk posed to this inhabitant if they lived in one part of the city compared to another. Exposure in this study therefore follows the epidemiological definition and refers to the presence of heat and cold. Under IPCC convention, this is more closely related to the presence of hazard (“the potential occurrence of a natural or human-induced physical event or trend that may cause loss of life…”36). Lastly, vulnerability is “the propensity or predisposition to be adversely affected…, including sensitivity or susceptibility to harm and lack of capacity to cope and adapt”36. Vulnerability determines the magnitude of impact of a given temperature exposure. In the current context, vulnerability may include, among others, physiological factors such as age and acclimatization to a particular climate, as well as societal and socioeconomic factors that hinder people’s ability to cope with heat or cold, such as access to air conditioning. In this study, vulnerability and exposure, jointly, determine risk.
Temperature-related mortality risk in European cities
Mortality risk generally increases towards the extreme ends of each city’s typical temperature range, with greater risks to older age groups (Fig. 1a–c). The degree of sensitivity to heat and cold differs between cities due to differences in population vulnerability, associated with local socioeconomic, infrastructural, and environmental characteristics37,38. The intensity of UHI, defined as the difference in average air temperature between urban areas and their rural surroundings, also varies between cities and across seasons (Fig. 1a–f, Supplementary Fig. S7)8,24,39. In this study, the differences in UHI intensity between cities are captured by the urban climate simulations through variations in the land surface, local climate, and human activities, while seasonality results from seasonal shifts in both climate (e.g., solar radiation) and anthropogenic heat emissions (e.g., central heating)40. The combined effects of varying both vulnerabilities and UHI (exposure) lead to variations in UHI impact on mortality risk across Europe. For instance, focusing on the heat extreme days, a greater impact is noted in Milan, which experiences more extreme heat conditions, than in London, despite the greater urban-rural temperature difference in the latter city. This is due to the strong non-linear increase in heat-related risk under Milan’s heat extreme days, which are notably warmer than in London (Fig. 1). Stockholm, on the other hand, has a relatively weak UHI and low vulnerability to heat, resulting in minimal impact of UHI on mortality risk.
a–c Age grouped exposure-response relationships, d–f temperature averaged over the warmest 2% days between 2015 and 2017, and g–i histograms of average temperature-related mortality estimates during the warmest 2% days shown separately for urban and rural areas for a, d, g Milan, b, e, h London, and c, f, i Stockholm. Vertical dashed lines in black/gray in a–c, g–i are the urban average over the warmest/coldest 2% days; vertical dotted lines, are the rural average. Hatching indicates rural areas in d–f, where areas excluded from analysis (water bodies and areas with elevation differential >100 m from the population-weighted average) are shown in white.
On an annual basis, UHIs lead to a general shift in the temperature distribution to warmer conditions in urban environments (Supplementary Fig. S8). This results in greater mortality risk associated with heat and lower risk associated with cold compared to rural surroundings. The annual net impact, therefore, further depends on the balance of warm and cold days per year, in addition to each city’s vulnerability to heat vs. cold.
Impact of urban heat islands on mortality risk
Despite the binary classification into urban and rural environments, a range of built and natural environments can be found within each. Examined as a function of land imperviousness, we find the largest change in mortality risk per unit change of land imperviousness around the urban-rural boundary (slopes in Fig. 2a–c), from rural conditions of minimal imperviousness (Δ imperviousness = 0%) to urban environments with slightly built-up areas (e.g., Δ imperviousness = 5%). This is followed by a continued but more gradual increase/decrease within the urban imperviousness gradient. The general shape of this relationship largely reflects that between imperviousness and air temperature UHI (Supplementary Fig. S12)9,41. The degree of increase/decrease in risk with imperviousness is further associated with each city’s UHI intensity and population vulnerability, as discussed in the next section.
The annual net impact per year across all days of the year is shown in a, followed by the average per day over the b warmest and c coldest 2% (22) days over the 2015–2017 time period. Cities are grouped by the Köppen-Geiger climate classification, as indicated in d and listed in Supplementary Table S3. Solid lines in a–c indicate the climate group median, with shading showing the interquartile range. Cities where the maximum urban-rural imperviousness difference is less than 80% are excluded (67 of 85 cities shown), and medians across less than four cities in a climate group, which can occur at high Δ imperviousness, are not shown to avoid misrepresentation.
Given that highly populated areas also tend to be highly built-up (statistically significant mean Spearman’s correlation of 0.47, Supplementary section 1), and that heat emissions from human activities are enhanced in densely populated areas, UHI’s impact on mortality risk also increases with population density (Supplementary Fig. S1). At the city level, this implies a biased exposure whereby greater fractions of the population live in areas of greater UHI. Given that risks in this study are assessed from the perspective of an individual inhabitant living in different parts of the city, this bias in exposure is not captured. Should it be included, however, the bias would result in a 6.4% median (interquartile range across 85 cities, IQR: 2.9–9.4%) increase in the urban average mortality risk estimate during heat days and a 1.3% (IQR: 0.7–2.3%) reduction during cold days (Supplementary section 1). For the outlier cases of Brussels, Dublin, Paris, and Lyon, the increase during heat days can reach over 25%.
Grouping cities by the Köppen–Geiger climate classification (Supplementary Table S3) and summing the impact over all days of the year, we note that temperate cities with no dry season and hot summers tend to experience an annual net adverse effect of UHI on mortality, while for all other climate groups, the net effect is generally protective (Fig. 2a). However, variability and outlier cities can be found within each climate group, as well as overlaps between climate groups (Supplementary Fig. S9). During days of extreme heat, the impact of UHI on mortality risk tends to be stronger for cities with temperate, hot summer climates (Fig. 2b), though this difference is removed and the impacts become similar for all climate groups during heat extremes when population age is standardized between cities (Supplementary Fig. S5b). On the other hand, consistently and notably weaker UHI protective effects are found for cold climate cities during cold extremes days, regardless of assumed population age structure (Fig. 2c, Supplementary Fig. S5c). The differences between climate groups are further discussed in Supplementary section 3.
In general, UHIs have a weak protective net annual impact on human mortality risk for most (70 out of 85; 90% confidence interval, CI: 60 to 75) cities examined in this study (Fig. 3a, b). Across all 85 cities, a median of 2.8 fewer deaths per 100,000 people per year (CI of the median estimate, representing uncertainty in exposure-response relationships: 1.7–3.5; IQR across 85 cities, representing differences between cities: 0.7–4.4) are attributed to temperature in urban versus rural areas. This is due to the high likelihood of cold days in most parts of Europe, defined as cooler than the city- and age-specific optimal temperature below which mortality risk generally decreases with increasing temperature. The most protective annual net effects of UHI are found in Glasgow, Porto, and London, with 16 (CI: 14–18), 12 (CI: 11–14), and 11 (CI: 9–13) fewer deaths per year per 100,000 people, respectively (Supplementary table S4). The most adverse annual net effects are found in the urban areas of Trieste, Genoa, and Bologna, with 5 (CI: 2–7), 4 (CI: 1–8), and 3 (CI: 1–5) additional temperature-related deaths per year per 100,000 inhabitants, respectively (Supplementary Table S4). On average, UHI has a protective effect in all cities for all seasons except summer (Fig. 3a). In some cities with cooler summer climates, such as those in Nordic countries and the United Kingdom, the average impact is protective even in summer.
Differences are calculated daily and then temporally averaged over the days/seasons indicated. Boxplots in a show the spread across 85 European cities and the variation between different temporal averages. Maps show geographical distributions of the average impact b annually, and over the c warmest and d coldest 2% days over the 2015–2017 period. Boxes in a indicate the median and the first and third quartiles, whiskers the minimum/maximum value within 1.5 times the interquartile range from the first/third quartiles, and dots the outliers. Cities, where UHI has an adverse annual net impact, are outlined in red in b.
The daily impact of UHI on mortality risk is greater during heat extreme days than during cold extreme days for all cities. Over the 85 European cities examined, UHI increases mortality risk during heat extreme days by a median of 0.25 (CI: 0.21–0.27; IQR: 0.18–0.29) additional deaths per 100,000 population per day (Fig. 3a, c), a median increase of 45% (IQR: 30–61%) from the risk in rural areas around the cities (Supplementary Fig. S10). This is in contrast to a reduction of 0.05 (CI: 0.04–0.07; IQR: 0.02–0.11) deaths per 100,000 population per day during cold extremes (Fig. 3a, d), a 7% (IQR: 3–15%) decrease from the rural average. Within the urban space, the impact is further enhanced in more built-up parts of the city. For instance, there is a 67% (IQR: 46–91%; 0.35, IQR: 0.24–0.44, additional deaths per 100,000 per day) median risk differential between the most and least built-up (≥90% vs. ≤10% imperviousness) areas during heat extreme days (Supplementary Fig. S11).
If population age structures are standardized across the cities according to the 2013 European standard population, the risk estimates would increase by 9% on median (Supplementary section 2, Fig. S4c), likely indicative of slightly younger populations in cities (Supplementary Fig. S4a).
To additionally consider the age dependence of UHIs’ impact on mortality, years of life lost (YLL) analyses are included in the Supplementary materials (Fig. S13, Tables S6 and S10). While the overall conclusion of a weakly protective annual net impact for most cities still holds, an annual adverse impact is found for more cities with the YLL approach (20 vs. 15 with the mortality counts-based analysis). As younger age groups are weighted more strongly in YLL analyses, this finding may be reflective of younger populations’ greater vulnerability to heat, which is more similar to that of older age groups, compared to their vulnerability to cold, which tends to be lower (Fig. 1a–c).
Factors controlling UHI’s impact on mortality
The magnitude of UHI impact on human mortality in this study is determined by four factors: vulnerability of the population (as captured by the relative risk, RR, at different temperatures) and population age structure, which together determine total vulnerability, and frequency of occurrence of different temperatures and magnitude of UHI, which together determine exposure. The relative importance of each in determining how the impact of UHI compares between different cities is discussed below.
Mortality risk tends to increase towards the extreme ends of a city’s typical temperature range, resulting in U- or J-shaped temperature-mortality exposure-response functions (ERFs). Subsequently, an indicator of a population’s vulnerability to heat and cold is the magnitude of the relative risk (RR) at the warmest and coldest extremes of the observed temperature range (RR\({}_{{{{{{{{\rm{Tmax}}}}}}}}}\) and RR\({}_{{{{{{{{\rm{Tmin}}}}}}}}}\), respectively). As expected given the approach we use for estimating mortality, statistically significant and notable rank correlations can be found between RR\({}_{{{{{{{{\rm{Tmax}}}}}}}}}\) and UHI’s impact on mortality during days of extreme heat, as well as between RR\({}_{{{{{{{{\rm{Tmin}}}}}}}}}\) and UHI’s impact during extreme cold (Fig. 4), indicating greater impact in more vulnerable cities. This vulnerability factor also partially explains the difference in UHI impact between cities of different Köppen–Geiger climate groups during days of temperature extremes (Supplementary section 3). Notably, the annual net impact of UHI on health is significantly correlated with RR\({}_{{{{{{{{\rm{Tmin}}}}}}}}}\) but not with RR\({}_{{{{{{{{\rm{Tmax}}}}}}}}}\), indicating an overall dominant role of impact during cold days across European cities.
Metrics examined include the relative risk (RR) at the extreme ends of the city’s temperature range, the number of days in a year warmer than the minimum mortality temperature for the 65–74 age group, the average magnitude of the urban heat island (UHI), the ratio of the population aged 85+ compared to those aged 20–44, the annual average temperature (Tavg), as well as the longitude and latitude. The size of the square corresponds to the magnitude of the correlation. Only correlations with statistical significance above 99% (p value <0.01) are shown. Negative correlations indicate a greater protective effect or lower adverse effect with a higher value of the metrics.
Not all summer days are considered warm (above the Minimum Mortality Temperature, MMT, above which UHI has an adverse effect on human health) for all cities (Supplementary Fig. S15), and during mild summer days, UHI has minimal or slightly protective effect on health. The number of warm days in a year therefore strongly determines UHI’s impact in summer, and to a lesser extent in spring, winter, and on the annual net balance (Fig. 4). The number of warm days in a year is correlated with the average temperature and anti-correlated with the latitude, but varies between cities independently of their population’s vulnerability to heat (Supplementary Fig. S14).
The magnitude of each city’s average UHI is correlated with the magnitude of its impact on health both during heat extreme days and on seasonal averages (Fig. 4). However, the magnitude of UHI is not a key factor in determining the relative ranking of each city’s annual net and cold extreme mortality impact. In those cases, the population’s vulnerability to cold plays the most important role.
Lastly, the proportion of older compared to younger adult populations in each city is a determining factor in the impact ranking between cities only during temperature extreme days and in winter (Fig. 4), though it affects the magnitude of the impact in each city, as discussed above and in Supplementary section 2.
Geographically, cities in eastern and northern parts of Europe, as well as those with colder average temperatures, tend to experience less protective effects of UHI on mortality during winter and cold days (Figs. 4, 3d, and 5b), possibly indicative of greater infrastructural and behavioral adaptation to cold weather. During summer, on the other hand, cities with warmer climates and in the southern and eastern parts of Europe tend to experience a stronger overall adverse impact, likely associated with a greater proportion of warm days. The daily impact during heat extremes, which is more correlated with population vulnerability and magnitude of UHI, does not follow this geographical pattern (Figs. 4 and 3c).
Economic assessment of UHI-related mortality
Economic assessments in this study mainly follow the value of statistical life (VSL) approach, which is commonly used in mortality risk valuations42. However, other valuation metrics may also be applied, and the value of life year (VOLY) approach is additionally assessed and discussed below at the end of this section.
Economic assessments of UHI heat- and cold-related mortalities in Europe following the VSL approach are on the order of a few tens to hundreds of Euros per adult city resident per year (Fig. 5a, b). The median economic impact of UHI heat-related mortality is €192 (IQR: €142 to 296) per adult resident per year, estimated in 2021 Euros, and that of cold-related mortality is €−314 (IQR: €−429 to −235) per adult resident per year. There is a weak correlation between the magnitudes of cities’ heat- and cold-related mortality impacts (Fig. 5d). The correlation persists even when population age is standardized (Supplementary Fig. S16d), though many cities represent exceptions. For instance, UHI yields considerable protection against cold in Glasgow, London, and Porto, while the annual heat-related mortality impact due to UHI remains low (lower left corner in Fig. 5d, Supplementary table S4). UHI-related mortality has an adverse net economic impact for 15 (CI: 10 to 25) of the 85 cities examined (circled in red in Fig. 5c, above the one-to-one line in Fig. 5d, Supplementary Table S4), with the greatest impacts for Turin and Bologna.
Annual impact of UHI-induced mortality associated with a heat, b cold, and c all temperatures. d shows the two-sided correlation between annual heat and cold UHI impacts. e shows two-sided correlations between heat-related mortality and air pollution-related mortality impacts for a subset of 70 cities. f shows comparisons of the economic impacts of mortality to costs of rent and public transport as obtained from Eurostat. The n-number in brackets indicates the number of cities represented by each boxplot. Note that due to limited data availability, rent and transport costs are representative of different subsets of European cities than heat- and air pollution-mortality and may be biased toward some countries (e.g., the majority of rent data is for German cities). Economic assessments of mortality are age-standardized according to the 2013 European standard population where indicated (std age). Only adult city inhabitants are accounted for all impacts except for ozone, which considers the entire population of all ages. Cities, where UHI has an adverse annual net economic impact, are outlined in red in c. Pearson correlation coefficients and associated p values are labeled in d, e. Boxes in f indicate the median and the first and third quartiles, whiskers the minimum/maximum value within 1.5 times the interquartile range from the first/third quartiles, and dots the outliers.
To put these figures in context, we compare these economic assessments with those for existing estimates of excess mortality due to air pollution43,44, which is another important human health hazard associated with urban environments. Across Europe, cities with higher economic impacts of UHI heat-related mortality tend to also bear greater impacts from air pollution-related mortality (Fig. 5e). In contrast, the correlation is weaker between UHIs’ protective effects against cold and each city’s air pollution-related risk (Supplementary Fig. S16e). On an annual basis, the median economic impact of UHI heat-related mortality is around one-fifth that of PM2.5-related mortality and ~1.2 times that of ozone-related mortality (Fig. 5f). However, it is important to note that UHI heat risk is a quantification of the urban-rural difference, while for the air pollution risks discussed above, city average pollutant concentrations are contrasted against an ideal threshold concentration that may also be exceeded in rural and peri-urban regions. This is especially the case for ozone, where higher concentrations can be observed in rural regions downwind of urban centers, e.g., ref. 45 and urban neighborhoods with lower imperviousness and more abundant vegetation46. Additionally, heat risk has a strong seasonality and is largely absent during colder seasons. Air pollution, on the other hand, is present throughout the year despite some seasonal variability. Moreover, mortality quantifications for air pollution used in this comparison43,44 are based on annual average estimates, while temperature-related risks are based on daily average estimates. The former, therefore, mainly represents the impact of chronic exposure while the latter captures acute effects (with lag periods of up to three weeks, see Methods). The variability in economic impacts of UHI heat-related mortality across Europe is relatively small compared to that of PM2.5 mortality (Fig. 5f).
Economic assessments by VSL as presented above do not explicitly consider age, therefore life expectancy at the time of death is assumed to be comparable to all causes of death. However, as can be observed from age-dependent temperature-mortality relationships (Fig. 1a–c), heat and cold exposure disproportionately affect older populations. The life expectancy at the time of death would therefore likely be shorter than the average for all-cause mortality. An approach to account for this is to consider the YLL through VOLY valuation, which is included in Supplementary Table S7. While there is a strong correlation between economic impacts as determined through VSL and that through VOLY (Supplementary Fig. S17a, b), the magnitude of the impact as assessed through VOLY is only ~14% (median of the multi-city median annual net, heat, and cold impacts) of that assessed through VSL (Supplementary Fig. S17c). This is mainly due to differences in valuation approaches which resulted in each VSL (3.91 million 2021-EUR per statistical life42) equating to 85 VOLY (46,000 2021-EUR per year47). However, there is currently no clear consensus on economic assessments by YLL, and at the higher end of the VOLY estimate (116,000 2021-EUR per year47), each VSL equates to around 34 VOLY, resulting in valuations by VOLY-based assessments to be around 35% of that using VSL (Supplementary Fig. S17d). Given limitations with the VOLY approach, including ethical concerns and lack of evidence in the assumption of a time-independent VOLY42, VSL-based assessments are the main focus of this study.