Climate change experiences raise environmental concerns and promote Green voting

Public support is fundamental in scaling up actions to limit global warming. Here, we analyse how the experience of climate extremes influences people’s environmental attitudes and willingness to vote for Green parties in Europe. To this end, we combined high-resolution climatological data with regionally aggregated, harmonized Eurobarometer data (34 countries) and European Parliamentary electoral data (28 countries). Our findings show a significant and sizeable effect of temperature anomalies, heat episodes and dry spells on environmental concern and voting for Green parties. The magnitude of the climate effect differs substantially across European regions. It is stronger in regions with a cooler Continental or temperate Atlantic climate and weaker in regions with a warmer Mediterranean climate. The relationships are moderated by regional income level suggesting that climate change experiences increase public support for climate action but only under favourable economic conditions. The findings have important implications for the current efforts to promote climate action in line with the Paris Agreement. Exposure to extreme weather events could increase environmental concerns and support for Green parties. With high-resolution data across European countries, the authors demonstrate the existence of such effect, then further discuss the heterogeneity and possible mechanisms.

concern about environmental and climate issues 7 . Between 2004 and 2019 the percentage of seats held by Green parties in the European Parliament increased by 74% from 5.7% to 9.9% (Fig. 1c,d).
Understanding the drivers of changes in public concern and support for Green parties is important to identify the mechanisms underlying transformations towards a greener economy and more sustainable society. Here, we empirically investigate the effect of more frequent and intense experiences with climate extremes on environmental concern and analyse to what extent changes in concerns translate into actual political support for Green parties [8][9][10][11] .We exploit time-series Eurobarometer data (42 survey waves, 2002-2019) and European Parliament election data (six elections, 1994-2019) to analyse changes in concerns and voting at the subnational level across 34 and 28 European countries, respectively (Supplementary  Tables 1 and 2). Our regional panel dataset allows us to causally test for the impacts of climatic extremes while controlling for unobserved heterogeneity and time trends using fixed effects models.
Experiencing the consequences of climate change can support the experiential processing and learning of information about climate risks and can thus influence the formation of environmental attitudes and concerns and ultimately the willingness to support climate action [12][13][14][15][16] (Box 1). Existing evidence shows that people who have experienced unusual weather and extreme climatic events are more likely to believe in the existence of global warming and its anthropogenic causes 17,18 , to express concern about climate change 19,20 , to show willingness to engage in mitigation actions 21 and to be in favour of climate policies 22,23 .
Our study provides three key contributions to the literature. First, we present evidence on the causal linkages between exposure to extreme climate events, environmental concerns and voting. There is limited empirical evidence on the links between climate change experiences and voting outcomes, especially for such major elections as the European Parliament 24,25 . Here, we overcome the common empirical difficulty of capturing how concerns are Climate change experiences raise environmental concerns and promote Green voting translated into action by using uniquely compiled longitudinal data. Second, our analysis is carried out at the subnational level, thus capturing heterogeneity at a more granular scale while still providing comprehensive insights for a broad number of countries and time periods. Third, using the comparative data, we are able to investigate how regional differences in local climatic and economic conditions shape the relationships between experiences, concerns and voting, complementing previous findings in the empirical literature 11,26 .

experiences influence both concerns and voting
Using fixed effects panel models, we regress the share of the environmentally concerned population in a region (columns 1-4) and the share of Green voters (columns 5-8) on indicators measuring extreme climate event and conditions (Table 1). These indicators capture temperature anomalies, heat episodes with unusually warm temperatures (at least three consecutive days >95th monthly percentile) and drought episodes in the past 12 months before the concern measurement or election date. The estimates are standardized 27 and corrected for spatial and temporal autocorrelation 28 (see Supplementary  Tables 3 and 4 for autocorrelation tests and Supplementary Tables 5  and 6 and Supplementary Figs. 1-4 for summary statistics). They are robust to different sensitivity tests (Supplementary Tables 7-17) including: dynamic models controlling for the lagged dependent variable (Supplementary Table 8); models using climate change concerns as an outcome (Supplementary Table 11); models with different climate measurements (Supplementary Tables 13 and 14); and models controlling for other environmental events, such as wildfires and flooding (Supplementary Tables 16 and 17).   The concern measure reflects the share of eurobarometer respondents in a NUTS region that considered environmental issues to be one of two priorities for national policy-making. b, Regional means of environmental concern 2004-2019. We find that exposure to temperature anomalies, heat episodes and drought events significantly increases environmental concerns and the vote share of Green parties. While there are some differences across models, with temperature anomalies and heat episodes exerting the strongest effects on concerns and dry spells on Green voting, all climate measures consistently have a positive relationship with the two outcomes.
The magnitude of the estimated effects is sizeable. For example, a temperature anomaly of one standard deviation over the past 12 months is estimated to increase environmental concerns on average by 0.167 (s.e. 0.028) and Green voting by 0.117 (s.e. 0.055) standard deviations within regions or by 0.8% and 0.3% in absolute terms (Supplementary Table 18). Likewise, if every month in a year had an additional unusually warm day (>95th monthly percentile), Green concerns and voting would increase by 0.8%, respectively (Supplementary Table 19).

Strong influence of positive temperature anomalies
Not only does climate change lead to higher temperatures and more extensive heat and drought episodes, it can also cause more extreme cold weather and temperature fluctuations, including cold snaps. Negative temperature anomalies and periods of extreme cold, however, have commonly been used by climate sceptics to spread misinformation about global warming 29 . We test for the impact of cold and wet extremes on concerns and voting in additional models (Supplementary Table 21) and analyse how their influence differs from the influence of positive temperature anomalies, heat episodes and dry spells, for which we observe a similar effect as in the baseline models (Table 1). For negative temperature anomalies, cold episodes and wet spells, on the other hand, no consistent patterns are found. Overall, the results suggest a stronger relevance of positive temperature extremes and heat-related events for environmental concerns and Green voting as people are likely to attribute positive Box 1 | theoretical links between experiences, concerns and voting An increasing number of studies have considered the role experience with climate change plays for the formation of environmental attitudes and concerns about environmental and climate issues 8 . Being directly exposed to climate extremes can reduce the psychological distance to climate change by making the impacts and related hazards appear more certain (hypothetical distance) and temporally closer (temporal distance) as opposed to an abstract threat in a distant future [12][13][14] . This is because experience can evoke strong feelings, which support the experiential processing and learning of information. This can make people better grasp the otherwise complex risks of climate change 15 . At the same time, experience can make people understand that climate change affects them personally and their neighbourhoods (spatial distance) and not a distant social group that they have no relations to (social distance). Ample evidence from psychology and cognitive sciences confirms that risk perceptions, beliefs and concerns are particularly influenced by recent or highly salient events that are cognitively more readily available than abstract statistical evidence (availability heuristics) [49][50][51][52] .
While the majority of empirical studies show that experience matters, its relevance and the magnitude of the influence differ widely across the study contexts and types of experiences considered 9,53 . Whether and how perceived changes become relevant to attitudes towards climate change is influenced by a range of individual characteristics and contextual factors. These include beliefs about local climate conditions and changes 11,24 and economic conditions that may compete with environmental concerns. Especially during times of economic uncertainty, such as in the aftermath of financial crises, individuals may opt to prioritize economic and financial needs over the environment 33,54 . Other influential factors are related to individual ideological predispositions, political worldviews and belief systems 11,24 as well as demographic factors including age, gender and education 55 .
The figure, which is adapted from ref. 16 , shows a simplified conceptual model of the links between climate change experiences, environmental concerns and Green voting. Blue boxes and arrows highlight the causal pathways considered in our empirical analysis. Experiences with climate change and its local impacts influence environmental perceptions and concerns together with indirect experiences shared by the media and social networks. The relationship is moderated by contextual influences, such as culture and belief systems or economic factors, which determine to what extent experiences are translated into concerns. Concerns can result in behavioural intentions, with the choice of actions being influenced by norms and habits. If different intrinsic and extrinsic conditions are met, intentions can lead to behavioural changes, such as increased political support and Green voting 16 .
While there is a broad literature connecting concerns, intentions and actions, few studies have directly considered climate change impacts on voting and electoral outcomes. Existing studies show that climatic factors can indeed influence voting behaviour such as voter turnout 56,57 , votes for the incumbent party 57 or pro-environmental voting in climate-related ballots 24 Fig. 1) 15,30 .

threshold and recency effects are relevant
In a further extension of our baseline models, we investigate whether more recent climate events are more influential than events that occurred longer ago (Supplementary Tables 22-25). When considering different time windows over which the climate variables are aggregated, an inverted U-shaped pattern is observable with effects sizes first increasing with the length of the time window and then decreasing. This pattern suggests the interplay of two effects: a threshold and a recency effect. On the one hand, extreme conditions exert an influence only if they are experienced for a sufficient length of time, representing a hurdle in the translation of experiences into concerns and ultimately into action. Once the hurdle is crossed (approximately after 12 months), more distant climate events tend to influence concerns and voting less strongly possibly due to the experiences becoming less salient (see ref. 31 for similar findings). For example, heat episodes that occurred in the past 12 months (Supplementary Table 23

impacts of extremes differ across europe
Regions across Europe are characterized by different climatic conditions and are differentially affected by climate change 32 . In additional interaction models (Fig. 2), we explore the differential impact of climate extremes on concerns and voting across the main climate zones of Europe (Supplementary Table 26). On the basis of the Köppen-Geiger typology (Extended Data Fig. 2), we distinguish between: (1) a hot, arid climate in Southern Europe (Mediterranean regions); (2) a temperate climate mainly in Western Europe; and (3) a colder climate, mainly in Northern and Central Europe.
We find that the impacts of climate extremes are not uniform but differ from region to region. Temperature anomalies, heat episodes and droughts have a consistently stronger effect on concerns and voting in regions with a temperate and colder climate compared to Mediterranean regions with a warm, arid climate, for which we find no significant effects. In the temperate and cold climate zones, an increase in temperature-based heat episodes by one standard deviation is estimated to increase concerns by 0.217 (s.e. 0.063) and 0.186 (s.e. 0.037) standard deviations and voting by 0.232 (s.e. 0.062) and 0.176 (s.e. 0.069) standard deviations, respectively. The patterns are less clear for cold-related events and wet spells, suggesting that changes in concerns and voting are mainly driven by experiences of extreme heat and dry spells, particularly in the context of a cool or temperate climate (Extended Data Fig. 3).
This evidence suggests that there are important differences across Europe in the way the public responds to extreme climatic conditions and impacts. The observed patterns are possibly due to the fact that the Mediterranean climate is already hot and dry in absolute terms, hence further deviations in temperature or rainfall may have little to no effect on environmental concerns and voting. Populations in these regions may have already adapted to the hotter baseline conditions, for instance with respect to housing and agriculture, or may be more used to temperature-related extremes because of habituation, making the extreme events become less noticeable. Other factors related to the socioeconomic, cultural and political conditions in a region can also play a role in explaining the observed heterogeneity.

economic conditions moderate climate impacts
To explore the underlying regional heterogeneity further, we estimate the influence of economic factors in moderating the relationship between climate change experiences, concerns and voting. Standardized regression coefficients with standard errors in parentheses. Standard errors are corrected for cross-sectional and serial correlation up to the indicated spatial and temporal cutoffs. All models control for regional and temporal fixed effects to account for unobserved heterogeneity and common time trends. Period fixed effects refer to three-yr periods in models 1-4 and election year fixed effects in models 5-8. Coefficients are standardized using the residual variance after applying the fixed effects. Accordingly, the coefficients refer to a marginal effect of a one standard deviation change of the covariates on the outcome within regions and periods. Temperature anomaly is defined as a standardized deviation from the long-run monthly temperature mean; heat episode (temp.) is defined as at least three consecutive days with a mean temperature above the local monthly 95th percentile; heat episode (UTCI) is defined as at least three consecutive days with a mean UTCI above the local monthly 95th percentile; dry spells are defined as mean of SPeI below -0.5. All measures are calculated using 1971-2000 as a reference period. P values: *<0.1, **<0.05, ***<0.01.
Previous research has suggested that people's economic interests can lead to a crowding out of concerns for the environment if there is a perceived trade-off between the two issues 33,34 . In the aftermath of the global financial crisis of 2007-2008, for example, a substantial reduction in environmental concerns was observable across all European regions (Fig. 1a). Hence, in times of economic difficulty, the impact of exposure to climatic extremes on public support for climate action may be reduced. Here, we analyse whether climate impacts on concerns and voting depend on (1) the relatively stable, general economic condition in a region, measured in form of the mean gross domestic product (GDP) per capita in the period from 2000 to 2019 and (2) variable changes in GDP that co-occurred with exposure to climate extremes. We consistently find that the effects of experiencing climate extremes on environmental concerns and voting are less pronounced in the regions with lower income level (Fig. 3). These effects also remain robust once other regional characteristics are controlled for, such as the level of education, age structure, urbanization, agricultural vulnerability, political conditions and the regional climate classification (Supplementary Table 27). Considering differences in mean GDP levels across regions, we find that the impact of temperature anomalies on environmental concerns is significantly reduced by 0.152 (s.e. 0.026) and on voting by 0.83 (s.e. 0.028) standard deviations for each one standard deviation decline in mean GDP (Supplementary Table 27, models 1 and 6). Changes in GDP over time, however, are not found to moderate the climate effects, possibly because such variation in income is not large enough to induce a change in the way people perceive climate extremes.
As illustrated in Fig. 3a,b, the impact of temperature anomalies on concerns and voting is much stronger in relatively rich regions (at the 75th percentile of the income distribution) than in poorer regions (at the 25th percentile of the distribution). The maps (Fig. 3c,d) highlight that differences exist not only between countries but also within countries with wealthier regions responding more strongly to the exposure to climate extremes. Across Europe, the effects of experiences on concerns and voting outcomes appear to be particularly pronounced in urban centres with their relatively wealthier populations.

Changes in concerns explain climate impacts on voting
Our conceptual model (Box 1) assumes that experiences of climate extremes activate environmental concerns, which in turn influence Green voting. In this section, we investigate to what extent climate-induced changes in concerns predict voting outcomes (Supplementary Table 28). In a first step, we regress the Green vote share in a region on the level of environmental concerns in that region one and two years before the election (lag). As a falsification test, we also regress the Green vote share on the level of environmental concerns one and two years after the election (lead), which should not have an effect. If environmental concerns in a region influence voting outcomes, we expect to see a positive effect in the first two and no effect in the latter two models.
The results suggest a sizeable influence of changes in environmental concerns on voting outcomes. A one standard deviation change in average concerns two years before an election is estimated to lead to a 0.203 (s.e. 0.071) standard deviation increase in Green votes and a similar change in concerns one year before an election leads to an even higher increase in Green votes of 0.272 (s.e. 0.071) standard deviations. In line with our expectation, the lead values of concerns are not found to exert any significant influence on voting outcomes, suggesting that voting for Green parties is influenced by environmental concerns and not just the realization of an unobserved underlying trend.
In the subsequent models (models 5-8), we use a two-stage instrumental variable approach to estimate the causal impact of changes in concerns on voting. The first stage estimates the variation in concerns that is driven by changes in climatic conditions as an exogenous and relevant instrumental variable; the second stage estimates the effect of this variation on Green voting. This reflects the full causal chain depicted in Box 1 from the experience of climate extremes which induces concerns about the environment to the change in voting behaviour. Also this estimation suggests a positive impact of concerns on voting of similar size as estimated in the previous models.

Conclusion
With the global temperature projected to rise to 1.5 °C above pre-industrial levels between the 2030 and 2050 35 Fig. 2 | effects of climate extremes on environmental concern and Green voting by region. Coefficients are standardized using the observed variance of the variables in the given climate classification after applying the fixed effects (see Supplementary Table 26 for the full models). Lines around the point estimates show the 95% confidence intervals. Regions are classified as hot, temperate, and cold on the basis of the Köppen-Geiger typology (extended Data Fig. 2).
actions will lead to many irreversible consequences. Accordingly, how to best increase citizens' environmental concerns and support for climate action is of great relevance. Using a new cross-country subnational dataset for Europe, we show that exposure to climate extremes, in particular heat and dry spells, activates environmental concerns and promotes Green voting. Studies in the United States also report a strong effect of experiencing warmer temperatures and dry spells on the public awareness of climate change 36,37 , possibly because they occur on larger scales affecting many sectors and groups and also last longer than short-term events, for example, flooding. That people tend to particularly associate warmer temperature and heat-related events with climate change is probably due to the early use of the term 'global warming' to describe the changing global climate conditions 38 .
With the issue of climate change becoming more concrete and salient, people's willingness to engage in and to support climate action increases 12 , including at the political level in the form of voting for pro-environmental parties. These changes can contribute to shifts in the political landscape at a larger scale, given the increased share of Green voters across countries in Europe in recent years. Our findings are in line with existing case studies on the role of climate experience on voting behaviour 24,25 providing new cross-country comparative evidence on the phenomenon and highlighting its broader implications.
Obviously, exposure to climate change impacts is not the ideal way to promote public concern and action. Climate communication and education can help fill the experience gap. Studies have shown that carefully designed messages can reduce the psychological  Table 27, models 1 and 6. c,d, Marginal effects of a one standard deviation temperature anomaly on environmental concern (c) and Green vote share (b), given regions' GDP level and climate zone. estimates are based on interaction models displayed in Supplementary Table 27, models 5 and 10.
distance and promote mitigation behaviours 14,39 . Information linked to concrete experiences or vivid descriptions is often better suited to induce behavioural changes than is purely analytical information, especially if salient and closely linked to people's everyday life 15 . Advancements in attribution science, which allow attributing an individual extreme weather event to climate change, for instance, can help scientists to strongly communicate to the public that climate change is happening 37 . Beyond personal experiences, peer groups and (social) media play an important role as indirect transmission channels. Extreme events have been shown to influence news reporting, leading to more media coverage about climate change 40 . This can contribute to an increased awareness and sensitivity about the issue, allowing people to put themselves into the position of others, even if they are not themselves affected. On the other hand, such indirect experience can also be subject to its own biases and distortions, including collective denial, selective perceptions and confirmation biases. Ideally, effective climate communication should be based on both analytical and experiential information combining facts with emotional and experience-based narratives 15 .
Our findings highlight the importance of increasing the salience of climate impacts in an inclusive way. There is a need to address the substantial geographic differences in public concerns and political support for climate action across regions in Europe, in particular with respect to differences by income levels. As suggested by Ronald Inglehart's postmaterialist theory, residents of wealthier nations whose basic needs for physical and economic security are met can afford to pursue other needs which are relevant for improving quality of life such as environmental quality 41 . This theory also hints that in difficult times, such as economic recessions, value orientations towards postmaterialist environmental and social preferences can be given lower priority due to a renewed prioritization of material needs 42 .
Aside from economic factors, heterogeneous effects may reflect differences in existing infrastructures and adaptation measures, for instance air-conditioning or well-insulated housing, which are influenced by the baseline climatic conditions. There is some evidence showing that adaptation may have happened in countries in hotter climatic zones such as Spain, Greece, and Italy. Here, for instance, a reduction in heat-related mortality is reported in the recent period despite the observed rise in temperatures 43,44 . Such adaptation and habituation processes may explain the differential effects of climate extremes across different regions in Europe. Apart from differential vulnerability to climate change impacts, the level of CO 2 emissions, the degree of reliance on fossil fuels, elite or expert cues and media coverage, among others have also been mentioned as possible underlying factors explaining cross-country variations in climate change attitudes 45 .
Our findings thus are of high relevance for the current debates on how to best promote and effectively implement further climate change mitigation measures in line with the Paris Agreement and the European Green Deal where the EU aims to take a leading position in tackling climate change. At the same time, economic challenges, social and political disruptions, and the switching balance of geopolitical and economic power, might hamper the Union's ability to fulfil its role as a policy innovator, pioneering solutions that tackle the climate emergency in a sustainable fashion. There is a need for an inclusive and equitable approach to climate protection that comprehensively highlights the potential threats of climate change while taking into account the needs and fears of local populations 46,47 .

Online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/ s41558-021-01263-8.

Methods
We make use of a range of georeferenced data sources to measure the main outcomes of interest. Our analyses are carried out at the subnational regional level, where we connect information on changes in climatic conditions to environmental concerns and voting outcomes over time. The resulting panel dataset allows us to test for climatic impacts while controlling for unobserved heterogeneity and time trends via the use of fixed effect.
Environmental concern data. Environmental concerns are measured using 42 waves of the Eurobarometer survey, which provides harmonized data for all EU member and candidate countries. The Eurobarometer is a repeated cross-sectional series of public opinion surveys based on a random, multistage sampling procedure. The surveys are carried out at regular intervals on behalf of the European Commission and other EU Institutions and cover various topics of thematic relevance for the EU (https://europa.eu/eurobarometer). Here, we use information gathered in the standard Eurobarometer trend questions series about issues perceived as the two most important problems in the respondents' country of residence.
By assigning Nomenclature of Territorial Units for Statistics (NUTS) codes to the region of residence of each of the 1,019,723 respondents, we construct a unique regional time series containing data for 34 countries and 264 subnational regions (mostly at NUTS 2 level) spanning 18 years (2002-2019). The standard trend questions are typically collected in the Eurobarometer surveys three times a year during different seasons, allowing us to derive a nuanced picture of trends in environmental concerns throughout the year. All our models control for seasonal effects in the form of season dummies.
As an indicator for environmental concerns, we use the share of respondents in each region who consider environmental issues to be among the two most important issues facing their country at the time of the survey. The answer categories to this question changed slightly over time. While until 2006 the questionnaires only listed an environment-related answer category, that is, 'protecting the environment' , the list was extended by adding another answer category 'energy related issues' afterwards. From 2011 onwards, the two separate answer categories were merged into a new category called 'the environment, climate, and energy issues' . As our goal was to create a harmonized time series for environmental concerns in Europe, we counted any responses referring to the environment as relevant irrespective of differences in the set of answer categories provided.
To account for potential differences in response behaviours by answer category types, we further tested whether any discontinuities in response behaviours were visible immediately before and after the changes in answer category types (Supplementary Table 29). We also reran our main models, restricting the data to different periods on the basis of the answer categories available (2007-2019 and 2011-2019) to test whether changes in the answer sets affect our estimates of the environmental concern models (Supplementary Tables 9 and 10). The additional analyses indicate no substantial changes in response behaviour over time and all results remain fully robust to restricting the data to different time periods.
The variable environmental concern is based on a broad measurement capturing concerns about a variety of environmental issues, including climate change, which may introduce noise in our estimation. To account for this, we have rerun our analysis using two alternative outcome measures reflecting the respondents' concerns about climate change more specifically (Supplementary Table  11). Even though the data on climate change concerns were collected only in a few Eurobarometer waves, mostly after 2010, they can be used to test for the robustness of our findings. Using the alternative outcomes, we find a significant positive impact of the exposure to temperature anomalies, heat episodes and dry spells on climate change concerns. This provides further support to our findings and suggests that we capture the relevant impacts of climate extremes in our baseline models. For this, we rely on the broader environmental concern indicator as a primary outcome as data for this indicator are available for a considerably longer time period.
Voting outcome data. To measure voting outcomes, we collected original data on electoral returns for European Parliament (EP) elections from national sources. The data cover 28 countries and 1,248 subnational regions (mostly at NUTS 3 level) and contain information for six EP election rounds spanning 25 years from 1994 to 2019. If major changes in regional boundaries occurred in a country, the election data were aggregated to NUTS 2 level.
In a first step, we collected the vote shares for all parties participating in the election across subnational regions. From this extensive list of parties, we classified parties as Green on the basis of their party family classification in the Manifesto Project electoral programme database 58 and their membership in the European Green Party, a federation of political parties across Europe supporting green politics, that forms the G-EFA parliamentary group in the European Parliament.
On the basis of this information, we calculated the Green vote share as a fraction of valid votes for Green parties in each NUTS region per election round. Each observation, then, is listed as a region-year election return.
Climatological data. The explanatory variables are constructed from gridded datasets of temperature, precipitation and evapotranspiration. Temperature data come from the ERA5 reanalysis product that uses a global climate model to interpolate the observed weather data to an even 0.1 ° raster 59 . The raster is aggregated temporally to the daily means of the hourly mean temperature and then spatially to the daily regional means. In the calculation of the regional means the grid cells are weighted with the fraction that is covered by the respective region.
On the basis of the region-day observations, we first calculate temperature anomalies as standardized deviations of monthly temperatures from the long-run monthly mean using 1971-2000 as a reference period. In the calculation of the positive (negative) anomaly the values below 0.5 (above -0.5) are set to zero before averaging.
As a second group of relative temperature measures, we define a heat episode as a period of at least three consecutive days with a mean temperature above the 95th percentile of the monthly long-run distribution of daily temperature values and a cold episode as a period of at least three consecutive days with a mean temperature below the 5th percentile. For each region-month the number of days classified as heat and cold episodes are counted and rolling averages are computed, similar to the temperature anomalies.
As a third group of measures, we use the Universal Thermal Climate Index (UTCI) instead of the temperature to calculate heat and cold episodes, again using the 95th and 5th percentile as thresholds to define extremes (Supplementary Table  14 gives alternative threshold specifications). The UTCI represents a thermal comfort indicator accounting for the human physiological response to temperature, humidity, wind and solar radiation 60 .
Finally, dry and wet spells are measured using the Standardized Precipitation-Evapotranspiration Index (SPEI) based on the gridded climate data (TS4.05) from the Climate Research Unit at the University of East Anglia 61 . The SPEI is the standardized water balance, defined as the difference between precipitation and potential evapotranspiration. Evapotranspiration captures the combined water loss of evaporation and transpiration by vegetation. Accordingly, positive SPEI values indicate a larger than usual water balance (wet spell) and negative values a smaller than usual water balance (dry spell). The water balance is accumulated over a rolling period of three months. Values are standardized using a log-logistic distribution based on 1971-2000 as a reference period.
Estimation methods. For our analysis, we combined the georeferenced concern and voting data at the NUTS level with the gridded climatological data to study the impact of variations in climatic conditions in a region. We test whether climate extremes affect environmental concerns and Green voting with a fixed effects panel model of the following form where y it captures the share of the environmentally concerned population or Green voters in a region i at time t. Here, t refers to the month, when the Eurobarometer respondents were interviewed or when the elections were held. C it is a set of climatic indicators capturing climate extremes that occurred before the concern and voting measurement. In our baseline, we consider the effects of extremes in the period 12 months prior, which allows us to broadly capture changes in the climatic conditions across all seasons. In additional sensitivity tests, the climate impact interval was broadened, showing an inverted U-shaped pattern in the relationship between climate extremes and concerns and voting over time (Supplementary  Tables 22-25).
We include a region-specific intercept α i to control for time-invariant factors (unobserved heterogeneity) that may confound the estimation, such as the general political orientation in a region, structural economic factors and the degree of urbanization. In addition, we include time period fixed effects δ t (three-yr periods for the concern data, elections for the voting data) and season fixed effects θ s (only for concern data) to control for time trends and seasonal changes that are common across all regions. As the occurrence of extremes within a region over time is plausibly exogenous conditional on geographic location and time trends, our model allows us to test for the causal impacts of climate extremes on concerns and voting.
By controlling for common underlying trends, the models rule out that the estimated effects are driven by other sociopolitical or economic changes and events that have occurred in Europe over the study period. While in general Europe has become more environmentally progressive in recent years, our estimates capture only the region-specific variations in concerns and voting over time that are related to within-region changes in the environmental conditions and the occurrence of extreme events. The estimates hence reflect climatic impacts over a relatively short time span (12 months for our baseline models) implicitly controlling for characteristics of the regions and European trends. In the tables, the within R² refers to the fraction of variance in the outcome which is explained by the climate variables after accounting for time-invariant and period-invariant factors.
All coefficient estimates are standardized making comparisons across models with different dependent and independent variables possible. We use the standard deviation of the fixed effects residuals for the standardization and thus consider only the variance that is observed within regions over time. This way, the coefficients can be interpreted as changes in concern and voting with a one standard deviation change in the respective climatic factor relative to its regional historical distribution 27 . Models using alternative standardization approaches are presented in Supplementary Tables 18-20. In additional exploratory models, we further extend the baseline model by including interaction terms to capture differences in climatic impacts by climate zones and economic conditions. Here, we rely on additional data provided by Beck et al. 62 for the construction of the climate zones as well as data from the Annual Regional Database of the European Commission 63 for the measurement of regional incomes. Furthermore, we test for the impact of changes in concerns on voting by (1) regressing the voting outcome in year t on changes in environmental concerns in the past one and two years and (2) by using a two-stage instrumental variable approach, where in the first stage climate variables are used as an instrument to predict changes in concerns and in the second stage the Green vote share is regressed on concerns predicted in the first stage.
Limitations. Our analysis comes with certain limitations that are important for the interpretation of our results. The main purpose of this study is to estimate the impacts of climate extremes on concerns and Green voting across subnational regions in Europe over time. While the level of aggregation allows us to study the relationships for a broad sample of regions and time periods and to compare the role of local conditions in moderating the effects, individual drivers of environmental concerns and voting such as values and attitudes are not captured. Further work, especially at more disaggregated or individual levels is therefore needed to fully grasp the underlying drivers and mechanisms beyond what we can examine with our data.
While our baseline models (Table 1) estimate the causal impacts of climate extremes on concerns and voting, the analyses of the heterogeneity by regional characteristics are explorative. The aim of these analyses is to highlight relevant patterns in the data and to suggest possible mechanisms explaining the observed effect. To account for this, the interaction models control for a range of additional demographic, agricultural and political background characteristics that are of relevance for the estimation, allowing us to explore the factors underlying regional differences (Supplementary Table 27).
Our analyses focus on explaining how voting behaviour is influenced by the experiences of climatic extremes through their impact on environmental concerns (Supplementary Table 28). Given the limitations of our data, we are unable to determine whether the reported effects are mainly due to personal experience with environmental events or whether they are driven by indirect experience channelled through the media, family networks or the peer group. While a closer inspection of these underlying mechanisms goes beyond the scope of our analysis, existing evidence suggests that these can play an important role in shaping perceptions of climate change [64][65][66][67] and the experience of extreme events 40 .
We rely on Eurobarometer and European Parliamentary election data to construct the main concern and voting outcomes. While these sources provide comparative longitudinal data for Europe over time, they may not capture all relevant aspects and facets of environmental concerns and Green voting decisions. Our concern measure was constructed on the basis of a priority assessment of the Eurobarometer respondents. Hence, it does not fully reflect the multidimensional nature of the concept of environmental concern, unlike more comprehensive indices, such as the new ecological paradigm (NEP) scale by ref. 68 or the environmental concern scale by ref. 69 . However, these measures are typically collected in case studies and are not available for comparative longitudinal analyses 70 .
Similarly, the regional share of Green voters reflects political support for climate action in a simplified way. In addition, like the results of any election, the outcomes of the European Parliamentary elections can be influenced by voter turnout and selection effects, which were partially accounted for by considering within-regional changes and by controlling for common underlying time trends. Moreover, we are not able to capture all aspects of the supply side and political dynamics of the party system. Some countries might have more credible environmentalist parties; while in other settings longer-term party attachments might prevent environmental concern from turning into Green voting. Again, these influences are captured by the regional fixed effects in our empirical design and hence are not expected to bias the estimation of climate impacts on voting.
Despite these limitations, this study adds important insights to the scientific literature on the climate-related experience, concern and behaviour nexus. The use of the harmonized Eurobarometer data on environmental concern and Europe-wide election measures enables us to achieve comparability across regions and to construct the unique cross-regional trend dataset required for the analysis. Our findings do not only show the role climate extremes play in influencing concerns and voting but also highlight the importance of regional factors, such as climatic and economic conditions. They can thus help to gain a comprehensive understanding of the underlying drivers of observed changes in concerns and voting patterns across Europe. Fig. 1 | Hypothetical relationship 20 . Fig. 3 | effects of heat and cold-related climate extremes on environmental concern and Green voting by climate zones. Coefficients are standardized using the observed variance of the variables in the given climate type after applying the fixed effects. Lines around the point estimates show the 95% confidence intervals. estimates are based on interaction models displayed in Supplementary Table 26.