Record-breaking and unprecedented compound hot and dry summers in Europe under different emission scenarios

After analysing observed summer compound hot and dry (CHD) events over Europe from 1950 to 2022, we employ a large ensemble of high-resolution regional climate model simulations to investigate CHD events under different emission scenarios. By the end of the century, even under a low-emission scenario, model results show a likely increase in the frequency and extension of CHD events over most (60%) of Europe. In particular, the fraction of land projected to be hit once every two years nearly doubles (at least 15%, likely range 6–21) compared to the historical period (8%, 6.5–10), and at least 5,3% (1–7) of land will be hit every year. Under a high-emission scenario, 50% of the Iberian Peninsula is projected to be hit at least twice every three years (20.3 times in 30 years, likely range 17.2–24.2), compared to 1 in ten years in the historical period, whereas 50% of the British Islands, France, and the Mediterranean will be hit more than once every two years. Moreover, 10% of European land will be hit nearly once every 7 years (4.2 times, 3.2–5.6) by CHD events whose intensity equalled or even surpassed the maximum recorded during 1950–2022, and 20% of the Iberian Peninsula once every 5 years. The increase in record-breaking or unprecedented CHD events is mostly related to the increase in record-breaking heatwaves, which is likely over most regions even for the low-emission scenario. In contrast, the increase in record-breaking drought events is limited to southern Europe under the medium- and high-emission scenarios.


Introduction
One of the most impactful consequences of global warming is the increase in the frequency and intensity of extreme events such as heatwaves, floods, and droughts (Seneviratne et al 2021).In particular, when two or more meteorological events (not necessarily extremes by themselves) happen at the same time (or in brief succession), the impact of this compound event can be much larger than the sum of the single events (Zscheischler et al 2018, Vogel et al 2021).For instance, the combined effect of very hot and dry conditions in Europe in recent years caused extensive damages on e.g.livestock farming and agricultural and forest productivity (Bednar-Friedl et al 2022), but also water supplies and hydroelectric power production (García-Herrera et al 2019) with large monetary and, ultimately, human losses (Grumm 2011, Zscheischler et al 2018, Ionita et al 2021).
Compound hot and dry (CHD) events are becoming more frequent at both global (Seneviratne et al 2021) and regional scales, such as in China (Yang et al 2023, Zhang et al 2023), Brazil (Geirinhas et al 2021), USA (Alizadeh et al 2020), Australia (Collins 2022) Europe and the Mediterranean (De Luca et al 2020, Ionita et al 2021, Markonis et al 2021, Hao et al 2022).
The physical mechanisms leading to CHD events are very complex, including both large-scale circulation patterns (blockings, see e.g.Kautz et al 2022) and land-atmosphere interactions and feedback (Zscheischler and Seneviratne 2017), and depend on the region and season.For instance, while in central and northern Europe, summer droughts are usually initiated by a lack of rainfall and high evapotranspiration in spring or early summer, over Southern Europe summer droughts are initiated by precipitation deficits during winter and spring (Russo et al 2019, Ribeiro et al 2020, Markonis et al 2021, Tuel and Eltahir 2021).In addition, although extreme droughts have happened in the long-term past (Hanel et al 2018), recent droughts at seasonal scale are exacerbated by declining soil moisture and increasing potential evapotranspiration caused by regional warming and increasing record-breaking temperatures (Stagge et al 2017, Ossó et al 2022, Schumacher et al 2023), and not necessarily by declining precipitation (Fink et al 2004, Hanel et al 2018, Vogel et al 2021).On the other hand, drought conditions in spring over the Mediterranean have been shown to precede summer heatwaves in continental Europe (Zampieri et al 2009, Stefanon et al 2012).
Under future warmer conditions, the dependence between temperature and precipitation can strengthen further, with higher evaporation causing soils to dry earlier and, ultimately, increasing the probability of CHD events (Zscheischler and Seneviratne 2017), even of those that are today considered extreme, such as the 2018-2019 CHD event in Germany (Toreti et al 2019, Zscheischler and Fischer 2020, Rousi et al 2023).
It is, therefore, essential to understand how CHD events will affect Europe under different emission scenarios, including those designed with the goal of stabilising global warming to below 2 • C above pre-industrial levels as agreed at the 21st Conference of the Parties in Paris (2015).
Although many studies exist analysing future projections for Europe of either droughts (Potopová et al 2018, Spinoni et al 2018, Toreti et al 2019, Hari et al 2020) or heatwaves (Molina et al 2020, Lin et al 2022), studies analysing future CHD events mainly focus on either the global scale (e.g.Liu et al 2021, Wu et al 2021, De Luca and Donat 2023, Tabari and Willems 2023, Tripathy et al 2023) or specific regions only, such as central Europe (Sedlmeier et al 2018, Zscheischler andFischer 2020).For instance, Wu et al (2021), Liu et al (2021), Tripathy et al (2023), Tabari and Willems (2023) and De Luca and Donat (2023) found that central Europe and the Mediterranean are amongst the regions showing stronger increase in frequency and intensity of CHD events through the late 21st century.They also note that under a low-emission scenario the CHD occurrences would stabilise, hence limiting the most adverse effects on the population.
These studies have been primarily carried out by means of Global Climate Models (GCMs), whose coarse horizontal spatial resolution makes them often inadequate in the representation of fine-scale regional processes, particularly those influenced by complex topography, land use heterogeneity, coastal lines and mesoscale convection (Doblas-Reyes et al 2021).In the framework of the COordinated Regional-climate Downscaling EXperiment, (CORDEX, Giorgi and Gutowski 2015), Regional Climate Models (RCMs) have been used to dynamically downscale the results of GCMs, with added value found especially in the representation of fine scale processes and in the ability of the RCMs to simulate extreme events especially for variables such as precipitation (Giorgi et al 2016, Ciarlo`et al 2021, Cardoso and Soares 2022, Careto et al 2022).For instance, Giorgi et al (2016) demonstrated that over the Alps, the summer convective precipitation response simulated by RCMS is opposite to that simulated by the driving GCMs because of better representation of the high-elevation surface heating.
Here we employ a large ensemble of high-resolution RCM simulations to investigate the future evaluation of CHD events in Europe.We first analyse the distribution and frequency of past observed heatwaves, droughts and CHD events from 1950 to 2022.Subsequently, we use the combination of the observed most severe drought and heatwave events as a benchmark for the analysis of the future evolution of record-breaking (or unprecedented) CHD events under different emission scenarios.
To the best of our knowledge, this is the first study investigating future characteristics of CHD events (and the possible occurrence of record-breaking and unprecedented ones) at European scale by means of the large EURO-CORDEX ensemble.In particular, the results at sub-regional scale (opposed to the macro-regions used in GCM-based results) can be particularly useful for the development of possible adaptation strategies at local level.
The paper is structured as follows: section 2 describes the data (both observed and modelled) used in the study and the methodology employed to define heatwave, drought and compound events.Section 3 discusses the results, starting with the analysis of observed CHD summer in Europe, model evaluation, and the analysis of projected events under different emission scenarios.The change in frequency and distribution of record-breaking and unprecedented events is also investigated.Finally, a summary and concluding remarks are presented in section 4.

Model simulations and observation dataset
Daily temperature and precipitation data for the period 1981-2100 was obtained from a large ensemble of model simulations listed in the Supplementary Information table S1.Historical simulations, forced by observed natural and anthropogenic atmospheric composition, cover the period until 2005, whereas projections  are forced by the Representative Concentration Pathways (RCP) 2.6, 4. 5 and 8.5 (Van Vuuren et al 2011), which are here referred to as low-, medium-and high-emission scenarios, respectively.In particular, RCP2.6 is designed as a mitigation scenario aiming to stabilise the increase of global mean temperature to around 2 • C ( Van Vuuren et al 2011).By the end of the century, GCM simulations project a warming relative to 1986-2005 by 1 • C (0.3-1.7), under RCP2.6,1.8 • C (1.1-2.6)under RCP4.5 and 3.7 • C (2.6-4.8)under RCP8.5, (IPCC, 2013).Compared to the pre-industrial period , the warming is estimated to range between 1.8 • C (1.3-2.4)under RCP2.6-4.4 • C (3.3-5.7)under RCP8.5 (IPCC 2021).According to the analysis of the ensemble of EURO-CORDEX simulations by Evin et al (2021Evin et al ( ), compared to 1981Evin et al ( -2010, by the end of the century Europe will be facing summer warming between 3.4 • C (northern Europe) and 4.4 • C (Mediterranean) under RCP8.5, reduced to 1.2 • C (northern Europe) and 1.5 These model simulations were bias-adjusted with a method developed by Lange (2019) and applied to regional climate simulations by e.g.Dosio et al (2022).The bias-adjustment is a trend preserving, parametric quantile mapping, designed to robustly adjust biases in all percentiles.Briefly, the method first generates pseudo-future observations by transferring, for each quantile, the simulated climate change signal to the historical observations, and, then, it uses these pseudo future observations as reference for correcting future model simulations with parametric quantile-mapping.Any trend in daily mean temperature is removed before and restored after these two steps.The method adjusts the wet-day frequency, and it does not consider any specific corrections for extremes preserving the trends in all percentiles.The method has been compared to several other bias-adjustment techniques by Casanueva et al (2020), showing overall satisfactory performance for indices of mean and extreme temperature and precipitation.
The ENSEMBLES daily gridded observational dataset for Europe (E-OBS, Cornes et al 2018), available for the period 1950-2022 (defined as observed period) at a spatial resolution of 0.1 • , is used as observational dataset for the analysis of past CHD events.
The period 1981-2010 is used as reference for the bias-adjustment and for the calculation of the drought and heat wave indices.The same period is also used as reference for model evaluation and for comparison with future projections (2071-2100).Here we focus on the projected change by the end of the century because at shorter time scales internal variability can delay the emergence of the anthropogenic signal in precipitation changes, especially at regional scale (Doblas-Reyes et al 2021).

Compound dry and hot event 2.2.1. Heat wave magnitude index
As an univocal and optimal definition of heat wave index is still under debate (Perkins and Alexander 2013), here we use the Heat Wave Magnitude Index daily (HWMId, Russo et al 2015), designed to take into account both heat wave duration and intensity.The HWMId is defined as the maximum magnitude (equation ( 1)) of all the heatwaves occurring in a period (summer in this case), where a heatwave is defined to occur when at least three consecutive days have maximum temperature Td above the calendar 90th percentile centred on a 31 d window for the reference period .The magnitude of a heatwave is defined as the sum of the daily magnitude Md(Td) of all the consecutive days composing a heatwave, and it is calculated as follows: Here, T30y25p and T30y75p are the 25th and 75th percentiles of the summer maximum temperatures over the reference period .The interquartile range T30y75p−T30y25p defines the heatwave daily magnitude unit: as a consequence, a daily magnitude Md(Td) equal to n indicates that the temperature anomaly on the day d with respect to T30yp25 is n times the climatological interquartile range.

Standardised precipitation index
As an index for drought, we use the standardised precipitation index (SPI; McKee et al 1993), based on the accumulated precipitation data fitted to a gamma distribution.Many definitions and indicators of drought exist (Ranasinghe et al 2021, Hao et al 2022), some of which take into account both precipitation and temperature (such as the SPEI e.g.Stagge et al 2017, Potopová et al 2018, or more complex indices, e.g. Spinoni et al 2015).As in this work it is important to decouple the effect of temperature and precipitation in the definition of CHD, hence we selected a drought index based on precipitation only.
As the focus of our study is the concurrent occurrence of heat waves and droughts in summer, we use the short-term scale SPI-3 index (an indicator of meteorological drought, see e.g.Spinoni et al 2020) rather than longer-term SPI (e.g.SPI-6 or SPI-12).The SPI is calculated monthly for the summer months (June-July-August).

CHD events
After defining the metrics for heatwaves and droughts, we describe the methodology for determining heatwave and drought events, and their concurrent occurrence.
To define the occurrence of an event, temperature and precipitation anomalies (or suitable indices/metrics) are commonly used with respect to fixed or percentile thresholds (for instance, the climatological 90th percentile), at different time scales (daily or monthly/seasonal e.g.Sedlmeier et al 2018, Ionita et al 2021, Vogel et al 2021, Bevacqua et al 2022, De Luca and Donat 2023, see also Hao et al 2022).Note that, for instance, commonly used definitions of heatwave require a further threshold for the minimum number of consecutive days when the condition is fulfilled (usually 3 or 6 d).
In this work, we define a heatwave (HW) event when HWMId is higher than 10, which can be considered a moderate heatwave.As discussed in Dosio et al (2018), the heat waves that hit the Balkans (2007), France (2003) and Russia (2010), had peak magnitudes of around 20, 40, and 80, respectively, which were used to define severe, extreme and exceptional heat waves.To ensure consistency with the definition of heatwave (where only the largest HWMId over the summer is considered), a drought event is defined when the minimum of the monthly (June-July-August) SPI-3 values is <−1 (moderate drought, see e.g.Spinoni et al 2018), meaning that a drought has occurred in at least one of the summer months.
This choice of these moderate thresholds is driven by the necessity of detecting a sufficiently large number of events, especially in the reference period, in particular when considering compound events (i.e. when both conditions need to be fulfilled at the same time).In fact, as noted by e.g.Bevacqua et al (2022), using more restrictive thresholds (such as the 99th percentile threshold instead of the 90th) drastically reduces the probability of occurrence in the observed period.
Finally, to define a CHD event we use the empirical approach based on the concurrence of exceedance of the two indicators (Hao et al 2022): namely, a CHD event is defined when both the above conditions (HW and drought events) occur in the same location in the same year.
Although more complex methods exist to compute the joint distribution of events (such as copulas) that can be useful to e.g.estimate the return period of rare events, in our analysis based on a large ensemble of RCM results, we empirically estimate, for each model, the frequency of univariate and compound events from the modelled time series.As shown by Zscheischler and Seneviratne (2017) this 'counting approach' (used in many other studies, e.g.Geirinhas et al 2021, Ionita et al 2021, Mukherjee and Mishra 2021, Vogel et al 2021, Zhang et al 2023) gives similar results to copula based methods especially in case when the variables are correlated (such as summer dry and hot conditions).
Analyses are performed for each model and RCP separately: geographical distribution (maps) of future events are shown as median of the RCM ensemble.Results based on cumulative distribution functions are shown as median and likely range (defined here as ± one standard deviation) of the RCM ensemble.
Sub-regions are defined according to the areas shown in figure 1, namely: Alps (AL), British Islands (IB), Eastern Europe (EA), Europe (EU), France (FR), Iberian Peninsula (IP), Mediterranean (MD), Middle Europe (ME) and Scandinavia (SC).Fractions of land area are computed by counting the number of grid points (weighted according to their latitude) in the selected sub-region.

Record-breaking and unprecedented events
We define the recorded most severe HW and drought events as those with the maximum HWMId and minimum SPI values for the observed period 1950-2022 (figures 1(a) and (b)).As a consequence, a future record-breaking HW (drought) event occurs if HWMId (SPI) is higher (lower) than the observed maximum (minimum) one.
Note that, in the observed period, the most severe droughts are not necessarily accompanied, in the same year, by the most intense HWs.Assume SPI_m and HWMId_M are the minimum and maximum recorded values in a given grid point.If SPI_m and HWMId_M occur in the same year, then any future event when, at the same time, both SPI and HWMId are equal or worse than SPI_m and HWMId_M will be defined as a record-breaking CHD event.In case SPI_m and HWMId_M occur in different years, any event with SPI and HWMId that are, at the same time, equal or worse than SPI_m and HWMId_M will be defined as unprecedented CHD event, as this particular SPI/HWMId combination has never occurred in the observed period.
It must be noted that the definition of the most intense event strongly depends on the definition of the metric used, which may greatly vary at local scale.For instance, although the heatwave occurred in 2003 has been regarded as one of the most intense (in terms of temperatures) and impactful (in terms of excess deaths) at country level in France, over the Paris area (and in general northern France), the 1976 HW had a longer duration (18 d compared to 8 d in 2003 over the Paris area, see Lemonsu et al 2014), which may result in a higher HWMId value.

Observed CHD events (1950-2022)
We first analyse the evolution of historical, observed CHD events over the entire E-OBS available period, i.e. 1950-2022.Generally, European summers have become hotter and, for some sub-regions, drier over the last The geographical distribution of the frequencies (i.e.number of observed events in the 1950-2022 period) of HW, droughts and CHD events differ greatly at local and sub-regional scale (figures 1(f)-(h)), with the majority of CHD events occurring over northern and eastern Europe.
Figure 2 shows the yearly evolution of the percentage of land area affected by HW, drought and CHD events for each sub-region.The positive trend in the fraction of EU land affected by CHD (red line and bars) events is evident, especially in the last decade or so, due mainly to the increase in the area affected by HWs after 2000 (yellow line), with the most impacting years being 2022 (with nearly 40% of European land affected by CHD) followed by 2015, 2003 and 2018.Results for other sub-regions show very large interannual variations, but, in general, present similar features i.e. small or nearly nihil trend in drought-affected land fraction and large increase in HW-affected land fraction, with the expectation of ME, AL and, partly, IP, showing an increase in the drought-affected land fraction after 2010, and a decline over SC (consistently with the trend in SPI, figure 1).
Although based on different metrics, our results agree with those reported by Ionita et al (2021) for CHD events and those by e.g.Spinoni et al (2015) for droughts.
The sub-regionally averaged, annual HWMId and SPI values are shown in figure 3. Years when HW, drought and CHD events have occurred (at sub-region level) are also shown.We note for instance that while most of the dry years in Scandinavia happened before 1990 (in accordance with the increase in SPI, figure 1(d) and the reduction in land fraction affected by droughts, figure 2), other regions do not show a clear tendency in drought frequency (MD, IB, EA, FR), whereas over AL and IP most of the drought events happened in the last 30 years.However, as the frequency of HW events increases steadily over all regions of Europe (compare the positive trend in both HWMId value, figure 1(e), and in the area fraction affected by HW events, figure 2), this results in CHD events clustering usually in the last decade or so in all sub-regions apart from BI and SC. Figure 3 also shows the years when the minimum SPI and maximum HWMId have Finally, the black box-and-whiskers plots show the median, interquartile and 95% ranges of the RCM ensemble over the historical period .

Model evaluation (1981-2010)
The model ensemble is evaluated over the reference period 1981-2010.When RCMs are driven by GCMs, a year-to-year correlation between the observed and modelled time series of a meteorological event (as a drought or HW) cannot be expected; hence, we evaluate the capability of the ensemble to capture the statistics of the geographical and temporal distribution of the events.At European scale, the ensemble captures very well (both in median and range) the yearly distribution of cumulative distribution functions (CDF) of the fraction of land for a range of SPI and HWMId values (figures 4(a) and (b)).The geographical distributions of drought, HW and CHD events are also well captured (figures 4(c)-(h)): although, locally, the spatial variability of the event frequency is less pronounced in the model ensemble (which is expected as observations are compared to the median value of a large model ensemble), the general geographical distribution is well captured, especially for HW events, with the largest values over the UK, Denmark and the southern parts of Sweden and Norway, and the lowest values over the Iberian peninsula, part of France and Germany, and Eastern Europe (with the exception of Russia where models underestimate the HW events).Also at sub-regional level (figure 3) the models are able to capture the range of observed SPI and HWMId values over the period 1981-2010, even for extreme HW events such as the 2003, 2007, and 2010 episodes (over specific sub-regions).

Projection of CHD events under different emission scenarios (2071-2100)
Figure 5 (top row) shows the geographical distribution of projected CHD events at the end of the 21st century (2071-2100) under low-, medium-and high-emission scenarios.Our results show that, under RCP2.6,most of Europe will face less than 6 CHD events in 30 years, apart from Ireland, Spain, Italy, Greece and Turkey where the number of events can reach up to 9.Under the medium-and high-emission scenarios, a north-south gradient in the CHD event geographical distribution appears, with most of the Mediterranean countries facing up to 15 CHD events in 30 years under RCP4.5, but up to 24 events (i.e. 4 every 5 years) under RCP8.5 in some areas of Spain and Portugal.
From these distribution maps, one can calculate the CDFs of how many times (at least) a certain fraction of land (at sub-regional level) is projected to face a CHD event in 2071-2100 under the three emission scenarios (see figure S1 for details on the methodology).Model results (sub-regional panels in figure 5) show a large difference between sub-regions and across emission scenarios: for instance, the fraction of SC land affected by CHD will hardly change with emission scenarios, and, for most of the region, the frequency of CHD events will be similar to the historical (modelled) and observed (E-OBS) one , with only less than 20% of land facing a discernible (as indicated by the non overlapping likely ranges), although small, change between the historical and the future scenarios.Over other regions, however, results are largely scenario dependent, especially for IP, MD and FR.In particular, the minimum number of CHD events hitting 50% of IP land grows from 2.7 (likely range 2.3-3.4) in the historical period to 7.1 (4.7-8.7)under RCP2.6,2)under RCP8.5.Similar large increases are projected for FR and MD (see table S2).IP also shows the highest number of CHD events projected to occur both at local scale (i.e. 5% of land), namely,11.7,17.7 and 25.3,under RCP2.6,RCP4.5,and RCP8.5 respectively, and large scale (95% of land), namely, 3.4, 7.8 and 14.9.In addition, all IP land points are projected to face a CHD event at least 12 times (panel IP and the RCP8.5 map in figure 5) although not necessarily in the same years (i.e.there will not necessarily be 12 years when a single CDH event covers the totality of IP land, see description in figure S1), compared to 6 for RCP4.5, 2 for RCP2.6 (figure 5) and 0 for the historical period (figures 4(e) and (h)).
It is worth noting that a discernible increase (compared to the historical period) in CHD frequency is likely to occur for the entire IP sub-region even under the low emission scenario RCP2.6.This is also true for around 60% of MD and EU land, meaning that even an emission scenario designed to stabilise global warming around 2 • C by the end of the Century (Van Vuuren et al 2011) will result in a likely increase in CHD events.
Figure 6 (top row) shows the yearly evolution of the area fraction (for the EU region) hit by a CHD event under the three RCPs in 2071-2100.The increase compared to the historical period (1981-2010, figure 2-EU) is evident even under the low-emission scenario RCP2.6.The 30 year average of the model ensemble median is 8.2% for the historical period (compared to 8.6% for the observed one in 1981-2010, see figure 2-EU), 15.5% under RCP2.6,24.8% under RCP4.5 and 38.3% under RCP8.5.Model uncertainty is however very large: for instance, under RCP8.5 we note that all models project at least 5% of EU land being hit by a CHD event every year, with some models projecting more than 80% of land hit by a CHD event at least one year.
From these time series, one can compute the CDF of land fraction (at sub-region level) that is projected to be hit N times by a CHD event (see methodology in figure S1).For instance, the fraction of IP land projected to be hit 15 times (i.e.once every two years) grows from at least 2% (likely range 1.3-3.5) in the historical period to at least 15% (6-21) under RCP2.6,36.5% (21-52.5)under RCP4.5 and 72% (54-81) under RCP8.5.Results for the other sub-regions are listed in table S3.
At EU level, the fraction of land projected to be hit once every two years grows from at least 8% (likely range 6.5-10) in the historical period to 15% (12-21.5)under RCP2.6,25% (20-30) under RCP4.5 and 38.5% (32-46.5)under RCP8.5.In addition 17% (6-24) of EU land will be hit by a CHD event every year, compared to 6% (1-10) under RCP4.5, 3% (1-7) under RCP2.6 and 1% (0.33-2) in the historical period (note that the fraction of land hit every year is not necessarily made up by the same grid points, see figure S1).Finally, we note that the likely ranges of the CDFs in the historical and RCP2.6 scenarios differ for CHD frequencies >8, meaning that the land fraction frequently affected by CHD events is likely to increase even under the low-emission scenario.

Record-breaking and unprecedented CHD events
Figure 7 shows the geographical distribution and the CDFs of area fraction (at sub-region level) projected to be hit by a certain number of record-breaking (or unprecedented) CHD events under the three emission scenarios.Under RCP4.5, in 2071-2100 up to two record-breaking CHD events are projected, locally, over parts of southern Europe.However, under RCP8.5 the vast majority of the Mediterranean countries (but also Switzerland, Ireland and areas in central and eastern Europe) will be hit by at least one record-breaking (or unprecedented) CHD event, with areas in Portugal and over the Pyrenees projected to face more than five events.
In particular, the number of record-breaking CHD events projected to hit 5% of IP land increases from at least 0.96 (0.95-0.98) in the historical period to 1.7 (0.99-2.19) under RCP2.6,3.3 (2.3-3.9)under RCP4.5 and 10.3 (8.3-11.8)under RCP8.5.This means that, under the high-emission scenario, a small but non-negligible part of IP will face, at least once every three years, CHD events whose intensity has equalled or even surpassed that of the historical record 1950-2022.In addition, a considerable land fraction (20%) will be hit at least once every 5 years (6.6 events, likely range 5.1-8.0).
Figure 8 (top row) shows the yearly evolution of the area fraction (for the EU region) hit by a record-breaking or unprecedented CHD event under the three RCPs in 2071-2100 and the corresponding CDFs at sub-region level.As from figure 7, it is clear that the regions most impacted will be southern Europe and the Mediterranean countries.In particular, nearly one quarter (at least 23%, likely range 14.5-35) of IP land will face record-breaking or unprecedented CHD events 5 times in 30 years under RCP8.5, at least 13% (6.5-22) of land will be impacted 10 times (i.e.once every three years), and at least 3% (1.3-8) 20 times (two out of three years).
Other regions will be affected, although at a lower extent: for instance, at least 14% (7-23) of MD land and 8% (2-18) of AL land will face record-breaking or unprecedented CHD events 5 times in 30 years.At EU level, under RCP8.5, 5 events in 30 years are projected over at least 6% (3.7-9.5) of land and 10 events over at least 4.2% (2.5-6.5) of land.
As record-breaking CHD events are defined by the combined occurrence of droughts and HWs that are at least as severe as the most intense ones observed during 1950-2022, it is interesting to investigate how these conditions will evolve, separately, under future emission scenarios.
The spatial distribution and CDFs of the number of record-breaking HW and drought events are shown in figures 9 and 10 respectively.The increase (compared to the historical period) in the occurrence of record-breaking HWs is likely even for the low-emission scenario over most (>50%) of land in most regions (apart ME, FR, and SC) with 50% of EU land projected to be hit by at least 5.6 (4.7-6.8)HW events under RCP2.6,7.6 (6.9-8.3)under RCP4.5 and 11.7 (10.3-13.1)under RCP8.5, i.e. more than once every three years in 2071-2100.
In addition, some sub-regions (BI) will have the near totality (>99%) of land hit by record-breaking HWs at least 7 times in 30 years under RCP8.5, rising to at least 10 times for MD and more than 12 times for FR and IP (i.e. more than once every three years); it is worth noting that in those regions (BI, IP, FR, MD) record-breaking HWs are projected to hit more than 95% of land at least 3 times even under the low-emission scenario.
On the contrary, the increase in frequency of record-breaking drought events is more limited although at European level, 5% of land is expected to face at least 3.0 (2.8-3.7)events under RCP2.6,3.9 (3.6-4.7)under RCP4.5 and 7.7 (5.7-9.2) under RCP8.5 (i.e.once every 4 years), and the increase in frequency is likely for up to 25% of land even under the low-emission scenario.Southern Europe will be particularly prone to record breaking droughts, with 50% of IP land projected to face at least 4.7 (3.6-6.1)record-breaking events under RCP8.5, and 20% of land at least 7.9 (6.5-10.0)events, which confirms the findings of the many studies highlighting the Mediterranean region as hotspot for future droughts (Russo et al 2019, Spinoni et al 2020, Ranasinghe et al 2021, Tabari and Willems 2023), and also corroborating the claims that future CHD events will be mainly constrained by the precipitation signal (Bevacqua et al 2022) which, at regional sale, can be often masked by natural variability especially for low-or medium-emission scenarios (Doblas-Reyes et al 2021).

Summary and conclusions
In this study we showed that even a low-emission scenario designed to stabilise global warming to 2 • C by the end of the century (RCP2.6)will result in a likely increase in the frequency and extension of CHD events over most (60%) of Europe.In particular, the fraction of European land projected to be hit once every two years nearly doubles under the low-emission scenarios (at least 15%, likely range 6-21) compared to the historical period (at least 8%, likely range 6.5-10) and a small but not negligible fraction (at least 5.3%, likely range 1-7) of land will be hit by a CHD event every year (compared to 1% in the historical period).
Under medium-and high-emission scenarios (RCP4.5 and RCP8.5), a large increase in the frequency of CHD events is projected especially over southern Europe (but also the British Islands): 50% of IP land is projected to be hit at least twice every three years (20.3, times likely range 17.2-24.2) under RCP8.5 (compared to 1 in ten years in the historical period) while 50% of BI, FR, and MD more than once every two years.In addition, all IP land points are projected to face a CHD event at least 12 times in 30 years.
With increasing warming Europe will face CHD events whose intensity has equalled or even surpassed that of the historical observed record , with the number of record-breaking or unprecedented CHD events hitting 10% of land projected to increase from at least 1.2 (0.9-1.5) under RCP2.6-1.7 (1.5-1.9)under RCP4.5 and 4.2 (3.2-5.6)under RCP8.5.In addition, 20% of IP land will be hit at least once every 5 years by record-breaking or unprecedented CHD events, under RCP8.5.
The increase in record-breaking CHD is mostly related to the increase in record-breaking HWs, which is likely over most regions even for the low-emission scenario, with 50% of European land projected to be hit by at least 5.6 (4.7-6.8)events under RCP2.6,7.6 (6.9-8.3)under RCP4.5 and 11.7 (10.3-13.1)i.e. once every three years, under RCP8.5.On the contrary, the increase in frequency of record-breaking drought events is more limited to southern Europe, with 50% of IP land projected to face at least 4.7 (3.6-6.1)record-breaking events under RCP8.5.Some caveats to our study that need to be mentioned are listed below: • (1) Extreme events such as heat waves and droughts (and consequently CHD events) can be quantified by several indices, and even definitions (e.g.meteorological vs. hydrological or agricultural droughts) which can lead to quantitatively different, but qualitatively similar, results (Cammalleri et al 2020, Ionita et al 2021).In addition, we only analysed the change in frequency and distribution of CHD events, and not their intensity (i.e. the change in HWMId or SPI).• (2) In addition, any study focusing on the analysis of the frequency of (extreme) events depends on a threshold to define the exceedance (e.g.90th percentile).In our study, we define a CHD event as the combination of relatively moderate extremes (HWMId >10 and SPI <−1).Although the results for the projected frequency of CHD events may depend on the choice of the threshold, the results for record-breaking and unprecedented CHD events are independent of this threshold, as they are defined relative to the most severe event occurred in the observed period.• (3) Our study analysed only 'short term' droughts (SPI-3) and we have not considered multi-year or consecutive events, which can have devastating impacts on e.g. the agricultural and forest sectors.To this extent, the impact of extreme events such as droughts and HWs can be very different depending on the region impacted and the socio-economic sector considered (Clarke et al 2022).In particular, we have not analysed other components of risk, namely exposure and vulnerability (Rohat et al 2019, Cammalleri et al 2020, Liu et al 2021, Spinoni et al 2021, Tabari and Willems 2023, Zhang et al 2023).• (4) Our estimates of the future uncertainty in the frequency and extension of CHD events are based on the concept of likelihood defined as one standard deviation.However, it must be noted that the full range of possible future events can be much larger (as shown for instance in figure 6 where all model results are shown); as a consequence, our findings may underestimate the risk of high-impact low-likelihood events, even under the low-emission scenario.• (5) Finally, our study does not investigate the physical mechanisms (or the change in the mechanisms) leading to (past and future) heat waves, droughts and, ultimately, CHD events (Ionita et al 2021, Herrera-Lormendez et al 2023, Rousi et al 2023).As mentioned, these mechanisms are very complex and include both dynamic and thermodynamic factors, at both large (circulation patterns) and local (e.g.land-atmosphere interaction) scales.Climate models may have difficulties in reproducing satisfactorily these mechanisms, their interaction and their change under future warming.For instance, Knist et al (2017) found that RCMs agree with the observational datasets in the large-scale pattern of the land-atmospheric coupling, characterised by strong coupling in southern Europe and weak coupling in northern Europe.However, over large parts of central Europe, in the transition zone from strong to weak coupling, many of the RCMs tend to overestimate the coupling strength.Lhotka et al (2018) found that most of the EURO-CORDEX RCMs struggle to simulate correctly the joint contribution of large-scale circulation and land-atmosphere interactions, which affects the magnitude of modelled heatwaves.This said, Ranasinghe et al (2021) show that RCMs strongly agree with GCMs (both CMIP5 and CMIP6) in the projected reduction in precipitation by mid-and end-century over the Mediterranean.In addition, our results confirm the findings in the existing literature based on GCMs (e.g.Liu et al 2021, Wu et al 2021, De Luca and Donat 2023, Tabari and Willems 2023, Tripathy et al 2023), highlighting the Mediterranean region as a hot-spot for future CHD events, which reinforce confidence in our results.
Despite these limitations, we believe that this work is useful for raising awareness, and clearly advocates for preparedness to a future where extreme events can occur, especially under mid-or high-emission scenarios, with a combined intensity higher than the most severe one observed in the last 70 years.

Figure 1 .
Figure 1.Panels (a), (b): spatial distribution of the minimum monthly (JJA) SPI value (a) and maximum HWMId (b) over the period 1950-2022 according to the E-OBS observational dataset.Panel (c): sub-regions used for the analysis, denied as follows: Alps (AL), British Islands (IB), Eastern Europe (EA), France (FR), Iberian Peninsula (IP), Mediterranean (MD), Middle Europe (ME), Scandinavia (SC), and Europe (EU).Panels (d), (e): observed (E-OBS) trend (per decade) in SPI and HWMId over the period 1950-2022.Hatching denotes areas where the trend is not statistically significant (least-squares linear trend at 5% significant level).Panels (f)-(h): observed number of drought, HW and CHD events over the period 1950-2022.Note the lack of data in the E-OBS dataset over most of Turkey and southern Greece (white areas).

Figure 2 .
Figure 2. Time evolution (1950-2022) of land area fraction (AF, %) affected by drought (purple bars), HW (yellow) and CHD (red) events for each European sub-region (see figure 1).For clarity, bold lines show the same data smoothed by a 10-year moving average.

Figure 3 .
Figure 3.For each of the sub-regions, the panels show the area-averaged, yearly values of SPI and HWMId (circles).Black circles denote years in the reference period 1981-2010 whereas red circles show the values in the remaining years of the entire 1950-2022 E-OBS dataset.The horizontal (vertical) purple (yellow) line marks the value used to define a drought (HW) event.The years when the minimum SPI and maximum HWMId occurred are also indicated.On top of each regional panel, the vertical bars show the years when a drought (purple), HW (yellow), and CHD (red) events have occurred.Finally, the black box-and-whiskers plots show the median, interquartile and 95% ranges of the RCM ensemble over the historical period.

Figure 4 .
Figure 4. Observed and modelled cumulative distribution function (CDF) of the fraction (%) of EU land area affected by droughts (a) and HW (b) during the period 1981-2010.Bold, thin and dashed black lines indicate the median, interquartile and 95% range of the observed (E-OBS) values over the period 1981-2010.Blue box-and-whiskers plots show, for selected SPI and HWMId values, the model results for the historical (1981-2010) period.Middle row (c)-(e): observed frequency (number of events) of droughts, HW and CHD events over the period 1981-2010.Bottom row (f)-(h): modelled frequency (median of the model ensemble) of droughts, HW and CHD events over the period 1981-2010.

Figure 5 .
Figure 5. Top row: geographical distribution of the number of CHD events in 2071-2100 under RCP8.5 (left), RCP4.5 (centre) and RCP2.6 (right) emission scenarios.Results are shown as the median of the model ensemble.Sub-region panels: fraction of land (AF, %) for each sub-region that is expected to be hit by at least N CHD events in 2071-2100 under RCP8.5 (red), RCP4.5 (yellow) and RCP2.6 (green).The black and blue lines indicate the observed (E-OBS) and modelled occurrence in 1981-2010, respectively.Bold lines and coloured areas denote the median and the likely range (i.e. ± one standard deviation) of the model ensemble, respectively.Details on how the CDFs are calculated are described in figure S1.

Figure 6 .
Figure 6.Top row: yearly evolution of the area fraction (AF, %) of EU land hit by a CHD event in 2071-2100 under the three RCPs.All model results are shown.Bold continuous and dashed lines indicate the model median and interquartile ranges, respectively.Sub-region panels: number of CHD events that are projected to hit a certain fraction X of land at sub-regional level in 2071-2100 under RCP8.5 (red), RCP4.5 (yellow) and RCP2.6 (green).The black and blue lines indicate the observed (E-OBS) and modelled occurrence in 1981-2010, respectively.Bold lines and coloured areas denote the median and the likely range (i.e. ± one standard deviation) of the model ensemble, respectively.Note that ensemble medians in the yearly evolution panels are not comparable to those in the CDF panels.In fact, CDFs are calculated for each individual model separately (i.e. for each individual line in the yearly time series) before computing median and likely range.The ensemble median of the CDFs is not the same as the CDF of the median.

Figure 7 .
Figure 7. Top row: geographical distribution of the number of record-breaking or unprecedented (compared to those observed in 1950-2022) CHD events in 2071-2100 under RCP8.5 (left), RCP4.5 (centre) and RCP2.6 (right) emission scenarios.Results are shown as median of the model ensemble.Sub-region panels: CDFs of the fraction of land (AF, %) for each sub-region that is expected to be hit by at least N record-breaking CHD events in 2071-2100 under RCP8.5 (red), RCP4.5 (yellow) and RCP2.6 (green).Bold lines and coloured areas denote the median and the likely range (i.e. ± one standard deviation) of the model ensemble, respectively.

Figure 8 .
Figure 8. Top row: yearly evolution of the area fraction (%) of EU land hit by a record-breaking or unprecedented CHD event in 2071-2100 under the three RCPs.All model results are shown.Bold continuous and dashed lines indicate the model median and interquartile ranges, respectively.Sub-region panels: number of record-breaking CHD events that are projected to hit a certain fraction X of land at sub-regional level in 2071-2100 under RCP8.5 (red), RCP4.5 (yellow) and RCP2.6 (green).Bold lines and coloured areas denote the median and the likely range (i.e. ± one standard deviation) of the model ensemble, respectively.

Figure 9 .
Figure 9. Top row: geographical distribution of the number of record-breaking (compared to those observed in 1950-2022) HW events in 2071-2100 under RCP8.5 (left), RCP4.5 (centre) and RCP2.6 (right) emission scenarios, shown as median of the model ensemble.Sub-region panels: fraction of land (AF, %) for each sub-region that is expected to be hit by at least N record-breaking HW events in 2071-2100 under RCP8.5 (red), RCP4.5 (yellow) and RCP2.6 (green).Bold lines and coloured areas denote the median and the likely range (i.e. ± one standard deviation) of the model ensemble, respectively.

Figure 10 .
Figure 10.Top row: geographical distribution of the number of record-breaking (compared to those observed in 1950-2022) drought events in 2071-2100 under RCP8.5 (left), RCP4.5 (centre) and RCP2.6 (right) emission scenarios, shown as median of the model ensemble.Sub-region panels: fraction of land (AF, %) for each sub-region that is expected to be hit by at least N record-breaking drought events in 2071-2100 under RCP8.5 (red), RCP4.5 (yellow) and RCP2.6 (green).Bold lines and coloured areas denote the median and the likely range (i.e. ± one standard deviation) of the model ensemble, respectively.