An endless summer: 2018 heat episodes in Europe in the context of secular temperature variability and change

The year 2018 was affected by very long‐lasting, stable summer conditions in vast parts of Europe, in many regions already starting in April and lasting until October. We investigate the thermal characteristics of this year in a secular time perspective, using a spatially well‐distributed dataset of 67 European stations (west of ~30°E) with daily long‐term air temperature observations. Our dataset comprises many of the longest and most reliable (homogenous) temperature measurements available in Europe, mainly starting already in the 19th or rarely even 18th century. Individual time series length is considered to analyse the summer 2018 temperatures into a more than two‐century time perspective, while European time series are presented for the period 1855–2018 and records of five European regions are considered from 1881 onwards. The extreme long duration of the 2018 summer most clearly manifested itself in pronounced new continental maxima of summer half year (April–September) temperature averages and the number of days with maximum temperatures ≥20 and ≥25°C. Furthermore, those indices reached new local maxima at about half of our investigated stations. Records of other temperature indices, rather representing intense heat conditions, were particularly broken in the extended Baltic Sea region, supported by distinct and long‐lasting anticyclonic conditions in this area. However, the extreme summer of 2003 still dominates the ranking of most of these indices on the continental level, even though it is now generally closely followed by summer 2018. We show that long‐term variations of teleconnection indices and the Atlantic Multidecadal Oscillation, as well as seasonal and anthropogenic effects, supported both the recent hot summers and the extreme summer of 2018.


| INTRODUCTION
The summer half year (SHY) of 2018 was characterized by unusually early, long lasting, very persistent and regionally record-breaking heat conditions, coupled with a pronounced precipitation deficit and sunshine surplus in large areas of Europe. According national reports include, among others, Sweden (SHMI, 2018), Finland (FMI, 2018), Estonia (EMHI, 2018), Great Britain (MetOffice, 2018), The Netherlands (KNMI, 2018), Germany (DWD, 2018a(DWD, , 2018b, Switzerland (MeteoSwiss, 2018), Austria (ZAMG, 2018) and Italy (ISPRA, 2018). Expectably, severe societal and environmental impacts were observed. They included unusual and widespread forest fires in Scandinavia, especially Sweden (Krisinformation, 2018), where NATO assistance was necessary to fight and contain the fires (NATO, 2018). Record-low river levels and very high water temperatures were witnessed in many areas, for example, Switzerland (BAFU, 2018) and Germany (BfG, 2019). Low soil moisture conditions triggered intense vegetation responses, like early leave fall or even dying of otherwise healthy trees (BMEL, 2019). It also negatively affected crop harvests, which were below average in most of Europe (JRC, 2018)for instance in Germany especially for grain and rapeseed .
2018 continues the series of very hot summers of recent years, whichsince the previously unprecedented summer of 2003 (Luterbacher et al., 2004) strongly accumulated in Europe (e.g., Christidis et al., 2015;Russo et al., 2015;Dong et al., 2017;Manning et al., 2019). The 2003 summer was up to five degrees warmer than  average conditions in the greater alpine area, with large anomalies in most parts of Europe, especially in a broad strip extending from North-Western to South-Eastern Europe (Schär et al., 2004). Seven years later the record-breaking summer of 2010 (e.g., Dole et al., 2011;Rahmstorf and Coumou, 2011;Otto et al., 2012;Trenberth and Fasullo, 2012;Hauser et al., 2016) even exceeded amplitude and spatial extent of 2003 (Barriopedro et al., 2011;Russo et al., 2015), with most strongly affected areas located in Western Russia and Eastern Europe. The summers of 2003 and 2010 are the most prominent examples of recent heat extremes, because of their extreme heat intensity, duration of heat waves and large-scale spatial relevance. Furthermore, they received wide-spread attention for their elevated levels of human morbidity and mortality (Åström et al., 2011;Li et al., 2015) with 10.000 s of excess deaths (Robine et al., 2007;Barriopedro et al., 2011), largely connected to their very long and uninterrupted heat waves.
Global warming is indeed the key factor of the high frequency of heat episodes occurring in the last decades all over the world (IPCC, 2013;Russo et al., 2014;Christidis et al., 2015), and further increasing heat conditions are expected by climate modellers within the 21st century (e.g., Barriopedro et al., 2011;Russo et al., 2014Russo et al., , 2015Christidis et al., 2015;Lhotka et al., 2018). However, also large scale (atmospheric and oceanic) circulation variations strongly influence extreme summer temperatures in Europe (e.g., Sutton and Hodson, 2005;Jones and Lister, 2009;van den Besselaar et al., 2010). Previously European summers were comparably warm from the 1930s to the 1950s, culminating in the very hot summers of 1946 (predominantly in South-East Europe) and 1947 (particularly in Western Europe, but also other areas; Hoy et al., 2017). That period was connected to a higher level of continentality, especially pronounced in the 1940s (Kożuchowski et al., 1994;Thompson, 1995). In contrast, warm summers were rather rare in the early 20th century and from the 1960s to the 1980s, connected to predominantly maritime air masses over Europe (Kyselý, 2002). Since the 1990s, much higher temperatures (alongside an increased level of continentality) were observed during summer, and during the early 21st century, temperatures settled around the level of the extreme 1947 summer in large parts of Europe (Hoy et al., 2017).
This study analyses duration, timing, intensity and spatial extent of the 2018 heat conditions in Europe in a secular time-perspective, and the possible contribution of atmospheric circulation and Atlantic Ocean temperatures.
It is based on a spatially well-distributed dataset comprising many of the longest and most reliable station time series with daily average and extreme temperature records available in Europe, with a special focus on data homogeneity. We involve a range of temperature indices and two heat wave definitions. We consider the length of individual station series to analyse the summer 2018 heat episodes into a more than two century time perspective, while we present time series for a European average for the period 1855-2018, completed by evaluations for five European regions from 1881 onwards. To explain observed variations in European summer heat, we employ the following datasets: (a) the original (manual) and one automated version of the well-known Grosswetterlagen classification (Hoy et al., 2013a(Hoy et al., , 2013b(Hoy et al., , 2013c, (b) a selection of teleconnection indices relevant for Europe (CPC, 2012) and (c) an index describing variations of the Atlantic sea surface temperatures, the Atlantic Multidecadal Oscillation (Sutton and Hodson, 2005;Sutton and Dong, 2012).
Our study is compiled the following: Section 2 introduces study area and data base and discusses data quality and homogeneity issues. It further presents temperature indices, classifications of atmospheric and oceanic circulation and methods. Section 3 gives an overview about the thermal conditions of the 2018 summer, relates them to centennial developments, and investigates the role of atmospheric and oceanic circulation. The paper finishes with a discussion of the results complemented by further seasonal and anthropogenic supporting factors of hot summers (Section 4), followed by the conclusions in Section 5.
2 | DATA AND METHODS 2.1 | Study area and data base

| Study area
The study area is the European continent apart of Russia east of its western borders, the eastern parts of the Ukraine and Iceland (Figure 1a). To account for regional specifics, five European regions are defined in this paper. In their compilation, we focussed on both achieving geographic consistency and a grouping of areas with similar climatic features (Figure 1b), using continentality (winter-summer temperature difference) and average summer temperatures (Supporting Information Data S1). We also looked at the distribution of subtypes of the Köppen-Geiger climate classification (according to Beck et al., 2018). All regions comprise 14 climate stations, apart of region South-East with 11 stations. The regions may be characterized the following: • NE (North-East): cool, rather continental climate (large differences between winter and summer temperature, summer temperature around 15 C; mainly Dfc and Dfb T A B L E 1 Characteristics of the station data set as described from headlines and additional information below Note: Data source abbreviations: national weather services (NMI, CHMI, DWD, RHSS); other names + institutions: personal communication; ECA&D hom: originating from homogenized dataset (Squintu et al., 2019) of the ECA&D daily dataset (see Section 2), ECA&D: original time series from ECA&D. Quality classes: (1) metadata-based homogenized data; (2) technically homogenized data or original data of high quality (no/small location changes); (3) data quality not assured (can be despite of technical homogenization of full or parts of the time series). reliable (homogenous) daily mean (TG), maximum (TX) and minimum temperature (TN) European station time series, while at the same time ensuring an as optimal as possible spatial distribution of stations within Europe. Station density is spatially similar within the regions except for region North-East, which has more stations in its southern part (≤60 N) than further north. Table 1 presents an overview of the included stations and their data characteristics including the start year of the records. Almost all stations are located at altitudesmostly considerablybelow 500 m (three exceptions up to 667 m), with an average elevation of 124 m. Average data availability is 98.2% for all three parameters, with only two stations dropping below 90% availability. Region Central comprises the longest data series, while time series in regions North-East and South-East are, on average, shorter ( Figure 2). Six stations with daily TG and three with TN/TX data (Brussels, Prague and Milan) start within the 18th century and five more stations before 1850. A clear increase of data availability gets apparent from about 1881 and again 1901, as well as a temporary drop in availability towards the end of both world wars (more pronounced for WWII). Data of all stations, with very few exceptions during single years, are available from 1951. All time series are updated until October 2018 or longer.

| Data quality and homogeneity
A high level of data homogeneity is crucial, because inhomogeneities may seriously affect results generated from the analysis of time series of meteorological data (Wijngaard et al., 2003;Böhm et al., 2010;Lundstad and Tveito, 2016). Supporting Information Data S2 illuminates potential inhomogeneity sources within time series of meteorological data and resulting problems for data analysis.
As we are investigating secular time frames, the quality of the (complete) underlying dataset determines the robustness of our study. Hence, the stations employed in this study were selected according to quality criteriaabout 90% of all station time series used in this paper underwent some form of homogeneity treatment (Table 1). We ranked data according to three different quality classes. Carefully documented homogenizations of full time series using local metadata form the best and most reliable Category 1, connected to a very high research effort per station. Only Stockholm (SE), Armagh, Oxford (both GB), De Bilt (NL), Basel, Geneva (both CH) and additionally for TG Tromso, Bergen and Oslo (all NO) fall in that categorythe relevant papers describing the performed quality controls and homogenization efforts are given in Table 1. On the other hand, only a few stations had to be allocated to the lowest Category 3, because their quality could not specifically be verified. They do not necessarily include (large) inhomogeneities, but their quality is not traceably enough for us to proof the opposite.
About 80% of our data fall in Category 2, which consists of technically homogenized data or quality-controlled original data with proven none or very little location changes. Most of the data in that category derives from a homogenized version of the original European Climate Assessment and Dataset (ECA&D) station data collective (see https://eca.knmi.nl//dailydata/predefinedseries.php), which itself can be considered to be the largest and most reliable source of long-term station data in Europe (Simolo et al., 2014;Cioffi et al., 2015;Hoy et al., 2017;Squintu et al., 2019). The station data relevant for this study were directly received by and discussed with A. Squintu (ECA&D). The homogenization of the ECA&D dataset was done using a quantile matching approach (Squintu et al., 2019). Such an automated procedure cannot match the quality of metadata-supported performances for individual stations as in Category 1. It is yet useful for large datasets with generally missing individual metadata (Squintu et al., 2019), which often would otherwise be biased by various unaccounted inhomogeneities. Be aware that ECA&D data are often blended (= added via synoptical messages) for recent months or years. Blended data have been included in the homogenization.
We checked all 67 station time series for cases of TN ≥ TX. That procedure helped us to find cases with F I G U R E 2 Increase of data availability over time within the study area (for TG, TX and TN) and the five regions (only TG) wrong 0.0 C values, partly mixed TN and TX series and other problems. Many of the cases could be solved by consultation with data providers or A. Squintu in case of the homogenized ECA&D dataset, yet a few remain for 27 stations. They should not affect any evaluation of this paper, because only a few cases per station normally occur and for all but one station such cases form (much) less than 0.1% of all data (exception of Palermo with 302 cases = 0.5%; Supporting Information Data S1). Another helpful quality check was a visual inspection of the annual cycle of daily maximum and minimum values, which additionally helped to find outliers and wrong zero cases, which could be corrected in consultation with the data providers or were deleted. Additional homogeneity information for individual stations can be found in the last column of Supporting Information Data S1.

| Temperature indices
The variability of (summer) heat conditions is explored via temperature indices described in Table 2. They mainly derive from publications of the World Meteorological Organisation (WMO; Klein Tank et al., 2009) and the ECA&D project (ECA&D, 2013). Percentile-based thresholds support a regionally comparable evaluation of seasonal characteristics, independent of climatic characteristics based on the geographical location or local specifics. Fixed thresholds impede comparisons in regions characterized by diverse climatic conditions (Shevchenko et al., 2014;Lhotka and Kyselý, 2015a), yet help understanding regional climatic peculiarities (Kyselý, 2002(Kyselý, , 2010. Therewith, the gradual change from cool and/or maritime conditions in Western and Northern Europe to the warm and/or continental climate of Southern and Eastern Europe gets visible. We use percentile-based indices as summation of all temperatures over the particular thresholds, while the days of occurrence are used for fixed thresholds. Heat wave characteristics are explored using percentiles of the local summer temperature distribution. Compared to other approaches (see summary in Lhotka and Kyselý, 2015a), typically using a three-day-onset with temperatures over a certain threshold, we decided for a slightly 'softer' onset, as we accepted a 3-day-average over

| Classifications of atmospheric and oceanic circulation
Indices and classifications condense the manifold variations of atmospheric circulation conditions into straightforward schemes of index values and circulation classes, determined by similar atmospheric processes like comparable (a) geographical distributions and spatial movements of driving pressure areas (anticyclones, cyclones, troughs and ridges), (b) air mass inflow directions and (c) air mass attributes. We use two versions of the well-known Grosswetterlagen classification: (a) the original manual Grosswetterlagen classification (GWLc; Baur, 1947;Hess and Brezowsky, 1977) and (b) an automated version, the SynopVis Grosswetterlagen classification (SVGc, version 2019) developed by Paul James (personal communication; see also Hoy et al., 2013b). The classification scheme is chosen for its proven large-scale relevance within Europe (James, 2007;Huth, 2010). For an enlarged overview about (a) the historical development of the classification, (b) the composition, character and trends of circulation types, (c) the spatial response to sea level pressure and (d) temperature in Europe and adjacent areas see Hoy et al. (2013aHoy et al. ( , 2013bHoy et al. ( , 2013c. We use 10 major circulation types, based on air mass inflow into Central Europe and cyclonicity in this area (James, 2007;Werner and Gerstengarbe, 2010). Therewith, we characterize the 2018 SHY. We employ GWLc data derived from Werner and Gerstengarbe (2010), updated until 2018 by monthly publications of the German Weather Service (DWD, 2019b). SVGc data were provided by Paul James.
The North Atlantic Oscillation is the most prominent teleconnection impacting the study area. It describes the intensity of westerlies as related to the pressure difference between Icelandic Low and Azores High (Hurrell, 1995). The North Atlantic Oscillation index (NAOI) used here derives from Li and Wang (2003), updated via http://ljp. gcess.cn/dct/page/65610. This NAOI shall provide a more faithful representation of the spatial-temporal variability associated with the NAO in all seasons compared to other (all-year) NAOIs, which are often relevant for the northhemispheric winter half year only (Li and Wang, 2003).
Five other teleconnection indices derive from the CPC webpage (http://www.cpc.ncep.noaa.gov/data/ teledoc/telecontents.shtml; see details there) and have been used in a number of recent publications (e.g., Bueh and Nakamura, 2007;Moore and Renfrew, 2012;Rust et al., 2015). These are: The EA is structurally similar to the NAO, but its two pressure centres are shifted south-eastwards compared to the typical NAO dipole. EAWR consists of four main circulation anomaly centres, located (a) over Europe and Northern China and (b) of opposite sign over the Central North Atlantic and north of the Caspian Sea. WP is the primary mode of low-frequency variability over the North Pacific, but affects European climate as well. SCAND inheres a primary circulation centre over Scandinavia and weaker centres of opposite sign over Western Europe/Eastern Russia. The POLEUR is characterized by negative geopotential height anomalies over the polar region and positive ones over Northern China/Mongolia. The positive (negative) phases reflect a stronger (weaker) circumpolar vortex.
The Atlantic Multidecadal Oscillation (AMO) is defined as a long-term cycle in North Atlantic sea surface temperatures (SST) with positive (warmer) and negative (cooler) phases of approximately 30-40 years duration, presumably driven by the Atlantic Thermohaline Circulation decadal-scale oscillation (Knight et al., 2005). Such changes have been occurring for at least the past 1,000 years and are natural. Data used here derive from the Earth System Research Laboratory of the US National Oceanic and Atmospheric Administration (http://www. esrl.noaa.gov/psd/data/timeseries/AMO/) and are available from 1856. The time series are calculated from the Kaplan SST data set in 5 × 5 resolution (Kaplan et al., 1998, updated monthly). The weighted area average SST are calculated over the North Atlantic (0-70 N) and time series are de-trended (Enfield et al., 2001).

| Methods
Peculiarities of 2018 and a comparison to earlier temperature variability is illustrated via a) maps of the record years since start of the individual time series given in Table 1 and b) time series calculated from the averages of all five regions for the period 1855/1881-2018. This approach of using twofold study periods for the station and the regional data enables us to assess the extremity of the 2018 summer at individual stations for the longest available instrumental datasets, while additionally allowing for a regional estimation of its extremeness with a comparatively dense and high-quality data set for a common 164/138 year period. Data from 1855-1880 are shown in light colours to account for their lower station data coverage. From 1855, all regions comprise at least one location with TG, TN and TX data, while from 1881 all regions are covered by at least three stations. The time series of the study area average are supplemented by the top/bottom 5 years of the five European regions for the period 1881-2018 (symbols in the same illustrations) to account for regional peculiarities.
The presented teleconnection indices have been correlated (Pearson correlation coefficient) with time series of our temperature indices for the entire study area and the five regions for the period of data availability 1951-2018 for summer and additionally the previous spring and winter. Such correlation values may be affected by similarcausal or non-causaltrend behaviour. In order to focus our correlation analyses on yearto-year variability only, linear trends have been removed before calculating the correlation coefficients (detrending). Note that other kinds of trends, as well as coherent variations on decadal scales, may still be present.

| THE LONG, HOT EUROPEAN SUMMER OF 2018
3.1 | The summer of 2018 -An overview Persistent and often large positive anomalies from reference climatology  characterized the year 2018 in Europe, which (typically) were more pronounced for maximum than minimum temperatures (Figure 3). After a cold March, April and May were (by far) the warmest since start of observations in our station collective, both setting new monthly records at about 30% of all included stations (not shown). During April, positive anomalies were extraordinary high in region South-East with a monthly anomaly of +5.2 K (previously largest anomaly was 2000 with +3.4 K, not shown), but also record-high in large areas of Central and Southern Europe. During May, positive anomalies were more evenly distributed, covering most of Europe apart of the Mediterranean. The extraordinary warm start into the extended summer season resulted in a considerable number of days over certain thresholds, like summer days (SU), already early in summer. Temperature anomalies were less severe in June, but strongly elevated again in July. Here, the longest and most intense heat wave of this summer started in region North-East (Section 3.2), later also severely affecting Central and Western Europe. At the end of July, new all-time records of maximum and highest minimum temperatures were recorded especially in Scandinavia (Supporting Information Data S3). A different short, but extremely intense heat wave struck the Mediterranean (mainly its western parts) at the beginning of August (Section 3.2). September was very warm in wide-spread areas apart of its final days, with a high number of SU especially in Central Europe. Lastly, mid-October provided a furious final of the 2018 summer, with high maximum temperatures and wide-spread SU in large areas of Europe.
Not surprisingly, positive temperature anomalies of the 2018 SHY (April-September) were more pronounced than those of the summer season (June-August), with 2018 by far the warmest SHY in Europe (2.7 K compared to 1.9 K in 2003; Figure 4a), but 'only' second-warmest summer (2.5 K compared to 2.6 K in 2003; Supporting Information Data S4). 2018 was the warmest SHY in four regions and second-warmest in region South. Anomalies were largest in Central Europe, with an average regional anomaly of +3.4 K, with higher local values (+4.0 K in Largest anomalies were observed in Prerov and Frankfurt (+3.8 K, both Central). Figure 4b shows the SHY temperature anomalies of the gridded E-OBS data and our 67 climate stations, with a very good agreement among both datasets. The areas with the strongest positive anomalies extend from central France, Switzerland and Northern Italy towards Southern Fennoscandia, the South-Eastern Baltic Sea area, Belarus and the North-Western Ukraine. Weakest signals appear near the Atlantic coast line around the British Isles and the Iberian Peninsula. Spatial patterns for summer show a similar spatial picture (Supporting Information Data S4). Figure 4c illustrates the record years of the locally warmest SHY since start of observations for our 67 stations. 2018 finished warmest at half (33)  Hot days with maximum temperatures above 30 C (HD) did not break as many records as the previously discussed two indices, connected to less intense heat than other record years in southern and central latitudes and because of their general rarity in Northern Europe. They reached new maxima at 10 stations in Scandinavia and Germany, with 8 days over the previous record in Stockholm (18 days, since 1859, North-East) and 7 in Berlin (28 days, since 1876, Central). 2018 scores third among the years with maximum number of hot days, being second in region Central, third in region North-East and fourth in region South.
Temperature sums of maximum temperatures over certain percentile-based thresholds (TX 90 and TX 99 ) reached new maxima at a number of stations in Scandinavia and Germany (Figure 5a,b). The new record values were regionally extreme, like for TX 90 in Stockholm (136 instead of 88 K, since 1859, North-East), Copenhagen (106 instead of 68 K, since 1874, Central) and Oslo (159 instead of 113 K, since 1937, North-East; Supporting Information Data S3). Within Europe, 2018 scored third for TX 90 and sixth for TX 99 , but largest for TX 90 in region North-East and Central, and second-largest in region North-East for TX 99 , (Figure 6a,b). In a long-term perspective, 3 of 4 (TX 90 ) and 2 of 3 (TX 99 ) of all extremes were recorded after the year 2000 (Figure 5a,b).
The three indices of high minimum temperatures used here (TNi, TN  Two approaches of heat wave definition (see Section 2.2) are used, one with high (over the 90th percentile; HW 90 ) and another with very high (over the 95th percentile; HW 95 ) threshold values. A very intense heat wave developed in the latter two decades of July and the first of August, spatially comprising large areas around the Baltic Sea region, where it was the most intensive heat wave since start of observations at 10 (HW 90 ) or 5 (HW 95 ) stations (Supporting Information Data S3). That heat wave manifested itself especially strongly at the very reliable and long-term observatory of Stockholm (TX data from 1859, North-East). On basis of HW 95 it was by farthe most intense (65 compared to 35 K in 1975) and longest (22 instead of 11 days) heat wave ever recorded there (Figure 7a) A separate, spatially and temporally much smaller, but extremely intense heat wave developed around the Western Iberian Peninsula in the first days of August. It was the most intense heat wave ever recorded in Lisbon, despite its length of only 6 days for both HW 90 and HW 95 . During its peak at fourth and fifth of August it topped all previous records of daily maximum TX, TG and TN at this station by a margin of 1 K or more (data since 1901; TX: 43.3 C compared to previous maximum of 41.8 C). Figure 7b shows the evolution of heat waves in Lisbon for HW 95 , with a strong increase in heat wave intensity (but not length) getting visible from the 1990s.
For all stations a list of the 2018 and (previously) highest values of all employed indices (Table 2), as well as the (current) record years, is provided in Supporting Information Data S3.

| Grosswetterlagen
The Grosswetterlagen classification was developed for Central Europe (Germany), but has a considerable largerscale relevance within Europe (Section 2.3). Comparing the number of certain patterns of the 2018 SHY to the average over the period 1951-2018, patterns with continental inflow from eastern and southern directions (NE, E, SE, S; GWLc 40% instead of 27%; SVGc 47% instead of 33%) as well as anticyclones centred over Central Europe (GWLc: 19% instead of 16%; SVGc: 21% instead of 15%) were prevailing. During the summer months (as a combined effect), they are typically associated with mainly warm and rather dry weather, especially towards regions North-East, West and Central (Hoy et al., 2013c). Considerably less situations with westerly to northerly air mass inflow into Central Europe (W, NW, N; GWLc 33% instead of 47%; SVGc 26% instead of F I G U R E 7 Heat wave evolution of the most intense annual HW 95 for (a) Stockholm (left) and (b) Lisbon (right; be aware that heat waves do not occur every year) 42%) occurred during the SHY. They typically bring rather cold and humid air masses into most parts of Europeagain with focus on previously mentioned regions (Hoy et al., 2013c). Additionally, during 2018 more days than usually were classified as anticyclonic in Central Europe (GWLc: 51% instead of 47%; SVGc: 62% instead of 48%), which during the SHY typically supports warm, dry and sunny conditions.

| Teleconnection indices
In the east of Europe, a positive EA during summer is associated to anticyclonic conditions and southerly inflow (Rust et al., 2015), supporting positive temperature anomalies there. In 2018, the EA index was third highest since 1951 (after 2017 and 2016) during both summer and SHY, with a new maximum value in July, where the strongest heat wave of this summer was observed. Correlation coefficients show that above-average EA values during spring and summer are related to warmer European summers, their relevance increasing from west (no signal) to east (strong signal; Supporting Information Data S5). Accordingly, the rather chilly summers of the 1961-1990 decade were accompanied by low EA summer values (average of −0.6), while the hot summers after the millennium occurred during high EA values (average of +0.7; Figure 8). In fact, the 10 warmest summers within the study region all occurred between 2002 and 2018.
Northerly air mass inflow towards Western Russia dominates the positive phase of the EAWR during summer (Rust et al., 2015), consequently leading to a negative relation with warm summers/hot temperature extremes in the very east of Europe (in our study this relation is mainly visible in region South-East; Supporting Information Data S5). The positive phase of WP is characterized by low pressure over most of study area during summer (CPC, 2012), hence supporting cooler European summers. During the summer of 2018 index values of both indices were low (WP third lowest together with 2002). Accordingly, the indices of both EAWR and WP during the cool summers of 1961-1990 were high (+0.6), while they were pretty low after the year 2000 (−0.4; Figure 8).
The positive phase of SCAND relates to high pressure and warmer summer temperatures over Scandinavia (Rust et al., 2015), which gets apparent in region North-East (Supporting Information Data S5). Stronger negative relations yet appear in the south, where low pressure is prevailing (regions South-East and South). The 2018 index values were unremarkable, but May and July, the months with the largest positive temperature anomalies in region North-East, displayed record high index values (in agreement with the very stable Scandinavian anticyclone during that time; WWA, 2018). Both months were very warm in Scandinavia and region North-East, but rather unremarkable in region South and (only July) South-East.
There is no explanatory power of the used NAOI related to warm summers in Europe (Supporting Information Data S5), for region Central in agreement with Kyselý (2002), who found only weak relations between a different NAOI and Prague summer mean and cumulated extreme temperatures. The only exception is a tendency towards less common high temperature extremes during the following summer in region North-East with prevailing positive winter NAOI values, possibly related to abundant precipitation and snow cover during winter in this region, which may impede early soil drying during spring and early summer. Also, relations between temperature indices and POLEUR are weak. Exception is, again, a tendency towards lower summer temperatures in region North-East with positive index values in spring and especially winter, which cannot be explained here.

| Atlantic multidecadal oscillation
We find a strong positive correlation of the European summer temperatures since the early 20th century with the AMO time series for our study area, which is specifically visible when smoothened over a number of years (a smoothing of 11 years results in a correlation of >0.5; from 1901 in >0.8; Figure 9a). Relations in early decades are weak, which considerably reduces the calculated correlations. The influence of time lags on the magnitude of correlations was checked on a yearly scale for various smoothing values. Results confirm the assumed multidecadal character of the AMO, with high correlation values of ≥0.5 (for 11-year smoothing) between the respective current summer temperatures with AMO values up to at least 10 years in the future. Correlations  Figure 10. Strongest signatures of the mid-summer heat wave appeared around Sweden, where all Swedish and nearby stations show strong new maxima for HW 90 . The spatial distribution of index records confirms that the region with the most unusual temperatures was centred over the South-Western Scandinavian Peninsula, with strong positive anomalies extending in all directions around the Baltic Sea ( Figure 10)supported by distinct and long-lasting anticyclonic conditions in this area (WWA, 2018).
Examining the continental/regional perspective since availability of such data, almost all indices show two maxima directly after 1855 (not robust as possibly biased by low data availability and limited data quality, respectively; see Sections 2.1 and 2.4) and around the 1940s. Yet, the high levels since the 1990s, which are further accelerated since the hot summer of 2003, are striking and unprecedented (Figures 4a and 6; Russo et al., 2015;Christidis et al., 2015;Dong et al., 2017;Manning et al., 2019). The last year within the regional low TOP5 (cold extremes) of any index is therefore the year 1980, while a large majority of all regional high TOP5 (warm extremes) of any index is found from 1992 onwards. That signal is specifically strong for high percentiles of minimum temperatures (TN 90 and TN 99 ), where nearly all regional TOP5 occurred after 1992.
Heat waves show similar temporal maxima like discussed before (after 1855, around 1940s Here, the year 2015 gets strongly visible, which was characterized by a number of very intense, but mainly short heat waves (Hoy et al., 2017).

| Atmospheric and oceanic circulation and hot summers
During the SHY 2018 both manual (GWLc) and automated (SVGc) Grosswetterlagen classifications often show the same (51% of all days) or related circulation patterns at a certain day, sometimes lagged by 1-2 days. The high level of similarity is remarkable, considering general difficulties to reproduce manual (subjective) classifications by automated methods, as well as methodical differences. We showed that the long and pronounced heat conditions of the 2018 SHY were supported by a combination of favourable circulation facts. Such conclusions are also confirmed by a study of Kyselý (2002) for Prague, who found that anticyclonic conditions and inflow from easterly to south-westerly direction (major types E, SE, S and SW) were responsible for 75% of all heat waves there. Within (2018 mostly affected) regions North-East and Central especially the very stable anticyclone over Scandinavia (WWA, 2018;Kornhuber et al., 2019) was responsible.
Earlier studies relate a significant part of summer temperature variability in Europe to the AMO (Sutton and Hodson, 2005;Sutton and Dong, 2012). Knight et al. (2006) show a strong relation between observed and simulated Central England temperatures (Parker and Horton, 2005) and the AMO for summer (and autumn). Kyselý (2010) found the positive/negative phase of the AMO almost perfectly corresponding to increased/ decreased heat wave severity in Prague and Della-Marta et al. (2007) that the AMO is a possible predictor of the heat wave frequency over Western Europe at decadal time scales. Dong et al. (2017) allocate almost two thirds of the seasonal mean warming in Western European summers since the mid-1990s to changes in SST, while changes in greenhouse gases and anthropogenic aerosols form the remaining third. As the occurrence of warm and cold periods is strongly linked between our five study regions at decadal scales, changes in Atlantic SST have impacts to regions further east as well. Knudsen et al. (2011) show that the current known mechanism of 55-70 year oscillations existed during most of the past 8.000 years. Hence, considering the AMO-effect on European summer temperatures alone, another 10-15 years of enhanced summer temperatures could follow in Europe, in case the current positive AMO phase will be as long as the previous. Afterwards, an AMO-related summer cooling could offset some of the anthropogenic effects in following decades. Such a transition may occur quite rapidly, like observed during the last transition from a warm to a cool AMO phase (Thompson et al., 2010). Yet, despite the striking link between the AMO and European summer temperature variability since the 20th century such forecasts are speculative, since the physical mechanisms of the AMO and its links to European summer temperatures are not adequately understood. In addition, correlation values before the 20th century are much lower for unclear reasons, possibly related to inaccuracies in early measurements or unknown differences in the underlying physical mechanisms. In that context, recent research suggests a much smaller role of internal ocean dynamics than previously assumed, arguing that the AMO largely follows external forcing (i.e., anthropogenic warming and aerosol changes), and therefore may not be used as a predictor of future summer conditions (Haustein et al., 2019). In any case, any decadal outlook of future summer temperatures is limited by the fact that the interplay of major teleconnection patterns and the role of oceanic circulation are just among a broad band of other contributors to the future of European summers.

| Seasonal supporting factors and anthropogenic effects on hot summers
This section summarizes the (currently known) additional natural drivers and the effects of global warming on occurrence and peculiarities of hot European summers. A positive feedback cycle of soil dryness, evaporation, F I G U R E 1 0 Graphical summary of number and characteristics of new record values since start of respective observations at our 67 stations precipitation and temperature contributes to summer heat waves in Europe (Fischer et al., 2007;Vautard et al., 2007;Haarsma et al., 2009;Quesada et al., 2012). Soil moisture depletion directly in summer or previously during winter and spring, caused by precipitation deficits and/or warm temperatures, leads to reduced evaporation and latent cooling, causing a reduction in cloud cover, resulting in increasing solar energy at the surface and further soil drying, causing higher summer temperatures and so on (Seneviratne et al., 2010). Dry (wet) Mediterranean winter and spring seasons cause a high (low) frequency of anomalously hot days during summer, often affecting other regions as well (Fischer et al., 2007;Quesada et al., 2012). Largest impacts are assumed for daily TX (Whan et al., 2015). According to Fischer et al. (2007), moisturetemperature-interactions account for 50-80% of hot summer days in Europe. Soil desiccation in combination with atmospheric heat accumulation is believed to have played a relevant role for both the 2003 Western European as well as the 2010 Eastern European/Russian mega heat waves (Miralles et al., 2014) and likely strongly contributed to the 2018 heat extremes as well.
Global warming plays indeed a major role in the observed and projected strong increase of heat conditions in Europe (IPCC, 2013;Christidis et al., 2015;Haustein et al., 2019). Continuously reduced aerosol emissions in combination with a reduction in cloudiness ('solar brightening') led to additional (specifically TX) temperature rises by increasing solar radiations at the surface since the 1980s (Wild, 2009;Tang et al., 2012;Dong et al., 2017). Further explanations reach above the background warming in Europe itself and include changes in atmospheric circulation patterns and blocking situations. A weakening atmospheric summer circulation over Europe is believed to contribute to more persistent and intensive heat waves, through a reduction in mean zonal wind speeds and an increase in quasi-stationary Rossby-waves because of the reduction in the Arctic-Tropics temperature gradient (Coumou et al., 2015;Mann et al., 2018). All aspects combined, researchers of the 'world weather attribution project' conclude that the probability of the 2018 summer heat wave in Scandinavia increased more than twofold compared to conditions without human contribution (WWA, 2018).

| SUMMARY AND CONCLUSIONS
We investigated the thermal characteristics of the 2018 summer, using a spatially well-distributed dataset of 67 European stations (west of~30 E) with daily air temperature observations mainly starting already in the 19th or rarely even 18th century. Compiling such a long-term (secular), quality-optimized (homogenized) and complete dataset (low number of gaps) enabled us to robustly assess the extremeness of the 2018 summer in a secular time-perspective. Individual time series length was considered for analysing station extremes and creating continental maps, while European time series were presented for the period of 1855-2018 and records of five subregions considered from 1881 onwards.
The extreme long duration of the 2018 summer most clearly manifested itself in pronounced new continental maxima of SHY (April-September) temperature averages and the number of days with maxima over 20 C (WD) and 25 C (SU). Those three indices also reached new local maxima at more than half of our investigated stations. Records of other temperature indices, rather representing intense heat conditions, were particularly broken in the extended Baltic Sea region. Our data show that the accumulation of hot European summers since the 1990s is unprecedented within the instrumental past, in all our investigated five regions. Considering our 12 analysed indices, the extreme summer of 2003 still dominates the ranking of most indices, but it is now generally closely followed by 2018 (Table 3).
We show that a combination of long-term variations of teleconnection indices and the AMO, as well as seasonal and anthropogenic effects, supported both the recent hot summers and the extreme summer of 2018. Recent decades saw positive index values of EA and negative of EAWR and WP teleconnections. The AMO is in its positive (warm) phase since the 1990s. Global warming in combination with reduced aerosol emissions fostered higher temperatures. Feedback mechanisms between (increasing) soil dryness in the Mediterranean affect other regions as well via the impact of atmospheric circulation. The combination of all mentioned factors strongly supported the observed summer warming in Europe since the 1990s.
Finally, the observed and projected developments described in this paper point to continuing or rather increasing heat conditions in near-future European summers. The role of the AMO may introduce some uncertainties in the decades from 2030, possibly offsetting some of the projected warming. Yet, global warming will further accelerate. In any case, additional negative health impacts are expectable with continuing summer warming, spatially increasing from comparably low influences in northern towards very high impacts in southern Europe (Gasparrini et al., 2017). Hence, a continuous monitoring of actual summer conditions and investigations of extraordinary events like 2018 is therefore necessary to (a) assess newly appearing extremes, (b) understand the underlying (new or known) mechanisms and (c) improve the interpretation of climate model outputs based on observed events.

ACKNOWLEDGEMENTS
A very special thank you goes to all the dedicated people at universities, national weather services and other institutions for their priceless efforts in data collection, digitalisation, processing and optimizing. Nowhere else is such abundant knowledge of metadata and station specifics gathered than locally. We also thank Antonello Squintu for his cooperativeness in sharing, discussing and optimizing the homogenized version of the ECA&D station data set. Finally, we thank Werner Partschefeld and Christoph Brendel for their technical support connected to data processing efforts.   WD  SU  HD  TX 90  TX 99  TN  TN 90  TN 99