Climate Dynamics https://doi.org/10.1007/s00382-023-06828-1 Analysis of synoptic weather patterns of heatwave events Sergi Ventura1 · Josep Ramon Miró2 · Juan Carlos Peña2 · Gara Villalba1,3 Received: 1 February 2023 / Accepted: 11 May 2023 © The Author(s) 2023 Abstract Heatwaves (HWs) are expected to increase both in duration and intensity in the next decades, but little is known about their synoptic and mesoscalar behavior, which is especially important in mid-latitude regions. Most climate research has focused on temperature analysis to characterize HWs. We propose that a combination of temperature and synoptic patterns is a better way to define and understand HWs because including atmospheric circulation patterns provides information about different HW structures that can irregularly affect the territory, and illustrate this approach at the regional and urban scales using the Iberian Peninsula and the Metropolitan Area of Barcelona as case studies. We first select HW events from 1950 to 2020 and apply a multivariate analysis to identify synoptic patterns based on mean sea level pressure, geopotential height at 500 hPa, and maximum daily 2 m temperature. The results indicate that four synoptic patterns reproduce at least 50% of the variance in HWs, namely, “stationary and stable”, “dynamic and advective”, “stationary and advective”, and “dynamic, advective and undulated”. Next, we apply the analysis to the Representative Concentration Pathway future scenarios (RCPs) 4.5 and 8.5 from the Coordinated Regional Climate Downscaling Experiment (CORDEX) to determine how these synoptic trends can change in the future. The analysis shows that the four synoptic patterns continue to explain 55 to 60% of the variance in HWs. Future HW events will be characterized by an increase in geopotential height at 500 hPa due to the northward shift of the anticyclonic ridge. This is especially true for RCP8.5, which simulates business as usual incrementing fossil fuel use and additionally shows an increase in atmospheric dynamism in north advections from all directions in comparison with RCP4.5. These findings point to the importance of considering the geopotential height in HW prediction, as well as the direction of advections. Keywords CORDEX · ERA5 reanalysis · Multivariate analysis · Climatic trends · Heatwaves 1 Introduction next 5 years (WMO 2021). This global warming is mainly attributed to the increase in greenhouse gas (GHG) con- According to the World Meteorological Organization, the centrations in the atmosphere, which mostly (78%) results annual mean global temperature is likely to rise between from fossil fuel burning and industrial processes (Blanco 0.9 and 1.8 °C above preindustrial levels (1850–1900) in the et al. 2014). The negative effects of global warming are not equally distributed but depend on geography and weather patterns (D’ippoliti et al. 2010). According to the Medi- * Gara Villalba terranean Experts on Climate and Environmental Change gara.villalba@uab.cat (MedECC), the Mediterranean region is warming 20% faster 1 Sostenipra Research Group (SGR 01412), Institute than the global average (MedECC 2020). The European of Environmental Science and Technology climate depends on mid-latitude atmospheric circulation, (MDM‑2015‑0552), Z Building, Universitat Autònoma which is mainly controlled by westerly flows from the Atlan- de Barcelona (UAB), Campus UAB, 08193 Bellaterra, Barcelona, Spain tic Ocean (Ozturk et al. 2021). The Mediterranean climate is 2 temperate with a dry summer season but suffers significant Department of Territory and Sustainability, Meteorological Service of Catalonia, Generalitat de Catalunya, Barcelona, variability due to the transition between cold mild latitudes Spain and the tropics, generating notorious circulation changes. 3 Department of Chemical, Biological and Environmental Wind flow anomalies in the upper troposphere and the half- Engineering, Universitat Autònoma de Barcelona (UAB), degree dip in the sea‒land temperature gradient may be the Campus UAB, 08193 Bellaterra, Barcelona, Spain 13 Vol.:(0123456789) S. Ventura et al. main causes of this atmospheric hotspot (Tuel and Eltahir increase in duration (by a factor of 1.4–2.0), frequency (by 2020). The Third Report on Climate Change in Catalonia a factor of 2.2–4.5) and number of HW days per year (by a (Government of Catalonia 2017) predicts temperature incre- factor of 3–7) in Europe. To simulate and interpret future ments ranging from + 1.1 to + 2.5 °C for 2031–2050 sum- predictions of HWs, some studies have used reanalysis data- mers in comparison to 1971–2000, which is higher than the sets (Bengtsson et al. 2004; Engdaw et al. 2021; Santer et al. annual increment, which ranges from + 1.0 to + 2.2 °C in 2004) together with climate models to statistically find cli- this region. mate trends focused on percentiles of temperature related In addition to an overall temperature increase, the con- to HWs. However, focusing solely on temperatures is insuf- sequences of global warming are also expected to include ficient to determine climate trends characterizing HWs; syn- an increase in the intensity and duration of heatwaves optic structures are also needed. The mechanisms that con- (HWs), a trend that has already been seen in recent years tribute to the formation of HWs do not occur independently (Perkins-Kirkpatrick & Lewis 2020). There are multiple of circulation conditions, and some configurations are more ways to define an HW. The World Meteorological Organi- likely to produce extremely warm periods (Sfîcă et al. 2017). zation (WMO 2021) considers an HW as a five-day episode An analysis of the behavior of synoptic weather patterns with temperatures 5 °C higher than the maximum mean (SWPs) could improve our understanding of HWs because of May–September calculated from the reference period including atmospheric circulation patterns provides informa- (1971–2000). (Peña et al. 2015) defined an HW as a period tion about the structures that generate these events. This bet- during which the 95th percentile of summer temperatures is ter understanding could improve HW forecasts, which could reached during at least three consecutive days. Such anoma- help alert authorities and decrease the negative effects of lous prolonged periods of excessive heat can cause wildfires extremely high temperatures (Della-Marta et al. 2007). Pre- and severe negative impacts on human health, agriculture vious studies have analyzed synoptic patterns to understand and nature (de Rigo et al. 2017; McMichael & Lindgren how extreme climate events are generated, such as wildfires 2011; Turco et al. 2014). For example, the HW of July 2003 (García-Ortega et al. 2011), Saharan dust intrusions (Díaz in Europe culminated in 30,000 deaths (14,800 deaths in et al. 2017), and heatwaves (Sousa et al. 2019).The latter France) (Bouchama 2004). shows that, in August 2018 and June 2019, a cyclonic cir- The effects of global warming and HWs are further exac- culation in the northeastern Atlantic and a subtropical ridge erbated in cities, which additionally suffer the urban heat pattern over the Iberian Peninsula advected an anomalously island (UHI) effect due to the heat accumulation of build- warm air mass which generated a HW episode. Other stud- ing materials and human activity (Liu et al. 2018; Morris ies have related HW events and synoptic patterns, such as et al. 2017). A UHI is defined as the temperature difference Choi et al. (2021) who found a positive linear relationship between the urban center and the rural surroundings due to between HW events and the synoptic stagnation index which the property alteration of the atmospheric boundary layer is an indicator of atmospheric stability and clear sky days. (Streutker 2002; Segura et al. 2021), including turbulence Another Asian study (An and Zuo 2021) found that regional (Grimmond and Oke 1999) and moisture (Hoffmann et al. dry heatwaves tend to happen when there is a high-pressure 2011). During HW episodes, UHIs can raise temperatures ridge situated to the northwest of North China and when by 0.5 °C in comparison with surrounding areas and by 2 °C the northern edge of the western North Pacific Subtropi- at night (Basara et al. 2010), which can lead to heat stress cal high is south of 30° N. Such synoptic behavior analysis (Guarino et al. 2014; López-Bueno et al. 2021). Further- is key in understanding heatwave generation. In this study more, an increase in temperature can increase the demand we aim to add to this knowledge by providing a long his- for energy and water for cooling, which is accompanied by torical statistical analysis to rigorously characterize HWs an increase in pollutants (Santamouris 2014). Currently, cit- using temperature and synoptic structure in a way that can ies concentrate more than half of the world´s population, and be applied to future climate predictions provided by regional by 2035, they are expected to hold 62.5% of the world popu- climate models (RCM). lation and 85% of the population in high-income countries To define the most representative SWPs related to HWs, (United Nations 2020). As cities plan adaptive and mitiga- we classify HW episodes based on the mean sea level pres- tion strategies for such events (Gilabert et al. 2021; Segura sure (MSLP) and geopotential height at 500 hPa (Z500) et al. 2022), a rigorous understanding of potential future using principal sequence pattern analysis (PSPA) (Peña et al. scenarios of HW episodes is highly relevant and necessary 2011), which is a variant of principal component analysis (United Nations 2016). (PCA) set in T-mode (correlation between fields) instead Future predictions of HW episodes state that extreme of S-mode (correlation between temporal series). PCA has episodes such as 2003 could occur every 15 years in the the advantages of reducing and interpreting massive data- 2020–2049 period (Barriopedro et al. 2011). Furthermore, sets while simultaneously minimizing information loss. according to (Lau and Nath 2014), HWs are expected to This method finds the most correlated input variables that 13 Analysis of synoptic weather patterns of heatwave events represent the highest number of total variances possible and Section 4 applies the historic analysis to future climate sce- generates new variables, which are linear functions of those narios to analyze possible trends in the atmospheric struc- in the original dataset. (Hotelling 1933; Pearson 1901) uti- ture. Finally, Sect. 5 presents a summary and conclusions. lized the first references of this method, which was not well known until computers obtained more computational power (Jolliffe et al. 2016). More recently, PCA has been applied in 2 Area of study and data different fields, such as urban traffic and meteorological data (Shiva Nagendra et al. 2003) or climate change assessment 2.1 Area of study (Tadić et al. 2019). With the overall objective of understanding HW devel- The region of interest is the AMB, located in the northeast- opment in terms of atmospheric structure and its effects in ern Iberian Peninsula (41–42° N/1.5–2.5° E), as shown in an urban area, we use the Metropolitan Area of Barcelona Fig. 1. (hereafter referred to as AMB) as a case study to develop a The climate is Mediterranean, characterized by warm and method to classify synoptic behavior in terms of HW events. dry summers influenced by sea breezes that regulate tem- We first classify past HW events using historical ERA5 rea- peratures. According to the Fabra Weather Station (Fabra nalysis data (1951–2020) and then analyze various simula- for short), the mean maximum and minimum temperatures tions of future climate scenarios (2011–2100) provided by in summer months (June, July and August: hereafter referred the Coordinated Regional Climate Downscaling Experiment to as JJA) are 27.7 °C and 18.8 °C, respectively, with a mean (CORDEX) to determine the SWPs that give rise to HW summer precipitation of 96 mm (15.3% of annual total, episodes. We attempt to answer two questions: (1) What which is 625 mm). HWs in Barcelona can reach values are the synoptic structures that give rise to HWs? (2) How of + 39 °C and are exacerbated by the high values of rela- well do models forecast the synoptic structures associated tive humidity generated by the Mediterranean Sea (summer with HWs? mean of 71%), causing a situation of thermal stress for the This article is organized as follows: Sect. 2 describes the inhabitants of Barcelona. area of study and data, including the datasets used for the The topography of the area is heterogeneous, with study, the typical weather, land use and geography. Section 3 a coastal mountain range 10 km from the sea reach- is dedicated to the methods, which include the selection of ing + 650 m.a.s.l. (meters above sea level), two important HW days, the PSPA, and the creation of mean maximum rivers, and the delta of the Llobregat River, which covers temperature at 2 m (TMAX) maps in relation to SWP trends. 98 ­km2. This region covers an area of 636 ­km2 and has a Fig. 1  On the left, the selected area from ERA5 and CORDEX simu- palities and two main rivers, as well as the location of the CORDEX lations, showing the surface altitude. The Metropolitan Area of Bar- pixel, coinciding with the meteorological station of Fabra Observa- celona (AMB) is marked in red. On the right, the topography of the tory, used for extracting the 95th percentile of the CORDEX datasets AMB, including altitude, urban fraction, the names of several munici- 13 S. Ventura et al. population of more than 3,300,000 inhabitants. From this ns-s​ ingle-l​ evels?t​ ab=f​ orm). The RCPs are pollutant concen- area, 48% is urbanized, while the rest is occupied by more tration pathways used in the Fifth Assessment Report of the than 250 ­km2 of green area. Intergovernmental Panel on Climate Change (IPCC). The values of the RCPs refer to the radiative forcing that would 2.2 Data occur due to the increase in anthropogenic emissions by 2100 (IPCC 2014). Thus, RCP4.5 is described as the most In this study, we used hourly observational data of 2 m likely scenario, with greenhouse gas emissions peaking in temperature available from the weather station of the Fabra 2040 followed by a decline. RCP8.5 is the worst-case sce- Observatory (41° 25′ 06″ N, 2° 07′ 27″ E, 415 m.a.s.l.) from nario in which emissions continue to rise throughout the 1951 to 2020. This database can be downloaded at https://​ twenty-first century. In this work, three RCM have been used apido​cs.​meteo​cat.​gencat.​cat/. We filtered this data selecting (Table 1) at 0.1° resolution to perform the analysis. the summer months (JJA) and it was used to find the HW days. The HW episodes are selected based on the definition of the Catalan Meteorological Service (SMC) adapted by 3 Methods (Peña et al. 2015), in which an HW is an episode of three or more consecutive days that reaches the 95th percentile of the The methods employed to classify HWs into synoptic and maximum summer temperatures. In addition to the obser- atmospheric structures consist of the three steps outlined in vational data from the SMC, we used the Spain02 dataset Fig. 2: (1) the classification of all the JJA days by synoptic providing daily temperature and precipitation from 1951 to weather types; (2) the extraction of HW days from observed 2020 and covering Spain. This dataset is downloaded in a data using the 95th percentile of maximum temperatures in regular grid of 5 km from Spain02 interpolated products the summer months; and (3) the application of the statisti- (data source: https://​www.​aemet.​es/​es/​servi​ciosc​limat​icos). cal PSPA to the HW days extracted in the previous step to Mean maximum temperature maps were used to support the reduce the dataset, improve the interpretability and find the results. SWP associated with HWs. In the following paragraphs, we Reanalysis gridded data at a 0.25° resolution from ERA5 describe each of these steps and the datasets used (all pub- was used for the historical period (1951–2020) analysis. This licly available) in further detail. data was necessary for the methods described in Sects. 3.1 and 3.2 and was downloaded from the Copernicus Climate 3.1 Analyzing synoptic weather types and trends Data Store available at https://​cds.​clima​te.​coper ​nicus.​eu/​ cdsapp#​!/​home. The region used for this analysis covers The JJA days from 1951 to 2020 are classified according to from 35° N, 14° W to 50° N, 6° E, as represented in Fig. 1. an objective synoptic classification appropriate in Mediterra- EURO-CORDEX data was used for the whole period nean climates as described by (Miró et al. 2020) and shown (1951–2100), considering two representative concentration in Table 2. This method is based on the classic Jenkin- pathways (RCPs) 4.5 and 8.5. The domain of the EURO- son–Collison classification (Jenkinson and Collison 1977) CORDEX simulations includes the entire European conti- but with an additional geopotential level of Z500 to account nent, from North Africa and the Atlantic in the south and for the important effect of mid-troposphere mechanisms in west, to the west of Russia and Turkey in the south and Mediterranean weather (Tuel and Eltahir 2020; Ward 1963), east (official source: https://​cordex.​org/​domai​ns/​cordex-​ resulting in 13 synoptic weather types for the northeastern region-​euro-​cordex/). The left hand panel of the Fig. 1 Iberian Peninsula. shows the selected area for the statistical analysis of the We used the Mann–Kendall test (Mann 1945; Kendall synoptic weather patterns over the Iberian Peninsula and 1975) to classify all summer days between 1951 and 2000 the AMB. The simulations of the historical period and both into the 13 synoptic weather types. This technique has scenarios are available from Copernicus (https://​cds.​clima​ been widely used for detecting trends in hydrometeoro- te.c​ opern​ icus.e​ u/c​ dsapp#!​ /d​ atase​ t/p​ rojec​ tions-c​ ordex-d​ omai​ logical series (Liu et al. 2013; Zhang et al. 2000). It is a Table 1  Description of the CMIP5 GCM Resolution GCM (º) EURO-CORDEX RCM Resolu- Institution (country) CORDEX data used in this tion RCM study (o) IPSL-CM5A-LR 1.9 × 3.75 WRF381P 0.11 IPSL (France) MPI-M-MPI-ESM-LR 1.9 × 1.9 REMO2009 0.11 MPI-CSC (Germany) MOHC-HadGEM2-ES 1.9 × 1.25 HIRHAM5 0.11 DMI (Denmark) 13 Analysis of synoptic weather patterns of heatwave events Fig. 2  Graphical workflow: Steps 1, 2, and 3 are defined at the top step. The upper section (in blue) corresponds to the historical analy- of the image. Rhomboids represent publicly available datasets needed sis based on 1951 through 2000, whereas in parallel but below and to apply methods (rectangles) at each step. Triangles represent ques- in red, the methods are applied to future (2011–2100) simulations tions, and ovals include the dataset or analysis resulting from each obtained from CORDEX for future climate scenarios RCP 4.5 and 8.5 Table 2  Synoptic weather types with a brief description as defined by or decreasing trend in the time series data. In this work, (Jenkinson & Collison 1977) and the frequency of HW events during the software provided by (Hussain and Mahmud 2019) is the period 1951–2020 in the AMB applied to generate the Mann–Kendall trends from the syn- Synoptic Description % Frequency optic weather types. If the output value has a significant weather type (1951–2020) trend, the Tau–Kendall value is used to check if it is an Type01 West advection 0 increasing (τ > 0) or a descending trend (τ < 0). However, Type02 Anticyclonic western advection 11.6 studies that consider a set of statistical inferences simulta- Type03 Northwest advection 0.5 neously could face the problem of multiple testing (Streiner Type04 North advection 1.1 and Geoffrey 2011). For this, we adjusted the p value using Type05 Northeast advection 2.1 the Holm-Bonferroni sequential correction (Holm 1979), Type06 East advection 2.1 which has been proved with better results than the Bonfer- Type07 East advection with cutoff low above 2.7 roni procedure (Aickin and Gensler 1996). In this paper, we Type08 South advection 0 calculated the adjusted p values using a Microsoft Excel Type09 Southwest advection 5.8 online calculator developed by (Gaetano 2013). Type10 Trough 9 The results of these methods can be categorized into Type11 Low or cyclone 13.2 four possibilities: a nonsignificant increasing trend (NSIT), Type12 Shallow cyclone or undetermined pres- 49.7 which shows an increasing trend that is not statistically sig- sure gradient nificant; a nonsignificant decreasing trend (NSDT), which Type13 Anticyclone 2.1 shows a decreasing trend that is not statistically significant; a significant increasing trend (SIT), which is statistically significant with a trend to increase; and a significant decreas- nonparametric method that is not influenced by extreme val- ing trend (SDT), which is statistically significant and has a ues due to its robustness to outliers in time series data. The trend to decrease. presence of a monotonic trend in the series has been esti- mated using the tau statistic at the 95% level (p value < 0.05). 3.2 Selection of heat wave episodes The Mann–Kendall test checks the null hypothesis, which indicates that there is no significant trend (p value > 0.05). The HW episodes can be selected following multiple defi- If this cannot be verified (p value < 0.05), the alterna- nitions, since there is not a universal definition for it. In tive hypothesis is accepted, which indicates an increasing this article, we used the definition of the SMC (Sect. 2.2), 13 S. Ventura et al. which has been elaborated for this region of study. The 95th the synoptic structure. However, we first need to analyze percentile has been calculated for each climatic period of how the climate models perform in representing synoptic 30 years separately and, consequently, has been changing behavior at the regional level. The objective of this sec- over time. The output of this step is a database of all the HW tion is to analyze the synoptic structures of the simulations episodes registered in each climatic period. available from CORDEX for the period 1951 to 2000 and determine if they are well represented when compared to 3.3 Applying principal sequence pattern analysis reanalyzed ERA5 data. to classify heat wave episodes Quantile‒quantile mapping transformation (Q–Q, Amen- gual et al. 2012) is used to normalize the model simula- We refer to this third step of the methods as the PSPA. tion grid. This is done to align the statistical distribution This process was applied for all days previously classified of ERA5 reanalysis and CORDEX datasets, which removes as HW days to obtain a synoptic classification related to technical variation from noisy data. In this work, we have these events, following the definition from 2.2. For this, applied this method for all three variables used, MSLP, we generated a database of HW episodes divided into five Z500 and T2M. This procedure consists of calculating the climatic periods of 30 years each (1951–1980, 1961–1990, changes, quantile by quantile, in the cumulative distribution 1971–2000, 1981–2010, 1991–2020), following the clima- frequency (CDF) of the daily MSLP and Z500 outputs and tological standard normal (WMO 2016). The PSPA was car- the observed data. The statistical adjustment is based on the ried out using gridded data at a 0.25° resolution from ERA5 relationship between the ranked value of the corresponding reanalysis (see Sect. 2.2.). CDFs for past calibrations (1951–2000), the control instru- Analyses of the synoptic history can be undertaken using mental or baseline (1971–2000), and the raw control simu- a multivariable classification of the synoptic sequences lated by CORDEX models (1951–2000). The Q–Q method related to the main atmospheric parameters, with an hourly is applied to the CORDEX data from simulations obtained or daily resolution. Methodologically, this classification is from three different regional climate models: WRF, REMO, supported as a variant of PCA and is known as PSPA (Aran and HIRHAM for the historical period 1951–2000 (Table 1). et al. 2011; Compagnucci et al. 2001; Escobar et al. 2004; Once the model simulation outputs for MSLP and Z500 are Esteban 2008; Jacobeit et al. 2006; Peña et al. 2015; Philipp corrected, steps 3.1, 3.2 and 3.3 are repeated for the period 2009). The analysis integrates different atmospheric levels 1951 through 2000 to obtain an analysis of historical simula- (MSLP and Z500) with the purpose of understanding the tions (see Fig. 2). main features that account for the dynamic atmospheric pro- cesses. The need for this has been recognized in previous 3.5 Analysis of CORDEX future scenarios of HW studies (Houssos et al. 2008; Peña et al. 2015; Sioutas and episodes Flocas 2003), which mention the interest of using this infor- mation to build climatological classifications and implement The same three steps 3.1, 3.2 and 3.3 are repeated to deter- them into forecast systems. mine the synoptic behavior of the HW events occurring in We applied the PSPA in T-Mode to the HW dataset two future scenarios for the period 2011 through 2100: RCPs resulting from the previous step using a correlation matrix 4.5 and 8.5. Simulations of these scenarios using WRF, (Escobar et al. 2004; Huth et al. 2008), a scree test to extract HIRHAM and REMO are available from CORDEX, which the most relevant components (Cattell 1966) and orthogonal details can be found at Table 1. The result of this step is a Varimax rotation to satisfy the orthogonality condition of the set of SWPs, as shown in Fig. 2. model (Richman 1986). More details about the data matrix can be found in the supplementary material S1. The analy- sis was conducted for the domain from 30° N to 70° N and 4 Results from 30° W to 30° E for the period 1951–2020. As a result, the output of the PSPA process is a set of SWPs (both from 4.1 Analyzing synoptic weather types and trends MSLP and Z500) that describe the main patterns associ- ated with HW days. We have developed a Python script to In this section, we classified all the summer days to differ- run this and have made it publicly available through GitHub ent weather types following the objective classification by [https://​github.​com/​URBAG-​ICTA/​PSPA_​HW.​git]. (Miró et al. 2020) and, next, we applied the Mann–Kend- all test to find the significant trends. This analysis for the 3.4 Analysis of CORDEX historical HW episodes historical reanalysis data (1951–2020) shows that synoptic weather types 11 (low or cyclone) and 12 (shallow cyclone One of the main objectives of this study is to analyze how or undetermined pressure gradient) are the most frequent future climate change simulations predict HWs in terms of during the summer months, dominating up to 18% and 40% 13 Analysis of synoptic weather patterns of heatwave events of the time, respectively, followed by types 2 (anticyclonic 4.2 Selection of heat wave episodes western advection), 5 (northeast advection) and 10 (trough), as shown in Table 3. This means that summers are mostly A rising trend in temperatures from 1951 until 2020 can characterized by undetermined pressure gradients (type 12), be noted in Fig. 3, where each boxplot shows the mean, which are periods without any dominant high or low pres- median and the 5th, 25th, 75th and 95th percentiles of maxi- sures. This pattern is commonly defined by a stationary high- mum summer temperatures for the historic five time periods pressure center located over the Azores that maintains low using observed data from the Fabra weather station. Due to pressures in northern regions, blocking any possible weather their insignificance in the scope of the study, outliers were front. Low pressures (type 11) can be defined by two differ- not included in the preparation of this figure. From these ent situations. In the first case, low pressure situations are boxplots, the 95th percentile is extracted to select the epi- structures with pressure levels lower than 1013 hPa, strong sodes of three or more days according to the definition of cyclonic winds and structures that generate weather fronts. HW adopted in this study. The result is a database of 139 However, in HW situations, the most common structure in HW days ranging from 31.9 to 39.8 °C. According to these the Iberian Peninsula is a thermal low resulting from the heating of the lower troposphere, generating weak cyclonic circulations. Table 3 also shows the trend determined by the Mann–Kendall method. Looking at the right of Table 3, only one significant trend exists in the sample, which is the decreasing trend of type 07. Although type 07 has a frequency of 2.5% in the first cli- matic period, its trend is to decrease, and in the last climatic period, it has a frequency of 1.56%. There are two synoptic weather trends that are close to having a significant trend according to the criterion used in this work (p value ≈ 0.05), which are types 04 and 12.Type 12 (which is a potential HW pattern according to the synop- tic classification) has an increase from 36.78% to 40.04%, and it is close to an increasing trend (p value = 0.07). Fig. 3  Boxplot from observations for the 1951–2020 JJA period. The boxes are defined by 25–75th quantiles, and whiskers are defined by the 5–95th quantiles. The 95th percentile is quantified (°C) Table 3  Percentage of times that synoptic events occur for every climatic period for JJA 1951–1980 1961–1990 1971–2000 1981–2010* 1991–2020* Tau–Kendall p value ** Trend TYPE01 1.38 1.12 1.30 1.52 1.41 − 0.05 0.59 NSDT TYPE02 9.64 9.71 9.75 11.09 11.12 0.05 0.58 NSIT TYPE03 2.75 2.36 2.54 2.14 2.50 0.06 0.51 NSIT TYPE04 6.20 5.14 4.93 4.35 4.93 − 0.16 0.07 NSDT TYPE05 7.97 8.19 8.41 8.22 8.70 0.12 0.16 NSIT TYPE06 1.92 2.17 2.36 2.57 2.10 − 0.05 0.56 NSDT TYPE07 2.50 2.21 1.67 1.67 1.56 − 0.18 0.05 SDT TYPE08 0.04 0.11 0.07 0.11 0.04 0.03 0.81 NSIT TYPE09 3.91 3.84 4.38 3.70 3.19 − 0.11 0.20 NSDT TYPE10 7.90 7.68 8.19 7.75 7.14 − 0.07 0.44 NSDT TYPE11 18.12 18.55 18.55 17.10 16.34 − 0.09 0.27 NSDT TYPE12 36.78 38.01 36.96 39.02 40.04 0.15 0.07 NSIT TYPE13 0.90 0.91 0.91 0.76 0.94 − 0.05 0.61 NSDT Mann–Kendall (Tau–Kendall, p value and trend NSDT nonsignificant decreasing trend, NSIT nonsignificant increasing trend, SDT significant decreasing trend, SIT significant increasing trend) *2003 not considered **p value adjusted using the Holm–Bonferroni method 13 S. Ventura et al. results, the Fabra weather station registered an increment In the next few paragraphs, we give a description of each of 1.8 °C for HW episodes between the first and the last SWP for each climatic period, which can be found in Fig. 4. climatic period under study. The amount of variance can be found in the upper right of each SWP and the short-name description in the bottom left. 4.3 Applying principal sequence pattern analysis This explained variance ratio is a metric commonly used to to classify HW episodes. evaluate the utility of SWPs and to choose the number of patterns considered (Jolliffe 2016). For the results, we have The PSPA method is applied for all HW days previously divided the four SWPs into four differentiated structures that identified to obtain a synoptic classification related to each are repeated in each climatic period. The different structures HW event. The results show that four SWPs are responsible take into account both MSLP and Z500 variables, locating for approximately 50% of the variance in MSLP and Z500 the mean elements such as low- and high-pressure centers, for all historic time periods, as shown in Table 3. These thermal lows, distance of the isolines or the distribution of four SWPs (named SWP1, SWP2, SWP3, and SWP4) are ridges and troughs. Dynamism (main meteorological struc- characterized by mean matrices of multiple HW days that tures that change over time) and stationarity (stationary pat- are constructed from linear combinations of the initial vari- terns that can remain permanent for multiple days) are also ables. The four resultant SWPs are sorted by the amount of considered and discussed in the results. explained variance, in which SWP1 always explains more SWP-S_S: stationary and stable pattern: MSLP variance than SWP2 and so forth. In this study, the explained (Fig. 4) shows a thermal low and a blocking anticyclone over variance refers to the statistical measure that explains the the Atlantic Ocean. There is a trend in surface level to an variation in a dataset attributable to each of the SWPs. increase in the dominance of the thermal low from the north of Africa and the south of the Iberian Peninsula. The Z500 Fig. 4  Mean sea level pressure (MSLP) variable (in hPa) for each patterns are found: S_S (stationary and stable), D_A (dynamic and SWP for the historical period 1951–2020. The variance % is given advective), S_A (stationary and advective) and D_AU (dynamic, in the upper-right corner and the pattern in the bottom-left. Four advective and undulated) 13 Analysis of synoptic weather patterns of heatwave events Fig. 5  SWP for the ERA5 historical period. 500 hPa geopotential tom-left. Four patterns are found: S_S (stationary and stable), D_A height (Z500) variable (in meters) plotted for 1951–2020. The vari- (dynamic and advective), S_A (stationary and advective) and D_AU ance is explained in the upper-right corner and the pattern in the bot- (dynamic, advective and undulated) level (Fig. 5) provides more information about the evolution low pressures with warm air masses that induce a thermal to a more stationary anticyclonic ridge due to the jet stream low in the Iberian Peninsula and northern Africa. There moving north. SWP-S_S is the first component in four out is a trend toward a decrease in high-pressure dominance of five periods. In the 1961–1990 period is SWP2 (15.3% in the NW. In Z500, there is an important undulation of of the total explained variance). There is a trend toward an the general circulation that advects winds from the south. increase in the variance of this pattern from 15.7% in the first There is a reduction trend for this pattern from 12.9 to period to 17.1% in the last period. 8.8%. SWP-D_AU is the third component in the first and SWP-S_A: stationary and advective: The MSLP vari- third climatic periods (12.9% and 9.2% of the variance, able indicates a stronger high-pressure center in the west- respectively), the second component during the 1961–1990 ern Iberian Peninsula in comparison with SWP-D_A and a period (15.3% of the variance), and the fourth component thermal low in northern Africa and the Iberian Peninsula. in the last two climatic periods (≈ 8% of the variance). Z500 indicates a SW flux due to the ridge in the S and the SWP-D_A: dynamic and advective: MSLP shows a trough in the NW of the Iberian Peninsula. The variance of thermal low in the Mediterranean area due to the warm this pattern decreases from 14.4 to 10.8%. SWP-S_A is the sea. The high pressures tend to lose importance due to the second component in 1951–1980 (14.4% of the variance), dominance of the thermal low. The Z500 variable shows the fourth component during the climatic period 1971–2000 SW flux, which is getting more undulations from the main- (8.6% of variance), and the third component in the rest of stream in comparison with the first climatic periods. Cold the climatic periods (9.8%, 9.8%, and 10.8% of the variance, air is restricted in the Atlantic Ocean. This pattern, which respectively). is becoming more advective in height, has increased in SWP-D_AU: dynamic, advective and undulated: frequency from 9.19 to 13.76%. SWP-D_A is the fourth MSLP indicates an area without the influence of high or component in the first climatic period (9.2%% of explained 13 S. Ventura et al. variance), the first component during the 1961–1990 In the AMB, the warmest HW patterns are SWP-D_A period (17.1% of variance), and the third component in and SWP-D_AU, which are dynamic and advective. the rest of the climatic periods (15.5%, 14.9%, and 13.8% Specifically, SWP-D_A of the 1981–2010 period is the of the variance, respectively). warmest (35.1 °C, mean daily maximum temperature in Once the HW events have been classified into the four the AMB), followed by SWP-S_A in 1981–2010 (35.0 °C) SWPs for the Iberian Peninsula, we next analyze the effect and SWP-S_A in the 1991–2020 period (34.9 °C). Accord- of these four SWPs on the intensity of HWs in the Met- ing to Fig. 6, prefrontal patterns are the warmest (D_AU) ropolitan Area of Barcelona. We use a dataset providing in the AMB, advecting air from the S‒SW at 500 hPa. daily temperature and precipitation from 1951 to 2020 in Moreover, SWP-D_A is the pattern with a thermal low a regular grid of 5 km from Spain02 interpolated products over the Mediterranean, indicating a warm sea influencing (data source: https://​www.​aemet.​es/​es/​servi​ciosc​limat​icos) the AMB temperatures. Intense ridges at 500 hPa with a to generate Fig. 6: a composite temperature map showing the thermal low at the surface (SWP-S_S) also generate warm mean of the daily maximum temperatures corresponding to conditions (34.6 °C in the last climatic period) in the AMB each SWP. Please see Supplementary Materials S2 and S3 but especially in the rest of the Iberian Peninsula. Maxi- for the tables with temperature values and S4 for the mean mum temperatures have increased by approximately 2 °C map covering all of Spain. between the first and last climatic periods analyzed. HWs registered in the AMB are mostly related to HW episodes in most of the Iberian Peninsula (see Fig. S4 in the Supplementary Material). The only SWP that differs Fig. 6  Mean maximum temperature maps (in °C) for the Metropolitan Area of Barcelona associated with SWPs by climatic period: a 1951– 1980, b 1961–1990 and c 1971–2000. Marked: Fabra Meteorological Station 13 Analysis of synoptic weather patterns of heatwave events from the rest is SWP-D_AU, in which a cold front gener- synoptic weather type (type 12), WRF and REMO forecast ates lower temperatures in the NW of the peninsula. a nonsignificant decrease in cases, while HIRHAM forecast a nonsignificant increase. 4.4 CORDEX historical Next, we analyze the differences among the HW days of the various models and compare them with the HW days The same analysis that was done for historical observed data from the observed data and the ERA5 reanalysis data. Box- (steps 1–3 of the methods section) is next repeated for the plots with percentiles of temperature are generated to study same historical period 1951–2000 JJA but with modeled the variability of every climatic period, shown in the sup- CORDEX (WRF, HIRAM and REMO) data to analyze the plementary material S6. The 95th percentile, represented as performance of the various models. Since one of the main the top of the whisker, is the HW temperature according to research objectives is to understand future HW events in the definition used in this study. terms of the SWPs in which they develop, it is important to The 95th percentile of Fabra recorded an increase first establish how well historical simulations of synoptic of + 0.81 °C between 1951–1980 and 1971–2000, while weather types are generated. First, all JJA days for the period ERA5 reflected an increase of + 0.55 °C. CORDEX datasets of 1951–2000 from the output of the WRF, HIRHAM, and register increases that range between 0.18 °C (HIRHAM), REMO models were classified into 13 synoptic weather 0.29 °C (WRF) and 0.32 °C (REMO). The adjustment of the types following the method described in Sect. 3.1, as shown CORDEX dataset corrects the overestimated values that all in Fig. 7. Most JJA days of both ERA5 reanalysis and COR- three models have, resulting in three datasets with similar DEX simulations fall into synoptic weather type 12 “shallow percentiles in comparison with ERA5. cyclone or undetermined pressure gradient”. HIRHAM esti- The PSPA (3.3 of methods section) for the three COR- mates more JJA days in type 12 than ERA5 (+ 5.05%), while DEX historical datasets reaches more than 50% of the vari- REMO and WRF underestimate type 12 days (− 9.45% and ance in four SWPs (variance represented in the upper-right − 14.39%, respectively). of the SWPs in Figs. 8, 9, 10). In this section, we show The main differences among model outputs can be found and describe the results of the WRF model due to the bet- in types 02, 11 and 12. REMO simulations show more dyna- ter representation of SWPs in comparison with ERA5. The mism at 500 hPa (due to advections from different direc- description for all the CORDEX models is in Supplementary tions) and stability at the surface level (high pressures), Material S7 and REMO-HIRHAM plots in S8–S9. which generates weather types 02 and 12. HIRHAM has The criteria used to select the model that best replicates the most similar general structure in comparison with ERA5, the SWPs found in ERA5 data have considered multiple with a similar percentage of cases in all the most repeated factors. WRF is the model with less overestimation for the patterns (02, 11 and 12). WRF has a significant increase in SWP1 (SWP-S_S) variance, according to Figs. 8 (MSLP), 9 synoptic weather type 11, which is related to low pressures (Z500) and 10 (TMAX). WRF shows a variance that ranges and thermal lows. between 20.24 and 22.52% for SWP-S_S, which is higher The results of the Mann–Kendall test are shown in than ERA5 (8.81–17.32%) but less than the rest of the COR- Table S5: the CORDEX models do not simulate any sig- DEX models (20.39–25.63% for REMO and 22.36–28.32% nificant trend in the 1951–2000 period, which is consist- for HIRHAM). The rest of the SWPs have similar variance ent with the ERA5 reanalysis. Analyzing the most frequent values (+ − 5%). WRF is the model that best shows the pres- ence of the general stream undulation, especially for D_AU, in comparison with ERA5 reanalysis. This fact has been considered due to the intensity of HWs in terms of the tem- perature that this pattern generates, as described in the previ- ous section. REMO and HIRHAM show the SWP-S_S with more advection at Z500 in comparison with ERA5. Due to the important amount of variance explained by this pattern (the most relevant), it is necessary to show it in a similar structure. The locations of the anticyclonic ridges (Z500) and high-pressure centers (MSLP) shown by WRF are well approximated, which is important to define the direction of wind advections to the AMB. The daily maximum temperature of each SWP is plot- ted for WRF in Fig. 10 (REMO and HIRHAM are found at Fig. 7  Synoptic weather types of both reanalysis (ERA5) and histori- cal simulations (CORDEX) for the period 1951–2000. Only JJA days the bottom of S8 and S9 for the Iberian Peninsula. Due to were selected the lack of resolution (0.1°), an analysis for the AMB scale 13 S. Ventura et al. Fig. 8  SWPs for the CORDEX historical period. Mean sea level pres- in the upper-right corner and the synoptic pattern in the bottom-left sure (MSLP) variable (in hPa) plotted by the climatic periods divided corner. Model: WRF by files. The columns represent the SWPs with the variance explained Fig. 9  SWPs for the CORDEX historical period. Geopotential height explained in the upper-right corner and the synoptic pattern in the at 500 hPa (Z5000) variable (in m) plotted by the climatic periods bottom-left corner. Model: WRF divided by files. The columns represent the SWPs with the variance 13 Analysis of synoptic weather patterns of heatwave events Fig. 10  SWPs for the CORDEX historical period. Daily maximum variance explained in the upper-right corner and the synoptic pattern temperature at 2 m (TMAX) variable (in °C) plotted by the climatic in the bottom-left corner. Model: WRF periods divided by files. The columns represent the SWPs with the HW episodes, we next apply the methods described above to future scenarios RCP4.5 and RCP8.5. Figure 11 shows the classification of JJA days from the model output into the thirteen synoptic weather patterns. There is no significant difference in how the days are classified between scenarios RCP4.5 and RCP8.5 for each model. The HIRHAM model, which best estimated the synoptic weather patterns in com- parison with the ERA5 reanalysis, simulates an increase in JJA days in type 12 in both scenarios, going from 42.09% in the historic CORDEX to 47.4% for RCP4.5 and 44.88% for RCP8.5. However, REMO and WRF do not simulate significant increases or decreases in type 12 with respect to Fig. 11  Synoptic weather types of CORDEX (scenarios 8.5 and 4.5) ERA5. CORDEX models agree that RCP8.5 increases types for the period 1951–2000. Only JJA months were selected 05 and 11 in comparison with RCP4.5, which are patterns with northeast advections and cyclones. On the other hand, cannot be performed. The mean temperatures for the Ibe- the CORDEX models simulate a decrease in the number of rian Peninsula range between 23.52 and 24.41 °C (WRF), days of types 06, 09 and 13 in the case of RCP8.5, which are 22.92–24.17 °C (REMO) and 22.73–23.91 °C (HIRHAM), east advections and pure anticyclones. which are slightly underestimated in comparison to ERA5 The Mann–Kendall test was run for scenarios 4.5 (23.85–25.3 °C). In all three models, the mean TMAX for (Table S10) and 8.5 (Table S11). A detailed description the SWP-S_S is the warmest pattern in the Iberian Penin- of all the model trends is found in supplementary material sula, which matches with ERA5. S12. Although in the CORDEX historical dataset, there are no significant trends, the CORDEX future datasets show 4.5 CORDEX future scenarios some possible significant increasing and decreasing trends. In RCP4.5, HIRHAM is the only model that shows signifi- Having understood how well the various models perform cant trends, which are on synoptic weather type 02 (SDT), in terms of simulating synoptic behaviors that give rise to type 12 (SIT) and type 13 (SDT). Type 12, which contains 13 S. Ventura et al. the highest number of days, shows an increasing trend from 4.5, the increase in HW thresholds by periods of 10 years 43.37 to 50.57% of the JJA days. This increasing trend is ranges from + 0.12 °C/10Y (WRF) to + 0.53 °C/10Y (HIR- not shown by WRF or REMO. RCP8.5 has more significant HAM), while in scenario 8.5, the 95th percentile ranges trends in all the CORDEX models. The most remarkable from + 0.53 °C/10Y (WRF) to + 0.95 °C/10Y (HIRHAM). ones (due to the high percentage of days) are the SDT for WRF is the model with the lowest heating, and HIRHAM is type 02 (shown both in HIRHAM and REMO), the SIT for the model that increases the most. type 11 (shown in HIRHAM dataset) and the significant Next, the PSPA method described in Sect. 3.3 is applied trends shown by type 12, which conflicts WRF (SDT) and to the three climatic model future simulations for both sce- REMO (SIT). narios. Due to the better representation of the WRF model Next, the temperature percentiles of both scenarios 4.5 described in the previous section, the following descrip- and 8.5 are analyzed to subsequently select the HW potential tion only considers this model. A detailed description of all days. CORDEX models show an increase in temperatures CORDEX models is found in supplementary material S14. in all their percentiles (more details in the Supplementary The WRF results are represented in Figs. 12, 13, 14 for the Materials S13), but there are differences among them. For RCP4.5 scenario and Figs. 15, 16, 17 for the RCP8.5 sce- scenario 4.5, WRF simulations result in a small increase in nario. To reduce the number of plots in the main description, the 95th percentile (+ 0.69 °C) in the last climatic period we have chosen only three climatic periods (2011–2040, (2071–2100), reaching 31.57 °C in comparison to the first 2041–2070 and 2071–2100) to represent the whole ana- period (2011–2040), which reaches 30.59 °C. For HIRHAM, lyzed period (2011–2100) despite the seven periods (e.g., the 95th percentile increases by + 3.17 °C (from 30.53 °C to 2011–2040, 2021–2050, …, 2071–2100). The complete 33.7 °C), and REMO increases by + 0.99 °C (from 30.58 °C sequence for WRF, HIRHAM and REMO can be found in to 31.57 °C). the Supplementary Material (S15–S23). In scenario 8.5, the CORDEX models simulate a higher In general, terms, RCP4.5 shows four patterns (SWP_S-S, increase in temperatures for all percentiles. The 95th per- SWP-D_A, SWP-S_A and SWP-D_AU) in the same manner centile for the WRF model rises 3.18 °C, ranging from as the historical dataset. However, there are some differences 30.9 to 34.08 °C, HIRHAM shows an increase of 5.7 °C and tendencies to mention. (from 30.79 to 36.49 °C) and REMO also considers an SWP-S_S: stationary and stable pattern: The variance increase of 3.89 °C (from 30.59 to 34.48 °C). All the COR- explained by this pattern does not have important changes, DEX models and scenarios agree in an increase in HW remaining the most important pattern (SWP1) with a vari- thresholds, which are higher in scenario 8.5. For scenario ance that ranges between 21.71 and 27.42%, while in the Fig. 12  SWPs for the CORDEX RCP4.5 scenario. The MSLP variable (in hPa) is represented. Resumed figure with the periods 2011–2040, 2041–2070 and 2071–2100. Variance explained in the upper-right corner and the pattern in the bottom-left. Model: WRF 13 Analysis of synoptic weather patterns of heatwave events Fig. 13  SWPs for the CORDEX RCP4.5 scenario. Geopotential height at 500 hPa (Z500) variable (in m) plotted by the climatic periods divided by files. Variance explained in the upper-right corner and the pattern in the bottom-left. Model: WRF Fig. 14  SWPs for the CORDEX RCP4.5 scenario. Daily maximum temperature at 2 m variable (in °C) plotted by the climatic periods divided by files. Variance explained in the upper-right corner and the pattern in the bottom-left. Model: WRF 13 S. Ventura et al. Fig. 15  SWPs for the CORDEX RCP8.5 scenario. The MSLP variable (in hPa) is represented. Resumed figure with the periods 2011–2040, 2041–2070 and 2071–2100. Variance explained in the upper-right corner and the pattern in the bottom-left. Model: WRF Fig. 16  SWPs for the CORDEX RCP8.5 scenario. Geopotential height at 500 hPa (Z5000) variable (in m) plotted by the climatic periods divided by files. Variance explained in the upper-right corner and the pattern in the bottom-left. Model: WRF historical WRF dataset, it ranges between 20.24 and 22.52%. latitudes. This fact generates higher geopotential heights in MSLP shows the same structure as the historical period the Iberian Peninsula, with maximum values that increase with a more intense thermal low. Z500 shows the same pat- from 5925 to 5975 m. tern with a more intense ridge, which is moving to higher 13 Analysis of synoptic weather patterns of heatwave events Fig. 17  SWPs for the CORDEX RCP8.5 scenario. Daily maximum temperature at 2 m variable (in °C) plotted by the climatic periods divided by files. Variance explained in the upper-right corner and the pattern in the bottom-left. Model: WRF SWP-D_A: dynamic and advective: The variance The daily maximum temperatures shown in Fig. 14 for explained by this pattern increases from 8.82–10.85% in the WRF4.5 result in a significant increase with respect to his- historical WRF dataset (SWP4) to 12.98–17.5% in WRF4.5 torical WRF. WRF4.5 simulates the SWP-S_S pattern as the (SWP2). The MSLP shows a less intense blocking anticy- warmest on average in the Iberian Peninsula (26.31 °C), fol- clone that generates a more intense thermal low, covering lowed by SWP-DA_U (25.95 °C) and SWP-D_A (25.91 °C), more terrain in southern Europe. Z500 shows a similar pat- which rise 1.93 °C, 2.03 °C and 1.75 °C, respectively, com- tern with increased advection due to the more important pared to the historic WRF. gradient generated by the north movement of the anticy- The PSPA has also been applied to scenario 8.5. Fig- clonic ridge. ures 15 (MSLP), 16 (Z500) and 17 (TMAX) show the SWPs SWP-S_A: stationary and advective: The explained for WRF8.5. A detailed description of WRF, HIRHAM and variance by this pattern remains similar during WRF4.5 REMO can be found in supplementary material S24. The (9.9–11.84%) in comparison with WRF historical complete associated charts can be found in Figures S25–S33. (10.87–11.81%). MSLP shows no differences in the general The MSLP synoptic structure does not present significant structure, although the blocking anticyclone is less intense, changes in comparison with WRF4.5, although there are which generates a stronger thermal low. Z500 shows higher remarkable differences at Z500. According to WRF8.5, D_ geopotential height values (+ 50 m in some cases) due to AU explains more variance (10.45–14.79%) than WRF4.5 the north displacement of the ridge, although the structure (6.59–9.19%), and at the end of the period, it is the second remains the same. most important pattern (SWP2). However, the structure of SWP-D_AU: dynamic, advective and undulated: this pattern at Z500 has differences. The S‒SW advection The explained variance by SWP-D_AU decreases from at the Iberian Peninsula that WRF4.5 and WRF historical 13.61–14.79% (WRF historical) to 6.59–9.19% (WRF4.5), simulate is not observed at WRF8.5, which simulates an which makes this pattern SWP4 (SWP2 in the historical intense anticyclonic ridge located at the west of the Ibe- dataset). In this sense, WRF4.5 shows less undulation in rian Peninsula. Z500 also shows westerly flows in the other the patterns. MSLP shows a similar structure with a slightly three SWPs, which differ from SW flows from WRF4.5. The increased blocking anticyclone. Z500 shows less advection maximum geopotential height values are increased, reach- due to the less undulated pattern and the north shift of the ing 6000 m in extended parts of the Iberian Peninsula. The anticyclonic ridge. MSLP shows more intense and Mediterranean-centered 13 S. Ventura et al. thermal lows, which indicates the intense warming of land periods in comparison with the middle of the XX century, and sea. which could potentially translate to more HW potential days The temperature values for scenario 8.5 are higher in the due to the reduction in less warm advections. The mid-1970s mid and late periods in comparison with WRF4.5. The pat- suffered moderate cooling (− 0.03 °C/year) that was fol- tern that shows the highest increase is SWP-DA_U, which lowed by significant warming (+ 0.07 °C/year) (Lionello reaches the highest values in the last period (28.12 °C in and Scarascia 2018). This trend is reflected in the SWPs we comparison to 26.26 °C in WRF4.5). The SWP-S_S shows a discuss next, in which late periods show warmer and more mean temperature value in 2071–2100 of 27.81 °C, which is stable patterns than the first periods. higher in comparison to the 25.81 °C simulated at WRF4.5. The rest of the patterns show less important increases. 5.2 Main synoptic patterns associated with heatwaves at the regional and urban levels 5 Discussion We determined that four SWPs explain more than 50% of the 5.1 Main synoptic types associated with summer HW statistical variability in the climatic periods covering days 1951–2020 at the AMB: two stationary patterns (SWP-S_S and SWP-S_A) that explain 27.9% of the variance and two It is important to determine the most common types of syn- dynamic patterns (SWP-D_A and SWP-D_AU) that explain optic events occurring during summer days to later classify 22.6%. The MSLP shows, in all four cases, an anticyclone the synoptic trends of HWs. The results of the classification over the Atlantic that blocks the possible flux of low-pres- for the synoptic weather types applied to historical ERA5 sure centers or weather fronts, generating stability with long data show that type 12 is dominant for summer days, fol- periods of cloudless conditions. Due to the latitude of the lowed by types 11 and 02. The rest of the types represent less Iberian Peninsula, cloudless conditions can intensify solar than 10% of the summer days. Type 12 is defined as an unde- radiation flux and generate positive temperature anomalies, termined pattern with no important advections or influence promoting HW conditions (Tomczyk et al. 2017). There is of a synoptic high- or low-pressure center (Jenkinson and also a general presence of thermal lows over the Iberian Pen- Collison 1977). In some cases, these patterns are reinforced insula, which are deeper in the case of intense solar radia- by strong irradiation, which generates mesoscalar thermal tion. The Z500 variable shows more information about gen- low pressures in the SW of the peninsula (Hoinka and Castro eral circulation, such as the presence of ridges and troughs, 2003). The intense solar radiation due to cloudy weather and which could indicate warm or cold intrusions. All four SWPs the scarce wind on the surface and lower layers is a potential show an anticyclonic ridge in the Iberian Peninsula, which, cause of HWs in the AMB (Gil et al. 2019). According to depending on its position, generates different directions and the Mann–Kendall test, this pattern shows a nonsignificant intensities of advections in height. increasing trend, accounting for 37.8% of the summer days The most stationary and static pattern associated with in 1951 to 40% in 2020. Types 11 and 02 are the next most HWs is SWP-S_S, with a ridge that covers the entire Iberian frequent patterns on summer days according to ERA5, which Peninsula for at least three days. This pattern reaches a mean are defined as a thermal low and anticyclonic western advec- temperature of 34.4 °C. MSLP shows a blocking anticyclone tion, respectively. Thermal lows are generated by intense over the Atlantic and a thermal low over the peninsula. The solar radiation and stability, which rises warm air from lower Z500 from SWPS-S_S presents an intense ridge centered atmospheric layers, generating a low pressure (Portela and on the peninsula, which due to its stationarity can remain Castro 1996), commonly produced after stationarity of type for long periods of time, causing a warm air mass to be 12. However, western advections are dynamic situations that generated by the peninsula itself. These synoptic conditions can advect warm and dry air masses from the interior of increase the surface sensible heat fluxes warming the air the peninsula (Mazon et al. 2014), generating potential HW parcels near the surface adiabatically (Zschenderlein et al. days in the Mediterranean basin. No significant trends are 2019), resulting in HW events. found for types 11 (18.1–16.3%) or 02 (9.6–11.1%) in the The SWP-S_A is also stationary, although it shows a 1951–2020 period. more advective pattern at Z500 with SW flux. There are We found a significant decreasing trend for type 07 (east two main differences between both stationary patterns advection with cutoff low above), a rare synoptic type in all (SWP-S_S and SWP-S_A). First, at MSLP, the Atlantic climatic periods (with a frequency of 2.5% in 1951–1980 anticyclone is located a few km to the south in case of to 1.56% in 1991–2020). North advections (type 04) show SWP-S_A. Second, at Z500, there is a greater geopotential a nonsignificant decreasing trend that ranges from 6.2% to height gradient due to the location of the low-pressure 4.93%. These results show more stability in late climate center in northern latitudes and the anticyclone ridge in 13 Analysis of synoptic weather patterns of heatwave events the south. These conditions make a potential HW pat- to intense solar radiation. This fact also indicates a notable tern in the E‒SE of the Iberian Peninsula, while in the warming of the sea temperature compared to the beginning NW, there are windy and colder conditions. Some studies of the analyzed period, which could result in more potential have discussed the relation between European HWs and HW days for the AMB. This warming of the sea coincides the tropical Atlantic conditions represented in this pat- with Lionello and Scarascia (2018), where the Mediterra- tern acting as a forcing, which amplify the residence of nean area is considered a climate change hot spot due to blocking regimes and therefore increase the possibility of the expected warming of the region compared to the rest of HWs in southern and central Europe (Cassou et al. 2005; the planet. Gil et al. 2019.). Our analysis of the SWPs associated with HWs in the The SWP-DA_U pattern is caused by a trough NW of AMB is limited due to the coarse resolution of reanalysis the Iberian Peninsula that undulates the general circulation data, which is available at 0.25° (approx. 25 km) and is thus and generates advection from S‒SW. This pattern not only unable to capture human activity and building materials advects warm air from the African continent but also gener- that exacerbate the effects of HWs (Liu et al. 2018; Morris ates intrusions of Saharan dust (Sousa et al. 2019), which et al. 2017). In this sense, different downscaling techniques worsens the air quality in the Iberian Peninsula. This pattern are being investigated at the urban scale, albeit with a high generates the warmest HWs registered in this study due to degree of uncertainty (Duchêne et al. 2020; le Roy et al. the major south component of advection. The SWP-D_A 2021), but further work is needed to improve mesoscale and case is characterized by the lowest surface pressures and urban scale meteorological simulations. thermal lows centered in the Mediterranean region. This is possibly generated by the temperature of the sea, which is 5.3 CORDEX historical simulations: analysis warming 20% more than the global average (Lionello and of limitations and advantages Scarascia 2018). This pattern also shows SW advection at Z500, which supports the importance of SW advections at Both model simulations and reanalysis data generally agree 500 hPa in HW episodes due to warm advection. in the summer’s overall synoptic classification, represent- Zooming in at the urban level, the AMB is significantly ing similar SWPs in terms of the locations of high pres- influenced by the dynamic patterns SWP-D_A and SWP- sures, thermal lows, ridges and troughs. However, we have DA_U (dynamic and advective and dynamic, advective and found some robustness and limitations on the performance undulated, respectively). SWP-D_AU for 1981–2010 reaches of climatic models in simulating the synoptic structure of the a mean value of 35.1 °C in the AMB, followed by SWP- HW episodes that are noteworthy. The CORDEX historical D_A for 1981–2010 (35 °C) and SWP-D_A for 1991–2020 simulations show discrepancies among the WRF, REMO (34.9 °C). Cases of S‒SW advections are potentially the and HIRHAM models that are worth mentioning due to the warmest patterns in the AMB. The mountain ranges, espe- possible overestimation of HW temperatures. Regarding the cially the Sistema Ibérico, which is located parallel to the TMAX variable, a positive BIAS of > 1.5 °C was found in Mediterranean coast, generate a foehn effect throughout all models compared to ERA5, even after the Q-Q technique the E‒NE of the Iberian Peninsula in the case of W‒SW for intermodel comparison was applied. This positive BIAS wind flow (Peña et al. 2016). In that context, prefrontal pat- agrees with what has been found in other studies such as terns such as SWP-D_AU intrinsically have an SW flow that Lhotka et al. (2018), in which the 90% quantiles were calcu- maintains warm temperatures and dry air in the AMB but lated separately for each simulation to remove the influence not in the W‒NW of the Iberian Peninsula. of the TMAX bias. Our analysis shows that there is an increasing trend of The MSLP and Z500 variables show similar patterns as HWs in the AMB associated with SWP-S_S, characterized ERA5, although the structures are shown in a less defined by an intense anticyclonic ridge that covers the Iberian Pen- way (general decrease in geopotential height gradient) as a insula and reaches central Europe. This SWP generates a result of the low resolution of the models. The Z500 variable stationary pattern that can last for multiple days, covering a results in an overestimation of the intensity of the anticy- large part of western and central Europe. SWP-D_AU lost clonic ridges, simulating values that range between 5900 some frequency of HWs with respect to the beginning of the and 5925 m, while ERA5 simulates values between 5800 period (1951–1980), contrary to other studies (Zschender- and 5825 m. In this way, the three analyzed models tend lein et al. 2019). This discrepancy could be because we are to shift circulation excessively to the north. On the other explaining only 50–55% of the variance in HWs with the hand, when analyzing the MSLP variable, the models tend to four SWPs and do not analyze the remaining 45–50% of the underestimate the pressure gradient, simulating more inde- information, which we consider noise (Wold et al. 1987). terminate patterns. Even so, the models are able to reflect The SWP-D_A pattern shows an increase in thermal the thermal lows and anti-cyclonic blockings in an approxi- lows in the Mediterranean regions affecting the AMB due mate way, which contrasts with other studies where the 13 S. Ventura et al. underestimation of these blockings has been demonstrated report and the results obtained in this article point to an (Scaife et al. 2010). increase in extreme temperatures that would result in more Several studies have analyzed the potential limitations of intense HWs. HW analysis with CORDEX. The elaboration of composites Next, we determine the synoptic weather types associ- can lead to multiple limitations that have been discussed in ated with summer days of future scenarios. This step is several studies (Lhotka et al. 2018). These limitations are necessary to quantify the tendency of potential HW types, related to the fact that some effects, such as advections from such as type 12, which in ERA5 represents 40% of the opposite directions, can be canceled when applying the com- summer days and 49.7% of the HW days in the 1991–2020 posite of different HW days. Since HWs can be produced by period. There are discrepancies between the CORDEX different synoptic patterns, performing a composite of differ- models for types 11 and 12. HIRHAM is the only model ent days that do not have synoptic similarity between them that shows a significant increasing trend for type 12, which (despite producing HWs) can be counterproductive and lim- would result in more potential HW days by the end of this iting. This is especially important for RCMs in which land‒ century. However, the most notable trends are the decrease atmosphere iteration plays a vital role in the development of in pure anticyclones (type 13) and east advections, such as HWs. In this study, this issue is largely corrected by using types 06 and 07, and the increase in north advections, such the PSPA, which groups the synoptic patterns for statistical as types 04 and 05, indicating a more extreme climate. In proximity, separating these patterns with different synoptic that sense, all models suggest a slight tendency toward an structures and avoiding this "canceling out" effect. Some increase in atmospheric dynamism in the summer months articles (Sfîcă et al. 2017; Zong et al. 2022) have remarked of JJA, generating a slightly more extreme climate than on the importance of the connections between extremely the current climate. This is especially the case for RCP8.5, warm periods and synoptic structures analyzed with PCA where we see more advective types, such as north and and PSPA, such as the Western Mediterranean Oscillation northeast advections. This change in climate extremes has found in Mohammed et al. (2018). been found by other studies: Wang et al. (2017) used Stu- Finally, we would like to point out the limitations in dent’s test to show that climate change features an increase analyzing HWs at the urban scale. Due to the CORDEX in warm extremes, and Donat et al. (2016) confirmed that resolution (0.11°), it is not possible to analyze the results at there is high confidence that temperature extremes have the urban scale. In that sense, some articles (Duchêne et al. been warming since the middle of the twentieth century. 2020; le Roy et al. 2021) have elaborated downscaling tech- In the mentioned works, they found a significant correla- niques to consider the influence of cities on the local climate tion between the changes in climate extremes and global (e.g., urban heat island). However, CORDEX simulations warming. In our study, we find a similar relation, espe- show the main HW synoptic patterns in a way that allows cially in RCP8.5. We see a major number of days classified an analysis of the main structures of HWs and the possible as type 12 (+ 1.75%) and type 11 (+ 1.66%), which are future trends, as we have shown in this study. In this sense, considered potential HW types but, at the same time, an WRF has been considered the model that best represents the increase in cold advections from the north (+ 2.48%) in synoptic structure of HWs in terms of similar variance, pres- the 2011–2100 period. sure centers and ridge locations when compared to ERA5. The PSPA for the CORDEX models gives us some insight into potential trends associated with HWs in the 2011–2100 5.4 Trends of future HW events based on CORDEX period. The Z500 variable shows an increase in geopotential simulations height, especially in scenario 8.5. The maximum geopoten- tial height ranges between + 50 and + 75 m in RCP4.5 and When we analyze the HW episodes of future simulations from + 75 to + 150 m in RCP8.5. This suggests that the anti- available from CORDEX based on temperature alone, we cyclonic ridge will shift to the north, especially in RCP8.5, find that the 95th percentile increases every period, result- advecting warm air from northern Africa. It has been proven ing in warmer HW episodes. For example, HIRHAM4.5 in multiple studies (Fischer et al. 2007; Serrano-Notivoli and WRF4.5 predict an increase in the 95th percentile et al. 2022) that positive 500 hPa height anomalies are by + 0.12 °C and + 0.53 °C every ten years, respectively, the main contributor to the increase in temperatures dur- for the RCP4.5 scenario. The same trend is more pro- ing HWs, which agrees with what we have found in this nounced for RCP8.5, with a potential increment ranging study, where geopotential heights are increased at the end from + 0.53 °C/10Y for WRF8.5 to + 0.95 °C/10Y for HIR- of the century. Some articles have related the increase in HAM8.5. This increase in temperature matches the findings extreme patterns, such as anomalous anticyclonic ridges, of the TICC report (Government of Catalonia 2017), which with the waviness of both polar and subtropical jets (Maher estimates an increase of 0.8 °C this decade and 1.4 °C by et al. 2020; Martin 2021). According to these studies, both 2050 compared to 1971–2000. In that sense, both the TICCC jets become wavier, while there are no significant trends in 13 Analysis of synoptic weather patterns of heatwave events their average speeds. This effect generates large-scale cir- • The historical CORDEX simulations, once corrected culation anomalies that are intrinsically linked to the HWs. with the Q-Q technique, provide SWPs similar to those Both ERA5 and CORDEX simulations show that the vari- of ERA5, showing good model performance in general. able MSLP indicates an increase in thermal lows not only This performance allows us to analyze possible future over the south of the peninsula, as pointed out by Jerez et al. trends according to the CORDEX models for both the (2012) but also over the Mediterranean region, further exac- RCP4.5 and 8.5 scenarios. However, we found that erbating HW events in the AMB. overestimation of the geopotential height could result in overestimation of the temperatures. • The evolution of the HWs, according to the PSPA 6 Conclusions applied to CORDEX RCP4.5 and RCP8.5, has a ten- dency toward the intensification of the anticyclonic This study analyzes the past and future HWs from multiple ridges, which reach geopotential values ​​higher than datasets: ERA5 reanalysis (historical period 1951–2020), 6050 m in the RCP8.5. This increase in the geopoten- CORDEX historical (1951–2000) and CORDEX future tial height could generate an increase in temperatures (RCP4.5 and RCP8.5 for the 2011–2100 period). We have in the lower layers of the atmosphere. Furthermore, we evaluated the synoptical structure and significant trends have found that the values of geopotential height reach associated with HW episodes to better understand the higher values in the case of the RCP8.5 scenario, com- potential HW patterns and the possible evolution not only pared with the RCP4.5, which may reflect the influence in the recent past but also in the future. The key findings of the anthropogenic warming. can be summarized as follows: • The PSPA analysis for CORDEX future simulations shows that the S‒SW advections caused by a trough in • The historical analysis (1950–2020) using ERA5 data the NW of the Iberian Peninsula will be less frequent in concluded that the most dominant synoptic weather the case of HWs at the end of the century in compari- type for summer days in the Metropolitan Area of Bar- son with the present. This fact contradicts the synoptic celona is type 12 (undetermined pattern without influ- weather type analysis, in which RCP8.5 has an increase ence of high or low pressures), followed by types 11 in atmospheric dynamism. Specifically, RCP8.5 shows (thermal low or cyclone) and 02 (anticyclonic western a decrease in anticyclones and an increase in north advection). advections. • The PSPA analysis based on historical ERA5 data from 1950 to 2020 identified some specific patterns associ- This study has evidenced the limitations of applying ated with the development of HW events. Four patterns RCMs at the urban scale for a more robust analysis of were found for each climatic period, representing at synoptic patterns of HWs and hopes to motivate future least 50% of the total HW variance. The four patterns work to improve methods to determine the development are divided into two groups: stationary patterns (S_S and effect of future HWs in urban areas. Future research and S_A) and dynamic patterns (D_A and D_AU). using high-resolution models with urban parameterizations • -The HW patterns with the highest temperatures on the is being planned, with the objective of further analyze the AMB are the dynamic cases, with maximum mean tem- effects of changing different urban land uses and param- peratures of 35 °C. The dynamic patterns show advec- eters in the future climate, focusing in HWs. In this work, tions from SW‒W that increase the temperature espe- we found that SW advections increase temperatures in cially in the Mediterranean region, possibly due to the coastal areas such as the city of Barcelona, possibly due inhibition of sea breezes, which is a thermal regulator to the inhibition of sea breezes. However, we would like in case of high temperatures. This pattern may be influ- to study it in more detail in future work. enced by the presence of a low-pressure in the NW of the Supplementary Information The online version contains supplemen- Iberian Peninsula, which generates an increase of S-SW tary material available at https://d​ oi.o​ rg/1​ 0.1​ 007/s​ 00382-0​ 23-0​ 6828-1. advection in the AMB. However, we have found that the stationary patterns with deep anticyclonic ridges increase Acknowledgements This work has been made possible thanks to the the temperatures in inland areas of the Iberian Peninsula. financial support of the ERC Consolidator, Integrated System Analysis of Urban Vegetation and Agriculture (818002-URBAG). The authors These stationary patterns can remain for multiple days would thank AEMET for the interpolated data from 1800 meteorologi- over the Iberian Peninsula, generating inland warming cal stations covering 1951-2020 provided for this work (Spain02 data- due to intense solar radiation. This pattern may be influ- set: https://w ​ ww.a​ emet.e​ s/e​ s/s​ ervic​ ioscl​ imati​ cos/c​ ambio_c​ limat/d​ atos_​ enced by the static presence of the Azores anticyclone, diari​os?w=​2&​w2=0). We acknowledge the World Climate Research Programme's Working Group on Regional Climate, and the Working which is commonly located over the Atlantic and influ- Group on Coupled Modelling, former coordinating body of CORDEX ences the meteorology of the Iberian Peninsula. 13 S. Ventura et al. and responsible panel for CMIP5. We also thank the climate model- Bengtsson L, Hagemann S, Hodges KI (2004) Can climate trends be ling groups (listed in Table 1 of this paper) for producing and making calculated from reanalysis data? J Geophys Res 109:D11111. available their model output. We also acknowledge the Earth System https://​doi.​org/​10.​1029/​2004J​D0045​36 Grid Federation infrastructure an international effort led by the U.S. Blanco G, Gerlagh R, Suh S, Barrett J, de Coninck HC, Morejon CFD, Department of Energy's Program for Climate Model Diagnosis and Mathur R, Nakicenovic N et al (2014) Chapter 5—Drivers trends Intercomparison, the European Network for Earth System Modelling and mitigation. Climate Change 2014: Mitigation of Climate and other partners in the Global Organization for Earth System Science Change. IPCC Working Group III Contribution to AR5. Cam- Portals (GO-ESSP). Finally, the authors would also thank to Servei bridge University Press Meteorològic de Catalunya for the meteorological datasets provided Bouchama A (2004) The 2003 European heat wave. Intensive Care by their Automatic Weather Stations. Med 30:1–3. https://​doi.​org/​10.​1007/​s00134-​003-​2062-y Cassou C, Terray L, Phillips AS (2005) Tropical Atlantic influence on Author contributions Conceptualization: all authors. Data collection: European heat waves. J Clim 18(15):2805–2811 SV. Data analysis: all authors. Visualizations: SV. Writing, review Cattell RB (1966) The scree test for the number of the factors. Multi- and editing of manuscript: all authors. Supervision: GV and JRM. All var Behav Res 1:245–276. https://​doi.​org/​10.​1207/​s1532​7906m​ authors read and approved the final manuscript. br0102_​10 Choi W, Ho CH, Jung J et al (2021) Synoptic conditions controlling Funding Open Access Funding provided by Universitat Autonoma de the seasonal onset and days of heatwaves over Korea. Clim Dyn Barcelona. This work has been made possible thanks to the financial 57:3045–3053 support of the ERC Consolidator, Integrated System Analysis of Urban Compagnucci RH, Araneo D, Canziani PO (2001) Principal sequence Vegetation and Agriculture (818002-URBAG). pattern analysis: a new approach to classifying the evolution of atmospheric systems. Int J Climatol 21:197–217. https://​doi.​org/​ Data availability The datasets generated during and/or analyzed during 10.​1002/​joc.​601 the current study are available in the URBAG-ICTA PSPA_HW reposi- de Rigo D, Libertà G (2017) Tracy Houston Durrant. Tomàs Artés tory, [https://​github.​com/​URBAG-​ICTA/​PSPA_​HW.​git]. Vivancos. Jesús San-MiguelAyanz. Forest fire danger extremes in Europe under climate change: variability and uncertainty. Declarations [Research Report] Publications Office of the European Union. ffhal-02906196f Conflict of interest The authors declare that they have no known com- Della-Marta PM, Luterbacher J, von Weissenfluh H, Xoplaki E, peting financial interests or personal relationships that could have ap- Brunet M, Wanner H (2007) Summer heat waves over western peared to influence the work reported in this paper. Europe 1880–2003, their relationship to large-scale forcings and predictability. Clim Dyn 29(2):251–275 Open Access This article is licensed under a Creative Commons Attri- Díaz J, Linares C, Carmona R, Russo A, Ortiz C, Salvador P, Trigo bution 4.0 International License, which permits use, sharing, adapta- RM (2017) Saharan dust intrusions in Spain: health impacts tion, distribution and reproduction in any medium or format, as long and associated synoptic conditions. Environ Res 156:455–467 as you give appropriate credit to the original author(s) and the source, D’Ippoliti D, Michelozzi P, Marino C et al (2010) The impact of provide a link to the Creative Commons licence, and indicate if changes heat waves on mortality in 9 European cities: results from the were made. The images or other third party material in this article are EuroHEAT project. Environ Health 9:37. https://​d oi.​o rg/​1 0.​ included in the article's Creative Commons licence, unless indicated 1186/​1476-​069X-9-​37 otherwise in a credit line to the material. If material is not included in Donat MG, Alexander LV, Herold N, Dittus AJ (2016) Temperature the article's Creative Commons licence and your intended use is not and precipitation extremes in century-long gridded observa- permitted by statutory regulation or exceeds the permitted use, you will tions, reanalyses, and atmospheric model simulations. J Geo- need to obtain permission directly from the copyright holder. To view a phys Res Atmos 121(19):11–174 copy of this licence, visit http://​creat​iveco​mmons.​org/​licen​ses/​by/4.​0/. Duchêne F, Van Schaeybroeck B, Caluwaerts S, De Troch R, Hamdi R, Termonia P (2020) A statistical-dynamical methodology to downscale regional climate projections to urban scale. J Appl Meteorol Climatol 59:1109–1123. https://​d oi.​o rg/​1 0.​1 175/​ JAMC-D-​19-​0104.1 References Engdaw M, Ballinger A, Hegerl G, Steiner A (2021) Changes in temperature and heat waves over Africa using observational Aickin M, Gensler H (1996) Adjusting for multiple testing when and reanalysis data sets. Int J Climatol. https://​doi.​org/​10.​1002/​ reporting research results: the Bonferroni vs Holm methods. Am joc.​7295 J Public Health 86(5):726–728 Escobar G, Compagnucci RH, Bishoff S (2004) Sequences patterns of Amengual A, Homar V, Romero R, Alonso S, Ramis C (2012) A statis- 1000 hPa and 500 hPa geopotential height fields associated with tical adjustment of regional climate model outputs to local scales: cold surges over Central Argentina. Atmósfera [online] 17: 69–89. application to Platja de Palma Spain. J Clim 25(3):939–957 Available in: https://​www.​redal​yc.​org/​artic​ulo.​oa? An N, Zuo Z (2021) Investigating the influence of synoptic circu- Esteban P (2008) Circulation pattern sequences applied to forest lation patterns on regional dry and moist heat waves in North fires and avalanches over the Pyrenees. Scientific report from China. Clim Dyn 57:1227–1240. https:// ​ d oi. ​ o rg/ ​ 1 0. ​ 1 007/​ STSM-COST733 s00382-​021-​05769-x Fischer EM, Seneviratne SI, Vidale PL, Lüthi D, Schär C (2007) Soil Aran M, Peña JC, Torà M (2011) Atmospheric circulation patterns moisture–atmosphere interactions during the 2003 European sum- associated with hail events in Lleida (Catalonia). Atmos Res mer heat wave. J Clim 20(20):5081–5099 100:428–438. https://​doi.​org/​10.​1016/j.​atmos​res.​2010.​10.​029 Gaetano J (2013) Holm–Bonferroni sequential correction: an EXCEL Barriopedro D, Fischer EM, Luterbacher J, Trigo RM, Garcia-Herrera calculator [Microsoft Excel workbook] R (2011) The hot summer of 2010: redrawing the temperature García-Ortega E, Trobajo MT, López L, Sánchez JL (2011) Synoptic record map of Europe. Science 332(6026):220–224. https://​doi.​ patterns associated with wildfires caused by lightning in Castile org/​10.​1126/​scien​ce.​12012​24 13 Analysis of synoptic weather patterns of heatwave events and Leon Spain. Nat Hazards Earth Syst Sci 11(3):851–863. Peninsula. J Geophys Res Atmos. https://​doi.​org/​10.​1029/​2011j​ https://​doi.​org/​10.​5194/​nhess-​11-​851-​2011 d0165​76 Gil V, Gaertner MA, Gutierrez C, Losada T (2019) Impact of cli- Jolliffe IT, Cadima J (2016) Principal component analysis: a review mate change on solar irradiation and variability over the Ibe- and recent developments. Philos Trans R Soc A Math Phys Eng rian Peninsula using regional climate models. Int J Climatol Sci 374(2065):2050202. https://​doi.​org/​10.​1098/​rsta.​2015.​0202 39(3):1733–1747 Kendall MG (1975) Rank correlation methods, 4th edn. Charles Grif- Gilabert J, Ventura S, Segura R, Martilli A, Badia A, Llasat C, Corbera fin, London J, Villalba G (2021) Abating heat waves in a coastal Mediterra- Lau N-C, Nath MJ (2014) Model Simulation and Projection of Euro- nean city: What can cool roofs and vegetation contribute? Urban pean Heat Waves in Present-Day and Future Climates. J Clim Clim 37:100863. https://​doi.​org/​10.​1016/j.​uclim.​2021.​100863 27(10):3713–3730. https://​doi.​org/​10.​1175/​JCLI-D-​13-​00284.1 Government of Catalonia (2017) Institute Of Catalan Studies. Climate Le Roy B, Lemonsu A, Schoetter R (2021) A statistical–dynamical change in Catalonia: Executive summary of the Third Report on downscaling methodology for the urban heat island applied to the Climate Change in Catalonia. Writing team: Xavier Duran, M. EURO-CORDEX ensemble. Clim Dyn 56:2487–2508. https://d​ oi.​ Josep Picó and Lluís Reales. Edited by Arnau Queralt. Translated org/​10.​1007/​s00382-​020-​05600-z by t&s – Traducciones y Tratamiento de la Documentación, S. Lhotka O, Kyselý J, Plavcová E (2018) Evaluation of major heat waves’ L. Barcelona: Government of Catalonia and Institute of Catalan mechanisms in EURO-CORDEX RCMs over Central Europe. Studies. Also available online at: cads.gencat.cat Clim Dyn 50(11):4249–4262 Grimmond CSB, Oke TR (1999) Aerodynamic properties of urban Lionello P, Scarascia L (2018) The relation between climate change areas derived from analysis of surface form. J Appl Meteorol in the Mediterranean region and global warming. Reg Environ 38:1262–1292 Change 18(5):1481–1493 Guarino MV, Martilli A, Di Sabatino S, Leo LS (2014) Modelling Liu YY, van Dijk AIJM, McCabe MF, Evans JP, de Jeu RAM (2013) the urban boundary-layer over a typical mediterranean city using Global vegetation biomass change 1988–2008 and attribution to WRF: assessment of UHI and thermal comfort. In: American environmental and human drivers. Global Ecol Biogeogr 22:692– Society of Mechanical Engineers, Fluids Engineering Divi- 705. https://​doi.​org/​10.​1111/​geb.​12024 sion (Publication) FEDSM. https://​doi.​org/​10.​1115/​FEDSM​ Liu X, Tian G, Feng J, Ma B, Wang J, Kong L (2018) Modeling the 2014-​21572. warming impact of urban land expansion on hot weather using the Hoffmann P, Krueger O, Schlünzen KH (2011) A statistical model weather Research and forecasting model: a case study of Beijing for the urban heat island and its application to a climate change China. Adv Atmos Sci 35(6):723–736. https://​doi.​org/​10.​1007/​ scenario. Int J Climatol 32(8):1238–2124 s00376-​017-​7137-8 Hoinka KP, Castro MD (2003) The Iberian Peninsula thermal low. Q López-Bueno JA, Díaz J, Follos F et al (2021) Evolution of the thresh- J R Meteorol Soc 129(590):1491–1511 old temperature definition of a heat wave vs evolution of the Holm S (1979) A simple sequentially rejective multiple test procedure. minimum mortality temperature: a case study in Spain during the Scand J Stat 6:65–70 1983–2018 period. Environ Sci Eur 33:101. https://​doi.​org/​10.​ Hotelling H (1933) Analysis of a complex of statistical variables into 1186/​s12302-​021-​00542-7 principal components. J Educ Psychol 24(6):417 Maher P, Kelleher ME, Sansom PG et al (2020) Is the subtropical jet Houssos EE, Lolis CJ, Bartzokas A (2008) Atmospheric circulation shifting poleward? Clim Dyn 54:1741–1759. https://​doi.​org/​10.​ patterns associated with extreme precipitation amounts in Greece. 1007/​s00382-​019-​05084-6 Adv Geosci 6:5–11. https://​doi.​org/​10.​5194/​adgeo-​17-5-​2008 Mann HB (1945) Non-parametric tests against trend. Econometrica Hussain et al (2019) pyMannKendall: a python package for non para- 13:163–171 metric Mann Kendall family of trend tests. J Open Sourc Softw Martin JE (2021) Recent trends in the Waviness of the Northern Hem- 4(39):1556. https://​doi.​org/​10.​21105/​joss.​01556 isphere Wintertime Polar and Subtropical Jets. J Geophys Res Huth R, Beck C, Philipp A, Demuzere M, Ustrnul Z, Cahynová M, Atmos. https://​doi.​org/​10.​1029/​2020J​D0336​68 Kyselý J, Tveito OE (2008) Classifications of atmospheric Cir- Mazon J, Pino D, Barriendos M (2014) Rapid and sudden advection culation Patterns. Recent advances and applications. Trends and of warm and dry air in the Mediterranean Basin. Nat Hazard directions in climate research. Ann N Y Acad Sci 1146:105–152. 14(2):235–246 https://​doi.​org/​10.​1196/​annals.​1446.​019 McMichael AJ, Lindgren E (2011) Climate change: present and future IPCC (2014) Climate Change 2014: Synthesis Report. Contribution risks to health and necessary responses. J Intern Med 270:401–413 of Working Groups I, II and III to the Fifth Assessment Report MedECC (2020) Climate and environmental change in the Mediter- of the Intergovernmental Panel on Climate Change [Core Writ- ranean Basin—current situation and risks for the future. First ing Team, R.K. Pachauri and L.A. Meyer (eds)]. IPCC, Geneva, Mediterranean Assessment Report [Cramer W, Guiot J, Marini Switzerland, p 151 K (eds)] Union for the Mediterranean. Plan Bleu. UNEP/MAP. Jacobeit J, Philipp A, Nonnenmacher M (2006) Atmospheric circula- Marseille. France. 632pp. ISBN: 978-2-9577416-0-1, https://​doi.​ tion dynamics linked with prominent discharge events in Central org/​10.​5281/​zenodo.​47688​33 Europe. Hydrol Sci J 51:946–965. https://​doi.​org/​10.​1623/​hysj.​ Miró JR, Pepin N, Peña JC, Martin-Vide J (2020) Daily atmospheric 51.5.​946 circulation patterns for Catalonia (northeast Iberian Peninsula) Basara JB, Basara HG, Illston BG, Crawford KC (2010) The Impact using a modified version of Jenkinson and Collison method. of the Urban Heat Island during an Intense Heat Wave in Okla- Atmos Res 231:104674. https://​doi.​org/​10.​1016/j.​atmos​res.​2019.​ homa City. Adv Meteorol 2010:10. https://​doi.​org/​10.​1155/​2010/​ 104674 230365. (Article ID 230365) Mohammed AJ, Alarcón M, Pino D (2018) Eventos de temperatura Jenkinson AF, Collison BP (1977) An initial climatology of gales over extrema en la península ibérica: análisis estadístico de trayecto- the North Sea. In: Synoptic Climatololgy Branch Memorandum. rias y patrones sinópticos. Revista Internacional De Climatología. vol. 62 Meteorological Office, London, UK, p 18 https://​doi.​org/​10.​1002/​joc.​5733 Jerez S, Montavez JP, Gomez-Navarro JJ, Jimenez PA, Jimenez-Guer- Morris KI, Chan A, Morris KJK, Ooi MCG, Oozeer MY, Abakr YA, rero P, Lorente R, Gonzalez-Rouco JF (2012) The role of the Al-Qrimli HF (2017) Impact of urbanization level on the interac- land-surface model for climate change projections over the Iberian tions of urban area the urban climate and human thermal comfort. 13 S. Ventura et al. Appl Geogr 79:50–72. https://​doi.​org/​10.​1016/j.​apgeog.​2016.​12.​ Shiva Nagendra SM, Khare M (2003) Principal component analysis of 007 urban traffic characteristics and meteorological data. Transp Res Ozturk T, Matte D, Christensen JH (2021) Robustness of future atmos- Part d: Transp Environ 8(4):285–297 pheric circulation changes over the EURO-CORDEX domain. Sioutas MV, Flocas HA (2003) Hailstorms in Northern Greece: syn- Clim Dyn. https://​doi.​org/​10.​1007/​s00382-​021-​06069-0 optic patterns and thermodynamic environment. Theoret Appl Pearson K (1901) XI. Mathematical contributions to the theory of Climatol 75(3):189–202 evolution—X. Supplement to a memoir on skew variation. Philos Sousa PM, Barriopedro D, Ramos AM, García-Herrera R, Espírito- Trans R Soc Lond Ser A 197(287–299):443–459 (Containing Santo F, Trigo RM (2019) Saharan air intrusions as a relevant papers of a mathematical or physical character) mechanism for Iberian heatwaves: the record-breaking events of Peña JC, Aran M, Cunillera J, Amaro J (2011) Atmospheric circula- August 2018 and June 2019. Weather Clim Extrem 26:100224 tion patterns associated with strong wind events in Catalonia. Nat Streiner DL, Norman GR (2011) Correction for multiple testing: is Hazard 11(1):145–155 there a resolution? Chest 140(1):16–18 Peña D, González Hidalgo JC, Trigo R, Cortesi N (2016) La variabili- Streutker DR (2002) A remote sensing study of the urban heat island dad espacial de las temperaturas en la Península Ibérica explicada of Houston Texas. Int J Remote Sens 23(13):2595–2608. https://​ por los tipos de tiempo de circulación zonal: este y oeste. Un doi.​org/​10.​1080/​01431​16011​01150​23 posible efecto Foehn a escala peninsular Tadić L, Bonacci O, Brleković T (2019) An example of principal Peña JC, Aran M, Raso JM et al (2015) Principal sequence pattern component analysis application on climate change assessment. analysis of episodes of excess mortality due to heat in the Barce- Theor Appl Climatol 138:1049–1062. https://​doi.​org/​10.​1007/​ lona metropolitan area. Int J Biometeorol 59:435–446. https://d​ oi.​ s00704-​019-​02887-9 org/​10.​1007/​s00484-​014-​0857-x Tomczyk AM, Półrolniczak M, Bednorz E (2017) Circulation con- Perkins-Kirkpatrick SE, Lewis SC (2020) Increasing trends in regional ditions’ effect on the occurrence of heat waves in western and heatwaves. Nat Commun 11:3357. https://​d oi.​o rg/​1 0.​1 038/​ southwestern Europe. Atmosphere 8(2):31 s41467-​020-​16970-7 Tuel A, Eltahir EA (2020) Why is the Mediterranean a climate change Philipp A (2009) Comparison of principal component and cluster anal- hot spot? J Clim 33(14):5829–5843 ysis for classifying circulation pattern sequences for the European Turco M, Llasat MC, von Hardenberg J et al (2014) Climate change domain. Theor Appl Climatol 96:31–41. https://​doi.​org/​10.​1007/​ impacts on wildfires in a Mediterranean environment. Clim s00704-​008-​0037-1 Change 125:369–380. https://d​ oi.o​ rg/1​ 0.1​ 007/s​ 10584-0​ 14-1​ 183-3 Portela A, Castro M (1996) Summer thermal lows in the Iberian United Nations (2020) World cities report 2020. The value of sustain- Peninsula: a three-dimensional simulation. Q J R Meteorol Soc able urbanization. https://​doi.​org/​10.​18356/​27bc3​1a5-​en. Urbani- 122(529):1–22 zation and Cities: Trends of a New Global Force: pp 1–42 Richman MB (1986) Rotation of principal components. J Climatol Wang X, Jiang D, Lang X (2017) Future extreme climate changes 6:293–335. https://​doi.​org/​10.​1002/​joc.​33700​60305 linked to global warming intensity. Sci Bull 62(24):1673–1680 Santamouris M (2014) Cooling the cities—a review of reflective and Ward JH Jr (1963) Hierarchical grouping to optimize an objective func- green roof mitigation technologies to fight heat island and improve tion. J Am Stat Assoc 58(301):236–244 comfort in urban environments. Sol Energy 103:682–703. https://​ WMO: World Meteorological Organization (2016) Technical Regula- doi.​org/​10.​1016/j.​solen​er.​2012.​07.​003 tions, Basic Documents No. 2, Volume I—General Meteorologi- Santer BD et al (2004) Identification of anthropogenic climate change cal Standards and Recommended Practices (WMO-No. 49). 2015 using a second-generation reanalysis. J Geophys Res 109:D21104. edition, updated in 2016. Geneva https://​doi.​org/​10.​1029/​2004J​D0050​75 WMO: World Meteorological Organization (2021) Global annual to Scaife AA, Woollings T, Knight J, Martin G, Hinton T (2010) Atmos- decadal climate update. Target years: 2021 and 2021–2025 pheric blocking and mean biases in climate models. J Clim Wold S, Esbensen K, Geladi P (1987) Principal component analysis. 23(23):6143–6152 Chemom Intell Lab Syst 2(1–3):37–52 Segura R, Badia A, Ventura S, Gilabert J, Martilli A, Villalba G (2021) Zhang X, Vincent LA, Hogg WD, Niitsoo A (2000) Temperature and Sensitivity study of PBL schemes and soil initialization using the precipitation trends in Canada during the 20th century. Atmos WRF-BEP-BEM model over a Mediterranean coastal city. Urban Ocean 38:395–429 Clim 39:100982. https://​doi.​org/​10.​1016/j.​uclim.​2021.​100982 Zong L, Yang Y, Xia H, Gao M, Sun Z, Zheng Z, Lolli S (2022) Joint Segura R, Scott Krayenhoff E, Martili A, Badia A, Estruch C, Ventura occurrence of heatwaves and ozone pollution and increased health S, Villalba G (2022) How do street trees affect urban temperatures risks in Beijing, China: role of synoptic weather pattern and and radiation exchange? Observations and numerical evaluation urbanization. Atmos Chem Phys 22(10):6523–6538 in a highly compact city. Urban Clim 46:101288. https://​doi.​org/​ Zschenderlein P, Fink AH, Pfahl S, Wernli H (2019) Processes deter- 10.​1016/j.​uclim.​2022.​101288 mining heat waves across different European climates. Q J R Serrano-Notivoli R, Lemus-Canovas M, Barrao S, Sarricolea P, Meseg- Meteorol Soc 145(724):2973–2989 uer-Ruiz O, Tejedor E (2022) Heat and cold waves in mainland Spain: origins, characteristics, and trends. Weather Clim Extrem Publisher's Note Springer Nature remains neutral with regard to 37:100471 jurisdictional claims in published maps and institutional affiliations. Sfica L, Croitoru AE, Iordache I, Ciupertea AF (2017) Synoptic condi- tions generating heat waves and warm spells in Romania. Atmos- phere 8(3):50 13