Mechanisms underlying temperature extremes in Iberia : a Lagrangian perspective

The mechanisms underlying the occurrence of temperature extremes in Iberia are analysed considering a Lagrangian perspective of the atmospheric flow, using 6-hourly ERA-Interim reanalysis data for the years 1979 2012. Daily 2-m minimum temperatures below the 1st percentile and 2-m maximum temperatures above the 99th percentile at each grid point over Iberia are selected separately for winter and summer. Four categories of extremes are analysed using 10-d backward trajectories initialized at the extreme temperature grid points close to the surface: winter cold (WCE) and warm extremes (WWE), and summer cold (SCE) and warm extremes (SWE). Air masses leading to temperature extremes are first transported from the North Atlantic towards Europe for all categories. While there is a clear relation to large-scale circulation patterns in winter, the Iberian thermal low is important in summer. Along the trajectories, air mass characteristics are significantly modified through adiabatic warming (air parcel descent), upper-air radiative cooling and near-surface warming (surface heat fluxes and radiation). High residence times over continental areas, such as over northern-central Europe forWCE and, to a lesser extent, over Iberia for SWE, significantly enhance these air mass modifications. Near-surface diabatic warming is particularly striking for SWE. WCE and SWE are responsible for the most extreme conditions in a given year. For WWE and SCE, strong temperature advection associated with important meridional air mass transports are the main driving mechanisms, accompanied by comparatively minor changes in the air mass properties. These results permit a better understanding of mechanisms leading to temperature extremes in Iberia.


Introduction
Temperature extremes, like heat waves and cold spells, have a strong impact on the environment and society (Alexander et al., 2006;Hanson et al., 2007;Seneviratne et al., 2012), and their occurrence in Europe is commonly associated with anomalous large-scale atmospheric conditions (e.g. Santos and Corte-Real, 2006;Cony et al., 2008;Efthymiadis et al., 2011;Andrade et al., 2012;Moore and Renfrew, 2012;Pfahl and Wernli, 2012). The occurrence of temperature extremes has been reported to increase in recent decades (e.g. Tank and Ko¨nnen, 2003;Seneviratne et al., 2014), a trend that is also projected for future decades (IPCC, 2013;Kharin et al., 2013), also for Europe (Cattiaux et al., 2012;Frı´as et al., 2012). In particular, the occurrence of heat waves is expected to increase in terms of their frequency, intensity and duration (Fischer and Scha¨r, 2010;Seneviratne et al., 2012). Owing to the strong impacts of weather extremes on many socio-economic sectors worldwide (Cooney, 2012;Field et al., 2012;Murray and Ebi, 2012;Orlowsky and Seneviratne, 2012), studying their underlying processes is of foremost relevance.
The Iberian Peninsula (IP) is an area frequently affected by such extremes. Moreover, an increase in frequency and intensity is projected for future decades (Ramos et al., 2011;Andrade et al., 2014). The occurrence of temperature extremes in the IP tends to be driven by large-scale circulation anomalies, particularity in the winter half of the year and over the western half of the peninsula, when and where a strong linkage exists to the North Atlantic Oscillation (NAO) and East Atlantic (EA) pattern (e.g. Ulbrich et al., 1999;Trigo et al., 2002Trigo et al., , 2004Garcı´a-Herrera et al., 2007;Santos et al., 2007). The location of the North Atlantic jet stream, with three preferred regimes (northern, central and southern regimes after Woollings et al., 2010), has been shown to strongly impact temperature extremes in Europe, also with significant signatures in Iberia (Mahlstein et al., 2012). For instance, the central regime is mostly unfavourable for temperature extremes in Iberia, while they occur primarily in the northern and southern regimes (e.g. winters of 2010 and 2012; cf. Santos et al., 2013b). Based on the aforementioned studies, temperature advection, radiation and surface fluxes are all expected to play a key role for the occurrence of temperature extremes over the IP. However, their relative weight has not yet been determined for the different types of extreme (cold/warm) and in the different seasons (winter/summer).
The Lagrangian perspective on the formation of temperature extremes permits a more quantitative understanding of the associated physical mechanisms (Bieli et al., 2014). Calculating backward trajectories of surface air masses associated with extreme temperature episodes (Lagrangian approach) enables not only to identify their origin, but also the physical mechanisms that occur along their pathway and the modification of their thermodynamic properties (Wernli and Davies, 1997). The temperature of a given air parcel can change through: (1) adiabatic cooling/warming, related to ascending/descending motion; and (2) diabatic fluxes that comprise latent and sensible heat fluxes, shortwave/solar radiation and long-wave/thermal radiation (Holton, 2004). The temperature at a given site (Eulerian approach) can also change through (horizontal and vertical) temperature advection (Holton, 2004).
The primary aim of this study is to identify the main physical mechanisms leading to temperature extremes affecting the IP. This provides not only a better understanding of the underlying physical processes, but also helps assess the skill of climate models in replicating the natural processes driving weather extremes, either in weather forecasting or decadal/long-term climate projections (Solomon et al., 2011). The main focus of the analysis is the continental area of the IP (Fig. 1). However, the strong contrast between the North Atlantic-facing (northern and western) and the Mediterranean-facing (eastern and southern) regions of the IP, which is significantly enhanced by several mountain ranges and by the latitude range from 35 to 448N (between subtropical highs and mid-latitude lows), leads to a large diversity of climatic conditions across the peninsula (Tullot, 2000;Santos et al., 2013a). Therefore, the regional climatic specificities of the IP also justify the study of subsectors, as the mechanisms underlying temperature extremes might be different between, for example, eastern and western Iberia. The structure of the paper is as follows: data and methods are described in Section 2, results are presented in Section 3 and a summary of the main conclusions and their discussion are provided in Section 4. Fig. 1. Orographic map of the Iberian Peninsula (altitude in m), along with the main mountain ranges, and the subsectors for regional assessments (see also Table 1).

Data and methods
Following Bieli et al. (2014), the Lagrangian analysis tool LAGRANTO (Wernli and Davies, 1997) is applied to calculate 10-d backward trajectories of surface air masses associated with temperature extremes in the IP. With this purpose, all atmospheric variables are retrieved every 6 hours from the ERA-Interim reanalysis dataset of the European Centre for Medium-Range Weather Forecasts (Dee et al., 2011), for the period from 1979 to 2012 (34 yr) on a 1.08 latitude )1.08 longitude grid (ca. 80Á100 km grid spacing). Temperature extremes are considered at 91 grid points in the Iberian sector, which we define as the region from 37 to 438N and 98W to 38E, that is, with seven grid points in the meridional and 13 in the zonal direction. Six-hourly cold/ warm extremes are defined at each grid point for 6-hourly minimum/maximum 2-m air temperature (T2MIN/T2MAX) below/above the 1st/99th percentile of the full empirical distribution at that grid point. The use of these percentilebased indices is widespread in climate research and particularly suitable for isolating extremes over regions with very heterogeneous climates (Zhang et al., 2011), such as the IP (Tullot, 2000;AEMET/IMP, 2011). Back-trajectories are then computed in order to identify the origin of air masses associated with the extreme events. They are started at the time of the extreme episode onset, and air parcels are traced backwards in time. Thus, the trajectory start time corresponds to the onset of the extreme event, while their ending corresponds to their earliest stage (several days before the onset of the event).
Owing to strong seasonality, cold and warm extremes are considered separately for winter (DJF) and summer (JJA). Hence, the number of extremes at each grid point is 1% of 34 yr)90/92 d )4 times a day, that is, 122/125 extremes of each type for winter/summer. Thus, for the whole sector (91 grid points), there is a sample of 11 102 local cold/warm extremes for winter and 11375 for summer. Trajectories were started at three pressure levels (10-, 30-and 50-hPa above the surface) at the grid points with an identified temperature extreme, which accounts for vertical variations in the boundary layer (Bieli et al., 2014). This approach inherently takes into account vertical mixing in the boundary layer at the location of the extreme event, which is not explicitly resolved in the 6-hourly ERA-Interim data. Because extremes typically occur simultaneously at multiple nearby locations and sometimes during several 6-hourly periods of the same day, the total number of non-simultaneous daily events is effectively much lower than the number of tracked trajectories: 498 for winter cold extremes (WCE), 562 for winter warm extremes (WWE), 490 for summer cold extremes (SCE) and 983 for summer warm extremes (SWE). Although some events are represented by a relatively large number of trajectories (large-scale cold/warm events that last for several time steps), the total number of independent events per season is high enough such that single days do not dominate the overall results. While the winter extremes (WCE and WWC) have a comparable number of distinct days, the number of distinct daily SWE episodes is almost three times as large as for SCE. Given the seasonality, WCE and SWE are typically responsible for the most extreme conditions in a given year.
Following Bieli et al. (2014), several variables are tracked (using linear interpolation in space and time) and analysed over the 10-d period preceding the extreme events in order to characterize the trajectories. Latitude, longitude and pressure jointly determine the location of the air mass at a given instant. For the thermodynamic characterization of the air parcel, temperature, potential temperature (theta), specific humidity and relative humidity are tracked. Furthermore, the following surface/near-surface variables beneath the trajectory are also considered: 2-m minimum (T2MIN) and maximum (T2MAX) temperatures, and anomaly indices (following Bieli et al., 2014) of surface sensible heat flux (SSHF), surface latent heat flux (SLHF), surface thermal radiation (STR) and surface solar radiation (SSR). In the present study, positive SLHF, SSHF and STR are directed from the surface to the atmosphere, while positive SSR is directed towards the surface. Daily anomalies of these surface fluxes with respect to the local climatology (dimensionless anomalies) are calculated by subtracting the local mean value of each 6-hourly time step and dividing through the sum of the 6-hourly standard deviations (cf. Bieli et al., 2014). Positive anomalies indicate more intense fluxes compared to climatology (SLHF, SSHF and STR) or enhanced solar radiation (SSR). Due to the construction of the anomaly indices, values between 0.5 and 1 already represent relatively strong anomalies compared to the climatology. The representation of surface fluxes in reanalysis data still has important limitations (Decker et al., 2012) and thus the results concerning the fluxes must be interpreted with care.
In order to assess regional heterogeneities in the temperature extremes, a sectorial analysis is also carried out for five subsectors ( Fig. 1 and Table 1). A higher number of subsectors could better capture regional climate heterogeneities, but the outcomes would become less robust due to the small sample sizes.

Winter cold extremes
The WCE episodes affecting the IP are fundamentally associated with northeasterly transport of very cold and dry air masses from central-northern Europe, reaching the IP mostly through the western-central Pyrenees and the northern (Cantabrian) coast (Fig. 2aÁc). Their paths exhibit strong similarity, which is highlighted by the high densities observed over the continent up to 5 d before the onset of the episodes in the IP. Many trajectories also reveal earlier northerly/northwesterly paths over the high-latitude North Atlantic and northern Europe (Fig. 2a). Streamlines of the wind at 850 hPa are used to analyse the air flow and the representative synoptic-scale features in the lowertroposphere. The composite streamlines (averaged over all WCE episodes) highlight the presence of a fast-growing and strong anticyclonic ridge over the eastern North Atlantic, indicating an omega-like blocking to the west of the British Isles (Fig. 2aÁc). A closed low-pressure system develops over the central Mediterranean, reinforcing the northerly flow towards Iberia. North Atlantic atmospheric blocking has been related to the occurrence of European winter cold air outbreaks (Walsh et al., 2001;Sillmann et al., 2011). Despite being composites, the streamlines suggest a frequent occurrence of anticyclonic upper-level wavebreaking over Europe during these episodes, which is also in agreement with the upstream blocking occurrence (Woollings et al., 2008). These synoptic features are consistent with the prevalent northeasterly paths of the parcel trajectories, though streamlines are not coincident with parcel trajectories in a non-stationary flow. These paths are also reflected in the corresponding latitudeÁlongitude plot of the median trajectory ( Fig. 2d), also for the subsectors of the IP (cf. Fig. 1 and Table 1). The air flows towards northern Europe at about 608N and then sharply bends southwestward, with a total zonal (meridional) displacement over Europe of ca. 158 (208), hinting at the highly advective nature of the WCE episodes. While the translation speed is quite low during the first half of the 10-d period, a rapid displacement is noteworthy in the last 3 d.
Despite some differences between sectors, there is a clear resemblance in the shape of their median paths. The more eastern median path for CE compared to SW can be largely explained by the geographical location of these two sectors, as trajectories for the former often pass the Pyrenees, whereas trajectories for the latter are mostly confined to the northern coast of the IP and western edge of the Pyrenees (not shown) and are thus moderated by a short period over the Bay of Biscay before entering the IP. As the continent is relatively cold compared to the ocean in winter, trajectories over the continent are warmed less by surface fluxes than over the ocean. Therefore, these continental northeasterly paths bring anomalously cold air masses into the IP.
In spite of inter-episode variability, a coherent descent of the air masses is observed along the trajectories (about 200 hPa in 10 d, Fig. 2e), resulting in a strong adiabatic warming of Â208C for the dry lapse rate, reinforcing the leading role played by horizontal advection in triggering these cold extremes. In addition, there is some diabatic cooling (steady decrease in potential temperature) not entirely offset by the abrupt but short-duration warming near the surface (Fig. 2f). For the WCE episodes, when the air parcels reach near-ground levels over the IP, 90% of the trajectories are associated with 2-m minimum temperatures approximately ranging from (108C to 58C (Fig. 2g), attesting the coldness of these episodes for Iberian standards. The median increase in the 2-m minimum temperature at the air parcel position over the 10-d period, from (108C (at high latitudes) to nearly 08C (in the IP), is coherent with the cold advection driven by the northeasterly flow. Note that Fig. 2f shows a strong warming of the considered air parcels due to adiabatic compression, whereas the associated 2-m temperature reveals a much weaker warming.
As the trajectories enter the boundary layer, they get in contact with the warmer surface and are affected by strong heat and radiation fluxes. This is supported by the highly positive anomalies of SSHF and STR (enhanced heat fluxes from the surface to the atmosphere) during the last 3 d (Fig. 2h). These strong STR anomalies are associated with cold nights with clear sky and strong radiative cooling that typically occur for WCE. The gradual diabatic cooling of the air masses (through net radiative deficits at high latitudes) is also manifested in negative anomalies of SSR during the first 5 d (Fig. 2h). The median air parcel temperature increases by about 208C in 10 d (Fig. 2f), which is in agreement with descent and adiabatic warming found for other European regions (Bieli et al., 2014). The final temperatures are much higher for SW (38C) than for CE ( (48C) (Fig. 2f), which can be explained by the stronger maritime influence in the SW compared to CE. Note the much weaker diabatic cooling in SW than in CE (Fig. 2f). In effect, most SW trajectories only enter the IP after travelling a short period over the North Atlantic, and typically have more westerly pathways (Fig. 2d), even though they experience an additional diabatic cooling (Fig. 2f) when travelling over inner Iberia. A similar near-surface diabatic cooling is also observed for CW, but much less pronounced. The continental path of the trajectories for WCE episodes is also manifested by very low specific humidity values (B3 g kg (1 ), even at near-surface levels (Fig. 6a). Furthermore, relative humidity undergoes a slight increase and presents relatively high values, despite some decrease at the latest stage of the trajectory due to near-surface  Fig. 1 and Table 1). The corresponding 5th/95th percentiles for the IS are also displayed in the chronograms (dashed lines in e and g). Black circles in (d) indicate days (from 10 days before onset to onset), while the grey circle in (f) indicates the trajectory at the episode onset. (h): Medians of SLHF, SSHF, STR and SSR anomalies (dimensionless) along the air parcel trajectories (positive SLHF, SSHF and STR directed from surface to atmosphere, positive SSR directed towards the surface).
warming. In general, the increase in specific humidity is largely balanced by the air parcel warming, resulting in only slight changes in relative humidity.

Winter warm extremes
The WWE episodes affecting the IP can be mainly attributed to westerly/southwesterly flow regimes (Fig. 3a), transporting warm air masses from the subtropical North Atlantic, mostly from the latitude belt of 25Á408N. Many of these trajectories also flow towards Northwestern Africa before entering the IP, as suggested by the high trajectory densities over that region (Fig. 3a). Despite the apparent advective nature of the WWE, in the latest stage the trajectories show clear stationarity (Fig. 3b and c) with important residence times over the inner IP. The corresponding streamlines reveal that a strong anticyclonic flow over Northwestern Africa plays a central role in generating these episodes, as they induce strong southwesterly winds over Iberia (Fig. 3aÁc). The large median longitudinal displacement (approximately 608) is much more pronounced than the corresponding meridional displacement (of only about 58, Fig. 3d). This result highlights that these air parcels are transported by the prevailing westerly winds over the North Atlantic at nearly constant velocity, and then decelerated over the IP. The paths for the extremes in the five subsectors are clearly similar, suggesting similar driving mechanisms of WWE episodes for the entire IP. The peculiar northward displacement at the latest stage of the trajectories for N and CW suggest that the air parcels are first transported to inner Iberia, as is in fact illustrated by the high density over that region 1 d before onset (Fig. 3c). The air parcel subsidence is relatively strong (250 hPa in 10 d, Fig. 3e), as well as the associated adiabatic warming (almost 308C). Over the 10 d of the median trajectories, there is significant diabatic cooling of almost 108C (Fig. 3f), which is, however, largely overcompensated by the aforementioned adiabatic warming through air parcel descent, eventually leading to a temperature increase of about 188C (Fig. 3f). The median of the collocated 2-m maximum temperature remains nearly constant and is relatively high (around 158C, Fig. 3g). Surface fluxes do not play a major role (Fig. 3h), except for the final stage of the trajectories, when they are nearly stationary over the IP at lowtropospheric levels. The strengthening of SLHF (enhanced evaporation) over Iberia is worth noting, as well as the enhanced fluxes of STR and SSR at the latest stage of the trajectories, indicating cloud-free conditions. For WWE episodes, there is a sharp increase in the specific humidity towards the latest stage of the trajectories, reaching median values as high as 6 g kg (1 at near-surface levels. However, owing to the relatively high temperatures during these episodes, the median of the relative humidity remains near 50% (Fig. 6b). This result also shows that the specific humidity increase is mostly compensated by air parcel warming.
Hence, the warm and cold winter extreme temperature episodes tend to arise due to totally different atmospheric flow regimes. While meridional transport plays a key role in triggering WCE episodes, zonal transport, followed by near stationarity over the IP, tends to favour WWE episodes. In both cases, adiabatic warming of the air parcels is predominant over the diabatic cooling along the trajectories, resulting in a net air parcel temperature increase of approximately 208C in 10 d. Stronger adiabatic warming for WWE is accompanied by stronger diabatic cooling. Near-surface diabatic warming through surface fluxes is more noticeable for WCE episodes, definitively contributing to a significant moderation of these cold spells, which are indeed much less severe in the IP than over the rest of Europe. WCE episodes are also characterized by cold and very dry (continental) air masses, whilst WWE episodes are associated with relatively warm and more humid (maritime) air masses.

Summer cold extremes
SCE episodes are typically associated with northwesterly flow from the high-latitude North Atlantic (Fig. 4aÁc). Unlike WCE episodes, the SCE trajectories are predominantly located over the ocean. Since the North Atlantic is colder than the IP in summer, cold extremes are naturally linked to maritime rather than continental pathways. Overall, synoptic-scale streamlines are in good agreement with trajectories, showing northwesterly winds towards Iberia (Fig. 4aÁc), in response to cyclonic troughs over Western Europe and westward shifted Azores high-pressure systems. In fact, cold-core upper tropospheric troughs over Western/ Northern Europe tend to be linked to these episodes (not shown), which is in agreement with the anomalously low near-surface temperatures during the SCE episodes. Both the meridional and zonal displacements of the trajectories are remarkable, roughly 208 and 408, respectively (Fig. 4d), also reflecting the strong advective origin of these extremes. Additionally, their motion is at nearly constant speed along the entire trajectory. Regarding the regional features, most trajectories come directly from the North Atlantic/Bay of Biscay, entering the IP through either the western coast (SW, CW and SE) or the northern coast (N and CE). A detailed analysis of the trajectory density patterns for each sector corroborates these results (not shown), but also highlights that an important number of trajectories for SE originate from the western Mediterranean. Two days before the SCE, the relatively cold air parcels surround the IP (Fig. 4b) and then gradually converge over the IP. This horizontal confluence of the air masses is driven by rising air motions over the innermost parts of the peninsula. Further, the relatively high trajectory density over the western Mediterranean 1Á2 d before the extreme event onset is mostly related to trajectories that recently entered the Mediterranean Basin from the North Atlantic over southern France (not shown).
The descent along the SCE trajectories is comparatively weak (100 hPa in 10 d), meaning that the adiabatic heating is also weak compared to the other categories (Fig. 4e), at most 158C in 10 d. This fact also contributes to establishing the SCE episodes. The median air parcel temperature increases by about 158C (Fig. 4f), which is due to the adiabatic warming enhanced by strong diabatic warming near the surface (Fig. 4f). The collocated 2-m minimum air tem-perature exhibits a steady increase, with median temperatures varying from nearly 2 to 158C, but with a strong cooling at the latest stage of the trajectories, when the air mass is over land (Fig. 4g). This evolution reflects the southward displacement of the air parcels over the North Atlantic, as sea surface temperatures progressively increase, emphasizing the key role of advection for the occurrence of SCE episodes. The nocturnal cooling over the IP at the latest stage of the trajectories also plays a role for the extremeness of the 2-m minimum temperature.
The aforementioned diabatic heating might be explained by important near-surface fluxes, which is clearly supported by positive anomalies in all fluxes (Fig. 4h), particularly from 5 to 1 d before SCE onset. Regarding the humidity parameters for the SCE episodes, the air parcel moistening is quite clear for the median trajectory, as it descends and approaches the surface, achieving a final (median) specific humidity of nearly 8 g kg (1 . Furthermore, relative humidity experiences a slow increase from 70% to 80%, followed by a steep decrease back to 70%. This behaviour is similar to the WCE episodes, but with slightly higher values ( Fig. 6a and c). Air parcel moistening is widely compensated by warming, explaining the weak changes in relative humidity. On the whole, the maritime characteristics of air parcels are noticeable, suggesting relatively cold and wet weather conditions during SCE episodes in the IP.

Summer warm extremes
The SWE episodes tend to be associated with westerly and northwesterly flow from the North Atlantic at very early stages, but mainly with easterly/northeasterly flow in the later stages of the trajectories, followed by comparatively high residence times over or near the IP (Fig. 5aÁc). The composite mean streamlines for these episodes demonstrate that anticyclonic circulation over Northwestern Africa generates southerly flow of continental tropical air masses towards the IP (Fig. 5aÁc). However, results of the Lagrangian approach do not corroborate this synoptic perspective: the highest trajectory densities are in fact found over surrounding seas, and very low densities actually over Northern Africa. It should be kept in mind that streamlines do not correspond to actual air parcel trajectories in transient flow regimes, indicating the added value of the Lagrangian approach for analysing these episodes.
The near stationarity of the air masses over the IP 1Á2 d before onset is clearly illustrated by the spiral trajectories for all subsectors (Fig. 5d). After moving 3Á4 d at nearly constant speed towards Iberia, they remain stationary over the peninsula during the last 6Á7 d before the episode onset. Further, most air masses associated with extreme heat over the IP approach the region from the north and not from the south. The high density of trajectories over the water surfaces surrounding the peninsula 1Á2 d before SWE onset is the footprint of shallow low-pressure systems (heat lows) that induce convergence (spiralling) flows towards the inner peninsula, similarly to the SCE episodes. The connection between these mesoscale systems and tem-perature extremes in Iberia has been discussed, for example, in Pfahl (2014). The increased trajectory density to the east and northeast is also consistent with the occurrence of extremes over Iberia in association with atmospheric blocking events over Europe. While in many European regions warm temperature extremes occur in the vicinity of a blocking anticyclone (Pfahl and Wernli, 2012), for the IP the anticyclones are typically shifted to the north-east, which may lead to easterly/northeasterly transport of warm, continental air masses (Pfahl, 2014). Furthermore, there is no clear southerly inflow from North Africa, but rather easterly/ southeasterly flow from the southern Mediterranean/ Gibraltar Straight area towards the inner IP, strengthened by the wind channelling effect. This is particularly clear for SW and CW (not shown).
Moderate to strong air mass subsidence contributes to important adiabatic heating of the air masses associated with SWE episodes (Fig. 5e). The temperature of the air parcel increases by about 258C (Fig. 5f), resulting from adiabatic warming in the first period (increased advective processes) and very strong and rapid diabatic warming during the 2 d before SWE onset (Fig. 5f). During this period of weak advection, the air masses acquire different thermodynamic characteristics, progressively warming, clearly expressed by the steep increase of the 2-m maximum temperature (Fig. 5g). Note also the increasing footprint of the diurnal cycle in the temperature evolution, as the trajectories enter the mixing layer (Fig. 5f). This behaviour is not surprising for 2-m maximum temperatures in summer. These results hint at strong surface diabatic heating during the last stage of the trajectories, also supported by the positive anomalies of STR and SSR (Fig. 5h).
The specific humidity evolution for SWE episodes is quite similar to that for SCE episodes ( Fig. 6c and d), as in both cases the trajectories are of maritime origin and are gradually moistened as they descend into the boundary layer. Nevertheless, specific humidity is higher for SWE than for SCE at the earliest stages of the trajectories, due to the lower latitudes and warmer air masses involved (Figs. 4 and 5). Conversely, their relative humidity evolution is radically different ( Fig. 6c and d). In fact, there is a remarkable decrease during the second half of the trajectory for the SWE episodes (last 5 d). This occurs in spite of the increase in specific humidity, stressing the strong warming during this final period.
Although the SCE and SWE episodes present fairly similar trajectory density patterns during the last days before the onset of the temperature extreme (a footprint of the Iberian summer heat low), the thermodynamic processes are fundamentally different. The origin of the trajectories is at much higher latitudes for SCE than for SWE, partially explaining the genesis of these episodes. Nonetheless, the much stronger adiabatic warming for SWE than for SCE clearly reinforces these differences in the air mass origins. Additionally, the longer residence times over the IP for SWE than for SCE yield much stronger diabatic warming by surface fluxes for SWE, enhancing the differences between these two contrasting categories. The SCE episodes are also associated with relatively cold and moist air masses, while the SWE episodes are characterized by very warm air masses, with high specific but low relative humidity.

Temporal variability
The occurrences of the extreme episodes are widely scattered over the entire period (1979Á2012; Fig. 7), except for SWE that shows a remarkable peak in summer 2003 (18% of all SWE occurrences). In fact, this summer, in particular August, was exceptionally warm, with a major heat wave and record-breaking temperatures over large areas of Europe (e.g. Scha¨r et al., 2004;Garcia-Herrera et al., 2010), including the IP. According to ERA-Interim 2-m temperatures, SWE episodes were particularly frequent in the period of 1Á13 August 2003, reaching a peak on 12 August, when nearly one third of the Iberian sector (cf. Section 2) recorded a SWE episode. The trajectory densities and corresponding streamline composites for all SWE episodes in summer 2003 display similar patterns to Fig. 5 (not shown), but with enhanced anticyclonic circulation over North Africa and the western-central Mediterranean, extending as a ridge towards northern France and diverting the westerly flow to northern Europe. Other summers with high daily SWE occurrences (not necessarily heat waves) were 1991 and 2012 (Fig. 7d). Relatively high daily occurrences are found during the winters of 1983, 1985, 1987, 2005 and 2012 for the WCE episodes (Fig. 7a), and in the winters of 1985, 1989, 1990, 1998 and 2000 for the WWE episodes (Fig. 7b). The SCE episodes were more frequent in the winters of 1984 and 1992 (Fig. 7c). Other extraordinary events include the 15 January 1985, when about 80% of the Iberian sector was under a WCE episode, and 6 June 1984, during which about 80% of the Iberian sector was under a SCE episode.

Summary and discussion
The present study explores the mechanisms underlying the occurrence of temperature extremes affecting the IP, using a Lagrangian investigation of the associated atmospheric flow. Only temperatures in the 1%-tails of the empirical distributions at each grid point are considered. Due to strong seasonality, winter and summer are analysed separately. Four temperature extremes are then selected: wintertime extremely cold and warm episodes (WCE and WWE, respectively) and summertime extremely cold and warm episodes (SCE and SWE). The WCE and SWE episodes correspond to the most extreme temperatures overall. The Lagrangian trajectories are isolated for each extreme and at each grid point in the IP domain, using the same methodology as in Bieli et al. (2014). For each episode type, their spatial densities (number of trajectories per km 2 ) are calculated over the North AtlanticÁEuropean sector up to 10 d before the episode onset. A complementary Eulerian description of the atmospheric flow is provided by the streamline composites of the 850 hPa wind fields for the temperature extreme days.
In winter, the WCE and WWE episodes are driven by clearly different atmospheric processes. Air parcel advection from the north governs the WCE episodes, whereas first eastward travelling air parcels that then become nearly stationary over the IP drive the WWE episodes. The 850 hPa streamlines for WCE hint at a quasi-stationary omega-like structure over the eastern North Atlantic that gradually strengthens, reaching its mature stage at the WCE onset, with an anticyclone over the British Isles and a cut-off low over Italy. These features suggest mid-latitude blocking over the eastern North Atlantic (e.g. Barriopedro et al., 2006), a negative phase of the EA pattern and anticyclonic Rossby wave-breaking over Europe (e.g. Woollings et al., 2008). The persistent southwestward transport to the IP driven by the blocking and downstream wave-breaking plays a key role in triggering these extremely cold episodes. Conversely, streamlines for the WWE episodes reveal strong southwesterly flow, due to a southwestward displaced Azores anticyclone and an anticyclonic circulation over Northwestern Africa. This pattern is clearly concomitant with the negative phase of the NAO (e.g. Hurrell et al., 2003).
For both extreme categories, adiabatic warming from air parcel descent tends to prevail over diabatic cooling along the trajectories. For WWE, the stronger adiabatic warming is partially offset by stronger diabatic cooling. For both categories, the net air parcel warming is approximately 208C in 10 d. For the WCE episodes, near-surface diabatic warming at the latest stage of the air parcel trajectories, mostly associated to SSHF and STR, considerably moderates the air mass extremeness. After the onset of the WCE, these processes further restore the extreme temperatures towards climatology. The humidity parameters suggest cold and dry continental air masses for WCE and warm and humid maritime air masses for WWE, which agrees with the respective trajectory paths.
For summer, both extreme episodes (SCE and SWE), particularly SWE, show clear footprints of the prevailing thermal low (low-tropospheric warm-core low-pressure system) over the IP. The heat low generates low-level wind jets along the coastal areas of the IP (Soares et al., 2014), which underlie the nearly circular shape of the trajectory densities. In contrast to winter, the association with teleconnection patterns, such as the NAO and EA, is less clear in summer, when the North Atlantic large-scale atmospheric circulation is generally much weaker ( Barnston and Livezey, 1987). However, while for SCE episodes the Azores anticyclone is clearly separated from the anticyclonic circulation over North Africa, depicting a trough in between over the IP, for SWE episodes these two anticyclonic circulations are almost merged (Fig. 5aÁc). As a result, northerly (southerly) flow prevails over the IP for SCE (SWE) episodes, with major implications in terms of the regional temperature. Furthermore, thermodynamic processes associated with SCE and SWE are quite different. First, trajectories originated at much higher latitudes for SCE than for SWE, which definitely contributes to the distinction between these two episodes. Second, there is a much stronger adiabatic warming for SWE (strong air parcel descent) than for SCE (weak descent). Third, the longer residence time of air parcels over land for SWE than for SCE significantly strengthens the near-surface diabatic warming, by surface fluxes and radiation, in the former. From the humidity parameters, it is shown that SCE episodes are characterized by cool maritime air masses, while the SWE episodes are characterized by very warm air masses with low relative humidity, despite the high moisture content. The results of the present study provide no evidence for trajectories coming from North Africa during SWE episodes, as suggested by the corresponding streamlines. In fact, their relatively high moisture content suggests maritime rather than continental origin, that is, high residence time over water surfaces surrounding the IP and not over North African dry land surfaces. These results also stress the significance of the Lagrangian description of the atmospheric flow for improving the current knowledge on temperature extremes.
The present study enabled a better understanding of the mechanisms underlying temperature extremes in Iberia from a large-scale perspective. The complex topography and resulting orographic effects in the IP, as well as the high landÁsea thermal contrasts in summer, play an important role in the development of regional mesoscale systems. The poor resolution of these systems by ERA-Interim (ca. 80Á100 km grid spacing) is an important limitation of the present study. Forthcoming studies could focus on a trajectory analysis on higher resolution grids (B10 km grid spacing), for example, as produced by Regional Climate Models (RCM) nested on ERA-Interim reanalysis (e.g. Soares et al., 2012) or on other higher resolution reanalysis datasets (Rienecker et al., 2011). Such a downscaling approach, which is very time consuming and beyond the scope of this first insight study, may enable more detailed path definitions and improve estimates of surface fluxes. They herein have shown to be critical for understanding the temperature extremes affecting the IP. Downscaling from global climate models may also be used to assess climate change signals on these mechanisms under future anthropogenic forcing (Kjellstrom et al., 2011), and thus also potentially contributing to a better risk management of climate extremes (Murray and Ebi, 2012) over Iberia.