The global mean near-surface air temperature has risen by 0.74 °C + 0.18 °C over the last century (1906–2005) (Trenberth et al., 2007). Whereas warming during the first half of the 20th century was moderate and partially explained by natural forcings (e.g. Stott et al., 2004), the rate of warming over the last 50 years has been high (Trenberth et al., 2007) and mainly attributed to human activity (Karoly et al., 2003; Meehl et al., 2004; Klein Tank et al., 2005; Solomon et al., 2007). This warming has not been temporally uniform, as a cooling phenomenon was observed from the 1950s to the 1970s. These low temperatures might have been related to a decrease in global solar radiation which masked the full magnitude of the greenhouse warming during that period (Wild et al., 2007). Subsequently, the sharpest rise in global temperature ever recorded was witnessed (Trenberth et al., 2007). Moreover, this warming was asymmetrical because minimum temperatures increased more rapidly than maximum ones, with a subsequent decrease in the global diurnal temperature range (Karl et al., 1993; Easterling et al., 1997; Vose et al., 2005).
Recently, there has been concern about the frequent and intense changes in extreme events resulting from anthropogenic influences on climate (Easterling et al., 2000; Karl and Trenberth, 2003; Trenberth et al., 2007). The study of the behaviour of these variations is of major interest if we consider the potential catastrophic impact they could have on ecosystems and human society (Karl and Easterling, 1999). During recent years, there has been much effort at international level aimed at developing a consensus on climate change indices primarily focussing on extremes (Alexander et al., 2006). At global scale, several studies have analysed temperature extremes, which have shown a decrease in intra-annual temperature variability associated with a decrease in the number of frost days and a rise in warm night-time temperatures (Christidis et al., 2005; Trenberth et al., 2007). Among others, similar results were observed in China, central, and southeast Asia (Manton et al., 2001; Zhai and Pan, 2003; Klein Tank et al., 2006), New Zealand, and Australia (Plummer et al., 1999; Salinger and Griffiths, 2001), Central and South America (Rusticucci and Barrucand, 2004; Vincent et al., 2005; Aguilar et al., 2005), the United States (DeGaetano and Allen, 2002); and for the European domain, in Scandinavia and the Nordic Seas (Tuomenvirta et al., 2000), Germany (Hundecha and Bárdossy, 2005) or peninsular Spain (Brunet et al., 2007).
Some European studies have detected a greater rise in warmer nights/days than a decrease in colder nights/days in summer during the 20th century (e.g. Klein Tank and Können, 2003; Moberg and Jones, 2005; Moberg et al., 2006). However, regional differences can be identified according to the magnitude of the trends of minimum and maximum temperatures (Moberg et al., 2006). For example, Brunet et al. (2007) observed higher rates of change in maximum temperatures than in minimum ones in Spain.
These changes could be partially linked with changes in the persistence of specific atmospheric circulation patterns. Thus, the dynamics of atmospheric circulation is an important factor with regard to understanding the variability of extreme warm temperatures and/or heat waves. Since the mid-1970s, the persistence of some specific atmospheric circulation patterns over Europe might be influencing the frequency and severity of heat and cold waves (Kysely, 2008). In general, the occurrence of heat waves in Europe is related to several types of anticyclone and southerly situations (e.g. Xoplaki et al., 2003b; Cassou et al., 2005; Maheras et al., 2006; Della-Marta et al., 2007). These situations are characterized by a positive radiative forcing resulting from low cloud cover, under a high-pressure system with warm air mass advection (Kysely and Huth, 2008; Kysely, 2008).
Extreme summer temperatures have a coercive impact upon human thermal biocomfort and health, but also on water and energy consumption, upon the spread of vector-borne infectious diseases, or on tourism, giving rise to substantial economic losses (e.g. Haines et al., 2000; McMichael and Woodruff, 2004; Garcia-Herrera et al., 2005; Matzarakis and Amelung, 2008). Furthermore, modelling studies clearly demonstrate the existence of a very likely future risk deriving from high temperature extremes associated with more extreme heat episodes (Clark et al., 2006; Meehl et al., 2007). The Mediterranean basin will be very vulnerable to climate changes resulting from the affluence of vast amounts of tourists in summer, especially in relation to water consumption and energy demand (e.g. Valor et al., 2001; Giannakopoulos et al., 2009).
Some evidence of the rise in extreme minimum temperatures involves the occurrence of tropical nights (TNs). Several consecutive extremely warm night-time temperatures might bring about the greatest impact on human health (Karl and Knight, 1997). Moreover, there is still a lack of studies relating climate indices such as TNs with atmospheric circulation patterns. Thus, the main objective of the present study is to investigate the temporal evolution of TNs for the whole Iberian Peninsula (IP) and for three subregions over the 1961–2007 period in summer, and the relationship thereof with changes in atmospheric circulation patterns over the Euro-Atlantic sector through Canonical Correlation Analysis (CCA). Section 2 shows the datasets used in this study, along with the methods used to perform the main analyses. In Section 3 we analyse the temporal evolution of TNs and discuss the trends in relation to atmospheric circulation changes. Finally, Section 4 presents the conclusions of our paper.
2. Data and methods
2.1. Tropical nights series
We selected a total of 17 meteorological stations throughout the IP with daily minimum temperatures during the extended summer (defined as June, July, August, and September, JJAS), as referred to by other authors (e.g. Xoplaki et al., 2003a, 2003b; Baldi et al., 2006), during the 1961–2007 period (Figure 1(d)). During this period of the year, a summer circulation type, with a thermal origin in the distribution of vorticity maxima and minima, has been found to be the dominant one in the Mediterranean region (for more details see Lolis et al., 2008). We obtained the series from the European Climate Assessment and Dataset (ECA&D) project (http://eca.knmi.nl/). The details of the quality control and homogenisation procedures are widely explained by Klein Tank et al. (2002) and Wijngaard et al. (2003).
The above mentioned TN is the index of daily temperature extremes identifying the number of days with minimum temperatures > 20 °C in each JJAS month for the 17 series. This threshold index has already been used in several studies (e.g. Miranda et al., 2002; Alexander et al., 2006).
In order to obtain a regionalisation of the meteorological stations according to their similar TN variability, an S-Mode Principal Component Analysis (PCA) was performed, based on a correlation matrix which considers all four JJAS months and makes use of the series of monthly TN anomalies (differences to the 1961–2007 mean) (Brunetti et al., 2006). We used monthly anomalies in order to obtain a more robust regionalisation, since the PCA was performed to a matrix comprising 188 rows (47 years × 4 months) rather than only 47 rows, which should be applied if seasonal JJAS means per year were to be considered. The PCA was applied for the whole 1961–2007 period, as all series are available without missing values. The results identified three Principal Components (PCs) with an eigenvalue > 1, and explained almost 70% of total dataset variance. The PCs selected were rotated with the Varimax orthogonal procedure in order to redistribute variance into stable components and physically meaningful patterns (e.g. von Storch and Zwiers, 1999). The geographical representation of the loadings obtained from the PCA (Figure 1(a–c)) enables us to identify three subregions across the IP, and each station was integrated into one of the described regions according to its maximum loading (Brunetti et al., 2006; Sanchez-Lorenzo et al., 2007). Figure 1(d) shows a schematic representation of the boundaries of the subregions established by the regionalisation method applied: the Mediterranean (M), the Atlantic (A), and the Centre-North of the Peninsula (CN), which explain 25.4, 22.2 and 20.7% of total variance, respectively. We then computed the JJAS mean series for the whole IP and the three subregions, by averaging all 17 JJAS TNs mean anomaly series and the available stations in each subregion, respectively. These series provide a more synthetic representation of the climate signal than the single station and permit a higher signal-to-noise ratio, enabling better identification of long-term trends (Brunetti et al., 2006).
The overall trends of all series were calculated over the 1961–2007 period by means of least-square fitting and their significance was estimated by the Mann-Kendall nonparametric test at the 95% confidence level. Finally, the temporal variations of JJAS TNs series were assessed by means of a Gaussian low-pass filter of 11 terms for better visualisation of long-term and decadal variability, and the sigma applied in the Gaussian filter was equal to 3 standard deviations (Brunetti et al., 2006; Sanchez-Lorenzo et al., 2007).
2.2. Connection between atmospheric circulation patterns and TNs series
In order to investigate the connection between the large-scale atmospheric circulation and TNs variability, a CCA in empirical orthogonal function space is used. CCA is a statistical technique that calculates linear combinations of a set of predictor variables which maximizes relationships, in a least-squares error sense, to many predictand variables (e.g. von Storch and Zwiers, 1999). CCA has been extensively used in climatological studies in order to investigate the large-scale features of atmospheric circulation and its connection with different climatic variables over Europe (e.g. Xoplaki et al., 2003a, 2003b; Della-Marta et al., 2007; Lolis, 2009). For a more detailed explanation of the methodology, see Haylock and Goodess (2004).
In order to apply the CCA in our study, the predictand variables are the original 17 JJAS TNs mean series (Section 2.1), and the predictor variables are mean sea level pressure (SLP) and 500 hPa geopotential height for JJAS provided by the NCEP/NCAR reanalysis on a regular 2.5° grid (Kalnay et al., 1996; Kistler et al., 2001), in a selected window over the Euro-Atlantic sector (50°W–40°E; 20°N–70°N) and during the 1961–2007 study period.
We followed recommendations in Della-Marta et al. (2007) prior to the CCA in order to avoid spurious results resulting from non-stationarities in the data, such as trends and autocorrelation. Thus, the long-term linear trend over the 1961–2007 period was previously removed from each predictand and predictor series. Equally, both predictor and predictand data were standardized by subtracting their mean and dividing them by their standard deviation for the whole study period. The predictor and predictand were then dimensionally reduced by means of a PCA in S-Mode with a Varimax rotation (Richman, 1986; Lolis et al., 2002; Lolis, 2009). Very similar results are obtained if non rotated PCs are considered, and consequently only the rotated solution will be shown in the next section. We retained a number of selected PCs determined by the break in the slope of the LEV-diagram (O'Lenic and Livezey, 1988). The PCA reduction is applied in order to avoid the quasi-degeneracy of the autocovariance matrices of the datasets, and to filter the data by eliminating the noise (e.g. Barnett and Preisendorfer, 1987). As two predictors were used, termed Multiple Predictor CCA (MPCCA, e.g. Della-Marta et al., 2007), a second PCA was applied to a matrix with the PC score series which was in turn obtained from the individual SLP and 500 hPa predictor fields. Finally, CCA was applied to a matrix which included the two reduced datasets for the predictand and predictor fields. The product of a CCA includes the canonical variate time series (canonical scores) for each variate separately. The significance of the canonical pairs was estimated by the chi-square test (Lolis, 2009) at the 95% level of confidence.
Lastly, we calculated the SLP and 500 hPa anomalies over the Euro-Atlantic sector for the years with the highest and lowest TNs, which are defined as the 75th and 25th percentile, respectively, over the whole IP and the 3 subregional series defined in Section 2.1. These results provide more direct information on the role of atmospheric circulation in TNs variability, and help to verify the reliability of the atmospheric circulation patterns previously obtained by means of the CCA (e.g. Lolis, 2009; Sanchez-Lorenzo et al., 2009).
3. Results and discussion
3.1. Temporal evolution and trends of the tropical nights
The mean JJAS TNs series for the whole IP and for all subregions show a significant (α< 0.05) positive trend during the period analysed. In general, the time evolution of JJAS TNs series shows a slight decrease during the 1960s–1970s period, with a sharp increase up to the present (Figure 2), in agreement with the evolution of mean temperatures at global scale (Trenberth et al., 2007) and in Spain (Brunet et al., 2007). Thus, in the mean IP series we observed a significant rise (4.5 days per decade) in the number of TNs, identifying a slight decrease therein from 1961 to 1977, a sharp increase from 1977 to 1991 and a moderate rise thereafter. The M subregion presents the highest increase (6.3 days per decade), with a temporal behaviour pattern similar to that of the whole IP in the occurrence of TNs. The CN subregion also shows a similar pattern, but the abrupt rise somehow begins later, around the 1980s, with an overall linear trend of 4.1 days per decade. The A subregion exhibits distinct behaviour in comparison with the others, as well as the lowest increase (1.7 days per decade). An increase starts later, in the mid-1980s, and the last decade even presents a slight decrease. Overall, we observed rising TNs trends over the IP for the summer season. Nonetheless, different rates of evolution were detected, mainly in the A subregion.
Our results are coherent with those found by Klein Tank and Können (2003). They detected the clearest positive trend in the warmest nights (minimum temperature > 90th percentile) in much of Europe during the 1976–1999 period, and a slight decrease between 1946 and 1975. Other studies analysing the Mediterranean basin also fit with the temporal evolution of TNs found in this research (Kostopoulou and Jones, 2005; Pereira and Morais, 2007; Nastos and Matzarakis, 2008; Toreti and Desiato, 2008). Brunet et al. (2007) identified a reduction of the warmest nights in some Spanish meteorological stations during the 1950–1972 period, followed by a sharp increase during the 1973–2005 period. Indeed, similar time evolution and trends of minimum temperatures over Spain (Brunet et al., 2007) are observed in our TN series over the IP (Figure 2). In order to verify their temporal similarity, we calculated the JJAS minimum temperature for the same 17 series used to calculate TNs and then performed the mean IP and subregional mean anomalies series following the same criteria as described above (Section 2.1). Results showed that minimum temperatures and TNs are highly correlated (α< 0.01) over the whole IP with a coefficient of 0.96, as well as in the different subregions: 0.95, 0.86, and 0.92 for the M, A, and CN, respectively. Thus, similar trends and time variability are shown by minimum temperatures and TNs on the IP, and consequently both variables might be controlled by the same climate forcings, including atmospheric circulation.
3.2. Tropical nights on the Iberian Peninsula and their relation with atmospheric circulation variability
By applying a PCA to the TNs series as described in Section 2.2, in order to reduce its dimensionality, we obtained 3 PCs which account for 69% of total variance and their PC time series are considered as the predictand dataset. The application of the PCA to the SLP and 500 hPa geopotential height leads to 10 and 11 PCs, accounting for 90 and 91% of total variance, respectively. As we are interested in a MPCCA, a second PCA is applied to a matrix which combines the 21 PC time series from the SLP and 500 hPa geopotential height. The results show that 11 PCs explain 89% of total variance, and their PC time series are considered as the final predictor dataset. Finally, a CCA is applied to both predictand and predictor PC time series datasets, and according to the results, two significant canonical pairs are found. In order to interpret the resultant canonical pairs, the predictor and predictand canonical series are correlated to all the original TNs and the SLP and 500 hPa fields, respectively (Lolis, 2009).
Figure 3 shows the results for the first canonical pair, accounting for 33.2% of the overlapping variance. For the predictor fields, the correlations values at the 500 hPa (Figure 3(b)) geopotential height are higher than those obtained in the SLP field (Figure 3(a)), which indicates that in this CCA mode the middle troposphere atmospheric circulation is more determinant in TNs variability than changes in surface pressure. Thus, for the 500 hPa geopotential field, the spatial pattern (Figure 3(b)) shows strong positive correlations centred over Central Europe and covering all the western sectors of continental Europe, including the IP. Two regions of negative anomalies, with lower magnitude than the positive ones, are centred over the Atlantic Ocean in southwestern Iceland and over Libya in North Africa. For the predictand field (Figure 3(c)), the CCA mode shows that the previously described atmospheric circulation is highly positively correlated with TNs frequency over the northeastern and central IP, mainly the areas defined by subregions M and CN. Figure 3(d) shows the first pair of canonical score series for the predictor (dashed lines) and predictand (solid lines) fields, with a canonical correlation (correlation between the two series) of 0.76 (α< 0.01). The inter-annual variation of the canonical scores, for a specific canonical variate, practically expresses the variation of the corresponding meteorological parameter in the high-correlation centre of the correlation pattern presented. Thus, Figure 3(d) approximately shows the variations of SLP and 500 hPa over central Europe (dashed line) and the associated variation of TNs in northeast and central Spain (solid line). The score series show a tendency towards negative values from the 1960s to the mid-1970s, with a greater frequency of higher scores from then to the end of the period analysed. Thus, there appears to be an evident contribution of this CCA mode to the increase in TNs frequency over the IP during the last three decades, especially in the central and northeastern areas.
The second canonical pair (Figure 4) accounts for 26.3% of the overlapping variance, and also shows a stronger signal at the 500 hPa geopotential height (Figure 4(b)) than in SLP (Figure 4(a)). Thus, in the middle troposphere, there are high positive correlations over the Atlantic Ocean, between the Azores and the IP. Another centre of action shows negative correlations over the Baltic countries. In the predictand field (Figure 4(c)), high positive correlations between the canonical score series and TNs frequencies are found over the southwestern and western IP. Consequently, when positive (negative) 500 hPa anomalies are present near the western coast of the IP, TNs values are high (low) on the IP, especially in the area comprising subregion A. The canonical correlation (Figure 4(d)) is 0.72 (α< 0.01), and shows negative values from the 1960s to the mid-1970s, with a sharp increase from then to the mid-1980s and a subsequent stabilisation. This time evolution resembles the JJAS TNs in subregion A (Figure 2).
Finally, in order to check the reliability of the previous circulation pattern analyses, we examined the mean JJAS anomaly patterns of the 500 hPa geopotential height with the highest and lowest number of TNs during JJAS, which are defined as the 75th and 25th percentile thresholds in the series, respectively. Results for the whole IP series (Figure 5) show that a high (low) frequency of TNs is associated with enhanced positive (negative) anomalies of the geopotential height over Western Europe and the Mediterranean Sea, which indicates anticyclonic (cyclonic) activity in this area. The spatial pattern of these anomalies fields resembles the spatial pattern of the first CCA mode (Figure 3(b)), and a similar agreement is detected when similar analysis are applied between TNs subregional series (not shown) and the results obtained with the two pairs of canonical circulation modes. Consequently, the circulation modes detected by means of the CCA appear to show a physical consistency and do not result from a mathematical artifact.
Della-Marta et al. (2007) observed the existence of two different circulation patterns responsible for the occurrence of heat waves in Europe. One of these (see Figure 6 in Della-Marta et al., 2007) is very similar to the one we detected in Figure 3, although their results were limited to SLP due to the non availability of 500hPa data for their study period. Indeed, Xoplaki et al. (2003b) also detected a circulation mode in the middle troposphere that resembles our first CCA mode, which has a strong influence on summer air temperature over the central and eastern sectors of the IP. The dynamical configuration for the positive (negative) phases of this circulation pattern increases (reduces) the stability over most of western Europe, which results in an increase (decrease) in persistent anticyclonic conditions, especially in the middle layers of the atmosphere (Xoplaki et al., 2003b).
To our knowledge, the second canonical mode has not previously been shown to constitute an atmospheric circulation pattern with a strong impact on minimum summer temperature or TNs variability over the IP. For example, Xoplaki et al. (2003b) and Della-Marta et al. (2007) focused on detecting the main atmospheric circulation patterns related to summer temperature variability over the whole Mediterranean basin and continental Europe, respectively. Thus, their results do not present this second canonical mode detected in our research. In fact, this second mode is probably derived from the regional character and low explained variance of our study area (the IP) if greater spatial resolution is considered. On the other hand, this second canonical mode resembles the Atlantic Ridge pattern detected during the summer period in Europe by means of cluster analyses (Cassou et al., 2005), and consequently its physical reliability is plausible. This pattern also resembles the negative phases of the summer North Atlantic Oscillation (NAO) and the Scandinavia pattern (SCAND) described in Barnston and Livezey (1987).
We observed a significant increasing TNs frequency trend on the IP in summer (JJAS) during the 1961–2007 period. We also identified three subregions (the M, the CN, and the A) with different temporal evolution of TNs occurrence. In the M and the CN subregions, we detected a slight decrease throughout the first period up to 1980; subsequently, an exceptional rise was observed. In the A subregion, we only detected a slight positive trend throughout the study period.
This increasing TNs occurrence might be partly explained by changes in atmospheric circulation over the North Atlantic. Thus, the CCA experiment with TNs as a predictand, and the SLP and the 500 hPa geopotential as predictors, detected two atmospheric circulation patterns which account for almost 60% of the JJAS TNs variability over the IP, and might explain part of the increase in the occurrence of TNs on the IP during the most recent period. The first circulation pattern mainly influences TN variability over the eastern and central sectors of the IP. This is linked to an increase (decrease) in the 500 hPa geopotential height over most of Western Europe in those summer seasons with a high (low) occurrence of TNs in these areas. This pattern shows a moderate decrease in its positive phases from the 1960s to the mid-1970s, and then a sharp increase up to the present. The second circulation pattern mainly affects the western sectors of the IP, and a high (low) occurrence of TNs in this area is linked to positive (negative) 500 hPa anomalies to the west of the IP.
These results lead us to conclude that the significant (α< 0.05) rise in TNs occurrence on the IP might be partially related to changes in specific atmospheric circulation patterns, the mid-troposphere being more important in these relationships than the SLP anomalies. Furthermore, we also highlight the need for more regional climate studies in order to improve our knowledge of surface climate variability and its relation with large-scale forcings.
There is a need for further research taking more climate extreme indices into account, and associating these with atmospheric circulation patterns. This will provide us with a better understanding of this relationship, which is of major interest to authors investigating climate change and impacts on society and in the environmental framework.
This research study was conducted within the framework of the Spanish RECABA Project (CGL2008-06129-C02-01/CLI) and the Group of Climatology of the University of Barcelona (2009 SGR 442, Catalonia Regional Govt.). We would like to thank Bruno Simoes and Nuno Jerónimo for their technical help, and the three anonymous reviewers for their helpful comments. Arturo Sanchez-Lorenzo was granted a Beatriu de Pinos postdoctoral position by the government of Catalonia (2009 BP-A 00035).