1. INTRODUCTION. Copyright 2002 Royal Meteorological Society

Transcription

1 INTERNATIONAL JOURNAL OF CLIMATOLOGY Int. J. Climatol. 22: (2002) Published online in Wiley InterScience ( DOI: /joc.759 SPATIAL AND TEMPORAL 850 hpa AIR TEMPERATURE AND SEA-SURFACE TEMPERATURE COVARIANCES IN THE MEDITERRANEAN REGION AND THEIR CONNECTION TO ATMOSPHERIC CIRCULATION C. J. LOLIS, A. BARTZOKAS* and B. D. KATSOULIS Laboratory of Meteorology, Department of Physics, University of Ioannina, Ioannina, Greece Received 15 March 2001 Revised 26 November 2001 Accepted 27 November 2001 ABSTRACT The spatial and temporal covariability between the lower troposphere and sea surface temperatures (SSTs) are studied in the Mediterranean basin for the period Monthly air temperature anomalies for the 850 hpa pressure level (T-850hPa) at grid points and SST anomalies in 5 5 grid boxes are utilized. As a first step, factor analysis is applied on both sets of data in order to reduce their dimensionality. Then, canonical correlation analysis is applied and this leads to one statistically significant pair of canonical variates for winter and to two pairs for summer. In winter, a teleconnection (see-saw) between western Europe and the eastern Mediterranean at the 850 hpa level is revealed, and a corresponding weaker one between the areas of central-west and eastern Mediterranean for SST. The correlation between T-850hPa and SST appears higher over the eastern Mediterranean. In summer, the first pair of canonical variates reveals a covariability between T-850hPa and SST in the western Mediterranean, and the second one shows a covariability in the eastern Mediterranean, without the existence of any strong spatial teleconnection. The analysis is repeated, using time lags of 1 month, or longer, in order to detect any possible non-synchronous relation. Statistically significant results are found only when T-850hPa leads SST with a time lag of 1 month. In particular, the results are statistically significant for winter only, and the findings are similar to those of the first analysis. Therefore, the existence of a 1 month time scale SST persistence is detected for winter months. Copyright 2002 Royal Meteorological Society. KEY WORDS: 850 hpa air temperature; sea-surface temperature (SST); Mediterranean Sea; Mediterranean oscillation; factor analysis; canonical correlation analysis 1. INTRODUCTION To assess the environmental problems associated with the potential impact of expected climatic changes on the marine environment and on adjacent coastal areas and to identify suitable policy options and response measures that may mitigate the negative consequences of the expected impacts, it is necessary to study both global- and regional-scale details of changes in climate variables such as air and sea-surface temperatures (SSTs). Such a case is the Mediterranean region, which is known throughout the world for its distinctive and sensitive climate. The Mediterranean Sea is located at a transitional geographical zone between the European and the African continents (Figure 1). It lies between 30 and 46 N, and between 5 W and 36 E. Its east west extent is approximately 4000 km, and its average north south extent is only about 800 km. The western outlet to the Atlantic through the Strait of Gibraltar is less than 15 km wide at the narrowest point. The sea is enclosed by mountains except along the North African coast east of Tunisia. The whole coastline (approximately km) is very irregular, with many indentations, and is characterized by geographical and topographical complexities. The complexity of the region may be seen in the many separate gulfs, seas, * Correspondence to: A. Bartzokas, Laboratory of Meteorology, Department of Physics, University of Ioannina, Ioannina, Greece. Copyright 2002 Royal Meteorological Society

2 664 C. J. LOLIS, A. BARTZOKAS AND B. D. KATSOULIS straits and channels. The topographical relief and land mass configurations show the three main basins into which the Mediterranean can be divided into (1) the western basin, from the Strait of Gibraltar to Corsica and Sardinia; (2) the central basin, the area between Corsica, Sardinia and Tunisia in the west and Greece and Cyrenaica on the east; and (3) the eastern basin, from the Central basin to the Syrian coast. The surface current of the Mediterranean shows a migration of Atlantic water, with salinity slightly above 36, towards the east, with numerous spin-off eddies along the way (Miller, 1983). The annual thermal changes of surface waters are very large and control their density and their basic characteristics. SST may range between 12 and 29 C, the lowest being in winter in the northwest basin and in the north Adriatic Sea and the highest in summer in the northeast Levant Sea. Deep sea water in the Mediterranean has a temperature between 12.5 and 13.5 C in the west and between 13.5 and 15 C in the east (1 1.5 C higher than that of the western basin). The basic nature of the Mediterranean circulation system contains components of strong deep vertical convections, specifically when winter storms lower the surface temperature in the western basin to 12 C. The estimated turnover time for its waters is 80 years. The Mediterranean Sea has a deficient hydrological balance, with loss through evaporation exceeding the input of water through run-off and precipitation. This deficiency is mainly compensated by the flow of Atlantic surface waters entering the basin through the Strait of Gibraltar, moving eastwards along the North African coast and being subjected to seasonal thermal variations (Miller, 1983; Bartzokas et al., 1991). Although the Mediterranean Sea is of a semi-enclosed nature, it affects significantly the climate of the neighbouring regions during the whole year. Climatologically, the Mediterranean region is located between the wet temperate climate of southern Europe and the warm and dry climate of north Africa. The Mediterranean basin is characterized by rainy winters, dry summers and a large variety of regional climates arising from the complex land sea alternation. The significant difference between winter and summer synoptic conditions is accompanied by a dominant annual cycle involving all the climatological parameters. The Mediterranean Sea, being located at the southern boundaries of Europe and the temperate latitudes, is affected by the westerlies, mainly during the cold period of the year. The eastern parts of the region represent the standard example of the subtropical summer dry weather regime, whereas the areas in the north and west can only be described as transitional zones between Mediterranean and continental weather systems. The western basin weather is Figure 1. Areas of the Mediterranean region discussed in text: (1) western Mediterranean; (2) central Mediterranean; (3) eastern Mediterranean; (4) Black Sea

3 TEMPERATURE-SST COVARIANCE AND ATMOSPHERIC CIRCULATION 665 dominated by the major synoptic feature of cyclogenesis in the Gulf of Genoa and the subsequent eastward movement of depressions, especially during the winter period (Katsoulis, 1982; Alpert et al., 1990). In the western Mediterranean, depressions are frequently steered along the Mediterranean front, formed when cold continental air moves over a warmer sea surface. This front is most pronounced in spring. Fronts formed in the eastern Mediterranean tend to follow a preferred path either to the northeast or the east. The Mediterranean Sea contains a high thermal capacity water mass, which interacts with the overlying atmospheric air mass in various ways. Because of its warm waters and the complicated topography, a complex interaction arises, causing interesting meteorological conditions in various sub-areas. Therefore, the Mediterranean basin affects the atmospheric circulation over the region to a significant degree, generating the main areas of cyclogenesis and modifying the depression trajectories (Alpert et al., 1990; Katsoulis et al., 1998; Trigo and Davies, 1999). The study of covariability between air temperature and SST is a difficult task because of the slower warming (or cooling) rate of water compared with the air mass. In recent years, the space time variations of air temperature and SSTs have been examined by a number of researchers, using various approaches and methodologies (Bartzokas and Metaxas, 1991; Sahsamanoglou and Makrogiannis, 1992; Bartzokas et al., 1994; Reddaway and Bigg, 1996; Maheras and Kutiel, 1999; Maheras et al., 1999). The aim of this study is to apply two statistical approaches in order to establish the interrelationship between air temperature and SSTs. Specifically, the synoptic-scale interaction between the Mediterranean SST and the lower troposphere air temperature is examined. As a representative temperature for the lower troposphere, the 850 hpa level air temperature (T-850hPa) is utilized. The main objectives are: (a) the revelation of covariability cases between SST and T-850hPa on a monthly basis, as well as the determination of the corresponding centres of action; (b) the examination of the influence of each parameter on the other, by considering a time lag of 1 month or longer; and (c) the revelation of the connection between the atmospheric circulation and the SST and T-850hPa parameters. The analysis is applied for winter (December February) and summer (June August) seasons separately. 2. DATA AND METHODS The basic data in the study consists of gridded values of mean monthly SST and air temperature at the 850 hpa level (T-850hPa) in the Mediterranean area. In particular, the data used are the mean monthly anomalies of (a) SST at grid boxes, and (b) T-850hPa at 189 grid points spaced by 2.5, confined by the 30 and 50 N parallels and the 10 W and 40 E meridians, for the period The SST data were kindly provided by the United Kingdom Meteorological Office (UKMO), and the T-850hPa data were obtained from the National Center for Environmental Prediction (NCEP) and the National Center for Atmospheric Research (NCAR) (Kalnay et al., 1996). The monthly anomalies of the two parameters were calculated, using the longterm period ( ) average at every grid box and every grid point for the above period. A small number of SST missing values were calculated from the values of the neighbouring grid boxes, by using linear regression analysis. Thus, monthly anomaly time series of 120 (3 40) values for winter (December February) and 123 (3 41) values for summer (June August) were constructed for both parameters. The statistical methods used in this work are factor analysis (FA) and canonical correlation analysis (CCA), briefly described in the following paragraphs. FA describes a set of p correlated variables X 1, X 2,...,X p in terms of a smaller number of new uncorrelated indices or factors. Hence, it elucidates the relationship between the original p variables. Each of the p initial variables can be expressed as a linear function of m (m <p) factors, i.e. X i = a i1 F 1 + a i2 F 2 + +a im F m,wheref 1, F 2,...,F m are the factors and a i1, a i2,...,a im are the factor loadings that express the correlation between the initial variables and the new ones (factors). The values of each factor are called factor scores and they are usually presented in standardized form, having zero mean and unit variance (Jolliffe, 1986; Manly, 1986). The number m of the retained factors has to be decided, by using various rules (Overland and Preisendorfer, 1982; Jolliffe, 1986) and considering the physical interpretation of the results (Bartzokas and Metaxas, 1993). A widely used process is the rotation of axes, which creates new

4 666 C. J. LOLIS, A. BARTZOKAS AND B. D. KATSOULIS factors with different variances, but it keeps the cumulative variance of the m factors unaffected. Rotation, by maximizing some factor loadings and minimizing some others, succeeds in a better separation among the initial variables and thus in a better interpretation of the results. There are various types of rotation. Varimax rotation is generally accepted as the most accurate orthogonal rotation. It maximizes the sum of the variances of the square factor loadings, keeping the factors uncorrelated (Richman, 1986). CCA has previously been described and/or applied in empirical climatological or geophysical contexts (Graham et al., 1987; Nichols, 1987; Chu and He, 1992), and is useful for both diagnosis and forecasting. It is a multivariate statistical technique that calculates linear combinations of a set of predictors that maximizes relationships, in a least-squares error sense, to similarly calculated linear combinations of a set of predictands. The terms predictor and predictand are sometimes arbitrary, as it is not apriori known which of the variables is the real predictor. That is, CCA investigates the relation between two sets of variables X 1,X 2,...,X p and Y 1,Y 2,...,Y q. The linear combinations W 1 = a 11 X 1 + a 12 X 2 + +a 1p X p V 1 = b 11 Y 1 + b 12 Y 2 + +b 1q Y q are formed and the coefficients a 11,a 12,...,a 1p and b 11,b 12,...,b 1q arecalculatedinsuchawaythatthe correlation coefficient, C 1 = cor(w 1,V 1 ),betweenw 1 and V 1 attains a maximum. W 1 and V 1 are called the canonical variates and C 1 is called the canonical correlation. In the next step, another set of canonical variates is identified W 2 = a 21 X 1 + a 22 X 2 + +a 2p X p V 2 = b 21 Y 1 + b 22 Y 2 + +b 2q Y q such that C 2 = cor(w 2,V 2 ) = maximum. The two sets of canonical variates W 1, V 1 and W 2, V 2 are uncorrelated. This procedure is continued up to the mth set of canonical variates, where m = min(p, q). Thus, m sets of canonical variates (W 1,V 1 ), (W 2,V 2 ),..., (W m,v m ) are created in a way that: (i) the corresponding canonical correlations C 1, C 2,..., C m are maximum and (ii) cor(v j,v k ) = cor(w j,w k ) = cor(w j,v k ) = 0,j k. The products of a completed CCA include: (i) canonical loading patterns for predictor and predictand variables separately; (ii) canonical variate time series (canonical scores) for each variate separately; and (iii) eigenvalues whose square roots are the canonical correlations. In practice, the number of the canonical pairs used never equals the total number of pairs. As with FA, only the statistically significant pairs are used, indicated by the chi-square test (Anderson, 1984; Sharma, 1995). In this work, FA (S-mode) (Richman, 1986) is applied separately for the time series of each parameter (T-850hPa and SST) in order to reduce the number of the initial variables. Many researchers have followed this process, i.e. the reduction of the dimensionality of the initial data sets before the application of CCA (Zorita et al., 1992; Knappenberger and Michaels, 1993; Corte-Real et al., 1995; Diaz et al., 1998). This is considered necessary in order to avoid, on the one hand, quasi-degeneracy of the autocovariance matrices of the data sets and, on the other hand, to filter the data by eliminating the noise (Barnett and Preisendorfer, 1987; Bretherton et al., 1992). Then, CCA is applied on the two sets of the resultant factor scores time series. The basic advantage of the above process is the fact that it takes into account both sets simultaneously, investigating the best interrelations between them. This would not have been the case if individual correlation coefficients had been calculated among the factor scores time series. Finally, in order to achieve an interpretation of the resultant canonical variates, each of them is correlated with all the original time series of the corresponding field, as they are modified by being projected to the factors. For each canonical pair, the correlation coefficients were plotted on two maps. By comparing the two maps, we are able to interpret the physical meaning of the canonical variates. 3. RESULTS AND DISCUSSION 3.1. Winter Winter season (December February) is characterized by cyclonic disturbances and low mean sea-level pressure in the Mediterranean (in comparison with the surrounding continental areas), with higher values

5 TEMPERATURE-SST COVARIANCE AND ATMOSPHERIC CIRCULATION 667 to the east associated with the extension of the Siberian anticyclone. For the winter season, FA led to four factors both for T-850hPa and SST, explaining 87% and 77% of the total variance respectively. We note that Bartzokas et al. (1994) and Bartzokas and Lolis (2000) also obtained four factors for winter SST in the Mediterranean for different time periods. FA results are not presented here, as this is not the main objective of the present work. It was applied only in order to reduce the dimensionality of both sets. CCA was applied on the two factor scores sets and revealed one statistically significant canonical pair (W,V) (99% confidence level) only. This canonical pair accounts for 64% of the overlapping variance and captures 10% of the variance in the T-850hPa field and 10% of the variance in the SST field. The correlation coefficient between the canonical variates W and V (canonical correlation) is r = The only statistically significant canonical pair corresponds to two coherent temperature see-saws. For T-850hPa, a see-saw between western Europe and the Middle East is detected (W Figure 2(a)), whereas for SST the see-saw appears between the west-central and eastern Mediterranean (V Figure 2(b)). For the T-850hPa field, the correlation coefficients between W and the filtered (projected to the factors) T-850hPa time series are above r = 0.8 over France and below r = 0.8 over the Middle East, whereas for the SST field the correlation coefficients between V and the filtered SST time series are about r = 0.6 inthestrait of Sicily region and below r = 0.7 in the Cyprus region. The canonical scores time series (Figure 2(c) and (d)) practically approximate the temporal variations of temperature at the centres of the above see-saws. According to these variations, there is, generally, an increasing temperature trend in the western Mediterranean area during the whole period. This trend is statistically significant at the 95% confidence level, both for W (T-850hPa) and V (SST) time series. The opposite (negative trend) is valid for the eastern Mediterranean, because of the opposite sign (negative) correlation coefficients there (see-saw). This negative SST trend for the eastern Mediterranean during the last 40 years is in accordance with the findings of many researchers, resulted from the application of various methodologies (e.g. Metaxas et al., 1991; Bartzokas and Lolis, 2000). In order to obtain a realistic/physical view of the positive and negative phases of these see-saws and to confirm the mathematical products of CCA, we sorted the scores of both canonical variates from the highest to the lowest value and we selected only the cases (months) belonging to the highest and the lowest deciles (one-tenth). Then, the temperature anomaly maps of the 10% highest and the 10% lowest value cases were constructed for Figure 2. Isopleths of correlation between the canonical variates (W,V) and (a) T-850hPa and (b) SST time series for winter. (c), (d) Canonical scores time series, smoothed (bold line) using 5 month moving averages with binomial coefficient weights

6 668 C. J. LOLIS, A. BARTZOKAS AND B. D. KATSOULIS Figure 3. (a), (b) T-850hPa anomalies; (c), (d) SST anomalies; (e), (f) 850 hpa geopotential height; (g), (h) 850 hpa geopotential height anomalies for the months corresponding to the highest and the lowest deciles (10%) of the canonical variates (W,V) for winter each canonical variate (Figure 3(a) (d)). These maps show that the Mediterranean is divided into two subregions with opposite signs in temperature anomalies for both T-850hPa and SST. For the T-850hPa field, the 10% highest score months present temperature anomalies above +2 C over France and below 3 C over the eastern Black Sea and Middle East (Figure 3(a)), whereas the 10% lowest score months present temperature anomalies below 3 C over central Europe and above +2 C over the eastern Black Sea and Middle East (Figure 3(b)). Similarly, for the SST field the 10% highest score months present SST anomalies of about +0.4 C in the southwestern part and below 0.8 C in the northeastern part of the basin (Figure 3(c)), whereas the 10% lowest score months present SST anomalies below 0.6 C in the southwestern part and above 0.6 C in the northeastern part of the basin (Figure 3(d)).

7 TEMPERATURE-SST COVARIANCE AND ATMOSPHERIC CIRCULATION 669 In order to seek out the circulation regimes favouring these extreme temperature anomaly patterns, we also constructed the 850 hpa geopotential height maps (Figure 3(e) and (f)) and the corresponding height anomaly maps (Figure 3(g) and (h)) (geopotential height data were also obtained from NCEP NCAR). From these maps it is apparent that the positive phase of the temperature see-saw is connected with an enhanced anticyclonic activity over western Europe (Figure 3(g)), due to the northeastward extension of the Azores subtropical anticyclone (Figure 3(e)). This situation leads to cold advection over the eastern part of the Mediterranean connected with frequent intrusions of cold air coming from the interior of Russia and to warm advection over the western part (Figure 3(a)). On the other hand, the negative phase is associated with enhanced depression activity over the central Mediterranean (Figure 3(h)), due to the attenuation of the subtropical anticyclone that allows the westerlies to prevail over the whole Mediterranean Sea (Figure 3(f)). This situation leads to depression formations over the central Mediterranean, causing warm advection over the eastern part and cold advection over the western part (Figure 3(b)). These two dominant circulation patterns affect not only temperature but also other climatic elements, such as precipitation over the Mediterranean basin (Metaxas et al., 1993; Xoplaki et al., 2000). The see-saw between the western and eastern Mediterranean was detected by Conte et al. (1989) for 500 hpa geopotential height and was called the Mediterranean oscillation. Since then, many researchers have examined this oscillation, by using circulation indices and temperature variability (Kutiel and Maheras, 1998; Maheras and Kutiel, 1999). The correlation between the T-850hPa W variable and the Mediterranean oscillation index (MOI), empirically defined by Conte et al. (1989), is equal to r = 0.60, confirming that there is a connection between the two oscillations. It is well known also that the North Atlantic oscillation (NAO) is connected with the temperature regime of many regions in Europe, North Africa and the Middle East (Hurrell and van Loon, 1997; Nichols, 1998). However, Conte et al. (1989) did not detect any statistically significant relation between the MOI and NAO index. By correlating the T-850hPa W variable with the NAO index given by Jones et al. (1997), we found a correlation coefficient equal to r = 0.33 (statistically significant at the 99% confidence level), indicating some dependence of the Mediterranean oscillation on the NAO, as is expected to occur. This means that, in general, a positive NAO index is significantly related to positive 850 hpa temperature anomalies over France and negative anomalies over the eastern Mediterranean and the Middle East and vice versa. In the following, the possible relation between the two parameters with a time lag of 1 month or longer is examined. It was found that the results of CCA are statistically significant only when T-850hPa leads SST Figure 4. As in Figure 2, but when T-850hPa leads SST with a time lag of 1 month

8 670 C. J. LOLIS, A. BARTZOKAS AND B. D. KATSOULIS Figure 5. As in Figure 3, but when T-850hPa leads SST with a time lag of 1 month with a time lag of 1 month; viz. the month sequence December January February for T-850hPa corresponds to the sequence January February March for SST. In this case, one statistically significant canonical pair (W,V ) arose, being almost identical to the one revealed by the analysis without a time lag (Figure 4(a) (d)). This canonical pair accounts for 63% of the overlapping variance and captures 10% of the variance in the T-850hPa field and 10% of the variance in the SST field. The canonical correlation is r = 0.63, practically equal to that of the previous analysis. Hence, we can indirectly infer the existence of a 1 month time scale persistence in SST, associated with the high thermal inertia of the sea water. The corresponding temperature and geopotential height maps for the 10% highest and lowest value cases are presented in Figure 5, where a similar situation to Figure 3 is seen. Thus, we can conclude that, in winter, air circulation provokes the Mediterranean oscillation, which is mainly responsible for the temperature see-saw in the lower troposphere.

9 TEMPERATURE-SST COVARIANCE AND ATMOSPHERIC CIRCULATION 671 This see-saw is also reflected in the SST with a time lag up to 1 month, because of the high thermal inertia of the sea water. In order to confirm the above conclusions, we calculated the correlation coefficients between W W, W V and V V. We found that, r(w,w ) = 0.97, r(w,v ) = 0.60 and r(v,v ) = The coefficient r(w,w ) is very high, as it ought to be, since it refers to the same parameter, areas and months. The coefficient r(v,v ) is greater than r(w,v ), supporting our argument that the 1 month time scale persistence of SST is mainly responsible for the time-lag association between the two parameters (their difference, 0.08, is statistically significant at the 70% confidence level). Also, we calculated the lag-one autocorrelation of V (taking into account the jump between February and December) and we found r = 0.72, thus providing one more piece of evidence for the SST persistence. Finally, for a further confirmation of our conclusions, we estimated the lag-one autocorrelation of the raw SST anomalies (December February, January March) and we found that the highest persistence (r >0.7) occurs in the eastern Mediterranean basin, where V and V are also referred to (figure not shown) Summer In the summer season (June August), the region experiences warm and dry conditions linked to the presence of a strong high-pressure ridge extending eastwards from the Azores subtropical anticyclone. For the summer season, FA led to six factors for T-850hPa and two factors for SST, explaining 85% and 75% of the total variance respectively. The analysis with two factors for summer Mediterranean SST is in accordance to the results of Bartzokas et al. (1994). In the next step, CCA resulted in two statistically significant (99% confidence level) canonical pairs. The first canonical pair (W 1,V 1 ) accounts for 65% of the overlapping variance and captures 8% of the variance in the T-850hPa field and 22% of the variance in the SST field. W 1 corresponds to a weak see-saw between the Gulfs of Genoa and Lions area and the eastern Mediterranean for T-850hPa. The pattern of the correlation isopleths shows that W 1 is mainly related to air temperature in the Gulf of Genoa and Gulf of Lions areas, because of the high (r >0.8) correlation values there, whereas it is weakly correlated to eastern Mediterranean air temperature (r <0.5, absolutely) (Figure 6(a)). The other canonical variate, V 1, is related to western Mediterranean SST only (Figure 6(b)). The canonical correlation between W 1 and V 1 is equal to r = Hence, it can be considered that there is a covariability between T-850hPa above the Gulf of Genoa area and the western Mediterranean SST. According to the scores of W 1 and V 1 (Figure 6(c) Figure 6. As in Figure 2, but for the first canonical pair of summer

10 672 C. J. LOLIS, A. BARTZOKAS AND B. D. KATSOULIS Figure 7. As in Figure 3, but for the first canonical pair of summer and (d)), the interesting features of the temporal variation of both parameters are the double minimum of late 1970s and early 1980s and the generally warming trend since then, which is more obvious for SST. These findings are in agreement with the results of Metaxas et al. (1991), who detected an increase of western Mediterranean SST for the summer season since the late 1970s. The highest and lowest decile maps are presented in Figure 7(a) (d), showing that T-850hPa anomalies of about C (absolutely) over the Gulf of Genoa are connected to greater than 1 C (absolutely) SST anomalies in the west. The geopotential height maps (Figure 7(e) and (f)) and the geopotential height anomaly maps (Figure 7(g) and (h)) show the specific circulation characteristics connected with the above temperature anomaly regime over this region; viz. when a subtropical anticyclonic ridge prevails over the western Mediterranean (Figure 7(e) and (g)), warm advection occurs over this area (Figure 7(a)). On the other hand, when a geopotential height trough prevails over the

11 TEMPERATURE-SST COVARIANCE AND ATMOSPHERIC CIRCULATION 673 Figure 8. As in Figure 2, but for the second canonical pair of summer same region (Figure 7(f) and (h)), some depression activity occurs in combination with cold advection from the northwest. The second canonical pair (W 2,V 2 ) accounts for 34% of the overlapping variance and captures 6% of the variance in the T-850hPa field and 15% of the variance in the SST field. It corresponds to the eastern Mediterranean area for both T-850hPa and SST. W 2 is highly correlated with T-850hPa over the southern Balkans, and V 2 is highly correlated with the eastern Mediterranean surface waters temperature (Figure 8(a) and (b)). The canonical correlation is equal to r = According to the W 2,V 2 scores (Figure 8(c) and (d)), there is a warming trend for both parameters during the last few years, following the minimum of the mid- 1970s. This trend is in agreement with that detected by Metaxas et al. (1991) for the eastern Mediterranean SST and Athens air temperature. The extreme warm and cold cases are presented in Figure 9(a) (d). It is shown that when T-850hPa is about 2 C warmer/colder than the normal over southern Italy and the Balkans, SST is also by about 1 C above/below the normal in the eastern Mediterranean. According to the geopotential height maps (Figure 9(e) and (f)) and the corresponding height anomaly maps (Figure 9(g) and (h)), the temperature over the southern Balkans is related to the origin of the air masses dominating over the region. It is well known that the major synoptic systems prevailing over the eastern Mediterranean during summer are the Azores subtropical anticyclone and the SW Asia heat low. The combination of these two synoptic systems provokes the Etesian winds that blow during summers over the Aegean Sea and east Balkans (Meteorological Office, 1962). Their intensity depends on the strength of the above combination, and especially on the atmospheric pressure over the northern Balkans (Metaxas and Bartzokas, 1994). The origin and the thermal characteristics of the air masses being transferred over this area via the Etesian winds are controlled by the circulation regime prevailing over the northern Balkans and the Black Sea. When a subtropical anticyclonic ridge prevails over this area (Figure 9(e) and (g)), warm advection occurs (Figure 9(a)) as the air masses are of subtropical origin. On the contrary, when a northwesterly flow prevails over the region (Figure 9(f) and (h)), the air masses come from central Europe and are cooler than normal. In order to reveal any possible connection between the NAO and the lower troposphere temperature regime of the above areas, W 1 and W 2 were correlated with the summer NAO index and the correlation coefficients were found to be r 1 = 0.12 and r 2 = 0.25 respectively. Both values are relatively low, but the second value (correlation coefficient between W 2 and the NAO index) is statistically significant at the 99% confidence level. This leads us to the conclusion that there is a weak relation between the NAO and the summer temperature regime over the Balkans; i.e. a positive NAO index is statistically significantly related to higher than normal

12 674 C. J. LOLIS, A. BARTZOKAS AND B. D. KATSOULIS Figure 9. As in Figure 3, but for the second canonical pair of summer summer temperature over the Balkans and vice versa. The relation of W 2 with the NAO can also be inferred by the difference in geopotential height gradient over southwest Europe between the highest and lowest decile patterns (Figure 9(e) and (f)), implying a different intensity of westerly winds there. There were no statistically significant results (at the 99% confidence level) for time lags of 1 month or longer. This indicates that SST persistence in the Mediterranean is lower in summer than in winter. This was also confirmed by the relatively low lag-one autocorrelation coefficients, estimated from the raw SST anomalies, which were found to be lower than the winter ones all over the Mediterranean basin (r <0.5) (figure not shown). This is due to the fact that, in summer, the vertical mixing is weaker and the sea surface mixed layer is thinner with lower thermal capacity. Hence, it is rapidly subjected to thermal variations caused by atmospheric variability (Namias et al., 1988).

13 TEMPERATURE-SST COVARIANCE AND ATMOSPHERIC CIRCULATION CONCLUSION In this work, the covariability between the lower troposphere (T-850hPa) and the SSTs of the Mediterranean region was investigated. The following conclusions and remarks can be made. (1) For winter, there is a statistically significant relation between the see-saw western Europe Middle East for T-850hPa and a weaker east west Mediterranean one for SST, corresponding to the so-called Mediterranean oscillation and being connected to the strength and the position of the Azores subtropical anticyclone. (2) For summer, there is no strong teleconnection between either of the two fields, but there was found to be a statistically significant relation between Genoa s T-850hPa and western Mediterranean SST, as well as between the south Balkans T-850hPa and eastern Mediterranean SST. The existence and the position of a subtropical anticyclone ridge is mainly responsible for the temperature variations over both areas. (3) A connection between T-850hPa and SST with a time lag of 1 month or greater is found only for winter and only when T-850hPa leads SST with a time lag of 1 month. The centres of action of the lower troposphere affecting SST remain the same, confirming indirectly the existence of an approximate 1 month SST persistence. From the above statements, it can be concluded that lower troposphere temperature anomalies control to a high degree the SST anomalies in the Mediterranean and the response of SST is of the order of 1 month. The correlation coefficients between the corresponding canonical variates are of the order of r = 0.6. This seems reasonable, as one would not expect very high correlation coefficients between the temperatures of the two different media (air and water), since apart from other factors affecting both parameters, there is a significant difference in thermal capacity. No synoptic-scale influence of Mediterranean SST on the next month s temperature of the lower troposphere is found. This is in agreement with the statement that, in the Mediterranean, which is a semi-enclosed sea, SST anomalies are to a high degree controlled by atmospheric circulation and to a lesser degree these anomalies influence atmospheric circulation. REFERENCES Alpert P, Neeman BU, Shay-El Y Inter-monthly variability of cyclone tracks in the Mediterranean. Journal of Climate 3: Anderson TW An Introduction to Multivariate Statistical Analysis. J. Wiley & Sons. Barnett T, Preisendorfer R Origins and levels of monthly and seasonal forecasts skill for the United States surface air temperatures determined by canonical correlation analysis. Monthly Weather Review 115: Bartzokas A, Lolis CJ Sea surface temperature 850hPa relative vorticity relations in the Mediterranean region during winter. In Proceedings of the International Colloquium The Mediterranean: Culture, Environment and Society, University of Haifa, Israel, May; Bartzokas A, Metaxas DA Climatic fluctuation of temperature and air circulation in the Mediterranean. In Proceedings of the European School of Climatology and Natural Hazards Course, Arles/Rhône, France; Bartzokas A, Metaxas DA Covariability and climatic changes of the lower-troposphere temperatures over the Northern Hemisphere. Il Nuovo Cimento 16(4): Bartzokas A, Metaxas DA, Kateri M, Exarchos N Sea surface temperature in the Mediterranean. Statistical properties. Rivista di Meteorologia Aeronautica 51(1 2): Bartzokas A, Metaxas DA, Ganas IS Spatial and temporal sea-surface temperature covariances in the Mediterranean. International Journal of Climatology 14: Bretherton CS, Smith S, Wallace J An intercomparison of methods for finding coupled patterns in climate data. Journal of Climate 5: Chu PS, He Y Prediction of Hawaiian winter rainfall using canonical correlation analysis. In 12th Conference on Probability and Statistics in the Atmospheric Sciences, June 22 26, Toronto, Ont., Canada. American Meteorological Society: Conte M, Giuffrida S, Tedesco S The Mediterranean oscillation: impact on precipitation and hydrology in Italy. In Conference on Climate and Water, vol. 1, Academy of Finland, September; Corte-Real J, Zhang X, Wang X Large-scale circulation regimes and surface climatic anomalies over the Mediterranean. International Journal of Climatology 15: Diaz AF, Studzinski CD, Mechoso CR Relationships between precipitation anomalies in Uruguay and southern Brazil and sea surface temperature in the Pacific and Atlantic oceans. Journal of Climate 11: Graham NE, Michaelsen J, Barnett TP An investigation of the El-Nino southern oscillation cycle with statistical models. Journal of Geophysical Research 92:

14 676 C. J. LOLIS, A. BARTZOKAS AND B. D. KATSOULIS Hurrell JW, van Loon H Decadal variations in climate associated with the North Atlantic oscillation. Climate Change 36: Jolliffe IT Principal Component Analysis. Springer-Verlag: New York. Jones PD, Jonsson T, Wheeler D Extension to the North Atlantic oscillation using early instrumental pressure observations from Gibraltar and south-west Iceland. International Journal of Climatology 17: Kalnay E, Kanamitsu M, Kistler R, Collins W, Deaven D, Gandin L, Iredell M, Saha S, White G, Woollen J, Zhu Y, Leetmaa A, Reynolds B, Chelliah M, Ebisuzaki W, Higgins W, Janowiak J, Mo KC, Ropelewski C, Wang J, Jenne R, Joseph D The NCEP/NCAR 40-year reanalysis project. Bulletin of the American Meteorological Society 77: Katsoulis BD Climatic and synoptic considerations of the Mediterranean depressions developing and passing over or near the Balkan Peninsula. In Proceedings of the 1st Hellenic British Climatic Congress, Athens, September 1980; Katsoulis BD, Makrogiannis TJ, Goutsidou YA Monthly anticyclonicity in southern Europe and the Mediterranean region. Theoretical and Applied Climatology 59: Knappenberger PC, Michaels PJ Cyclone tracks and wintertime climate in the mid-atlantic region of the USA. International Journal of Climatology 13: Kutiel H, Maheras P Variations in the temperature regime across the Mediterranean during the last century and their relationship with circulation indices. Theoretical and Applied Climatology 61: Maheras P, Kutiel H Spatial and temporal variations in the temperature regime in the Mediterranean and their relationship with circulation during the last century. International Journal of Climatology 19: Maheras P, Xoplaki E, Davies T, Martin-Vide J, Bariendos M, Alcoforado MJ Warm and cold monthly anomalies across the Mediterranean basin and their relationship with circulation; International Journal of Climatology 19: Manly BFJ Multivariate Statistical Methods: A Primer. Chapman & Hall: London. Metaxas DA, Bartzokas A Pressure covariability over the Atlantic, Europe and N. Africa. Application: centers of action for temperature, winter precipitation and summer winds in Athens, Greece. Theoretical and Applied Climatology 49: Metaxas DA, Bartzokas A, Vitsas A Temperature fluctuations in the Mediterranean area during the last 120 years. International Journal of Climatology 11: Metaxas DA, Bartzokas A, Repapis CC, Dalezios NR Atmospheric circulation anomalies in dry and wet winters in Greece. Meteorologische Zeitschrift 2(3): Meteorological Office Weather in the Mediterranean. HMSO: London; 362 pp. Miller A The Mediterranean Sea, A. Physical aspects. In Estuaries and Enclosed Seas. Elsevier. Namias J, Yuan X, Cayan DR Persistence of North Pacific sea surface temperature and atmospheric flow patterns. Journal of Climate 1: Nichols JM (ed.) The Global Climate System Review, December 1993 May World Climate and Monitoring Programme, WMO-No 856, 99 pp. Nichols N The use of canonical correlation to study teleconnections. Monthly Weather Review 112: Overland JE, Preisendorfer RW A significance test for principal components applied to a cyclone climatology. Monthly Weather Review 110: 1 4. Reddaway JM, Bigg GR Climatic change over the Mediterranean and links to the general atmospheric circulation. International Journal of Climatology 16: Richman MB Rotation of principal components. Journal of Climatology 6: Sahsamanoglou HS, Makrogiannis TJ Temperature trends over the Mediterranean region, Theoretical and Applied Climatology 45: Sharma S Applied Multivariate Techniques. John Wiley & Sons. Trigo IF, Davies TD Objective climatology in the Mediterranean region. Journal of Climate 12: Xoplaki E, Luterbacher J, Burkard B, Patrikas I, Maheras P Connection between the large-scale 500hPa geopotential height fields and precipitation over Greece during wintertime. Climate Research 14: Zorita E, Viacheslav K, von Storch H The atmospheric circulation and sea surface temperature in the North Atlantic area in winter: their interaction and relevance for Iberian precipitation. Journal of Climate 5: