Trends in temperature indices over Serbia: relationships to large-scale circulation patterns

Authors


ABSTRACT

In this work, the trends of six climate indices based on the daily maximum and minimum temperatures during the period 1949–2009 at 15 stations distributed across Serbia were analysed. The results showed seasonal changes in the minimum and maximum temperature extremes. An analysis of the extreme temperature indices suggested that the Serbian climate generally tended to become warmer in the last 61 years. The most significant temperature trends were revealed for the summer season. The influence of large-scale variables on the temperature was examined by means of the empirical orthogonal function and correlation. It was found that the East Atlantic pattern dominated during the winter, spring and summer, while the East Atlantic/West Russia pattern governed during the autumn. In addition, the North Atlantic Oscillation dominated the Serbian extreme temperature variability during the winter.

1. Introduction

Studies performed on the global scale, based on observational (Easterling et al., 1997) or model data (Intergovernmental Panel on Climate Change-IPCC, 2007), revealed tendencies towards warming, mainly due to increased minimum temperatures. Heino et al. (1999) in Finland and Brázdil et al. (1996) in Central and Southeast Europe showed that the minimum temperatures had a greater increase than the maximum temperatures on an annual basis. During the 20th century, in most European regions, an increase in the surface air temperature was observed (Brunetti et al., 2000; Houghton et al., 2001). Beniston (2004) found a strong increase in the mean annual minimum temperatures of about 2 °C 100 year−1 during the 20th century in Switzerland, while the increase in the mean annual maximum temperatures was much smaller. Goubanova and Li (2006) expected the largest warming of the maximum temperature over Southern Europe for the IPCC-A2 emission scenario.

According to Houghton et al. (2001), an increase in the frequency of extreme warm days and a decrease in the frequency of cold nights during the summer were observed over many areas. The number of frost days decreased in many areas of the world (Karl et al., 1999), but many areas of the eastern Mediterranean showed significant increasing trends in the frost day index at the annual scale (Kostopoulou and Jones, 2005). Efthymiadis et al. (2011) found decrease of cold extremes (in winter) and the increase of warm/hot extremes (in summer) over the central and eastern Mediterranean. Averaged across the eastern Mediterranean region, Kuglitsch et al. (2010) established that, since the 1960s, the hot summer daytime and nighttime temperature have increased by +0.38 ± 0.04 °C decade−1 and +0.30 ± 0.02 °C decade−1, respectively.

In Serbia, the extreme temperature increased at Belgrade (Unkašević et al., 2005; Unkašević and Tošić, 2009a). The warmest summers with regards to heat wave duration and severity occurred within the periods 1951–1952, 1987–1998 (especially 1994) and 2000–2007 (Unkašević and Tošić, 2009b). Because the Serbian region is not always covered in European studies, the analysis of temperature tendencies can contribute to better understanding of the temperature changes.

This study focuses on an analysis of the changes and trends in the extreme temperature indices over Serbia, as well as on the connection between the temperature indices and large-scale circulation patterns. The studied indices were selected from the list of climate change indices recommended by the World Meteorological Organization (WMO) – Commission for Climatology and the Research Programme on Climate Variability and Predictability (CLIVAR). The influences of large-scale patterns, such as the North Atlantic Oscillation (NAO), the East Atlantic (EA) and East Atlantic/West Russia (EA–WR) patterns, on the extreme indices were examined by means of the empirical orthogonal function (EOF) and correlation.

The paper is organized as follows: Section 'Data and methods' presents a description of the station data and methods, temperature indices and large-scale atmospheric circulation patterns. The obtained results are presented in Section 'Results'. Discussion and conclusions are summarized in Section 'Discussion and conclusions'.

2. Data and methods

2.1. Data used

This study is based on the collection of the daily minimum (Tn) and maximum (Tx) temperatures from the observational network of the Serbian Meteorological Service during the period 1949–2009. The spring (MAM), summer (JJA) and autumn (SON) seasons correspond to the calendar year but the winter season (DJF) corresponds to January-February of the calendar year and to December of the previous year. The lowest value (about −29.5 °C) of the absolute annual minimum temperatures was registered in the plain part of northeastern Serbia. The highest value (about 43.5 °C) of the absolute annual maximum temperatures was observed in the central part of Serbia, caused by the continental effect.

Stations with less than 61 years of daily data (37.5%) were discarded from the analysis. The locations of used stations did not change during the observation period. Measurements were performed every day without a break using the same type of instruments. The technical and critical control of these measurements was made by the Serbian Meteorological Service. For the quality control of the temperature data, we checked whether there were any cases where Tx < Tn. We did not found inconsistent temperature values. Fifteen stations without gaps were selected for the analysis. All the stations are located at elevations from 81 to 215 m (Figure 1), except for the one at Vranje (432 m).

Figure 1.

Map of Serbia with stations used (list of stations with their abbreviations, latitudes, longitudes and altitudes).

The basic assessment of station homogeneity was performed by examining a series of the mean monthly minimum and maximum temperature anomalies and by comparing them with those from neighbouring stations to confirm that the trends were not due to inhomogeneities (Alexandersson and Moberg, 1997). Figure 2 shows the statistic T for Tn and Tx during the period from January 1949 to December 2009 for two selected stations. It is seen that values for variable T are less than the critical value 10.5 of T for the significance level at 5%. From the homogeneity analysis of the mean monthly minimum and maximum temperatures for 15 stations we can conclude that the monthly temperature time series are homogeneous. On the basis of the daily maximum and minimum summer air temperature series, Kuglitsch et al. (2010) found that many instrumental measurements in the eastern Mediterranean were warm-biased in the 1960s, leading to heat wave trends up to 8% higher. Hence, more detailed work should be done for testing the homogeneity of the daily data.

Figure 2.

The sequence of statistic T at station: (a) Kragujevac for Tn and (b) Belgrade for Tx from January 1949 to December 2009. The critical value of T for significance level at 5% (T5) is presented by dashed line.

2.2. Methods used

The influence of large-scale patterns was examined by means of the EOF and correlation. We used the Pearson's correlation, the most common measure of correlation, which reflects the degree of linear relationship between two variables (Wilks, 2006).

2.2.1. The EOF

One of the main purposes of EOF is to reduce the number of variables to be studied while retaining most of the information contained in the original set of variables in order to understand and interpret the structure of the data. In this analysis the singular value decomposition (SVD) approach (Wilks, 2006) is used. SVD provides a one-step method for computing all the components of the eigenvalue problem, without having to compute and store large covariance matrices. The sum of the squared singular values (SVs) equals the variance of the original series so that the ratio between each squared SV and their sum is the fraction of variance explained by that SV. To uniqely define an EOF, a normalization to its SV is chosen. The temporal variability of the time series associated to the main EOF configurations, i.e. the PCs are examined. The main variability modes of the seasonal temperatures in Serbia are identified with this technique.

2.2.2. The Mann–Kendall test

The nonparametric Mann–Kendall (M-K) test (WMO, 1966) was used to detect any possible trend in temperature series, and to test whether such trends are statistically significant. The M-K rank correlation statistic τ is defined as

display math

where ni is the number of values larger than the ith value in the series subsequent to its position in the series of N values. To apply this statistic to evaluate significance, a comparison is made with

display math

where, tg is the desired probability point of the normal distribution with a two-sided test, which is equal to 1.96 and 2.58 for the 5 and 1% levels of significance, respectively. Using a two-sided test of the normal distribution, null hypothesis of absence of any trend in the series is rejected if |τ| > |(τ)t| for the desired level of significance.

2.3. Temperature indices

Climatologists have suggested a number of indices of climate extremes that can be easily calculated and can be applied in different parts of the world (Frich et al., 2002).

There are two main categories of extremes indices: those based on absolute thresholds and those based on percentiles. The first category refers to counts of the days crossing a specified absolute value, whereas the second category of indices is based on the statistics of a climate variable (i.e. percentiles). Percentiles are generally preferable to absolute thresholds because they can be extended globally and across areas with varying elevation, although absolute thresholds (e.g. 0 °C) can be particularly helpful in some applications (Folland et al., 1999). Warm days and cold nights are percentile-based indicators, while the frost day index is the threshold-based indicator. All of them give information how frequently extreme events occur.

This paper concentrates on indices that refer to cold and warm extremes, i.e. which investigate the characteristics of the Serbian climate with respect to intense events and changes in the daily temperature. A list of the indices is shown in Table 1. The period 1961–1990 was set as the base period for determining the frequency distribution for the following indices: Tn10, Tx10, Tn90 and Tx90. Linear regression analysis was used to calculate possible trends of the indices.

Table 1. Summary of the seasonal trend analysis of the extreme temperature indices based on 15 stations in Serbia during the period 1949–2009
Extreme indicesWinterSpringSummerAutumn
  1. Sign in parentheses indicates the tendency of regional mean coefficients being not significant at the 5% level. Stations with positive or negative trend coefficients being significant at the 5% level are indicated by numbers.

Tn10: number of cold nights(−)

2–

(−)

11–

(−)

2–

Tx10: number of cold days(−)

15–

(−)

3–

(+)
Tn90: number of warm nights(+)(+)

1–, 7+

+

1–, 11+

(+)

1–, 1+

Tx90: number of warm days(+)(+)+

11+

(+)

1+

FD: number of frost days(−)(−)

9−

/(+)
DTR: daily extreme temperature range(+)(−)(+)

3+

(+)

2.4. Large-scale atmospheric circulation patterns

To describe the link of the extreme temperature indices with the large-scale atmospheric circulations, indices of teleconnection patterns such as the NAO, EA and EA–WR were investigated.

The NAO is one of the major modes of variability of the Northern Hemisphere atmosphere that exhibits considerable interseasonal and interannual variability. It is particularly important in the winter, when it exerts strong control on the climate of the Northern Hemisphere (Hurrell, 1995). In the positive NAO phase there is stronger zonal flow over the North Atlantic which brings warmer moister air to Europe. The EA consist of a north–south dipole of anomaly centres but displaced southeastward with respect to the NAO (Barnston and Livezey, 1987). The EA index is different from the NAO because it contains more subtropical influences. The positive phase of the EA pattern is associated with above-average surface temperatures in Europe in all months. The EA–WR pattern (Barnston and Livezey, 1987) is one of three prominent teleconnection patterns that affect Eurasia throughout the year. The positive phase is associated with positive height anomalies located over Europe and northern China, and negative height anomalies located over the central North Atlantic and north of the Caspian Sea.

The index values were taken from the NOAA Climate Prediction Center (CPC) Website http://www.cpc.ncep.noaa.gov/data/teledoc/.

3. Results

Six indices were calculated to assess changes in the trends of temperature extremes.

The increasing (+) and decreasing (−) seasonal tendencies of the indices for the entire territory of Serbia during the period of 61 years are summarized in Table 1. The sign of the trend coefficients is not directly indicative of a warming or cooling tendency. For instance, a negative trend of the number of cold days (Tx10) and positive trend of the number of warm nights (Tn90) both indicate warming. Therefore, the warming is shown by light grey, while cooling by dark grey.

3.1. Tn10 – cold nights

Tn10 represents the number of days where Tn is smaller than the 10th percentile of the daily minimum temperatures occurring during the base period. These days correspond to the lower Tn values recorded throughout a season, and their trends provide evidence regarding potential changes in the number of cold nights. All four seasons in Serbia are dominated by negative trends, which indicate decreases in the number of cold nights of the order from 3 days (in spring) to 4 days (in winter) during the 61-year period. These values were computed as the number of years multiplied by the regression coefficient.

During the summer, a pronounced decrease in the number of cold nights was found at 11 meteorological stations with the trend coefficients being significant at the 5% level (Table 1). The number of cold summer nights in Serbia during the period 1949–2009 with the linear regression and the EA index are presented in Figure 3(a). It can be seen that the number of cold nights in summer decreased (by 3.6 days) during the study period.

Figure 3.

Number of days smaller than the 10th percentile of the: (a) minimum temperature (Tn10) and (b) maximum temperature (Tx10) against the EA index and the correlation coefficient (r).

3.2. Tx10 – cold days

Tx10 represents the number of days where Tx is smaller than the 10th percentile of the daily maximum temperatures occurring during the base period. This indicator identifies trends in very cold days. The winter, spring and summer seasons were dominated by negative trend, revealing a trend to warmer conditions. During the spring, a decrease in the number of cold days (Figure 3(b)) of about 6 days during the study period was found in Serbia with a trend coefficient significant at the 1% level. During the autumn, 13 stations had a non-significant positive trend of the number of cold days, indicating a trend to colder conditions (not shown). This indicates that in Serbia the spring days became warmer and the autumn days colder.

3.3. Tn90 – warm nights

Tn90 represents the number of days where Tn is greater than the 90th percentile of the daily minimum temperatures occurring during the base period. All seasons showed positive trends (Table 1), revealing a general tendency for warmer nights in Serbia. Researchers (Heino et al., 1999; Jones et al., 1999; IPCC, 2007) have connected the rise in the mean global temperature with increases of Tn, which caused an increase of Tn90. The number of warm nights during the summer and the EA index are presented in Figure 4(a). The largest increase of about 12 days during the 61 years was found in the summer, with the trend coefficient being significant at the 1% level.

Figure 4.

Number of days greater than the 90th percentile of the (a) minimum temperature (Tn90) and (b) maximum temperature (Tx90) against the EA index and the correlation coefficient (r).

3.4. Tx90 – warm days

Tx90 represents the number of days where Tx is greater than the 90th percentile of the daily maximum temperatures occurring during the base period. Again, the seasonal indices show positive trends. The summer index revealed a significant positive trend at the 5% level towards an increased number of warm days in Serbia. This increase (of 11 days during the 61 years) is presented in Figure 4(b). The results from this index are consistent with those of the Tn90, which revealed an increased number of warm nights over Serbia.

Most of the warmest summers in Serbia during the examined period occurred within the periods 1950–1952 and after 1991 (Unkašević and Tošić, 2009b). These periods coincide with the number of warm days and with the results obtained for the Czech Republic (Kyselý, 2002), for the central European stations (Moberg and Jones, 2005) and for Athens, Greece (Efthymiadis et al., 2011).

According to Figures 3 and 4 we could not conclude that the circulation trends explained the temperature trends completely. The influence of other factors needs to be considered in future studies.

3.5. FD – number of frost days

FD represents the number of days with Tn below 0 °C. No significant trends in the seasonal pattern were revealed. During the winter and spring, negative trends in the number of frost days were obtained, while in the autumn, a positive one was found. A significant decreasing trend in the number of FDs was obtained at nine stations during the spring (Table 1) of the order of 4–7 days for the 61-year period.

3.6. DTR – daily temperature range

The DTR represents difference between the observed daily maximum and minimum temperatures. Most of the stations in Serbia showed a positive trend in the summer, while a negative one in the spring. A significant positive trend of the DTR at the 5% level was observed only at three stations during the summer (Table 1). This was caused by the greater increase of Tx than Tn during the summer in the period after 1975 (Unkašević et al., 2005). The positive trend coefficients of the DTR were very small during the autumn and winter; hence, they do not imply a warming tendency and they are assigned by white (Table 1). A reduction in DTR during the spring can be related to the greater increase in the minimum temperature than in the maximum temperature, as was found in other regions (Karl et al., 1993; Brázdil et al., 1996; Heino et al., 1999).

3.7. Relationships between extreme temperature indices and large-scale atmospheric pattern

In this section, correlation between the temperature indices and the large-scale circulation patterns are examined. It was found that the connections existed between the temperature indices and the EA index during the winter, spring and summer, the NAO index during the winter and with the EA–WR index during the autumn. All temperature indices were correlated with the EA index at least at the 5% significance level during the winter (Table 2). The negative phase of the EA caused more cold nights and frost days. In addition, the cold nights, cold days and the DTR were negatively correlated at the 5% significance level with the NAO index during the winter, reflecting the association with anomalously warm conditions over southern Europe (Table 2). According to Table 2, the summer correlation of Tn10 (Figure 3(a)) and Tx10 with the EA index was negative, while a positive one was found for Tn90 (Figure 4(a)) and Tx90 (Figure 4(b)). The weather regimes that bring dry and warm air from North Africa to the Balkan Peninsula (Unkašević and Tošić, 2009b) are the main reasons for the increase in the number of warm days and nights. Such a synoptic situation caused record values of the maximum temperatures across Serbia during the summer of 2007 (Unkašević and Tošić, 2011). Similar results for correlation of Tx10, Tn90 and Tx90 with the EA index were obtained for the winter and spring. The positive correlations of Tn10 and FD and the negative one of Tn90 with the EA–WR teleconnection pattern were obtained during the autumn. These were caused by penetration of cold air from western Russia into the Balkans.

Table 2. Seasonal correlation coefficients between the extreme temperature indices and the East Atlantic (EA) index during the winter, spring and summer, the North Atlantic Oscillation (NAO) index during the winter (in parenthesis) and the East Atlantic/West Russia (EA-WR) index during the autumn
Extreme indicesWinterSpringSummerAutumn
  1. Coefficients being significant at the 1 and 5% level are indicated by bold and italic, respectively.

Tn10−0.316 (−0.312) −0.3490.544
Tx10−0.357 (0.346)−0.365−0.460 
Tn900.4410.5120.526−0.351
Tx900.5350.3910.526 
FD−0.422  0.308
DTR−0.294 (−0.297) 0.274 

3.8. Leading modes of interannual variability of extreme air temperatures

In order to investigate the characteristics of the seasonal variability of the occurrence of extreme events over Serbia, EOF analysis was applied to the minimum and maximum temperature time series. The results of this analysis enabled the identification of possible leading modes in the interannual variability of the extreme events.

The explained variances of the first three EOF patterns of the seasonal minimum (Tn_EOFs) and maximum (Tx_EOFs) temperatures are presented in Table 3. For all four seasons, the first EOF mode of Tn and Tx is represented by the monopole pattern, accounting for 68.7–82.1% and 77.4%–87.1% of the total variance, respectively. The first two EOFs for the winter Tn and summer Tx are represented in Figure 5. The second EOF, which explains 7.2% (Tn, winter) and 6.9% (Tx, summer) of the total variance (Table 3), shows a dipolar structure (Figure 5) and indicates the influence of orography on the temperature regime. The zero line divides the area considered into two regions: southern (mountainous) with positive values and northern (plain) with negative values. The first leading mode of Tn and Tx and their associated time series or principal component (Tn_PC1 and Tx_PC1) can be used to represent the variability and to obtain their links with large-scale fields, as the EOF2 and EOF3 explained less than 10% of the variance. The correlation coefficients of the Tn_PC1 and Tx_PC1 with the EA and NAO indices are given in Table 4, from which it can be seen that the summer Tx_PC1 is positively linked to the summer EA pattern and negatively linked to the summer NAO. The NAO has a significant influence on the minimum temperatures in the winter because of the more intense pressure anomaly patterns.

Table 3. Explained variances (%) of the first three EOF patterns of the seasonal minimum (Tn) and maximum (Tx) temperatures
EOF patternsWinterSpringSummerAutumn
Tn_EOF176.882.168.779.1
Tn_EOF27.25.27.36.0
Tn_EOF34.53.84.73.2
Tx_EOF177.487.178.486.1
Tx_EOF27.44.36.94.4
Tx_EOF33.12.04.52.8
Figure 5.

The first two EOFs for the (a,b) minimum winter and (c, d) maximum summer temperatures.

Table 4. Correlation coefficients between the first mode of the minimum (Tn_PC1) and maximum (Tx_PC1) temperature and the large-scale modes of the North Atlantic Oscillation (NAO) index and East Atlantic (EA) index during the winter and summer
Mode of temperatureWinterSummer
 NAOEANAOEA
  1. Coefficients being significant at the 1 and 5% level are indicated by bold and italic, respectively.

Tn_PC10.3820.301  
Tx_PC1 0.422−0.3220.375

The time series of the principal components, as well as the NAO and the EA indices, are presented in Figure 6(a) during the winter and in Figure 6(b) during the summer. Figure 6(a) shows that the winter Tn_PC1 is positively correlated with the NAO index (Table 4), reflecting the association of a high NAO index with warm conditions over Serbia. It is shown that a positive phase of the NAO and EA was prevailing after 1980 during the winter (Figure 6(a)). The negative phase of the EA and the positive one of the NAO caused negative Tx_PC1 during the period 1965–1985 in the summer (Figure 6(b)). During the spring and autumn, the PCs do not show significant links with the major large-scale modes of atmospheric circulation, potentially implying the role of different circulation modes throughout the transitional seasons.

Figure 6.

Time series of the East Atlantic index, the NAO index and the PC1 of the (a) winter minimum temperature and (b) summer maximum temperature.

3.9. Summer climate analysis

In this section, a brief analysis of Tn90 and Tx90 during the summer is given. The greatest numbers of Tn90 (22.2) and Tx90 (24.0) during the summer in Serbia (Table 5) were observed in the last decade (2000–2009). The maximum values of Tn90 and Tx90 were recorded in 2003 (31.3) and 2000 (40.4), respectively. In addition, high values of Tn90 (27.8) and Tx90 (36.8) were observed in 2007, when the record values of the maximum temperatures were observed across Serbia (Unkašević and Tošić, 2011).

Table 5. Mean decadal number of warm nights (Tn90) and warm days (Tx90) during the summer in Serbia
DecadeTn90Tx90
1950–195914.715.8
1960–19698.99.6
1970–19797.05.9
1980–198910.610.3
1990–199917.317.2
2000–200922.224.0

A record heat wave affected many parts of Europe during the course of summer 2003. The largest monthly mean summertime temperature anomaly (of about 5 °C) occurred over the Euro-Mediterranean region in June and over southern France, Switzerland and southwest Germany in August 2003 (Schär et al., 2004). The strongest anomalies were centred over the Alps in June, extending as far as Scotland in August 2003. This exceptional behaviour was also observed for the 500-hPa geopotential height throughout the summer months (Beniston, 2004).

In Serbia, very intensive heat waves were observed in July 2007 (Unkašević and Tošić, 2009b). The 850- and 500-hPa geopotential anomalies (according to the reference period 1961–1990) exceeded 5 and 35 gpm, respectively, over Serbia during the summer 2007.

The summer of 2000 was also very warm. On average 40 warm days and 21 warm nights (Figure 4) were detected across Serbia. That summer was characterized by a strong positive geopotential anomaly at 500 hPa, based on the 1961–1990 reference period of about 40 gpm, over Serbia with a maximum over the Black Sea (Figure 7(a)). The summer 850-hPa geopotential anomaly reached 27 gpm over Serbia with the maximum located southeast of the Black Sea (Figure 7(b)). The sea level pressure showed a positive pressure anomaly (about 2.5 hPa) over Serbia (not shown). It is supposed that the one possible reason for this extremely warm summer could be the positive geopotential anomaly. In addition, the air flow from lower latitudes (Figure 7), i.e. SE flow at 850 hPa and S/SW flow at 500 hPa, would be a factor too.

Figure 7.

Geopotential height anomalies (gpm) at (a) 500 hPa and (b) 850 hPa in the summer 2000 based on the reference period 1961–1990.

4. Discussion and conclusions

Six temperature indices were calculated for the territory of Serbia using the extreme temperatures at 15 meteorological stations during the period 1949–2009. Similar to the global and European trends (Frich et al., 2002; Klein Tank et al., 2002; Kostopoulou and Jones, 2005; Bartholy and Pongrácz, 2007), analysis of the extreme temperature indices suggested that the Serbian climate generally tended to become warmer over the last 61 years.

According to the seasonal analysis of six temperature extreme indices, it was found that a warming tendency was dominant. It was mentioned that tendencies significant at the 5% level were obtained during the spring (Tx10) and the summer (Tn10, Tn90 and Tx90). The largest warming tendencies of greater than 1 day decade−1 were found in a number of warm days and nights (Tx90 and Tn90) in the summer. Similar to the results obtained in this study, the most significant trends identified for both the minimum and maximum temperature, exceeding the 90th percentile, especially during the warm season of the year, were found by Kostopoulou and Jones (2005), Moberg et al. (2006) and Rodríguez-Puebla et al. (2010). These indices imply increases in the summer nighttime and daytime temperatures with the beginning of these increases starting roughly during the late 1970s.

A cooling tendency of Tx10 and FD was revealed only during the autumn, suggesting a rise in the number of cold and frost days. A negative trend of Tx10 during the summer season and an increase in the number of frost days during the autumn throughout the eastern Mediterranean area were also found by Kostopoulou and Jones (2005). Previous studies (e.g. Frich et al., 2002; Kiktev et al., 2003) showed that these temperature indices exhibited coherent trends over the mid-latitudes during the second half of the 20th century.

To examine the links of the extreme temperature indices with the large-scale atmospheric circulations, indices of teleconnection patterns such as the EA and EA–WR were investigated. It was found that the Tn90 and Tx90 values were highly positively correlated with the EA index during the winter, spring and summer, while the Tx10 and Tn10 values were negatively correlated. The Tn10 values probably decreased because of the trend to more positive phases of the EA. Opposite correlations with the EA–WR index were found for the Tn10 and Tn90 values in the autumn. In the summer, Tx90 (Tn10) is negatively (positively) correlated with NAO over Serbia. It is unlikely that atmospheric circulation variability alone can explain the observed trends in extremes. The influence of other factors such as relationship between soil-moisture deficit and summer hot extremes in southeastern Europe (Hirschi et al., 2011) also needs to be considered in future studies.

The characteristics of the seasonal variability of the occurrence of extreme events over Serbia applying an EOF to the minimum and maximum temperatures were analysed. Considering the first mode of extreme temperatures, the EA pattern and the NAO dominated the Serbian extreme temperature variability during the winter and summer. According to Efthymiadis et al. (2011), the summer NAO positive phase (with anticyclonic conditions in NW Europe) caused northeasterly winds and lower-than-normal temperatures in the Eastern Mediterranean and vice versa for the negative phase of summer NAO. This could account, in part, for the observed positive correlation with cold extremes and negative correlation with hot extremes. The NAO was found to have a significant influence on the extreme winter daily temperatures in many areas in the world (Brown et al., 2008). In addition, Brown et al. (2008) concluded that the similarities in the influence of the NAO between the minimum tails of the daily minimum and maximum temperatures suggest that similar weather types were causing the extremes.

We consider that the results of this regional study could contribute to a better understanding of the warming tendencies and extreme variations in southeastern Europe, a region not always covered in European studies.

Acknowledgements

This study was supported by the Serbian Ministry of Science and Education, under grant no. 176013. The authors highly appreciate comments and suggestions of reviewers that led to a considerable improvement of the paper.

Ancillary