• Open Access

On the relationship between the Mediterranean Oscillation and winter precipitation in the Southern Levant

Authors


Abstract

Empirical Orthogonal Function (EOF) analysis was applied to 25 homogenous precipitation series in the Southern Levant covering the years 1960–1993. The EOF-1 explained 60–71% of variance and exhibited a significant correlation with a particular Mediterranean Oscillation Index (MOI) in December–February. It is shown that winter precipitation is associated with positive MOI phases and Cyprus lows. By fitting gamma distributions to monthly precipitation; it could furthermore be shown that during negative MOI phases, the probability of above average winter precipitation is 22%. During positive MOI phases, the corresponding probability is much higher, at around 59%.

1. Introduction

Many studies have focused on the synoptic systems associated with precipitation in the Mediterranean region (Maheras et al., 1999; Dünkeloh and Jacobeit, 2003; Xoplaki et al., 2004). In the East Mediterranean (EM), most precipitation is associated with mid-latitude cyclones (Black, 2011). Low pressure and rainfall in the EM is often accompanied by high pressure in the West Mediterranean (WM); together these systems comprise a pattern referred to as the Mediterranean Oscillation (Conte et al., 1989; Palutikof et al., 1996).

Well-known synoptic patterns can be expressed in terms of a teleconnection index based on the pressure differences between two distant locations; the prevailing wind direction can be determined by simply knowing the sign and the magnitude of the index. In the EM, most related studies have focused on the positive linkage between winter precipitation and either the North Atlantic Oscillation (NAO) or the Eastern Atlantic/Western Russian (EA/WR) pattern (Krichak and Alpert, 2005a, 2005b; Black, 2011). Considering that the Mediterranean Oscillation is linked to the NAO especially in winter (Dünkeloh and Jacobeit, 2003) it is therefore also logical to examine the region's precipitation regime in terms of the Mediterranean Oscillation Index (MOI) which is expressed as differences in sea level pressure (SLP) or geopotential heights between the WM and EM. Although winter precipitation in the WM (Gonzalez-Hidalgo et al., 2009) and EM (Ramadan et al., 2011) has been shown to have a negative and positive correlation with the MOI, respectively, this relationship has not been addressed in detail.

This study focuses on the Southern Levant located in the South-eastern Mediterranean region. The relationship between a particular MOI and the region's rainfall regime is addressed and considerations are taken into account regarding data homogeneity and local variations. The study more specifically aims (1) to identify homogeneous precipitation series and coherent spatiotemporal precipitation patterns, (2) to plot the large scale pressure anomalies associated with winter precipitation, (3) to derive the correlation between a particular MOI and winter precipitation and (4) to determine the probabilities of low and high winter precipitation during negative, normal and positive MOI phases.

2. Study region

The Southern Levant includes Israel, Jordan and the Palestinian Authority (Figure 1). The region has a Mediterranean climate with hot, dry summers and cool wet winters. Several distinct precipitation gradients are present which are influenced by orographic factors, a north to south gradient, proximity to the sea and the lee-side effect (Goldreich, 1994). The north-western elevated region receives moist winds from the Mediterranean Sea and is thus characterized by sub-humid conditions with an annual precipitation exceeding 700 mm, the mountainous region west of the Dead Sea stretching from the Golan Heights in the north to the Negev in the south acts as a barrier and creates an eastern lee side. The southern part of the study region experiences arid to hyper arid conditions with annual precipitation of less than 50 mm.

Figure 1.

Map of study region showing elevation, locations of precipitation gauges, as well as the average annual precipitation in mm. As the subsequent figures, the figure is based on the years 1960–1993.

The Southern Levant is one of the water scarcest in the world, with the strong inter-annual variation in precipitation frequently giving rise to drought conditions. This study focuses on the winter months of December–February in which about 64% of annual precipitation falls. Precipitation data obtained at a total of 15 Israeli and 15 Jordanian precipitation gauges were analysed, covering the years between 1960 and 1993. For later years, the continuous data availability was lower. The daily precipitation data were aggregated to monthly totals by considering December, January and February separately.

3. Methods and data

3.1. Homogeneity analysis and Empirical Orthogonal Function analysis

On the basis of a method described in Wijngaard et al. (2003), three homogeneity tests were applied to annual precipitation data (hydrological year October–September): the Standard Normal Homogeneity Test (SNHT) for a single break (Alexandersson, 1986), the Pettitt test (Pettitt, 1979) and the von Neumann ratio test (von Neumann, 1941). Each precipitation series was compared to a reference series defined as a weighted average of observations made at correlated (correlation coefficient > 0.70) surrounding stations. The weighting was carried out according to the correlation coefficient. The precipitation series were classified as homogenous if one or zero tests rejected the null hypothesis at the 95% confidence level. The series were also tested for serial correlation according to Haan (2002).

In the next step, Empirical Orthogonal Function (EOF) analyses (von Storch and Zwiers, 1999) were employed to extract two unrotated EOFs (EOF-1 and EOF-2) from the homogenous precipitation series for December–February. Additional EOFs only explained 6% or less of the total variance and were not considered. The eigenvector components were plotted as a map, and thereafter the correlations between a particular MOI and the coefficient time series related to the first two EOF modes were derived.

Before conducting the EOF analyses, both precipitation data and MOI values were standardized by subtracting the mean and dividing by the standard deviation. The data were also detrended by removing the best linear fit.

3.2. Mediterranean Oscillation and SLP

Several versions of the MOI exist; these are based on differences in either normalized SLP or 500 hPa geopotential height between the Western and Eastern Mediterranean. The MOI applied in this study is based on the normalized pressure difference between the dipoles Algiers and Cairo (Conte et al., 1989; Palutikof et al., 1996). This version of the index has been shown to correlate with moderate to heavy precipitation in Israel (Yosef et al., 2009) and will hereafter be referred to as MO1. Another version of the index (MO2) which is obtained from observation in Gibraltar and Israel will in addition be shortly addressed. The MO1 data were downloaded from the Climatic Research Unit (http://www.cru.uea.ac.uk/data) and tested for normal distribution by applying the Shapiro–Wilk test. The full length data series has a standard normal distribution N(0,1) with mean 0 and standard deviation 1. The data were classified into three MO1 phases: negative, normal and positive according to the following conditions:

display math(1)
display math(2)
display math(3)

where MO1 is the December, January and February MO1 considered separately.

Although this study focuses on the MO1, correlation analyses were also conducted between the EOF coefficient time series and four additional well-known teleconnection indices; NAO and EA/WR downloaded from the NOAA Climate Prediction Centre (ftp://ftp.cpc.ncep.noaa.gov/wd52dg/data/indices/tele_index.nh) as well as the North Sea Caspian pattern (NCP) and the MO2 available from the Climatic Research Unit.

SLP is an important parameter affecting atmospheric circulation and weather patterns. In this study, monthly 2.5° gridded SLP values were obtained from the NCEP reanalysis derived data provided by the NOAA/OAR/ESRL PSD (Kalnay et al., 1996). To obtain the SLP pattern associated with precipitation in the Southern Levant, Pearson correlation analyses were conducted between the EOF coefficient time series and the SLP of each grid points between ca 23°W to 70°E and 5°N to 67°N. By plotting the correlation coefficients, regions with above normal (positive correlation) and below normal (negative correlation) SLP by the time of rainfall in the study region could be identified. In the Northern Hemisphere, anomaly circulation is clockwise (anticyclonic) around centres of above normal SLP and counter clockwise (cyclonic) around centres of below normal SLP.

3.3. Precipitation probabilities

The probability of rainfall was derived for each MO1 phase based on a method employed by Muñoz-Díaz and Rodrigo (2003). Initially, December, January and February precipitation was calculated for each station and winter between 1960 and 1993. A gamma distribution, skewed to the right and defined for values above zero, was then fitted to the precipitation data of each station and MO1 phase. The gamma parameters were estimated with the methods of moments and the Kolmogorov–Smirnov test was applied to examine whether the observed and fitted data shared the same distribution. If the hypothesis could be supported, the probability of monthly precipitation below the 25th (P25), 50th (P50) and 75th (P75) percentile was derived for every single station and MO1 phase. This was carried out according to the cumulative distribution function of the fitted gamma distribution.

4. Results and discussion

4.1. Homogeneity analysis and EOF analysis

In this study, 30 precipitation series were available. Of these, 25 were identified as homogenous and having no serial correlation (Figure 1). EOF analyses were applied to the homogeneous precipitation data for the months of December, January and February separately. The calculated eigenvalues give that EOF-1 accounts for 60–71% and EOF-2 an additional 10–17% of total variance. Eigenvector plotting revealed all EOF-1 eigenvector components to be positive; every single precipitation series is therefore positively correlated with the coefficient time series of the first EOF (Figure 2, upper row). Eigenvector components associated with the central part of the study region are between 0.80 and 0.90, indicating a strong relationship between the EOF-1 coefficient time series and the original precipitation data. This pattern of values forms an obvious gradient to that of the south, where annual precipitation is also much lower. The eigenvector components for the Gulf of Aqaba are below 0.40, which illustrates how this area's temporal variation in rainfall is less related to that of the rest of the region. Furthermore, the climates of the Jordan Rift Valley and the Negev are also known to be more associated with the Red Sea trough (Kutiel and Paz, 1998).

Figure 2.

EOF-1 (upper row) and EOF-2 (lower row) eigenvector components represented as isolines, explaining 60–71 and 10–17% of total variance in monthly precipitation, respectively. In the upper row, the correlation coefficient between single precipitation series and the MO1 is also shown.

EOF-2 eigenvector components are characterized by both positive and negative values; for positive EOF-2 time coefficients, analysis of Figure 2 (lower row) reveals the presence of a spatial pattern associated with precipitation (positive eigenvector components) in the South-eastern arid region and a lack of precipitation (negative eigenvector components) in the northern sub-humid region in December and February. Since values increase inland, it is clear that the second EOF cannot be related to moist westerly winds as seems to be the case with EOF-1. In January, the EOF-2 eigenvector components reveal a pattern associated with precipitation in the north and drier conditions in the central and southern parts during a positive MO1 phase.

4.2. Mediterranean Oscillation and precipitation

The correlation coefficients between the MO1 and single precipitation series reach values up to 0.60 (Figure 2). Further analyses give a significant correlation between the EOF-1 coefficient time series and the MO1; the obtained correlation coefficient is 0.48, 0.39 and 0.42 in December, January and February, respectively (not shown). No correlation is apparent between the EOF-2 coefficient time series and MO1 values. The results also identify the NAO, EA/WR, NCP and MO2 as highly related to precipitation. A significant correlation between the EOF-1 coefficient time series and NAO is obtained in December whereas the EA/WR and NCP are significantly correlated to the EOF-1 in both December and February. For the MO2, a significant correlation is obtained for all three winter months. The correlation however, is weaker than for the MO1. The linkage between the NAO, EA/WR, NCP and MO2 and precipitation agrees with studies conducted by Kutiel and Benaroch (2002), Dünkeloh and Jacobeit (2003) and Krichak and Alpert (2005a, 2005b), among others.

To identify the synoptic systems associated with precipitation in the study region, both the EOF-1 and EOF-2 coefficient time series were correlated with SLP values of single grid points. This process resulted in correlation coefficients ranging from −0.5 to 0.5 (Figure 3). For positive time coefficients, the results of SLP/EOF-1 analysis show above normal SLP over central Europe and the WM. In December and February, the high pressure is centred over United Kingdom and central Europe and extends from northern Africa to southern Scandinavia. In January the pattern is weaker. At the time of rainfall, a low pressure tends to penetrate the EM and form a Cyprus low; the resultant cyclone favours the formation and introduction of moist westerly winds into the study region. The large scale pattern is similar to the results observed in several earlier studies (Kutiel and Paz, 1998; Krichak et al., 2000; Tolika et al., 2007); the location of the cyclone centred south or east of Cyprus is known to be associated with above normal levels of rainfall in southern parts of the EM (Saaroni et al., 2010). Since the SLP in Algiers is above normal and that in Cairo below normal, the figure also represents a positive MO1 phase. This result is in agreement with the previously identified positive correlation between the MO1 and the EOF-1 coefficient time series.

Figure 3.

Correlation coefficient between the EOF-1 (upper row) and EOF-2 (lower row) coefficient time series and the SLP of single grid points. The MO1 dipoles in Algiers and Cairo are also marked.

The results of SLP/EOF-2 analysis show a less coherent pattern in which the area of above normal SLP has moved eastwards in December and February and weakened in January. No clear gradient exists between the WM and EM or the MO1 dipoles (Figure 3). Hence, as in the previous analysis, no linkage between the MO1 and EOF-2 can be identified.

4.3. Precipitation probabilities

The MO1 phases and their associated probability of precipitation were derived based on the P25, P50 and P75 thresholds. As a first step, the normal distribution of the MO1 was ensured by applying the Shapiro–Wilk test with a confidence level of 95%. Following this process, the MO1 could then be divided into negative, normal and positive phases. Of the 102 months (34 years × 3 months) for which data were available, 29 were classified as MO1 negative, 48 as MO1 normal and further 25 as MO1 positive. After fitting a gamma distribution to each station and MO1 phase, the Kolmogorov–Smirnov test showed the observed and fitted values to share the same distribution at the 95% confidence level. Only the station in Eilat failed the test and was excluded from the calculation of the precipitation probabilities and the visualization of the results. Figure 4 displays the observed precipitation and fitted gamma distribution for all MO1 phases for a single selected station (Poriya) located in the northern part of the study region. It can be seen how the fitted distribution is shifted towards higher precipitation values during positive MO1 phases.

Figure 4.

MO1 phases and observed December, January and February precipitation (bars), as well as the precipitation percentiles P25, P50 and P75 for the Poriya station. Also shown are the fitted gamma distributions, including the number of months (n), fitting parameters (Alpha and Beta) and the Kolmogorov–Smirnov test statistic D.

On the basis of cumulative distribution function of the fitted gamma distributions, the probabilities of abnormally low and abnormally high precipitation were derived separately for each station and MO1 phase, with the results then spatially interpolated (Figure 5). Figure 5 shows that negative MO1 phases are associated with low (precipitation < P25) and below average (precipitation < P50) December, January and February precipitation. During negative MO1 phases, the station probability of low precipitation is on average 39%, whereas during normal and positive MO1 phases the corresponding probabilities are 25 and 10%, respectively. The same tendency can be observed with respect to below average winter precipitation, with calculated probabilities of 78, 61 and 41% for negative, normal and positive MO1 phases, respectively. In contrast, the probabilities for above average (precipitation > P50) and high (precipitation > P75) precipitation are characterized by the opposite relationship. During negative MO1 phases, the probability of receiving above average precipitation is 22%, compared to 39% during normal MO1 phases. Since the corresponding value is 59% during positive MO1 phases, it is shown how only the latter are associated with above average precipitation. The probability of receiving high precipitation during negative MO1 phases is 11%, during normal MO1 phases 24% and during positive MO1 phases 41%. Positive MO1 phases are therefore also associated with high precipitation.

Figure 5.

The probability (%) of low (precipitation < P25), below average (precipitation < P50), above average (precipitation > P50) and high (precipitation > P75) precipitation in December–February during negative, normal and positive MO1 phases. P25, P50 as well as P75 denote percentiles of monthly precipitation. This means that the average station value (avg.) of precipitation < P50 and precipitation > P50 sums to 100% for every given MO1 phase.

5. Summary and conclusions

An EOF analysis was conducted on homogeneous precipitation series in the Southern Levant; the resulting EOF-1 and EOF-2 accounted for 60–71% and 10–17% of total variance, respectively. The EOF-1 coefficient time series was furthermore significantly correlated with the MO1, whereas no correlation was observed for EOF-2. In order to determine the synoptic systems associated with precipitation in the study region, the EOF coefficient time series were then correlated with gridded SLP. The SLP/EOF-1 pattern was associated with a positive MO1 phase characterized by high pressure above central Europe and the WM, and a Cyprus low in the EM. By fitting gamma distributions to observed precipitation values, it could furthermore be shown that negative MO1 phases are associated with low and below average winter precipitation, whereas positive MO1 phases are associated with above average and high precipitation. Although not extensively addressed in this article, the same methodology was applied to four additional well-known teleconnection indices; all were less associated with winter precipitation than the MO1 was.

Acknowledgements

This study was funded by the German Federal Ministry of Science and Education (BMBF) within the GLOWA Jordan River project. The precipitation data were provided through cooperation within the project. The author is thankful to two anonymous reviewers for their helpful comments.

Ancillary