INTERNATIONAL JOURNAL OF CLIMATOLOGYInt. J. Climatol. 31: 2055–2067 (2011)Published online 13 September 2010 in Wiley Online Library(wileyonlinelibrary.com) DOI: 10.1002/joc.2212

Use of the spectral clustering to determine coherentprecipitation regions in Turkey for the period 1929–2007

Murat Turkes and Hasan Tatlı*Physical Geography Division, Department of Geography, Canakkale Onsekiz Mart University, 17020 Canakkale, Turkey

ABSTRACT: In this study, we suggest the spectral clustering (SC), a hybrid clustering technique based on singular valuedecomposition (SVD) and K-means for grouping features of precipitation totals of 96 stations in Turkey. Clustering processestablishes an exhaustive set of occupied regimes into distinct climatic zones. Results of the SC satisfactorily represent theinfluences of the synoptic-scale weather systems including such as the mid-latitude and Mediterranean frontal cyclones, andthe mid-latitude travelling and eastern Europe high pressures in winter, sub-tropical Azores high pressure and monsoon lowin summer. Results of the SC also well display the influences of local-scale atmospheric disturbances, and direct influencesof physical geographical features of Turkey (i.e. exposure, topography, orography, land-sea distribution, continentality andthe high Anatolian peninsula) on the geographical variability and coherent distribution of the annual precipitation totalsover Turkey. Finally, based on the results of the SC method employed to annual precipitation totals of 96 stations in Turkeyfor the period of 1929–2007, eight clusters of precipitation coherent zones are determined, namely Black Sea, NorthwestTurkey, southern Aegean and western Mediterranean, Mediterranean, West Continental Central Anatolia, East ContinentalCentral Anatolia, Continental eastern and south-eastern Anatolia. Copyright 2010 Royal Meteorological Society

KEY WORDS K-means; precipitation; spectral clustering singular value decomposition; Turkey

Received 20 November 2009; Revised 8 July 2010; Accepted 27 July 2010

1. Introduction

In most cases, for the climatological and meteorologi-cal applications, comprehensive information is requiredwith a regionalized spatial variability of climate data.One of the regionalization methods is clustering tech-nique. Cluster analysis is based on some form of sim-ilarity matrix for the clustering of climatic time-seriesattempts to find the groups of datasets having similarcharacteristics. These groups can then be further analysedin detail to gain insight from the common characteristicsof the datasets in each group of climatic sub-regions. Forthat reason, cluster analysis in climatology/meteorologyis to define classes of synoptic types or climate regimes,or stations and/or grid points to define climatic sub-regions (Lund, 1963; Anderberg, 1973; Richman andLamb, 1985; El-Kadi and Smithson, 1992; Fovell andFovell, 1993; Wilks, 1995; Bunkers et al., 1996; vonStorch and Zwiers, 1999; Spellman, 2000; Unal et al.,2003; Gordon et al., 2005; Donna et al., 2007; Bisginand Dalfes, 2008; van Cooten et al., 2009; Kucken et al.,2009).

In many fields, spatial distribution of precipitationamounts is required. The spatial distribution of precip-itation treats clustering data. For example, hydrologistsand groundwater hydrologists ask precipitation values

* Correspondence to: Hasan Tatlı, Physical Geography Division, De-partment of Geography, Canakkale Onsekiz Mart University, 17020Canakkale, Turkey. E-mail: tatli@comu.edu.tr; tatli.hasan@gmail.com

corresponding to their catchments areas of interest. Theecologists who are studying the dynamics and responsesof forests and various species in the mountainous terrainneed to know features of precipitation. It is evident toincrease such examples.

The climate of Turkey and its nearby regions is char-acterized mainly with the Mediterranean macro climategoverned by an unequivocal seasonal alteration betweenthe mid-latitudes and sub-tropical/tropical pressure andwind systems (circulation patterns) from winter to sum-mer (Erinc, 1969; Trigo et al., 2006; Turkes, 1996, 1998;Tatli et al., 2004, 2005; Turkes and Erlat, 2003, 2005).There is, therefore, a highly seasonal precipitation regimein the western and southern regions, continental semi-aridclimate in the Central and eastern Anatolia regions, andhumid-temperate, uniformly rainy climate in the northernMarmara and the Black Sea regions of Turkey (Erinc,1969; Turkes, 1996, 1998, 1999).

The precipitation climatology, drought events andtheir spatial distribution and long-term variability, trendsand changes in precipitation series over Turkey havebeen comprehensively analysed previously, for exampleby Turkes (1996, 1998, 1999, 2003), Turkes et al.(2002), Tatli et al. (2004), Aksoy et al. (2007); Tatli andTurkes (2008), Turkes and Tatli (2009), Turkes et al.(2009) and Saris et al. (2010). Many studies indicatethat precipitation regimes and variability in Turkey withthe various time-scales are related with the locations,variations and activity of the atmospheric systems overthe Mediterranean and beyond, except in summer months

Copyright 2010 Royal Meteorological Society

2056 M. TURKES AND H. TATLI

(e.g. Turkes, 1998; Kutiel et al., 2001; Turkes et al.,2002; Turkes and Erlat, 2003, 2005, 2006; Tatli et al.,2004; Kutiel and Turkes, 2005; Tatli, 2006; Turkes et al.,2009).

Then again, some recent studies revealed that theNorth Atlantic Oscillation (NAO) could be one of themajor atmospheric controls for the spatial and temporalvariability of precipitation series over Turkey includingthe significant wet periods and meteorological droughts(Turkes and Erlat, 2003, 2006; Tatli, 2006; Trigo et al.,2006).

In a recent study, Turkes et al. (2009) have investigatedspatial relations among the precipitation variations ofTurkish stations by principal component analysis (PCA),making use of the knowledge that explains influencesof the major physical geographical control mechanismsof Turkey, which are consist of the regional-scale circu-lation, topography, land-sea distribution, exposition andcontinentality, on regional rainfall regime types as thebasis for spatial assessments over Turkey.

The main results of Turkes et al. (2009) are brieflygiven in the following in order to bridge them withthis study. In the relevant study, they found that theeigenvalue of the first principal component (PC1) isthe highest in winter, and generally explains the varia-tions of spatial relationships in precipitation series thatare closely influenced by large-scale weather systemsin west and relatively smaller-scale systems in the eastcontrolled by the regional physical geographical factors.They have also underlined a relatively weak correlationpattern seen over the middle-east and north-eastern Ana-tolia sub-regions could be a result of the influence ofdominant weather systems associated with the large-scaleatmospheric circulation at the west half of the coun-try, and forcing effects of nearly west to east trendinghigh ranges of the northern Anatolian Mountains to oro-graphically lift of the maritime polar air masses overthese mountain ranges. On the other hand, the secondPC pattern of the winter precipitation characterized withthe highest loadings over the Continental Mediterraneanand Continental Eastern Anatolia rainfall regions couldbe explained by the influences of humid-temperate andthus conditionally unstable Mediterranean weather sys-tems on the variability of spatial relations. Turkes et al.(2009) have found that the PC3 pattern treats humid-temperate influences of the Mediterranean Sea on win-ter precipitation totals by making stronger direct influ-ence of the mid-latitude cyclones over the Mediterraneanregion.

They have stated that the PC1 of summer explainsdirect influences of mid-latitude cyclones related withthe large-scale atmospheric circulations reaching fromeastern Europe to Turkey. Accordingly, the loadings ofPC2 established over western Black Sea sub-region couldexplain orographic rainfalls well over the region causedby the forcing of north-westerly air flows carrying fromthe Atlantic sourced polar air to the north of Turkey.However, the loadings, mainly over the continentaltype rainfall regions of Turkey, very likely explain the

influences of regional and/or local convective instabilityover these regions in summer. They have also revealedthat the loadings of PC3 over the eastern Black Seaand north-eastern Anatolia sub-regions of Turkey, whichare the rainiest districts in summer, satisfactorily explainthe influences of coastal orographic and continentalconvective instability showers and thunderstorms, both ofwhich are closely associated with northerly sector surfaceand upper-air flows and atmospheric disturbances; andthey have concluded that the regional surface warmingcontributes to strengthen that mechanism in the summermonths.

Even though this study involves an objective classifica-tion of Turkish precipitation series by using the spectralclustering (SC) method, the information and maps of pre-cipitation climatology of Turkey such as the spatial vari-ations of long-term precipitation averages, coefficientsof variation, seasonality of precipitation totals, droughtevents and regions and drought probabilities are not givenin this study, because the detailed assessments for sea-sonal and annual precipitation climatologies had beenperformed previously, and one may find in the worksof Turkes (1996, 1998, 1999, 2003), Tatli et al. (2004),Turkes and Tatli (2009) and Turkes et al. (2009).

This paper includes the following aspects: after Section1, the major features of SC method and data used in thestudy are described in Section 2. Results of the studyare presented in Section 3, and some comments anddiscussion are given in Section 4, and finally conclusionsare given in Section 5.

2. Data and methodology

2.1. Precipitation data

The data set used in this study includes yearly precipi-tation totals (millimetres) recorded at the stations of theTurkish State Meteorological Service. The precipitationdata set has been developed by Turkes (1996, 1998)for 99 stations covering the period 1929–1993, and 96of these stations were updated to 2007 for this study(Figure 1). This station-based precipitation data set hasthe most continuous/homogeneous and longest precipita-tion time-series of Turkey, and they mostly have 65 to77-years’ records and represent a very good spatial dis-tribution over Turkey. The required information for thedata set and homogeneity analyses applied to the long-term precipitation series can be found in Turkes (1996,1998, 1999).

2.2. Methodology

The employed technique of SC in this paper builds onobservations presented in (Ng et al., 2002; Von Luxburg,2007). Here, we briefly review the algorithm of theso-called Ng–Jordan–Weiss (NJW algorithm) as in thefollowing:

Given a set of X = {x1, . . . , xm} that we want toclassify them into k clusters then

Copyright 2010 Royal Meteorological Society Int. J. Climatol. 31: 2055–2067 (2011)

DETERMINING COHERENT PRECIPITATION REGIONS IN TURKEY 2057

Figure 1. Spatial distribution of 96 stations used in the study (names of the meteorology stations are abbreviated) over the rainfall regimeregions of Turkey. BLS: Black Sea; MRT: Marmara Transition; MED: Mediterranean; MEDT: Mediterranean Transition; CMED: Continental

Mediterranean; CCAN: Continental Central Anatolia and CEAN: Continental Eastern Anatolia.

1. Form the affinity (similarity) matrix Cm×m to bedefined by

Cij = exp(− [

xi − xj

]2/2s2

)for i �= j,

and Cii = 0 (1)

where the scaling parameter s given in Equation (1)controls how rapidly the affinity matrix C falls offwith the distance between x1 and x2. Since we havestandardized the precipitation data to have zero mean andvariance one, we have selected s = 1.2. Define E being the diagonal matrix whose (j ,j )-

element is the sum of C’s j-th rows defined by

Eij =∑j

Cij (2)

3. Construct matrix of L as

L = E−1/2CE−1/2 (3)

4. Find u1, u2, . . . , uk, the first k largest eigenvectorsby singular value decomposition (SVD) of the matrixL (k < m), and assume to form the matrix Um×k =[u1 u2 . . . uk] by ordering the eigenvectors incolumns.

5. Form the matrix P from U by renormalization of eachrow of U’s to have unit length as

Pij = uij

/ ∑

j

u2ij

12

(4)

6. Treating each row of P as a point in k-dimensionalspace, and cluster them into k clusters by using the tra-ditional clustering techniques of K-means (MacQueen,1967).

7. Finally, assign the original point xi to cluster j if andonly if row i of the matrix P was assigned to cluster j .

Technically, there are some methods for deciding onthe number of clusters. Moreover, nearly all techniquesfor visualizing multivariate data can also be used forcluster visualization. Among the separation measures,validity index (VI ) and cophenetic correlation (C) arewell-known measures (Sokal and Rohlf, 1962; Dunn,1974; Davies and Bouldin, 1979; Bezdek and Pal, 1998;Ray and Turi, 1999). Besides, Volkovich et al. (2008,2010) have suggested another external cluster validationindex with the misclassified quantities obtained in theprocess of repeated clustering based on the measuring ofcluster stability.

For instance, in the K-means problem, assume that aset of X = {x1, . . . , xm} in m-dimensions is given. Thegoal is to arrange these points into K clusters, with eachcluster having a representative point Z = {z1, . . . , zk},usually chosen as the centroid of the points in the cluster.The energy of each cluster (Ek) is then obtained as

Ek = 1

n

∑m

i=1(xi − zk)

2 (5)

For a given set of clusters, the total energy is thensimply the sum of the cluster energies Eks. The goal isto choose the clusters in such a way that the total energyis minimized. Usually, a point xi goes into the clusterwith the closest representative point zk . Hence, to definethe clusters, it is enough simply to specify the locations ofthe cluster representatives. On the other hand, similar tothe measure of cluster energies, the energy within clusters(Pj s) could be defined, where the clusters being as far aspossible away from each other as

Pj =∑K

i=1(zj − zi)

2, i �= j (6)

Copyright 2010 Royal Meteorological Society Int. J. Climatol. 31: 2055–2067 (2011)

2058 M. TURKES AND H. TATLI

Ray and Turi (1999) suggest a measure of the so-calledvalidity index (VI ) based on cluster energies:

V I =

k∑1

Ek

min(Pk)(7)

where Ray and Turi (1999) have named the term of1m

∑k1 Ek as intra and min (Pk) as inter, respectively.

That is, by the notations of Ray and Turi (1999), VI isdefined by

V I = intra

inter=

1

m

k∑j=1

∥∥x − zj∥∥

min(∥∥zi − zj

∥∥2) (8)

where ‖·‖ indicates the metric of norm. Obviously,VI defined above is going to be minimized where wewill have minimized cluster energies (i.e. Ek’s) andmaximized within cluster energies (i.e. Pk’s) values.Figure 2 shows a comparison of the performances ofthe SC with K-means. In this figure, the original patternis formed from nested two circles, while SC is able toextract these two circles perfectly; however, the K-meansclassifies the pattern into two half circles, on the otherhand.

3. Results

The names and spatial distribution patterns of major rain-fall regime regions of Turkey developed by Turkes (1996,1998) were used in spatially assessing results of the SCapplied to annual precipitation total series measured at96 stations over Turkey (Figure 1). This previous well-known classification of Turkish rainfall regime regionswas realized with respect to the seasonality and physi-cal geographical characteristics of monthly and seasonalprecipitation totals. In this study, our aim is to extractthe major spatial dependency of precipitation by SC inthe following analyses. We have applied the VI mea-sure for deciding the number of precipitation clusters.After employing the VI measure for clustering precipita-tion data, we have had eight clusters, but here the resultsof clusters are interpreted separately from 2 to 8 in order

(a) (b)

Figure 2. Clustering example of two circles, with clusters indicatedwith different symbols. (a) Spectral clustering, (b) K-means cluster.

to bridge the rainfall regimes with the obtained clustersunder view of not only statistical measure but also usingcommon climatological and physical geographical fea-tures. The abbreviations of cluster-2, cluster-3,. . . andcluster-8 used in the following text indicate clustering theprecipitation data into 2, 3, . . ., 8 clusters, respectively.

3.1. Cluster-2

According to the results of the regionalization of Turkishannual precipitation totals by the SC method, the ter-ritory of Turkey first separates into two large rainfallregions with the exception of a small area over the Isk-enderun Bay and the Amanos Mountain of the easternMediterranean sub-region of the Turkish Mediterraneancoastal region (Figure 3). The resulting main western partof Turkey with the number one cluster region includeswestern Black Sea sub-region, both western sub-regionsof the Continental Central Anatolia and the ContinentalMediterranean rainfall regions along with the MarmaraTransition, the true Mediterranean and the MediterraneanTransition rainfall regime regions (Figure 3). The spa-tially coherent western precipitation region with the firstcluster of the cluster-2 is mainly explained with precip-itation occurrences and variations controlled closely bythe Northeast Atlantic and the Mediterranean originatedlarge-scale pressure and wind circulation patterns. Thesepatterns are connected to those of synoptic-scale weathersystems mostly including mid-latitude cyclones and anti-cyclones, sub-tropical anticyclones and circulation-basedinfluence of tropical weather systems such as north-western extension of the southern Asiatic monsoon lowassociated with the northward seasonal migration of theinter-tropical convergence zone (ITCZ) during the sum-mer months, and so on. Consequently, this precipitationregion is closely characterized mostly by the mid-latitudeand the Mediterranean cyclones and associated frontalrainfall events over Turkey (Turkes, 1996, 1998; Kutielet al., 2001; Turkes et al., 2002, 2009; Turkes and Erlat,2003, 2005, 2006; Tatli et al., 2004; Kutiel and Turkes,2005; Tatli, 2006, and so on). Furthermore, orographiclifting of the air masses over mountainous areas of thewestern Black Sea sub-region and the Mediterraneanregion have a great importance for precipitation occur-rence, while precipitation occurrence and variations overthe continental inner regions are also dominated mainlyby the local and/or regional convective instability rainfallevents in spring and early summer months of the year.

On the other hand, the ensuing main eastern partof Turkey with the second cluster of the cluster-2 isdetermined strongly by influences of the weather systemsoriginated from the further northern and the easternEuropean regions along with the western Siberia regionof the great Asia continent. The rainfall events over thateastern region of Turkey are developed by controls ofthe mid-latitude cyclones, particularly their cold frontsextending towards further southern regions of Turkey,northerly circulation types and the convective instabilityevents over the continental inner regions of the high

Copyright 2010 Royal Meteorological Society Int. J. Climatol. 31: 2055–2067 (2011)

DETERMINING COHERENT PRECIPITATION REGIONS IN TURKEY 2059

Figure 3. Geographical distribution patterns of the cluster-2 results extracted from the spectral clustering approach applied to annual precipitationtotal series measured at 96 stations over Turkey. This figure is available in colour online at wileyonlinelibrary.com/journal/joc

Figure 4. As in Figure 3, but for the cluster-3 results of annual precipitation totals. This figure is available in colour online at wileyonlinelibrary.com/journal/joc

Anatolian peninsula (Turkes, 1996, 1998, 1999; Kutielet al., 2001; Tatli et al., 2004; Turkes and Erlat, 2005,2006; Tatli, 2006; Turkes et al., 2009).

The precipitation occurrence and variability over thehumid-temperate Black Sea region are significantly con-tributed by orographically induced rainfall events relatedwith the northerly circulations mostly ignoring whetherthey are cyclonic or anticyclonic flows in nature. Thisregion, at the same time, except the Mediterranean Tau-rus Mountains at the west, is the highest elevation area ofTurkey’s territory. With the exception of the Black Searainfall region of eastern Turkey region, the mountain-ously higher elevation Northeast Anatolia sub- region ofthe region receives most of the precipitation amount in thelate spring and summer months. The primary reason ofthis clear situation is of the higher territory, late warmingof the surface soils of the area due to the continentality

and higher elevation of the area with snow occurrencesheavily in winter, and convective instability events hav-ing controlled by the lateness activity of upper-air atmo-spheric disturbances (e.g. troughs or lows). This is theprimary and leading atmospheric factor.

3.2. Cluster-3

In the spatial pattern of cluster-3, the previous region-alization pattern derived from the cluster-2 increases tothree rainfall regions including mainly the continentalinner region of Turkey (Figure 4). With respect to thisspatial pattern, the region of the first cluster is taken as aregion of reflecting influences of the large-scale weathersystems particularly related to those of the NortheastAtlantic, the Mediterranean and the northern and easternEuropean originated systems. Second cluster precipitationregion is influenced by both of the eastern Mediterranean

Copyright 2010 Royal Meteorological Society Int. J. Climatol. 31: 2055–2067 (2011)

2060 M. TURKES AND H. TATLI

Figure 5. As in Figure 3, but for the cluster-4 results of annual precipitation totals. This figure is available in colour online atwileyonlinelibrary.com/journal/joc

cyclones and the eastern European–western Siberia highpressures and their associated circulation types in winter,and the monsoon low-pressure related tropical circula-tion via dry and hot Mesopotamia region in the summermonths. The third cluster region of the cluster-3 is mainlythe result of spring and early summer time rain showersand thundery showers (precipitation producing thunder-storms) controlled by the regional scale convective insta-bility events over that region with the exception of theBlack Sea rainfall region.

3.3. Cluster-4

The regional pattern of cluster-4 is the one that is spatiallydivided into two parts of the previously extracted westernand eastern regions of Turkey in the clustering level-twoof regionalization (Figure 5). In other words, this pat-tern gives a facility to make a detailed examination ofthe pattern of these two rainfall regions. Consequently,number four of the SC rainfall region that correspondsto Northwest Turkey basically reflects the influences ofthe Northeast Atlantic and partially Mediterranean origi-nated weather systems. Most of the precipitation amountsover that region are mostly produced under the control ofthe mid-latitude cyclonic variability. Effect of the orogra-phy is evident on the western Black Sea sub-region. Thefirst cluster rainfall region of the cluster-4, which can becalled as the south and south-west Turkey, generally cor-responds to the true Mediterranean rainfall regime regionwith a hot and dry long summer and rainy and temperatewinter along with a high climatic seasonality as a whole.Precipitation occurrences and variability over this regionare closely controlled by the eastern Mediterranean basinoriginated frontal low-pressure systems and the Azoreshigh-pressure systems. Some of the precipitation occur-rences and variability are associated with frontal lifting,while some particularly over the Mediterranean TaurusMountains and coastal belt of the Anatolian peninsula

are explained with cloud and precipitation occurrencesrelated to orographic lifting. Meteorologically, precipita-tion occurrences are mostly characterized by the frontalshowers and thundery showers associated relatively large-scale frontal thunderstorms over the region.

The inner-southwest area at north of the Antalya Bayand the southern-west of the Continental Central Anatoliarainfall regime region have the precipitation occurrenceand variability that is also significantly supported by therainfall showers due to the convective instability over thatregion. The third cluster of precipitation region, whichcovers the western and eastern Black Sea sub-region,eastern and northern parts of the Continental CentralAnatolia, and so on can be explained by both the northernand eastern European originated weather systems at thenorth (i.e. the Black Sea Region) and orography andconvective instability events in the Central Anatoliaregion. These activities are also facilitated by northerlyupper-air troughs and low centres. Precipitation eventsare mostly in the forms of rain showers and thunderyshowers with the thunderstorms at the inner regions, andin the forms of strong and long-lasting rainfall and rainshowers in the Black Sea region. Snowfalls are of greatimportance over the high elevation continental regionsand the North Anatolia Mountains.

The second cluster of precipitation region of thecluster-4 pattern, which is called as eastern and south-eastern Turkey, mainly reflects the influences of theweather systems that originated from the eastern Euro-pean – western Siberia – Caspian Basin in winter, and ofthe weather systems from the south-western Asia and themonsoonal Asia. This region, with the exception of lowplateaus and the Tigris River plains staying at the south-ern part of the south-eastern Taurus Mountains, is of thehighest and the most mountainous area of Turkey. In thenorthern part of this region, snowfalls and snowy condi-tions in winter and the rain showers and thundery showers

Copyright 2010 Royal Meteorological Society Int. J. Climatol. 31: 2055–2067 (2011)

DETERMINING COHERENT PRECIPITATION REGIONS IN TURKEY 2061

Figure 6. As in Figure 3, but for the cluster-5 results of annual precipitation totals. This figure is available in colour online atwileyonlinelibrary.com/journal/joc

Figure 7. As in Figure 3, but for the cluster-6 results of annual precipitation totals. This figure is available in colour online atwileyonlinelibrary.com/journal/joc

in summer take an important aspect of this precipitationregion.

3.4. Cluster-5

The most attractive aspect of the results of spectralcluster-5 (Figure 6) of the precipitation regionalizationof Turkey is that the number one Mediterranean pre-cipitation region extracted from the cluster-4 level isdivided into two parts with a small change occurs in theIzmir district: the western and the eastern Mediterraneanregions. Western Mediterranean precipitation region isclosely characterized by the westerly and south-westerlyweather systems, while the eastern Mediterranean regionreflects the influences of southerly and south-easterlyweather systems. The eastern Mediterranean region hasthe stronger summer dryness (long-lasting climatological

drought) with the higher maximum temperatures thanthose in the western Mediterranean region.

3.5. Cluster-6

The resultant pattern of the spectral cluster-6 (Figure 7)appears as the fractions of the number four region (i.e.Northwest Turkey) of the cluster-4 and the re-occurrenceof the Black Sea precipitation region except its westernpart. This distribution pattern may be accepted to re-appear at the Mediterranean to Black Sea transition fromthe Marmara and the northern Aegean regions.

3.6. Cluster-7

Regionalization results of the spectral cluster-7 (Figure 8)give a precipitation region pattern of Turkey that hasthe clear similarity with both Turkey’s conventional geo-graphical regions and Turkes’s (1996, 1998) precipitation

Copyright 2010 Royal Meteorological Society Int. J. Climatol. 31: 2055–2067 (2011)

2062 M. TURKES AND H. TATLI

Figure 8. As in Figure 3, but for the cluster-7 results of annual precipitation totals. This figure is available in colour online atwileyonlinelibrary.com/journal/joc

Figure 9. As in Figure 3, but for the cluster-8 results of annual precipitation totals. This figure is available in colour online atwileyonlinelibrary.com/journal/joc

regime regions of Turkey. Consequently, in addition tothe influences of the synoptic- or regional-scale weathersystems, this pattern from cluster-7 analysis of Turk-ish annual precipitation data is closely associated withthe direct influences of physical geographical featuresof Turkey (i.e. exposure, topography, orography, land-sea distribution, continentality and the higher peninsula).On the basis of the results of cluster-7 analysis, and thenames of Turkey’s seven geographical and rainfall regimeregions and their spatial patterns, cluster-7 precipitationregions are given in the following names.

1. Black Sea precipitation region (the 6th cluster ofcluster-7);

2. Northwest Anatolia – Thrace precipitation region(western Black Sea – Marmara – northern Aegean)(the 4th and 7th clusters of cluster-7);

3. Southern Aegean and western Mediterraneanprecipitation region (the 1st cluster of cluster-7);

4. Mediterranean precipitation region (the 5th cluster ofcluster-7);

5. Continental eastern and south-eastern Anatolia precip-itation region (the 2nd cluster of cluster-7).

3.7. Cluster-8

This cluster-8 precipitation pattern (Figure 9) is verysimilar with the cluster-7 pattern of Turkey’s precipitationtotals with some differences. In the cluster-8 pattern,the Continental Central Anatolia precipitation region isdivided into its western and eastern parts, and it isinteresting to see that the Black Sea region is delimitedmainly with the coastal region as in Turkes (1996,1998). Meanwhile, it is also observed that the north-western Anatolia – Thrace (shortly Northwest Turkey)

Copyright 2010 Royal Meteorological Society Int. J. Climatol. 31: 2055–2067 (2011)

DETERMINING COHERENT PRECIPITATION REGIONS IN TURKEY 2063

Figure 10. Resultant precipitation regions of Turkey re-drawn by simplifying spatial variations of the cluster-8 results of the spectral clusteringanalysis of annual precipitation totals measured at 96 stations in Turkey: BLS: Black Sea; NWTR: Northwest Turkey; SAEG-WMED: southernAegean and western Mediterranean; MED: Mediterranean; WCCAN; West Continental Central Anatolia; ECCAN: East Continental CentralAnatolia and CEAN-CSEAN: Continental Eastern and South-eastern Anatolia are the special regional names in Turkey. This figure is available

in colour online at wileyonlinelibrary.com/journal/joc

precipitation region has now much more particles. Thewestern Continental Central Anatolia precipitation regionis mostly characterized by the influences of variability ofthe westerly and south-westerly weather systems from theEurope, Balkans and the Mediterranean regions, whereasthe Eastern Continental Central Anatolia precipitationregion is mostly explained by the influences of variabilityof the northerly weather systems and further convectiveinstability events and snowfalls over the region.

Finally, in this study, following precipitation regionsof Turkey are extracted from the SC method applied tothe annual precipitation totals of 96 stations of Turkeyfor the period of 1929–2007 (Figures 9 and 10):

1. Black Sea (BLS) precipitation region (the 6th clusterof cluster-8);

2. Northwest Turkey (NWTR) precipitation region (the7th and 8th clusters of cluster-8);

3. Southern Aegean and western Mediterranean (SAEG-WMED) precipitation region (the 1st cluster ofcluster-8);

4. Mediterranean (MED) precipitation region (the 5thcluster of cluster-8);

5. West Continental Central Anatolia (WCCAN) precip-itation region (the 4th cluster of cluster-8);

6. East Continental Central Anatolia (ECCAN) precipi-tation region (the 3rd cluster of cluster-8);

7. Continental eastern and south-eastern Anatolia(CEAN-CSEAN) precipitation region (the 2nd ofcluster-8).

By evidently regarding the similarity of its spatialdistribution pattern to the precipitation regime regions ofTurkey, represented mainly by the seasonal precipitationvariability and physical geographical features of Turkey,these newly extracted regions of Turkey (Figure 10) mayalso be considered as the precipitation regime regions ofTurkey produced by the SC method.

4. Discussion

Among others, precipitation is the main driver of thevariations in the water balance in space and time,and changes in precipitation amounts and variabilityhave very important implications for hydrology andwater resources (Compagnucci et al., 2001). Hydro-climatological variability over time in a catchment ora geographical region is mainly influenced by daily,seasonal, annual and decadal time-scale variations inprecipitation amounts and intensity. Furthermore, riverflood frequency is influenced by changes in inter-annualvariability in precipitation and by changes in short-termrainfall properties (such as storm rainfall intensity). Thefrequency of low or high streamflows is also affectedmainly by the changes in seasonal distribution patternsand year-to-year variability of precipitation totals, andoccurrence of severe prolonged drought events.

The impacts of climate change on hydrology and waterresources are usually estimated by defining scenarios forchanges in climatic inputs to a hydrological model fromthe output of general circulation models (GCMs) (Com-pagnucci et al., 2001; Tatli et al., 2004). Accordingly,hydrological impact assessments, developing and usingrealistic hydrological models, and understanding betterthe relations and feedback mechanisms between the cli-mate and hydrological systems are the key developmentsfor constructing scenarios. Understanding of the contem-porary hydro-climatological systems and their variabilityin space and time is also of great importance for cli-matic model projections, their meaningful interpretationand assessments, and adaptation measures to impacts ofthe climate change on hydrological systems and waterresources.

On the other hand, many downscaling techniques havebeen developed (e.g. Wilby and Wigley, 1997; Wilbyet al., 1998, 1999; Tatli et al., 2004, 2005;, and so on)and widely used in climatological and hydrological stud-ies. These techniques range from simple interpolation of

Copyright 2010 Royal Meteorological Society Int. J. Climatol. 31: 2055–2067 (2011)

2064 M. TURKES AND H. TATLI

climate model output (e.g. Felzer and Heard, 1999; Sen,2009), through the use of empirical/statistical relation-ships between the catchment and regional climate (e.g.Crane and Hewitson, 1998; Wilby et al., 1998, 1999), tothe use of nested regional climate models (e.g. Chris-tensen and Christensen, 1998; Compagnucci et al., 2001;Onol and Semazzi, 2009).

Increased awareness of influences of the climatic vari-ability on hydrology and water resources and relatedsocioeconomic sectors and services including forestry,agriculture, energy and drinking water has increasedfor about 20 years (Kundzewicz et al., 2007). Variousstudies have investigated the linkages between recog-nizable oscillation and teleconnection patterns of large-scale climatic variability including particularly the ElNino – Southern Oscillation (ENSO), North AtlanticOscillation (NAO), Arctic Oscillation (AO), North Sea-Caspian Pattern (NCP) and variations of the hydro-logical variables in time. These studies, for exampleMcCabe (1996), Piechota et al. (1997), Vogel et al.(1997), Olsen et al. (1999) for North America, Marengo(1995), Marengo et al. (1998) and Compagnucci and Var-gas (1998) for South America, Chiew et al. (1998) andPiechota et al. (1998) for Australasia, Shorthouse andArnell (1997), Rodo et al. (1997), Cullen et al. (2002),Xoplaki et al. (2004), Mariotti et al. (2002), Tatli (2007)and Beranova and Huth (2008) for the Europe andMediterranean region, Serreze et al. (1997) and Wanget al. (2005) for the Northern Hemisphere, Cullen et al.(2002) for the Middle East, Schulze (1997) for southernAfrica and for Turkey, for example Tatli et al. (2004),Turkes and Erlat (2003, 2005, 2006) revealed types ofvariability from inter-seasonal and year-to-year variabil-ity to decadal fluctuations. It is evidently seen that theseobserved patterns of variability also vary noticeably fromregion to region. Such linkage studies are appreciablyvaluable because they help in understanding atmosphericcauses of observed hydrological and hydro-climatologicalchanges and variability not only over time but also inspace.

Water management including drought managementplans is mainly dealing with the minimization of riskand adaptation to changing circumstances arising fromboth increased and/or altered demands in the watersector and spatially coherent significant precipitationand groundwater changes over time. Various adaptationtechniques has been developed and applied in the watersector for decades (Compagnucci et al., 2001; Adgeret al., 2007). One widely used classification differentiatesbetween increasing capacity, changing operating rulesfor existing structures and systems, managing demandand changing institutional practices. The first two oftenare termed supply-side strategies, whereas the latter twoare demand-side (World Bank, 1993; Kindler, 2000)strategies. Well-organized adaptation to climate changein the water sector requires endeavouring the followingfive main areas (Compagnucci et al., 2001):

• Data for monitoring : Adaptive water managementrequires reliable data on which to make decisions,calibrate models and develop projections for the future.

• Understanding patterns of variability : An understand-ing of the patterns of variability is important formedium-term water management, because, even in theabsence of climate change, the recent past may not bea reliable guide to the hydrological resource base ofthe near future.

• Analytical tools: Effective water management requiresvarious analytical tools to assess options and the future,which should consist of the scenario and risk analyses.

• Decision tools: Scenario and risk analyses provideinformation on the characteristics of possible futuresand their consequences for policy makers and allrelated end users.

• Management techniques: These are the techniquesthat are actually implemented to meet managementobjectives.

Nevertheless, such assessments generally do notinclude the understanding of contemporary hydro-climatological (i.e. existing and natural) characteristics ofthe forests, agricultural and hydrological systems, energyproduction, water resources and drought events withrespect to the efficient and functional adaptation strategiesto climate change. As we have already discussed pre-viously, hydro-climatological aspects of the hydrology,groundwater hydrology and water resources in a regionor in a country, particularly in the drought-prone regionsfaced water-scarcity under the present climatic condi-tions, such as arid and semi-arid African countries, andthe semi-arid and dry sub-humid Mediterranean coun-tries, predominantly characterized with high seasonal andyear-to-year precipitation variability should be consid-ered. One of these highlighted hydrological and hydro-climatological aspects is of precipitation regime types ofa region or a country. These should clearly show themeaningful spatial distribution patterns of precipitationregime regions arranged by mainly physical geographi-cal, seasonal and/or estimated inter-annual variations ofprecipitation amounts for a sufficient period of observ-ing time. Therefore, as we have proposed in this paper,clustering of observed precipitation time series, for pro-ducing well-defined distribution maps of the precipitationseasonality (or precipitation regime) regions; in terms ofthe year-to-year variability of precipitation amounts isalso of great importance for adapting to the impacts ofclimate change. This should be devoted particularly tomitigate the negative impacts of global warming that isone of the significant large-scale results of the increasedhuman-induced green-house effect.

5. Conclusions

In this paper, the SC approach is used to determine thecoherent precipitation regime regions in Turkey. Clus-tering techniques have been frequently applied in various

Copyright 2010 Royal Meteorological Society Int. J. Climatol. 31: 2055–2067 (2011)

DETERMINING COHERENT PRECIPITATION REGIONS IN TURKEY 2065

disciplines. Generally, hierarchical clustering algorithm isused in climatology (e.g. Lund, 1963; El-Kadi and Smith-son, 1992), but in this study we have applied the SC tech-nique for sorting different meteorological stations intogroups in a way that the degree of association betweentheir precipitation values is maximal if they belong to thesame group (or cluster) and minimal otherwise. The SCis a hybrid method having advantages over other cluster-ing techniques that the example of the two circles givenin Section 2 indicates its ability while determining thecorrect coherent structures.

In its traditional applications, cluster analysis is simplyused to discover structures in data without explainingwhy they exist; however, in this study, we have givensome comments in association with the atmospheric andphysical geographical controls according to mostly theprevious studies (e.g. Turkes, 1996, 1998; Kutiel et al.,2001; Turkes et al., 2002; Turkes and Erlat, 2003, 2005and 2006; Tatli et al., 2004; Turkes et al., 2009). From2- to 8-clusters of the precipitation coherent regionsover Turkey were bridged with upper-air circulations andthe well-known precipitation climatological aspects beingfrequently observable, though the cluster-8 is enough tointerpret the precipitation clusters according to the VIgiven in Section 2.

On the other hand, the obtained spatial patterns of theprecipitation regions by using the SC algorithm give anew hydro-climatological insight. As a result, not onlyin Turkey but also in other countries that have simi-lar hydro-climatological conditions and socioeconomicfeatures, policy makers and/or policy-making processof the various socioeconomic sectors, and experts onwater management and other fields of expertise in agri-culture and forestry, natural resources, decision makingin all areas of private and public life related with theweather and climate and hydrology and water resourcesshould make use of this kind of precipitation regimeregions by considering the scientific frame of our discus-sion. For that reason, the main implications of our studymight give them a strong facility and a unique chancefor understanding and planning their hydrological andhydro-climatological systems and water resources includ-ing drinking waters, groundwater, rivers, hydroelectricproduction, forestry, agricultural and irrigation activitiesand others. Such hydro-climatological studies, droughtand precipitation severity indices could also be taken asignificant part in a region’s or a country’s medium andlong-term range water resources and drought manage-ment plans along with other social and economic policies(Turkes and Tatli, 2009, 2010; Turkes, 2010).

Consequently, for future studies, besides the precip-itation values, for example other atmospheric surfacefluxes and upper circulation variables, together of three-dimensional atmospheric structure, could also be used tocluster at the same time in order to extract an evidentpicture of the micro-climate regions. This kind of studiesalso increases our understanding of climate change andvariability, and adaptation capacities for projected futureclimate changes, particularly the spatial and temporal

patterns of precipitation variability varying from smallscale to large scale, and from a seasonal and year-to-yearto centennial precipitation over time.

Acknowledgements

We would like to thank the anonymous reviewers whooffered valuable suggestions and constructive comments.We are also grateful to the research assistant M. ZeynelOZTURK in our Department for his help in preparing ageneralized map for the precipitation regions of Turkey.

References

Adger WN, Agrawala S, Mirza MMQ, Conde C, O’Brien K, Pulhin J,Pulwarty R, Smit B, Takahashi K. 2007. Assessment of adaptationpractices, options, constraints and capacity. In Climate Change 2007:Impacts, Adaptation and Vulnerability, Contribution of WorkingGroup II to the Fourth Assessment Report of the IntergovernmentalPanel on Climate Change, Parry ML, Canziani OF, Palutikof JP,van der Linden PJ, Hanson CE (eds). Cambridge University Press:Cambridge; 717–743.

Aksoy H, Unal NE, Alexandrov V, Dakova S, Yoon J. 2007. Hydrom-eteorological analysis of northwestern Turkey with links to climatechange. International Journal of Climatology 28: 1047–1060.

Anderberg MR. 1973. Cluster Analysis for Applications, AcademicPress: New York.

Beranova R, Huth R. 2008. Time variations of the effects of circulationvariability modes on European temperature and precipitation inwinter. International Journal of Climatology 28: 139–158.

Bezdek JC, Pal NR. 1998. Some new indexes of cluster validity. IEEETransactions on Systems, Man, and Cybernetics, Part B 28: 301–315.

Bisgin H, Dalfes HN. 2008. Parallel clustering algorithms withapplication to climatology. Geophysical Research Abstracts 10:EGU2008–A-00000.

Bunkers MJ, Miller JR, DeGaetano AT. 1996. Definition of climateregions in the northern plains using an objective cluster modificationtechnique. Journal of Climate 9: 130–146.

Chiew FHS, Piechota TC, Dracup JA, McMahon TA. 1998. El NinoSouthern Oscillation and Australian rainfall, streamflow anddrought – links and potential for forecasting. Journal of Hydrology204: 138–149.

Christensen O, Christensen J. 1998. Climate simulations with theHIRHAM limited area regional climate model over Scandinavia.In Proceedings of the Second International Conference on Climateand Water, Espoo, Finland , August 1998, Helsinki University ofTechnology, Helsinki; 10–19.

Compagnucci R, Da Cunha L, Hanaki K, Howe C, Mailu G, Shiklo-manov I, Stakhiv E, (Lead Authors). 2001. Hydrology and waterresources. In Climate Change 2001: Impacts, Adaptation and Vul-nerability, Contribution of Working Group II to the Third Assess-ment Report of the Intergovernmental Panel on Climate Change.Becker A, Zhang J (eds). Cambridge University Press: Cambridge;191–233.

Compagnucci RH, Vargas WM. 1998. Interannual variability of theCuyo River’s streamflow in the Argentinian Andean mountains andENSO events. International Journal of Climatology 18: 1593–1609.

van Cooten S, Elmore KL, Barbe DE, McCorquodale JA, Reed DJ.2009. A statistical methodology to discover precipitation microcli-mates in Southeast Louisiana: implications for coastal watersheds.Journal of Hydrometeorology 10: 1184–1202.

Crane RG, Hewitson BC. 1998. Doubled CO2 precipitation changesfor the Susquehanna River Basin: downscaling from the GENESISgeneral circulation model. International Journal of Climatology 18:65–66.

Cullen HM, Kaplan A, Arkin PA, DeMenocal PB. 2002. Impact of theNorth Atlantic oscillation on Middle Eastern climate and streamflow.Climatic Change 55: 315–338.

Davies DL, Bouldin DW. 1979. A cluster separation measure. IEEETransactions on Pattern Analysis and Machine Intelligence 1:224–227.

Donna B, McGregor GR, Hannah DM, Sheridan SC. 2007. Develop-ment of a spatial synoptic classification scheme for Western Europe.International Journal of Climatology 27: 2017–2040.

Copyright 2010 Royal Meteorological Society Int. J. Climatol. 31: 2055–2067 (2011)

2066 M. TURKES AND H. TATLI

Dunn J. 1974. A fuzzy relative of the ISODATA process and its use indetecting compact well-separated clusters. Journal of Cybernetics 3:32–57.

El-Kadi A, Smithson PA. 1992. Atmospheric classifications andsynoptic climatology. Progress in Physical Geography 16: 432–455.

Erinc S. 1969. Climatology and Its Methods, Istanbul University,Institute of Geography Press: Istanbul (in Turkish).

Felzer B, Heard P. 1999. Precipitation differences amongst GCMs usedfor the U.S. national assessment. Journal of the American WaterResources Association 35: 1327–1339.

Fovell RG, Fovell MYC. 1993. Climate zones of the ConterminousUnited States defined using cluster analysis. Journal of Climate 6:2103–2135.

Gordon ND, Norris JR, Weaver CP, Klein SA. 2005. Cluster analysisof cloud regimes and characteristic dynamics of midlatitude synopticsystems in observations and a model. Journal of GeophysicalResearch 110: D15S17, DOI:0.1029/2004JD005027.

Kindler J. 2000. Integrated water resources management: the meanders.Water International 25: 312–319.

Kucken M, Gerstengarbe FW, Orlowsky B. 2009. A combination ofcluster analysis and kappa statistic for the evaluation of climatemodel results. Journal of Applied Meteorology and Climatology 48:1757–1765.

Kundzewicz ZW, Mata LJ, Arnell NW, Doll P, Kabat P, Jimenez B,Miller KA, Oki T, Sen Z, Shiklomanov IA. 2007. Freshwaterresources and their management. In Climate Change 2007: Impacts,Adaptation and Vulnerability, Contribution of Working Group IIto the Fourth Assessment Report of the Intergovernmental Panelon Climate Change, Parry ML, Canziani OF, Palutikof JP, vander Linden PJ, Hanson CE (eds). Cambridge University Press:Cambridge; 173–210.

Kutiel H, Hirsch-Eshkol TR, Turkes M. 2001. Sea level pressurepatterns associated with dry or wet monthly rainfall conditions inTurkey. Theoretical and Applied Climatology 69: 39–67.

Kutiel H, Turkes M. 2005. New evidence for the role of the NorthSea – Caspian Pattern on the temperature and precipitation regimesin continental central Turkey. Geografiska Annaler: Series A,Physical Geography 87: 501–513.

Lund IA. 1963. Map-pattern classification by statistical methods.Journal of Applied Meteorology 2: 56–65.

MacQueen JB. 1967. Some methods for classification and analysis ofmultivariate observations. In Proceedings of 5th Berkeley Symposiumon Mathematical Statistics and Probability, University of CaliforniaPress: California; 281–297.

Marengo JA. 1995. Variations and change in South Americanstreamflow. Climatic Change 31: 99–117.

Marengo JA, Tomasella J, Uvo CR. 1998. Trends in streamflow andrainfall in tropical South America: Amazonia, eastern Brazil andnorthwestern Peru. Journal of Geophysical Research – Atmospheres103: 1775–1783.

Mariotti A, Zeng N, Lau K-M. 2002. Euro-Mediterranean rainfall andENSO – a seasonally varying relationship. Geophysical ResearchLetters 29: 1621.

McCabe GJ. 1996. Effects of winter atmospheric circulation ontemporal and spatial variability in annual streamflow in the westernUnited States. Hydrological Sciences Journal 41: 873–888.

Ng AY, Jordan MI, Weiss Y. 2002. On spectral clustering: analysisand an algorithm. In Advances in Neural Information ProcessingSystems 14, Dietterich T, Becker S, Ghahramani Z (eds). MIT Press:Cambridge.

Olsen JR, Stedinger JR, Matalas NC, Stakhiv EZ. 1999. Climatevariability and flood frequency estimation for the Upper Mississippiand Lower Missouri Rivers. Journal of the American WaterResources Association 35: 1509–1523.

Onol B, Semazzi FHM. 2009. Regionalization of climate changesimulations over the Eastern Mediterranean. Journal of Climate 22:1944–1961.

Piechota TC, Dracup JA, Chiew FHS, McMahon TA. 1998. Seasonalstreamflow forecasting in eastern Australia and the El Nino-SouthernOscillation. Water Resources Research 34: 3035–3044.

Piechota TC, Dracup JA, Fovell RG. 1997. Western U.S. streamflowand atmospheric circulation patterns during El Nino-SouthernOscillation. Journal of Hydrology 201: 249–271.

Ray S, Turi RH. 1999. Determination of number of clusters inK-Means clustering and application in colour image segmentation.In Proceedings of the 4th International Conference on Advancesin Pattern Recognition and Digital Techniques . Calcutta, India;137–143.

Richman MJ, Lamb PJ. 1985. Climatic pattern analysis of three- andseven-day summer rainfall in the central United States: Somemethodological considerations and a regionalization. Journal ofClimate and Applied Meteorology 24: 1325–1343.

Rodo X, Baert E, Comin FA. 1997. Variations in seasonal rainfallin southern Europe during the present century: relationships withthe North Atlantic oscillation and the El Nino southern oscillation.Climate Dynamics 13: 275–284.

Saris F, Hannah DM, Eastwood WJ. 2010. Spatial variability ofprecipitation regimes over Turkey. Hydrological Sciences Journal-Journal Des Sciences Hydrologiques 55: 234–249.

Schulze RE. 1997. Impacts of global climate change in a hydrologicallyvulnerable region: challenges to South African hydrologists.Progress in Physical Geography 21: 113–136.

Serreze MC, Carse F, Barry RG, Rogers JC. 1997. Icelandic lowcyclone activity: climatological features, linkages with the NAO,and relationships with recent changes in the Northern Hemispherecirculation. Journal of Climate 10: 453–464.

Shorthouse C, Arnell NW. 1997. Spatial and temporal variabilityin European river flows and the North Atlantic Oscillation.FRIEND’97: International Association of Hydrological SciencePublications 246: 77–85.

Sokal RR, Rohlf FJ. 1962. The comparison of dendrograms byobjective methods. Taxon 11: 33–40.

Spellman G. 2000. The application of an objective weather-typingsystem to the Iberian Peninsula. Weather 55: 375–385.

Sen Z. 2009. Precipitation downscaling in climate modelling usinga spatial dependence function. International Journal of GlobalWarming 1: 29–42.

von Storch H, Zwiers FW. 1999. Statistical Analysis in ClimateResearch, Cambridge University Press: Cambridge.

Tatli H. 2006. Analysis Mediterranean precipitation associated withthe North Atlantic Oscillation Index (NAOI) via Hilbert-Huangtransformation. Presented at the Conference on Water Observationand Information System for Decision Support (BALWOIS-2006).Ohrid; Paper No. A-329, 11 pp.

Tatli H. 2007. Synchronization between the North Sea-Caspian pattern(NCP) and surface air temperatures in NCEP. International Journalof Climatology 27: 1171–1187.

Tatli H, Dalfes HN, Mentes SS. 2004. A statistical downscalingmethod for monthly total precipitation over Turkey. InternationalJournal of Climatology 24: 161–180.

Tatli H, Dalfes HN, Mentes SS. 2005. Surface air temperaturevariability over Turkey and its connection to large scale upperair circulation via multivariate techniques. International Journal ofClimatology 25: 331–350.

Tatli H, Turkes M. 2008. Drought events of 2006/2007 in Turkeyand determination of its association with large-scale atmosphericvariables by logistic regression. In Proceedings of 4th AtmosphericSciences Symposium, Technical University of Istanbul: Istanbul;516–527 (in Turkish).

Trigo R, Xoplaki E, Zorita E, Luterbacher J, Krichak S, Alpert P,Jacobeit J, Saenz J, Fernandez J, Gonzalez-Rouco F, Garcia-Herrera R, Rodo X, Brunetti M, Nanni T, Maugeri M, Turkes M,Gimeno L, Ribera P, Brunet M, Trigo I, Crepon M, Mariotti A.2006. Relations between variability in the Mediterranean regionand mid-latitude variability. In Mediterranean Climate Variability,Developments in Earth and Environmental Sciences, 4, Chapter3, Lionello P, Malanotte-Rizzoli P, Boscolo R (eds). Elsevier:Amsterdam, Nederland; 179–226.

Turkes M. 1996. Spatial and temporal analysis of annual rainfallvariations in Turkey. International Journal of Climatology 16:1057–1076.

Turkes M. 1998. Influence of geopotential heights, cyclone frequencyand Southern Oscillation on rainfall variations in Turkey.International Journal of Climatology 18: 649–680.

Turkes M. 1999. Vulnerability of Turkey to desertification with respectto precipitation and aridity conditions. Turkish Journal of theEngineering and Environmental Sciences 23: 363–380.

Turkes M. 2003. Spatial and temporal variations in precipita-tion and aridity index series of Turkey. In Mediterranean Cli-mate – Variability and Trends, Regional Climate Studies , Bolle HJ(ed). Springer Verlag: Heidelberg; 181–213.

Turkes M. 2010. Analysis of the UN Convention to CombatDesertification with respect to the climate, climate change anddrought, and applications in Turkey. In Proceeding of Symposiumon Combating Desertification . Corum; 601–616 (in Turkish).

Copyright 2010 Royal Meteorological Society Int. J. Climatol. 31: 2055–2067 (2011)

DETERMINING COHERENT PRECIPITATION REGIONS IN TURKEY 2067

Turkes M, Erlat E. 2003. Precipitation changes and variability inTurkey linked to the North Atlantic Oscillation during the period1930–2000. International Journal of Climatology 23: 1771–1796.

Turkes M, Erlat E. 2005. Climatological responses of winterprecipitation in Turkey to variability of the North Atlantic Oscillationduring the period 1930–2001. Theoretical and Applied Climatology81: 45–69.

Turkes M, Erlat E. 2006. Influences of the North Atlantic Oscillationon precipitation variability and changes in Turkey. Nuovo CimentoDella Societa Italiana Di Fisica C-Geophysics and Space Physics 29:117–135.

Turkes M, Koc T, Saris F. 2009. Spatiotemporal variability ofprecipitation total series over Turkey. International Journal ofClimatology 29: 1056–1074.

Turkes M, Sumer UM, Kılıc G. 2002. Persistence and periodicity inthe precipitation series of Turkey and associations with 500 hPageopotential heights. Climate Research 21: 59–81.

Turkes M, Tatli H. 2009. Use of the standardized precipitation index(SPI) and modified SPI for shaping the drought probabilities overTurkey. International Journal of Climatology 29: 2270–2282.

Turkes M, Tatli H. 2010. The Role of drought and precipitationseverity indices for determination, characterization and monitoringof the desertification. In Proceeding of Symposium on CombatingDesertification . Corum; 245–263 (in Turkish).

Unal Y, Kindap T, Karaca M. 2003. Redefining the climate zonesof Turkey by using cluster analysis. International Journal ofClimatology 23: 1045–1055.

Vogel RM, Bell CJ, Fennessey NM. 1997. Climate, streamflow andwater supply in the northeastern United States. Journal of Hydrology198: 42–68.

Volkovich Z, Kirzhne V, Barzily Z, Hosid S, Korenbla K. 2010. Alinguistic approach to classification of bacterial genomes. PatternRecognition 43: 1083–1093.

Volkovich Z, Kirzhne V, Barzily Z, Morozensky L. 2008. A sta-tistical model of cluster stability. Pattern Recognition 41:2174–2188.

Von Luxburg U. 2007. A tutorial on spectral clustering. Statistics andComputing 17: 395–416.

Wang D, Wang C, Yang X, Lu J. 2005. Winter Northern Hemispheresurface air temperature variability associated with the ArcticOscillation and North Atlantic Oscillation. Geophysical ResearchLetters 32: L16706, DOI:10.0292005GL022952.

Wilby RL, Hay LE, Leavesley GH. 1999. A comparison of downscaledand raw GCM output: implications for climate change scenariosin the San Juan River Basin, Colorado. Journal of Hydrology 225:67–91.

Wilby RL, Wigley TML, Conway D, Jones PD, Hewitson BC, Main J,Wilks DS. 1998. Statistical downscaling of general circulation modeloutput: a comparison of methods. Water Resources Research 34:2995–3008.

Wilby RL, Wigley TML. 1997. Downscaling general circulation modeloutput: a review of methods and limitations. Progress in PhysicalGeography 21: 530–548.

Wilks DS. 1995. Statistical Methods in the Atmospheric Sciences: AnIntroduction, Academic Press: New York.

World Bank. 1993. Water Resources Management, World Bank PolicyPaper, World Bank: Washington, DC.

Xoplaki E, Gonz’alez-Rouco JF, Luterbacher J, Wanner H. 2004. Wetseason Mediterranean precipitation variability: influence of largescale dynamics. Climate Dynamics 23: 63–78.

Copyright 2010 Royal Meteorological Society Int. J. Climatol. 31: 2055–2067 (2011)