Adv. Geosci., 12, 127–135, 2007www.adv-geosci.net/12/127/2007/© Author(s) 2007. This work is licensedunder a Creative Commons License.

Advances inGeosciences

The atmospheric conditions over Europe and the Mediterranean,favoring snow events in Athens, Greece

E. E. Houssos, C. J. Lolis, and A. Bartzokas

Laboratory of Meteorology, Department of Physics, University of Ioannina, 45110 Ioannina, Greece

Received: 1 March 2007 – Revised: 21 May 2007 – Accepted: 6 August 2007 – Published: 21 September 2007

Abstract. The 3-dimensional structure and the evolution ofatmospheric circulation favoring snowfall in Athens are ex-amined. The study refers to 61 snow events, which occurredduring the period 1958–2001. For each one of the events, thepatterns of MSL pressure, 850 hPa and 500 hPa air tempera-tures, 500 hPa geopotential height and 1000–500 hPa thick-ness are constructed for the European region, for the daybefore (D-1), the first day (D) and the day after the end ofthe event (END). A statistical methodology involving FactorAnalysis and Cluster Analysis is applied to the above datasets and the 61 cases are finally classified into five clusters.These clusters are generally characterized by a north-easterlyflow in the lower troposphere over the Athens area. Thisflow is associated with the presence of a low pressure systemaround Cyprus and an anticyclone over Europe. The posi-tion, the intensity and the trajectories of the surface and theupper air systems during D-1, D and END days are generallydifferent among the five clusters.

1 Introduction

Scientific research in meteorological forecasting during thelast decades has mainly focused on the construction andthe improvement of numerical models for weather predic-tion. These models aim at the prediction of the oncomingweather and the possible occurrence of extreme meteorolog-ical events that may cause inconvenience in human activi-ties (e.g. Keller, 2002; Hervella et al., 2003). On the otherhand, some researchers have examined the connection be-tween extreme meteorological events and atmospheric circu-lation by using statistical methodologies (e.g. Metaxas et al.,1993; Romero et al., 1998; Jansa et al., 2001; Kutiel et al.,2001; Kostopoulou and Jones, 2005). In many regions of the

Correspondence to:A. Bartzokas(abartzok@uoi.gr)

Mediterranean basin, snowfall can be considered as an ex-treme event and it is responsible for significant problems intransportation, cutback on industrial and trade activity, dam-ages to the power supply network and sometimes fatal acci-dents. The synoptic conditions associated with snowfall invarious Mediterranean regions, including Athens, have beenstudied in the past, by using various approaches (e.g. Prez-erakos and Angouridakis, 1984; Tayanc et al., 1998; Este-ban et al., 2005). In the present study, the issue of snow-fall in Athens is approached by introducing a new objectivemethodology scheme, in order to identify not only the maincirculation patterns associated with the event, but also the de-tails referring to the position, the intensity and the evolutionof the associated circulation systems from the day before thebeginning till the day after the end of the event.

2 Data and methodology

The data used consists of: i) 3-hourly meteorological obser-vations recorded at the meteorological station of Hellenikonairport (Athens area), provided by the Hellenic National Me-teorological Service and ii) 12:00 UTC 2.5×2.5 grid pointdata of MSL pressure, 850 hPa and 500 hPa air tempera-tures, 500 hPa geopotential height and 1000–500 hPa thick-ness over Europe (10◦ W–40◦ E and 30◦–60◦ N) for the pe-riod 1958–2001, obtained from the ECMWF ERA40 Reanal-ysis Project. When at least one, of the 3-hourly observationsin a day reports snowfall, this day is considered as a snowday. The sequence of successive snow days is then consid-ered as a snow event. The first day of a snow event will bementioned as D day, the day before D day as D–1 day andthe day that follows the last day as END day. By analyzingthe dataset, 61 snow events are extracted. Three matrices areconstructed (one for each of D-1, D and END days), con-taining the grid point values of MSL pressure, 850 hPa and500 hPa air temperatures, 500 hPa geopotential height and

Published by Copernicus Publications on behalf of the European Geosciences Union.

128 E. E. Houssos et al.: Atmospheric conditions over Europe and the Mediterranean

Fig. 1. Distribution of snow events according to their duration. 3

33

22

51

048

12162024283236

1 2 3 4Duration of events (days)

Num

ber o

f day

s

Figure 1 Distribution of snow events according to their duration.

048

12162024283236

Jul

Sep Nov Jan

Mar

May

Num

ber o

f day

s

4 days3 days2 days1 day

Figure 2 Number of snow days and duration of snow events per month.

0

2

4

6

8

10

12

14

1957

-195

8

1960

-196

1

1963

-196

4

1966

-196

7

1969

-197

0

1972

-197

3

1975

-197

6

1978

-197

9

1981

-198

2

1984

-198

5

1987

-198

8

1990

-199

1

1993

-199

4

1996

-199

7

1999

-200

0

Num

ber o

f day

s

7th event6th event5th event4th event3nd event2nd event1st event

Figure 3 Number and duration of snow events per winter (November – March).

3.2. Classification of snow events

The 61 snow events are classified into 5 clusters. The monthly distribution of the snow events classified in the 5 clusters (Fig. 4a) shows, that clusters 1, 2 and 5 can be characterized as late winter clusters, as they mainly appear from January to March. On the other hand, clusters 3 and 4 appear mainly during the conventional winter (December – February). According to the duration of snow events of each cluster (Fig. 4b), it is shown that the events classified in clusters 3, 4 and 5 appear to last longer than these of clusters 1 and 2. By examining the inter-annual variation of the events of each cluster (Fig. 4c), it is shown that the events of cluster 1 appear frequently during the whole under study period, but especially during the period 1971-1975. The events of cluster 3 appear relatively frequently before 1981, but they do not appear at all since 1981, except from the winter of 1991-1992, which is the snowiest winter of the whole period. Also, the events of cluster 5 appear frequently before 1977, but since 1977, they appear only during the winters of 1987-1988 and 1991-1992. Finally, the events of cluster 4 seem to be evenly distributed in the under study period.

Fig. 2. Number of snow days and duration of snow events permonth.

1000–500 hPa thickness, for the 61 snow events. Thus, eachone of the three matrices consists of 61 rows correspondingto the snow events and 1365 columns corresponding to the273 grid point values of the 5 parameters.

At first, Factor Analysis(FA) (Jolliffe, 1986; Manly, 1986)is applied to each one of the three matrices in order to reducetheir dimensionality. 11 factors are retained for each case, ac-counting for at least 85% of the total variance. By unifyingthe resulted 61×11 matrices, a new 61×33 matrix is con-structed. Each row represents the evolution (D-1, D, ENDdays) of the 3-dimensional atmospheric structure associatedwith the corresponding snow event. Then,K-Means Clus-ter Analysis(CA) (Manly, 1986; Sharma, 1996) is applied tothe above 61×33 matrix, classifying the 61 cases into 5 dis-tinct clusters. The number of clusters is selected by testingan hierarchical cluster analysis using Ward’s method in orderto obtain a general view of the clustering step by step in thecorresponding dendrogram. Also, a technique introduced bySugar and James (2003), based on distortion, a quantity thatmeasures the average distance between each observation andits closest cluster center, is applied confirming our selection.We note, that our purpose is to identify not only the generalcharacteristics of the synoptic conditions favoring snowfallin Athens, but also the details referring to the position, theintensity and the trajectories of the associated circulation sys-

3

33

22

51

048

12162024283236

1 2 3 4Duration of events (days)

Num

ber o

f day

s

Figure 1 Distribution of snow events according to their duration.

048

12162024283236

Jul

Sep Nov Jan

Mar

May

Num

ber o

f day

s

4 days3 days2 days1 day

Figure 2 Number of snow days and duration of snow events per month.

0

2

4

6

8

10

12

14

1957

-195

8

1960

-196

1

1963

-196

4

1966

-196

7

1969

-197

0

1972

-197

3

1975

-197

6

1978

-197

9

1981

-198

2

1984

-198

5

1987

-198

8

1990

-199

1

1993

-199

4

1996

-199

7

1999

-200

0

Num

ber o

f day

s

7th event6th event5th event4th event3nd event2nd event1st event

Figure 3 Number and duration of snow events per winter (November – March).

3.2. Classification of snow events The 61 snow events are classified into 5 clusters. The monthly distribution of the snow events classified in the 5 clusters (Fig. 4a) shows, that clusters 1, 2 and 5 can be characterized as late winter clusters, as they mainly appear from January to March. On the other hand, clusters 3 and 4 appear mainly during the conventional winter (December – February). According to the duration of snow events of each cluster (Fig. 4b), it is shown that the events classified in clusters 3, 4 and 5 appear to last longer than these of clusters 1 and 2. By examining the inter-annual variation of the events of each cluster (Fig. 4c), it is shown that the events of cluster 1 appear frequently during the whole under study period, but especially during the period 1971-1975. The events of cluster 3 appear relatively frequently before 1981, but they do not appear at all since 1981, except from the winter of 1991-1992, which is the snowiest winter of the whole period. Also, the events of cluster 5 appear frequently before 1977, but since 1977, they appear only during the winters of 1987-1988 and 1991-1992. Finally, the events of cluster 4 seem to be evenly distributed in the under study period.

Fig. 3. Number and duration of snow events per winter (November–March).

tems. For each of the 5 clusters revealed, the mean patternsof the meteorological parameters are constructed for D-1, Dand END days, describing the specific evolution of the atmo-spheric structure, associated with a snow event in Athens.

3 Results

3.1 General characteristics of snow events

More than 50% of snow events that occurred in Athens dur-ing the period 1958–2001 lasted for 1 day, while only oncedid the event last for 4 days (Fig. 1). Only one snow eventoccurred in November while the others occurred during themonths December, January, February and March (Fig. 2).Before 1982, it snowed almost every year, but the yearly fre-quency and duration of the events were low and short respec-tively. Since 1982, it does not generally snow every year,but the yearly frequency and duration are higher and longerrespectively, especially during the winters of 1982–1983,1986–1987 and 1991–1992 (Fig. 3). It is noted, that the var-ious climate change models predict a general increase in thefrequency and the intensity of the extreme weather events.However, a possible association between climate change andthe snowfall change in Athens during the recent decades hasto be further investigated.

3.2 Classification of snow events

The 61 snow events are classified into 5 clusters. Themonthly distribution of these snow events classified in the5 clusters (Fig. 4a) shows, that clusters 1, 2 and 5 can becharacterized as late winter clusters, as they mainly appearfrom January to March. On the other hand, clusters 3 and4 appear mainly during the conventional winter (December–February). According to the duration of snow events of eachcluster (Fig. 4b), it is shown that the events classified in clus-ters 3, 4 and 5 appear to last longer than these of clusters 1and 2. By examining the inter-annual variation of the eventsof each cluster (Fig. 4c), it is shown that the events of clus-ter 1 appear frequently during the whole under study period,but especially during the period 1971–1975. The events of

Adv. Geosci., 12, 127–135, 2007 www.adv-geosci.net/12/127/2007/

E. E. Houssos et al.: Atmospheric conditions over Europe and the Mediterranean 129 4

0

4

8

12

16

20

24

Nov Dec Jan Feb Mar

Num

ber o

f day

s Cluster 5Cluster 4Cluster 3Cluster 2Cluster 1

(a)

0

4

8

12

16

20

Clu

ster

1

Clu

ster

2

Clu

ster

3

Clu

ster

4

Clu

ster

5

Num

ber o

f day

s

4 days3 days2 days1 day

(b)

0

1

2

3

4

5

6

7

8

1957

-195

8

1959

-196

0

1961

-196

2

1963

-196

4

1965

-196

6

1967

-196

8

1969

-197

0

1971

-197

2

1973

-197

4

1975

-197

6

1977

-197

8

1979

-198

0

1981

-198

2

1983

-198

4

1985

-198

6

1987

-198

8

1989

-199

0

1991

-199

2

1993

-199

4

1995

-199

6

1997

-199

8

1999

-200

0

Num

ber o

f day

s Cluster 5Cluster 4Cluster 3Cluster 2Cluster 1

(c)

Figure 4 (a) Monthly distribution, (b) duration and (c) inter-annual variation of snow events for the 5 clusters.

In order to examine the time of the initiation of the events, the number of snow events versus the first hour of their report is plotted (Fig. 5). It is shown, that the initiation of the events generally appears to occur in the morning. For a statistical confirmation of this finding, Chi-square test is applied in order to examine whether the diurnal distributions of the initiation hour differ significantly from the uniform one. It is found, that this is valid only for cluster 1 and the total number of the events (95% confidence level). This morning maximum may be connected to the fact that, in the morning, air temperature is lowest and relative humidity is highest, favoring increased condensation. It has to be mentioned that the maximum in diurnal variation of winter precipitation and cloudiness in Athens occurs in the morning also (Karapiperis, 1963; Metaxas, 1970; Catsoulis et al., 1976).

0

24

68

1012

1416

18

0 3 6 9 12 15 18 21 0UTC

Num

ber o

f eve

nts cluster 1

cluster 2

cluster 3

cluster 4

cluster 5

Total

Figure 5 The number of snow events versus the first hour of their report for each cluster.

Fig. 4. (a)Monthly distribution,(b) duration and(c) inter-annual variation of snow events for the 5 clusters.

cluster 3 appear relatively frequently before 1981, but theydo not appear at all since 1981, except from the winter of1991–1992, which is the snowiest winter of the whole pe-riod. Also, the events of cluster 5 appear frequently before1977, but since 1977, they appear only during the winters of1987–1988 and 1991–1992. Finally, the events of cluster 4seem to be evenly distributed in the under study period.

In order to examine the time of the initiation of the events,the number of snow events versus the first hour of their re-port is plotted (Fig. 5). It is shown, that the initiation of theevents generally appears to occur in the morning. For a sta-tistical confirmation of this finding, Chi-square test is appliedin order to examine whether the diurnal distributions of theinitiation hour differ significantly from the uniform one. Itis found, that this is valid only for cluster 1 and the totalnumber of the events (95% confidence level). This morningmaximum may be connected to the fact that, in the morn-ing, air temperature is lowest and relative humidity is high-est, favoring increased condensation. It has to be mentionedthat the maximum in diurnal variation of winter precipitationand cloudiness in Athens occurs in the morning also (Kara-piperis, 1963; Metaxas, 1970; Catsoulis et al., 1976).

For each one of the 5 clusters, the mean patterns of MSLpressure (hPa), 850 hPa air temperature (◦C), 500 hPa airtemperature (◦C), 1000–500 hPa thickness (m) and 500 hPageopotential height (m) for D-1, D and END days are pre-sented in Figs. 6, 7, 8, 9, 10.

4

0

4

8

12

16

20

24

Nov Dec Jan Feb Mar

Num

ber o

f day

s Cluster 5Cluster 4Cluster 3Cluster 2Cluster 1

(a)

0

4

8

12

16

20

Clu

ster

1

Clu

ster

2

Clu

ster

3

Clu

ster

4

Clu

ster

5

Num

ber o

f day

s

4 days3 days2 days1 day

(b)

0

1

2

3

4

5

6

7

8

1957

-195

8

1959

-196

0

1961

-196

2

1963

-196

4

1965

-196

6

1967

-196

8

1969

-197

0

1971

-197

2

1973

-197

4

1975

-197

6

1977

-197

8

1979

-198

0

1981

-198

2

1983

-198

4

1985

-198

6

1987

-198

8

1989

-199

0

1991

-199

2

1993

-199

4

1995

-199

6

1997

-199

8

1999

-200

0

Num

ber o

f day

s Cluster 5Cluster 4Cluster 3Cluster 2Cluster 1

(c)

Figure 4 (a) Monthly distribution, (b) duration and (c) inter-annual variation of snow events for the 5 clusters.

In order to examine the time of the initiation of the events, the number of snow events versus the first hour of their report is plotted (Fig. 5). It is shown, that the initiation of the events generally appears to occur in the morning. For a statistical confirmation of this finding, Chi-square test is applied in order to examine whether the diurnal distributions of the initiation hour differ significantly from the uniform one. It is found, that this is valid only for cluster 1 and the total number of the events (95% confidence level). This morning maximum may be connected to the fact that, in the morning, air temperature is lowest and relative humidity is highest, favoring increased condensation. It has to be mentioned that the maximum in diurnal variation of winter precipitation and cloudiness in Athens occurs in the morning also (Karapiperis, 1963; Metaxas, 1970; Catsoulis et al., 1976).

02468

1012141618

0 3 6 9 12 15 18 21 0UTC

Num

ber o

f eve

nts cluster 1

cluster 2

cluster 3

cluster 4

cluster 5

Total

Figure 5 The number of snow events versus the first hour of their report for each cluster.

Fig. 5. The number of snow events versus the first hour of theirreport for each cluster.

Cluster 1 (Fig. 6): During D-1 day, an upper air troughis shown over the eastern Balkans and the Black Sea anda ridge is shown over the western Europe. At the surface,the low pressure center is located over Cyprus and the an-ticyclone is located over central Europe (depressions tilt to-wards low temperatures and anticyclones tilt towards hightemperatures). The combination between the two systemsinduces a cold advection over the eastern Balkans, in themiddle and the lower troposphere. On D day, the upperair trough moves over the eastern Aegean, the anticyclonemoves over the northwestern Balkans and the north-easterlyflow is enhanced, leading to a strong cold invasion over

www.adv-geosci.net/12/127/2007/ Adv. Geosci., 12, 127–135, 2007

130 E. E. Houssos et al.: Atmospheric conditions over Europe and the Mediterranean

5

Cluster 1 D-1 day D day End day

500h

Pa H

eigh

t

1000

-500

Th.

850h

Pa T

emp.

MSL

pre

ssur

e

500h

Pa T

emp.

Figure 6 Mean patterns of MSL pressure (hPa), 850hPa air temperature (oC), 500hPa air temperature (oC), 1000 – 500hPa thickness (m) and 500hPa geopotential height (m) for D-1, D and END days, for cluster 1 (18 snowfall events).

For each one of the 5 clusters, the mean patterns of MSL pressure (hPa), 850hPa air temperature (oC), 500hPa air temperature (oC), 1000–500hPa thickness (m) and 500hPa geopotential height (m) for D-1, D and END days are presented in Fig. 6, 7, 8, 9, 10. Cluster 1 (Fig. 6): During D-1 day, an upper air trough is shown over the eastern Balkans and the Black Sea and a ridge is shown over the western Europe. At the surface, the low pressure center is located over Cyprus and the anticyclone is located over central Europe (depressions tilt towards low temperatures and anticyclones tilt towards high temperatures). The combination between the two systems induces a cold advection over the eastern Balkans, in the middle and the lower troposphere. On D day, the upper air trough moves over the eastern Aegean, the anticyclone moves over the northwestern Balkans and the north-easterly flow is enhanced, leading to a strong cold invasion over Athens, associated with a temperature decrease of about 4C at 850hPa level. On END day, the upper air trough and the surface low move eastwards and the surface flow is weakened.

Fig. 6. Mean patterns of MSL pressure (hPa), 850 hPa air temperature (◦C), 500 hPa air temperature (◦C), 1000–500 hPa thickness (m) and500 hPa geopotential height (m) for D-1, D and END days, for cluster 1 (18 snowfall events).

Athens, associated with a temperature decrease of about 4◦Cat 850 hPa level. On END day, the upper air trough and thesurface low move eastwards and the surface flow is weak-ened.

Cluster 2 (Fig. 7): This is the most infrequent cluster,as it comprises only 3 late winter days. During D-1 day,an upper air trough affects Italy, the Adriatic Sea and thenorthern Balkans and it is associated with a strong north-easterly flow in the lower troposphere over these areas. OnD day, the whole system and the associated cold upper airmass move over the eastern Balkans and the Aegean, wherethe very strong northerly flow is responsible for the advec-tion of very cold air masses in the lower troposphere. Thisflow is attributed to the combination between an anticycloneover France and a very deep depression over Cyprus. The D

day of this cluster is characterized by the lowest air tempera-tures over Athens among the five clusters (500 hPa:−33◦C,850 hPa:−9◦C, 1000–500 hPa thickness: 5080 m), as wellas by the deepest Cyprus depression (below 1000 hPa). OnEND day, the trough and the associated cold mass move overAsia Minor and the Black Sea, while anticyclonic conditionsprevail over Greece.

Cluster 3 (Fig. 8): During D-1 day, an upper air trough anda cold air mass are shown over the Balkans, while a combi-nation between an anticyclone over central Europe and a de-pression between Crete and Cyprus in the lower troposphereis responsible for a north-easterly flow over the same area.During D day, an upper low is formed over the Aegean andwest Asia Minor, while the surface anticyclone and the de-pression move over the northern Balkans and the Cyprus area

Adv. Geosci., 12, 127–135, 2007 www.adv-geosci.net/12/127/2007/

E. E. Houssos et al.: Atmospheric conditions over Europe and the Mediterranean 131 6

Cluster 2

D-1 day D day End day

500h

Pa H

eigh

t

1000

-500

Th.

850h

Pa T

emp.

MSL

pre

ssur

e

500h

Pa T

emp.

Figure 7 As in Fig. 6, but for cluster 2 (3 snow events).

Cluster 2 (Fig. 7): This is the most infrequent cluster, as it comprises only 3 late winter days. During D-1 day, an upper air trough affects Italy, the Adriatic Sea and the northern Balkans and it is associated with a strong north-easterly flow in the lower troposphere over these areas. On D day, the whole system and the associated cold upper air mass move over the eastern Balkans and the Aegean, where the very strong northerly flow is responsible for the advection of very cold air masses in the lower troposphere. This flow is attributed to the combination between an anticyclone over France and a very deep depression over Cyprus. The D day of this cluster is characterized by the lowest air temperatures over Athens among the five clusters (500hPa: -33C, 850hPa: -9C, 1000-500hPa thickness: 5080m), as well as by the deepest Cyprus depression (below 1000hPa). On END day, the trough and the associated cold mass move over Asia Minor and the Black Sea, while anticyclonic conditions prevail over Greece.

Fig. 7. As in Fig. 6, but for cluster 2 (3 snow events).

respectively causing a very strong north-easterly flow overGreece. On END day, the cold air mass becomes warmer andmoves eastwards along with the associated upper air troughand the surface low.

Cluster 4 (Fig. 9): On D-1 day, an upper air trough ap-pears over the northern Balkans, an anticyclone is centeredover western Europe and a low pressure system is centeredover the area between Crete and Cyprus. During D day,the upper air trough moves over the Aegean, the anticyclonemoves eastwards over central Europe and the north-easterlyflow over Athens is intensified, transferring cold air masses.This cluster presents the highest temperature in the lower tro-posphere over Athens, during D day, among all the clusters(850 hPa:−6◦C, 1000–500 hPa thickness: 5170 m). DuringEND day, the surface and upper air systems and the north-easterly flow are weakened.

Cluster 5 (Fig. 10): During D-1 and D days, a 500 hPatrough and a cold air mass exist over the Balkans, whilea north-easterly flow, associated with an anticyclone overcentral Europe and a depression between Crete and Cyprus,prevails over the Aegean and the Athens region. On ENDday, an upper low is formed over the north-eastern Balkans,while the high pressure gradient area in the lower tropospheremoves eastwards affecting north-west Asia Minor. This clus-ter prevails mainly in late winter.

Despite the fact that the D day synoptic conditions of thefive clusters seem to be quite similar to each other, it has to bementioned that each cluster is not characterized by the pat-terns of D day only, but also by these of D-1 and END days.This means, that each cluster is described by the whole set of15 patterns corresponding to the five meteorological param-eters of the three days. Thus, the individual characteristics

www.adv-geosci.net/12/127/2007/ Adv. Geosci., 12, 127–135, 2007

132 E. E. Houssos et al.: Atmospheric conditions over Europe and the Mediterranean 7

Cluster 3

D-1 day D day End day

500h

Pa H

eigh

t

1000

-500

Th.

850h

Pa T

emp.

MSL

pre

ssur

e

500h

Pa T

emp.

Figure 8 As in Fig. 6 but for cluster 3 (10 snow events).

Cluster 3 (Fig. 8): During D-1 day, an upper air trough and a cold air mass are shown over the Balkans, while a combination between an anticyclone over central Europe and a depression between Crete and Cyprus in the lower troposphere is responsible for a north-easterly flow over the same area. During D day, an upper low is formed over the Aegean and west Asia Minor, while the surface anticyclone and the depression move over the northern Balkans and the Cyprus area respectively causing a very strong north-easterly flow over Greece. On END day, the cold air mass becomes warmer and moves eastwards along with the associated upper air trough and the surface low.

Fig. 8. As in Fig. 6 but for cluster 3 (10 snow events).

Table 1. Average D day values for specific meteorological characteristics of the 5 clusters.

Cluster 500 hPa Temperature 850 hPa Temperature 1000–500 hPa Thickness Anticyclone center Depression centerover Athens (C) over Athens (C) over Athens (m) SLP (hPa) SLP (hPa)

1 −29 −8 5160 >1028 <10122 −33 −9 5080 >1028 <10003 −29 −7 5160 >1028 <10104 −29 −6 5170 >1030 <10105 −31 −7 5160 >1028 <1006

of each cluster may refer to one or two of the three days, oreven to some specific patterns of one of the three days only.For example, a specific characteristic of cluster 2 is the verystrong MSL pressure gradient over eastern Europe and the

Balkans during D-1 and D days, while a specific character-istic of cluster 5 is the formation of an upper low over theeastern Balkans between D and END days. In Table 1, somequantitative features of D day are presented for each cluster.

Adv. Geosci., 12, 127–135, 2007 www.adv-geosci.net/12/127/2007/

E. E. Houssos et al.: Atmospheric conditions over Europe and the Mediterranean 133 8

Cluster 4

D-1 day D day End day

500h

Pa H

eigh

t

1000

-500

Th.

850h

Pa T

emp.

MSL

pre

ssur

e

500h

Pa T

emp.

Figure 9 As in Fig. 6, but for cluster 4 (18 snow events).

Cluster 4 (Fig. 9): On D-1 day, an upper air trough appears over the northern Balkans, an anticyclone is centered over western Europe and a low pressure system is centered over the area between Crete and Cyprus. During D day, the upper air trough moves over the Aegean, the anticyclone moves eastwards over central Europe and the north-easterly flow over Athens is intensified, transferring cold air masses. This cluster presents the highest temperature in the lower troposphere over Athens, during D day, among all the clusters (850hPa: -6C, 1000-500hPa thickness: 5170m). During END day, the surface and upper air systems and the north-easterly flow are weakened.

Fig. 9. As in Fig. 6, but for cluster 4 (18 snow events).

4 Conclusions

Snowfall in Athens can be generally considered as a rareevent and snowing for more 2 consecutive days is excep-tional. The winters of 1982–1983, 1986–1987, and 1991–1992 are the snowiest winters of the period 1958–2001.

The 61 snow events can be classified into 5 clusters.Each distinct cluster describes a specific evolution of the 3-dimensional atmospheric structure over Europe, associatedwith snowfall in Athens. The main common characteristicamong the 5 clusters is a north-easterly surface flow overAthens during D day, associated with the presence of a lowpressure system around Cyprus and an anticyclone over Eu-rope. The corresponding air masses arriving in Athens arecold and humid, as they are originated from the continen-tal areas of Eastern Europe and they have passed over the

Aegean Sea. The differences among the 5 clusters refer tothe specific positions, the intensity and the trajectories of thesynoptic systems and the associated cold air masses in themiddle and the lower troposphere.

Acknowledgements.The present study has been funded by thePENED Programme 2003 for the support of research potential(Ministry of Development/GSRT – 3rd EU Framework Pro-gramme/Operational Programme of Competitiveness).

Edited by: P. Alpert, H. Saaroni, and E. HeifetzReviewed by: two anonymous referees

www.adv-geosci.net/12/127/2007/ Adv. Geosci., 12, 127–135, 2007

134 E. E. Houssos et al.: Atmospheric conditions over Europe and the Mediterranean 9

Cluster 5

D-1 day D day End day

500h

Pa H

eigh

t

500h

Pa T

emp.

1000

-500

Th.

850h

Pa T

emp.

MSL

pre

ssur

e

Figure 10 As in Fig. 6, but for cluster 5 (12 snow events).

Cluster 5 (Fig. 10): During D-1 and D days, a 500hPa trough and a cold air mass exist over the Balkans, while a north-easterly flow, associated with an anticyclone over central Europe and a depression between Crete and Cyprus, prevails over the Aegean and the Athens region. On END day, an upper low is formed over the north-eastern Balkans, while the high pressure gradient area in the lower troposphere moves eastwards affecting north-west Asia Minor. This cluster prevails mainly in late winter.

Despite the fact that the D day synoptic conditions of the five clusters seem to be quite similar to each other, it has to be mentioned that each cluster is not characterized by the patterns of D day only, but also by these of D-1 and END days. This means, that each cluster is described by the whole set of 15 patterns corresponding to the five meteorological parameters of the three days. Thus, the individual characteristics of each cluster may refer to one or two of the three days, or even to some specific patterns of one of the three days only. For example, a specific characteristic of cluster 2 is the very strong MSL pressure gradient over eastern Europe and the Balkans during D-1 and D days, while

Fig. 10. As in Fig. 6, but for cluster 5 (12 snow events).

References

Catsoulis, V., Tselepidakis, H., and Theocharatos, G.: Character-istics of precipitation in Athens, Bull. Hellenic Meteorol. Soc.,1(2), 1–19, 1976 (in Greek).

Esteban, P., Jones, P. D., Martin-Vide, J., and Mases, M.: Atmo-spheric Circulation Patterns Related to Heavy Snowfall Days inAndorra, Pyrenees, Int. J. Climatol., 25, 319–329, 2005.

Hervella, B., Perez-Munuzuri, V., Balseiro, C. F., and Montero, P.:Application of the Non-Hydrostatic ARPS Model to the 21st–23rd December 2000 Extreme Weather Event. Proceedings of the4th EGS Plinius Conference held at Mallorca, Spain, October2002.

Jansa, A., Genoves, A., Picornell, M. A., Campins, J., Riosalido, R.,and Carretero, O.: Western Mediterranean cyclones and heavyrain. Part 2: Statistical approach, Meteorol. Appl., 8, 43–56,2001.

Jolliffe, I. T.: Principal Component Analysis, Springer, New York,1986.

Karapiperis, Ph.: Diurnal variation and probabilities of rainfall atAthens. Hypomnemata of National Observatory of Athens, Ser.II, Meteorology, No. 12, 1963.

Keller, J. L.: Weather risk assessment at the dawn of ensemble nu-merical weather prediction, Proceedings of the American Mete-orological Society: 3rd Symposium on Environmental Applica-tions, Orlando, J9–J15, 2002.

Kostopoulou, E. and Jones, P. D.: Assessment of climate extremesin the Eastern Mediterranean, Meteorol. Atmos. Phys., 89, 69–85, 2005.

Kutiel, H., Hirsch-Eshkol, T. R., and Turkes, M.: Sea level pressurepatterns associated with dry or wet monthly rainfall conditions inTurkey, Theor. Appl. Climatol., 69, 39–67, 2001.

Manly, B. F. J.: Multivariate statistical methods: A primer, Chap-man & Hall, London, 159 pp., 1986.

Adv. Geosci., 12, 127–135, 2007 www.adv-geosci.net/12/127/2007/

E. E. Houssos et al.: Atmospheric conditions over Europe and the Mediterranean 135

Metaxas, D. A.: Diurnal variation of cloudiness in Attica, Bull. Hel-lenic Military Geographical Service, 98, 8–15, 1970 (in Greek).

Metaxas, D. A., Bartzokas, A., Repapis, C. C., and Dalezios, N.R.: Atmospheric circulation anomalies in dry and wet winters inGreece, Meteorol. Z., 2, 127–131, 1993.

Prezerakos, N. G. and Angouridakis, V. E.: Synoptic Considerationof Snowfall in Athens, J. Climatol., 4, 269–285, 1984.

Romero, R., Guijarro, J. A., Ramis, C., and Alonso, S.: A 30-year(1964–1993) daily rainfall data base for the Spanish Mediter-ranean regions: first exploratory study, Int. J. Climatol., 18, 541–560, 1998.

Sharma, S.: Applied Multivariate Techniques, John Wiley, NewYork, 493 pp., 1996.

Sugar, C. A. and James, G. M.: Finding the Number of Clusters ina Dataset: An Information – Theoretic Approach, J. Am. Stat.Assoc., 98(463), 750–763, 2003.

Tayanc, M., Karaca, M., and Dalfes, H. N.: March 1987 cyclone(blizzard) over the eastern Mediterranean and Balkan region as-sociated with blocking, Mon. Wea. Rev., 126(11), 3036–3047,1998.

www.adv-geosci.net/12/127/2007/ Adv. Geosci., 12, 127–135, 2007