Research Article Trends of Dust Transport Episodes in ...downloads.hindawi.com/archive/2013/280248.pdf · dust events to synoptic types. Having performed this, a second objective

Hindawi Publishing CorporationJournal of ClimatologyVolume 2013 Article ID 280248 8 pageshttpdxdoiorg1011552013280248

Research ArticleTrends of Dust Transport Episodes in CyprusUsing a Classification of Synoptic Types Established withArtificial Neural Networks

S Michaelides F Tymvios S Athanasatos and M Papadakis

Cyprus Meteorological Service 28 Nikis Avenue 1086 Nicosia Cyprus

Correspondence should be addressed to S Michaelides silasucyaccy

Received 15 March 2013 Revised 22 May 2013 Accepted 15 July 2013

Academic Editors E Paoletti and A Rutgersson

Copyright copy 2013 S Michaelides et al This is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

The relationship between dust episodes over Cyprus and specific synoptic patterns has long been considered but also furthersupported in recent studies by the authors Having defined a dust episode as a day when the average PM10 measurement exceedsthe threshold of 50mg(m3 day) the authors have utilized Artificial Neural Networks and synoptic charts together with satelliteand ground measurements in order to establish a scheme which links specific synoptic patterns with the appearance of dusttransport over Cyprus In an effort to understand better these complicated synoptic-scale phenomena and their associations withdust transport episodes the authors attempt in the present paper a followup of the previous tasks with the objective to furtherinvestigate dust episodes from the point of view of their time trends The results have shown a tendency for the synoptic situationsfavoring dust events to increase in the last decades whereas the synoptic situations not favoring such events tend to decrease withtime

1 Introduction

It is common knowledge that dust transport fromdesert areasis not confined to the regions adjacent to the desert sourceitself but dust can be transported extensively and the finaldeposition can be thousands of kilometers away from theoriginating source (see [1]) According to Marelli [2] thisnatural contribution to Particulate Matter (PM) may rangefrom 5 to 50 in different European countries The Medi-terranean Basin is generally recognized as a major recipientof desert dust originating from the Sahara and Saudi Arabiadeserts several tkm2 are deposited each year in theMediter-ranean Sea [3] profoundly affecting its coastal regions (see [4ndash15])

Dust transport can be defined as a three-stage processinitially dust is lifted and suspended in the atmospheric airresulting in the occurrence of high level concentrations ofdust in desert areas and then transported to great distancesfrom the initial source where it finally settles on the groundThis is a quite complex phenomenon since a large numberof factors contribute to its intensity and frequency Particu-larly in countries of the southeastern Mediterranean region

which is particularly affected there is an acute interest inunderstanding the phenomenon itself as well as the variousmechanisms which govern important attributes like theseverity duration and time trends of dust transportation(see [16]) The present study focuses on the third stage ofdust transportation namely the deposition of desert dust atground level

The importance of such a task is undoubted since dusttransport episodes affect a number of human activities rang-ing frompublic health to ground and air transportation safetyand renewable energy efficiency (to name but a few) Restrict-ing the exemplars of dust implications to these three areas itis worth elaborating that implications of high dust levels tohuman health include respiratory disorders eye inflamma-tions heart disease and lung cancer (see [17 18]) A deepunderstanding of dust events and their associated processes isessential for the application of renewable energy systems thatutilize solar radiation (eg photovoltaic systems for the gen-eration of electricity) These systems are affected profoundlyby dust deposition (wet or dry) which can lead to seriousproblems such as system degradation or inefficiency due tofrequent cleaning requirements [19] Also aviation safety can

2 Journal of Climatology

be severely affected in some circumstances as the abrasiveeffect of dust on aircraft hull and engines can have an accu-mulating adverse impact (see [20]) Dust events can lead to aconsiderable reduction of visibility at airports thus resultingin delays and cancellations (see [21]) Under severe condi-tions road transportation can also be affected by reducedvisibility [22]

A novel approach has recently been proposed contribut-ing to the diagnosis and prediction of dust events in theeasternMediterraneanThe proposedmethodology incorpo-rates the adoption of Artificial Neural Networks (ANN) inorder to diagnose and predict atmospheric pollutant levels inthe eastern Mediterranean resulting from the transportationof dust from the originating sources located in adjacentregions (see [23]) In that study synoptic circulation typesand surface measurements were employed along with neu-ral methodologies [24] and satellite measurements In thepresent paper the association between synoptic patterns thatwas established by using the ANN methodology and dustevents is presented Thus the first objective of the researchis to identify favorable synoptic situations that lead to dustevents over Cyprus in order to justify objectively the longexisting empirical knowledge of preference of occurrence ofdust events to synoptic types

Having performed this a second objective of this paperis to investigate whether the frequency of appearance ofsuch favorable or unfavorable synoptic patterns exhibits anytime tendencies which can subsequently be used to explainthe increase with time of dust episodes this increase hasbeen (qualitatively and subjectively) registered at least duringthe last decade at the weather observing station networkmaintained by the Meteorological Service of Cyprus

Section 2 summarizes the methodologies and data usedin this study also a brief account of the known relationshipbetween dust episodes and prevailing weather conditions isgiven The results section (Section 3) is separated into twoparts in the first the relationship between synoptic patternsand dust events is established (first objective above) and inthe second the trends with respect to time of the synoptic pat-terns are investigated (second objective above) Conclusionsare presented in Section 4

2 Methodologies and Data

Michaelides et al [23] employed digital meteorologicalcharts satellite data and surface measurements of dust inorder to determine which synoptic circulation types are mostlikely to produce conditions which favor dust transport epi-sodes over the island of Cyprus For that study a numberof integral factors were investigated and discussed namelythe atmospheric conditions leading to dust transportationand the meteorological conditions associated with the phe-nomenon along with a presentation of the surface measure-ments of PM10 (particles that are less than 10 120583m in aero-dynamic diameter) used for the definition of a dust eventAlso the methodology for using ANN in order to construct aclassification of synoptic patterns as well as the identificationof those selected types favoring dust transportation was dis-cussed Furthermore the exploitation of satellite technology

in estimating dust load in the atmosphere (using the Atmo-spheric Optical Thickness determined by the ModerateResolution Imaging Spectrometer (MODIS) sensor onboardthe Aqua-Terra satellites) was presented The application ofmultiple regression in combination with the synoptic clas-sification for the prediction of dust episodes was discussedas well as a neural network prediction methodology Finallyan integrated approach for the prediction of dust episodesthat makes use of either the multiple regression or the neuralapproaches was considered

In the following a brief outline of the ANNmethodologyadopted is given The reader is referred to Michaelides et al[23 25] for more details on this methodology the mathemat-ical formulations and the results obtained for the synopticclassification

The neural network architecture used in this research isKohonenrsquos Self Organizing Maps (SOM) [26 27] These net-works provide a way of representing multidimensional datainmuch lower dimensional spaces usually one or two dimen-sions An advantage of the SOM networks over other neuralnetwork classification techniques is that the Kohonen tech-nique creates a network that stores information in such away that any topological relationships within the training setare maintained for example even if the Kohonen networkassociates weather patterns with dust events inaccurately theerror obtained will not be of great amplitude since the resultwill be a class with similar characteristics For a recent reviewof the advantages in using SOM as a tool in synoptic clima-tology the reader is referred to Sheridan and Lee [28] Alsothe work by Tsakovski et al [29] stresses the importance andusefulness of Kohonenrsquos Self Organizing Maps in reachingsuch classification goals as well as their role in decision-making processes

Kohonenrsquos SOM algorithm in its unsupervised mode waschosen for building the neural network models becauseneither the number of output classes nor the desired output isknown a prioriThis is a typical example where unsupervisedlearning is more appropriate since the domain expert will begiven the chance to see the results and decide which modelgives the best resultsThe expertrsquos guidance can help to decidethe number of output classes that better represent the system[30]

In unsupervised learning there are no target values as inthe case of other methods of ANN Given a training set ofdata the objective is to discover significant features or regu-larities in the training data (input data) The neural networkattempts to map the input feature vectors onto an array ofneurons (usually one- or two-dimensional) thereby it com-presses information while preserving the most importanttopological andmetric relationships of the primary data itemson the display By doing so the input feature vectors can beclustered into 119899 clusters where 119899 is less or equal to the numberof neurons used Input vectors are presented sequentially intime without specifying the desired output (see [31 42]) Thetwo-dimensional rectangular grid architecture of KohonenrsquosSOM was adopted in the present research

The input vector is connected with each unit of the net-work through weights whose number is equal to the numberof grid points for the area in study The training procedure

Journal of Climatology 3

utilizes competitive learning When a training example is fedto the network its Euclidean distance to all weight vectorsis computed The neuron whose weight vector is closestto the input vector (in terms of Euclidean distance) is thewinner The weight vectors of the winner neuron as well asits neighborhood neurons are updated in such a way thatthey become closer to the input pattern The radius of theneighborhood around the winner unit is relatively large tostart with in order to include all neurons As the learningprocess continues the neighborhood is consecutively shrunkdown to the point where only the winner unit is updated [32]As more input vectors are represented to the network thesize of the neighborhood decreases until it includes only thewinning unit or the winning unit and some of its neighborsInitially the values of the weights are selected at random

The number of outputs is not a priori determined but anldquooptimumrdquo can be adopted by experimentation in relation tothe specific application under study For several applicationsit appears that the optimum number of outputs is around30 as this exhibits the level of discretization required for thesynoptic scale phenomena examined (see [33])

Local weather forecasters have long identified specificatmospheric circulations as favoring the overall process fordust transportation over the area of eastern MediterraneanThe type of synoptic-scale atmospheric circulation whichfavors wet or dry dust deposits over the eastern Mediterra-nean is comprehensively described as one yielding a southerlyto south-westerly flow throughout the entire troposphere(see [34]) The first step in the process is the elevation andsuspension of dust in the atmosphere which starts with thedevelopment of a north African low pressure system givingrise to a dust stormThis low pressure is primarily initiated intwoways either by an upper-level troughwhich occurs on thepolar front jet when it overlies a heat low or by the presenceof a low level frontal system southeast of the Atlas Mountains[35 36] Both ldquoinitial conditionsrdquo are more frequent in latewinter and spring (see [37]) indeed this is the time of the yearwhen dust events aremost frequent over the easternMediter-ranean [1] The second step the transition of the dust ldquocloudrdquofrom its point of origin to its final destination (which can spana considerable distance) is supported by a south-westerly tro-pospheric flow Finally the third step of dust deposition canoccur under dry conditions (gradual sedimentation due togravity) or under conditions of increased humidity (wheredust particles mix with rain droplets and fall on the groundas colored precipitation) Spring and autumn are the twoseasonal periods favoring dust episodes whereas summerappears to be suppressing these events Dust episodes arealso present in winter although not as much as duringspring or fall

The ability of ANN to group synoptic patterns intoseasonally dependent clusters was originally noted by Mich-aelides et al [25] Since the number of distinctive synopticpatterns used in order to classify synoptic patters over anyparticular geographical region is by no means fixed it wasdecided to run a number of experiments and build clas-sification models with different numbers of output nodes(ie classes) For the present analysis 35 output nodes wereused For the construction of the synoptic patterns data from

the National Centers for Environmental Prediction (NCEP)were retrieved The data consist of the 500 hPa isobaric levelvalues distributed on a 25∘ times 25∘ grid and valid at 1200UTCof each day Such data for the 25-year period 1980ndash2005were exploited in order to establish the synoptic classificationwhich is based on ANN as explained above

A dust transport episode is considered as a day whenthe average PM10 measurement exceeds the threshold of50mg(m3 day) (which is the threshold used by CyprusrsquoMinistry of Labor and Social Insurance) For the needs of thisstudy ground measurements of PM10 were used obtainedfrom the Background Representative Station at Ayia MarinaXyliatou in Cyprus (35 021015840 1710158401015840N 33 031015840 2810158401015840E) and operatedby the Cyprus Ministry of Labor and Social Insurance Thelocation has been selected because it has negligible localparticulate sources so the pollution levels registered areascribed to a large extent to external sources The dataretrieved for this station cover the three-year period 2003ndash2005 which consists of the available data provided by theministry

Having established a relationship between synoptic pat-terns and dust events (by using the available data for thethree-year period as explained above) the synoptic clas-sification over the entire 25-year period (1980ndash2005) wasexploited in an attempt to identify trends in those synopticconditions which favor or not dust transport episodes for thisperiod For each year the total of days belonging to each ofthe 35 classes was counted and the trends were exposed byusing linear regression This task is performed assuming thatthe three-year period used for the identification of classes thatfavor or not dust events is representative enough to establishsuch a relationship

3 Results

31 Synoptic Patterns and Dust Events In the three-yearperiod 2003ndash2005 85 dust deposition events (as definedabove) were recorded (out of a total of 1096 days) Figure 1displays the monthly distribution of these episodes duringthe three-year period 2003ndash2005 It is evident from this figurethat there is a seasonal preference for dust events to occur

It has long been realized that there is a strong associationbetween large scale atmospheric circulation patterns andregional meteorological phenomena that are observed atEarthrsquos surface making synoptic charts a valuable tool for theoperational weather forecaster to predict qualitatively occur-rences of certain weather phenomena over particular areas(see [33 38]) One such typical example is the close associa-tion between the atmospheric circulation and the onset andmaintenance of desert dust transport episodes

Several techniques for weather type classification can befound in the literature automatic objective and consistentmethodologies each of which being applicable to differentcircumstances and different problems (eg [39ndash41]) Eachmethod has its strong and weak points For the present studya new methodology with the adoption of ANN for the classi-fication of synoptic circulations as it was originally proposedby Michaelides et al [25] is utilized The methodologyemploysKohonenrsquos Self OrganizingMaps (SOM) architecture


Janu

ary

Febr

uary

Mar

ch

April

May

June July

Augu

st

Sept

embe

r

Oct

ober

Nov

embe

r

Dec

embe

r0

5

10

15

Month

Num

ber o

f eve

nts

Figure 1 Monthly totals of dust events in the three-year period2003ndash2005

0 5 10 15 20 25 30 350

1

2

3

4

5

6

7

Class number

Num

ber o

f day

s (

)

Figure 2 Percentage of days per class for the 35 classes in thesynoptic classification during 2003ndash2005

[27] this is a neural network method with unsupervisedlearning (see also [25 26 42]) for the classification ofdistribution of isobaric patterns on 500 hPa chartsThe resultof this classification is the formulation of 35 synoptic pro-totypes (classes) Using these results a correspondence wasmade between dust events and particular synoptic classes asexplained below

Figure 2 shows the frequency of appearance of the 35 syn-optic patterns for the three-year period 2003ndash2005 with class35 being themost frequently encountered followed by classes1 25 and 34

Figure 3 shows the way in which the 85 dust transportevents are distributed among the 35 classes in the synopticclassification adopted above There appears to be a certainpreference of classes associated with these dust events mostprone to dust events is class 1 followed by class 31 For the sake

Table 1 Categorization of classes according to the associationbetween dust events

Category Classes

Classes associated with dust events(ordered in diminishing importance)

1 31 11 15 20 23 6 7 1032 16 17 35 13 14 26 28 18 19 22 24 29 30

33 and 34Classes with no association to dustevents

3 4 5 9 12 21 25 27and 28

0

2

4

6

8

10

12

Num

ber o

f eve

nts

0 5 10 15 20 25 30 35Class number

Figure 3 Distribution of dust deposition events per class for the 35classes in the synoptic classification during 2003ndash2005

of brevity representative synoptic situations for these twoclasses are selectively shown in Figure 4 Figure 4(a) refers to1200UTC February 1 2003 and Figure 4(b) at 1200UTCMay10 2004 in the former a central Mediterranean upper troughextends well into the north African desert in the later thetrough axis extends southwards from the Iberian PeninsulaIn both cases typical patterns are identified favoring dustraising and its transfer eastwards with the prevailing south-westerly airflow over the eastern Mediterranean

Table 1 summarizes the findings of the above analysis 26classes appear to be associated with dust events whereas 9classes appear not to be associated with any dust events at all

32 Trends in Synoptic Classes and Dust Deposition EventsFigure 5 shows the frequency of appearance of the synopticpatterns in the classification adopted here for the 25-yearperiod 1980ndash2005 Class 35 is again the most frequentlyencountered followed by classes 30 and 25

Figure 6 presents the time evolution of Classes 1 31 and11 in the 25-year period considered here the first two (ieclasses 1 and 31) exhibit an overall negative trend while thethird (ie class 11) exhibits a positive one These classes werefound above to represent the synoptic conditions that mostlyfavor dust deposition in the 3-year period

Table 2 summarizes the trends for each of the 35 classesin the classification It is evident from this table that while


5840 5840

5840

5760

5760 5760

5760

5680

5680

5680

5600

5600

5600

5520

5520

5520

5440

5440

5440

5360

5360

5360

52805200

60∘N

50∘N

40∘N

30∘N

20∘N

0∘

20∘E

40∘E20

∘W

(a)

5920

58405840

58405760

5760

5760

5680

5680

5680

5600

5600

5600

5520

5520

5520

5440

60∘N

50∘N

40∘N

30∘N

20∘N

0∘

20∘E

40∘E20

∘W

(b)

Figure 4 Representative synoptic situations at 500 hPa corresponding to (a) class 1 (b) class 31 Isolines are drawn for every 60 geopotentialmeters

0 5 10 15 20 25 30 350

50

100

150

200

250

300

350

400

450

500

Class number

Num

ber o

f day

s

Figure 5 Number of days per class for the 35 classes in the synopticclassification during 1980ndash2005

for both categories (dust and no dust) an overall prevalenceof negative trends was found this is more notable in the nodust Classes Indeed 8 out of the 9 no dust classes exhibit anegative trend regarding the dust related cases for 14 out of26 classes the trend is negative too Furthermore for the lattercategory if we consider only the classes that are associatedwithmore than twodust events then 6 out of 13 classes exhibitnegative trends Bearing the above in mind we can postulatethat there is an increased presence (with time) of Classes thatfavor the presence of dust however this postulation surelyneeds to be investigated further by involving a larger periodthan the one used for this research work

4 Concluding Remarks

As mentioned in the introduction local weather forecastershave traditionally identified specific atmospheric circulations

as favoring the overall process for dust transportation overthe area of eastern Mediterranean However such a tradi-tional association of atmospheric circulations and dust eventsis purely qualitative and subjective The present researchpresents an objective methodology which may be applied inestablishing an association between the prevailing synopticcirculation and the occurrence of dust events Howeverbecause of the limited time span of the ground dust collectiondata used further research is required in order to reach a firmconclusion

In this respect an objective classification of synoptic typesas they are portrayed by the 500 hPa isobaric analyses wasperformed Adopting this classification an association bet-ween dust transportation and prevailing weather conditionsin the middle troposphere is established These results cansubsequently be utilized in studying dust transport trendsfor the island of Cyprus Despite the fact that the results aresite specific and the ANN methodologies used are generallyhighly data demanding the methodology has some meritsince it allows expanding our knowledge on the phenome-non Indeed by utilizing the results of this study one canachieve some idea of how these synoptic types are associatedwith dust episodes The findings can also be used to note thetemporal evolution of the synoptic patterns established andalso reveal their trends

The findings of the research performed can be summa-rized as follows

(i) There seems to be a preference of synoptic patternsfavoring dust event occurrence over Cyprus also it isrevealed that there are some other synoptic situationsthat do not favor such events at all

(ii) There seems to be a tendency for the synoptic sit-uations favoring dust events to increase with timewhereas the synoptic situations not favoring suchevents tend to decrease with time


1980 1985 1990 1995 2000 20050

5

10

15

20

25

30

35

Year

Num

ber o

f eve

nts

(a) Class 31

1980 1985 1990 1995 2000 20050

5

10

15

20

25

30

35

Year

Num

ber o

f eve

nts

(b) Class 1

1980 1985 1990 1995 2000 20050

5

10

15

20

25

Year

Num

ber o

f eve

nts

(c) Class 11

Figure 6 The time evolution of classes 31 (a) 1 (b) and 11 (c) in the 25-year period (1980ndash2005)

The fundamental mechanism behind the initiation andfurther evolution of a dust episode over the study area is oneof synoptic scale bearing this in mind the above outcomesof this research can be used to explain the observed increasein dust episodes over the area as revealed from synopticobservations at weather observing stations in the country(independently recorded from the data used in this study)

Although as mentioned above the findings in this studyare based on a limited time span of dust ground mea-surements the above conclusions are qualitatively justifiedusing the opinion knowledge and experience of operationalweather forecasters working in the area for many decadesThis qualitative validation of the findings focuses mainlyon the justification of the relationship between dust eventsand specific synoptic situations and the trend for increasingnumber of dust events over the area over the past decade asmentioned in the introduction

The use of an objective methodology for synoptic patternclassification is appropriate Any subjective classificationmethod can only be used as a first guess approach as it canassist to identify important favorable patterns but cannot pro-vide an objective quantitative background for an investigationsuch as the one presented here (see [43]) ANN appears tobe appropriate for the purpose as it comprises a nonlinearapproach to tackling a very complex and definitely non-linearproblem (see [44])

Plans for further research by the group include the expan-sion of the ground dust measurements once it becomes avail-able and experimentation with other synoptic classificationmethodologies (see [41 45])

Acknowledgments

Parts of this work have been carried outwithin the frameworkof the Extreme Weather impacts on European Networks


Table 2 Synoptic classes and their trends for the 25-year period1980ndash2005 Classes associated with dust events are in bold whereasclasses with no association to dust events are not in bold positivetrends are shown as POS and negative trends as NEG The 1198772 foreach class is also shown

Class Slope Trend 1198772

1 minus01019 NEG 000832 minus00588 NEG 000853 minus00537 NEG 00064 minus01255 NEG 00335 minus03484 NEG 01486 01036 POS 00367 minus00803 NEG 00258 00581 POS 0019 minus00058 NEG 000110 0147 POS 005211 02574 POS 0112 minus01015 NEG 001813 minus00147 NEG 000114 minus00291 NEG 000115 minus00472 NEG 000416 minus00817 NEG 002217 minus00574 NEG 000918 minus00229 NEG 000219 00131 POS 004520 00499 POS 000621 minus00048 NEG 000122 0012 POS 000123 00961 POS 002524 minus0121 NEG 002225 00834 POS 001226 00602 POS 002127 minus00048 NEG 000128 minus01959 NEG 009229 minus00178 NEG 000130 01084 POS 001331 minus01364 NEG 00232 0026 POS 000133 minus01159 NEG 001334 minus01338 NEG 000235 06376 POS 026

of Transport (EWENT) project that was funded by theEuropean Commission under its 7th Framework Programme(Transport Horizontal Activities)The authors wish to thankthe anonymous reviewers for their constructive commentsthat have led to important improvement of the paper

References

[1] U Dayan J Heffter J Miller and G Gutman ldquoDust intrusionevents into the Mediterranean basinrdquo Journal of Applied Meteo-rology vol 30 no 8 pp 1185ndash1199 1991

[2] L Marelli ldquoContribution of natural sources to air pollutionlevels in the EU a technical basis for the development ofguidance for the member statesrdquo in Post Workshop Report FromldquoContribution of Natural Sources to PM Levels in Europerdquo EUR22779 EN JRC Ispra 2007

[3] E Ganor and YMamane ldquoTransport of Saharan dust across theeastern Mediterraneanrdquo Atmospheric Environment vol 16 no3 pp 581ndash587 1982

[4] G Bergametti L Gomes M N Le Coustumer G Coude-Gaussen and P Rognon ldquoAfrican dust observed over CanaryIslands source-regions identification and transport pattern forsome summer situationsrdquo Journal of Geophysical Research vol94 no 12 pp 14ndash864 1989

[5] S Guerzoni and R Chester The Impact of the Desert DustAcross the Mediterranean Kluwer Academic Publishers Nor-well Mass USA 1996

[6] X Querol A Alastuey J A Puicercus et al ldquoSeasonal evolutionof suspended particles around a large coal-fired power stationparticulate levels and sourcesrdquo Atmospheric Environment vol32 no 11 pp 1963ndash1978 1998

[7] S Rodrıguez X Querol A Alastuey G Kallos and O Kaka-liagou ldquoSaharan dust contributions to PM10 and TSP levels inSouthern and Eastern SpainrdquoAtmospheric Environment vol 35no 14 pp 2433ndash2447 2001

[8] J Barkan P Alpert H Kutiel and P Kishcha ldquoSynoptics ofdust transportation days from Africa toward Italy and centralEuroperdquo Journal of Geophysical ResearchD vol 110 no 7 ArticleID D07208 pp 1ndash14 2005

[9] M Escudero S Castillo X Querol et al ldquoWet and dry Africandust episodes over eastern Spainrdquo Journal of GeophysicalResearch D vol 110 no 18 Article ID D18S08 pp 1ndash15 2005

[10] E Gerasopoulos G Kouvarakis P Babasakalis M VrekoussisJ P Putaud and N Mihalopoulos ldquoOrigin and variability ofparticulatematter (PM10)mass concentrations over the EasternMediterraneanrdquo Atmospheric Environment vol 40 no 25 pp4679ndash4690 2006

[11] G Kallos A Papadopoulos P Katsafados and S NickovicldquoTransatlantic Saharan dust transport model simulation andresultsrdquo Journal of Geophysical Research D vol 111 no 9 ArticleID D09204 2006

[12] G Kallos M Astitha P Katsafados and C Spyrou ldquoLong-range transport of anthropogenically and naturally producedparticulate matter in the Mediterranean and North Atlanticcurrent state of knowledgerdquo Journal of Applied Meteorology andClimatology vol 46 no 8 pp 1230ndash1251 2007

[13] M Kocak N Mihalopoulos and N Kubilay ldquoContributions ofnatural sources to high PM10 and PM25 events in the easternMediterraneanrdquo Atmospheric Environment vol 41 no 18 pp3806ndash3818 2007

[14] C Mitsakou G Kallos N Papantoniou et al ldquoSaharan dustlevels in Greece and received inhalation dosesrdquo AtmosphericChemistry and Physics vol 8 no 23 pp 7181ndash7192 2008

[15] C D Papadimas N Hatzianastassiou N Mihalopoulos XQuerol and I Vardavas ldquoSpatial and temporal variability inaerosol properties over the Mediterranean basin based on 6-year (2000ndash2006)MODIS datardquo Journal of Geophysical ResearchD vol 113 no 11 Article ID D11205 2008


[16] S Michaelides P Evripidou and G Kallos ldquoMonitoring andpredicting Saharan desert dust events in the eastern Mediter-raneanrdquoWeather vol 54 no 11 pp 359ndash365 1999

[17] D W Griffin V H Garrison J R Herman and E A ShinnldquoAfrican desert dust in the Caribbean atmosphere microbiol-ogy and public healthrdquo Aerobiologia vol 17 no 3 pp 203ndash2132001

[18] T Sandstrom and B Forsberg ldquoDesert dust an unrecognizedsource of dangerous air pollutionrdquo Epidemiology vol 19 no 6pp 808ndash809 2008

[19] H K Elminir A E Ghitas R H Hamid F El-Hussainy MM Beheary and K M Abdel-Moneim ldquoEffect of dust on thetransparent cover of solar collectorsrdquo Energy Conversion andManagement vol 47 no 18-19 pp 3192ndash3203 2006

[20] J J Simpson G L Hufford R Servranckx J Berg and D PierildquoAirborne Asian dust case study of long-range transport andimplications for the detection of volcanic ashrdquoWeather amp Fore-casting vol 18 no 2 pp 121ndash141 2003

[21] A Sunnu F Resch andG Afetia ldquoBack-trajectorymodel of theSaharan dust flux and particlemass distribution inWest AfricardquoAeolian Research vol 9 pp 125ndash132 2013

[22] LMona Z Liu DMuller et al ldquoLidarmeasurements for desertdust characterization an overviewrdquo Advances in Meteorologyvol 2012 Article ID 356265 36 pages 2012

[23] S Michaelides F Tymvios D Paronis and A Retalis ldquoArtificialneural networks for the diagnosis and prediction of desert dusttransport episodesrdquo Studies in Fuzziness and Soft Computingvol 269 pp 285ndash304 2011

[24] M H Beale M T Hagan and H B Demuth Neural NetworkToolbo Userrsquos Guide The MathWorks 2010 httpwwwmath-workscomhelppdf docnnetnnetpdf

[25] S C Michaelides F Liassidou and C N Schizas ldquoSynopticclassification and establishment of analogueswith artificial neu-ral networksrdquo Pure and Applied Geophysics vol 164 no 6-7 pp1347ndash1364 2007

[26] T Kohonen ldquoThe self-organizingmaprdquo Proceedings of the IEEEvol 78 no 9 pp 1464ndash1480 1990

[27] T Kohonen Self-Organizing Maps vol 30 of Series in Informa-tion Sciences Springer Heidelberg Germany 2nd edition 1997

[28] S C Sheridan and C C Lee ldquoThe self-organizing map in syn-optic climatological researchrdquo Progress in Physical Geographyvol 35 no 1 pp 109ndash119 2011

[29] S Tsakovski P Simeonova V Simeonov M C Freitas IDionısio and A M G Pacheco ldquoAir-quality assessment ofPico-mountain environment (Azores) by using chemometricand trajectory analysesrdquo Journal of Radioanalytical and NuclearChemistry vol 281 no 1 pp 17ndash22 2009

[30] C S Pattichis C N Schizas and L T Middleton ldquoNeuralnetwork models in EMG diagnosisrdquo IEEE Transactions onBiomedical Engineering vol 42 no 5 pp 486ndash496 1995

[31] F Schnorrenberg C S Pattichis C N Schizas K Kyriacou andM Vassiliou ldquoComputer-aided classification of breast cancernucleirdquo Technology and Health Care vol 4 no 2 pp 147ndash1611996

[32] D W Patterson Artificial Neural Networks Theory and Appli-cations Prentice Hall 1995

[33] F Tymvios K Savvidou and S C Michaelides ldquoAssociationof geopotential height patterns with heavy rainfall events inCyprusrdquo Advances in Geosciences vol 23 pp 73ndash78 2010

[34] N G Prezerakos A G Paliatsos and K V Koukoulet-sos ldquoDiagnosis of the relationship between dust storms over

the Sahara desert and dust deposit or coloured rain in the SouthBalkansrdquoAdvances inMeteorology vol 2010 Article ID 76054614 pages 2010

[35] N G Prezerakos ldquoSynoptic flow patterns leading to the gener-ation of north-west African depressionsrdquo International Journalof Climatology vol 10 no 1 pp 33ndash48 1990

[36] N G Prezerakos S C Michaelides and A S Vlassi ldquoAtmo-spheric synoptic conditions associated with the initiation ofnorth-west African depressionsrdquo International Journal of Cli-matology vol 10 no 7 pp 711ndash729 1990

[37] N Kubilay S Nickovic CMoulin and FDulac ldquoAn illustrationof the transport and deposition of mineral dust onto the easternMediterraneanrdquo Atmospheric Environment vol 34 no 8 pp1293ndash1303 2000

[38] S Michaelides F Tymvios and D Charalambous ldquoInvestiga-tion of trends in synoptic patterns over Europe with artificialneural networksrdquo Advances in Geosciences vol 23 pp 107ndash1122010

[39] A K A El-Kadi and P A Smithson ldquoAtmospheric classifica-tions and synoptic climatologyrdquo Progress in Physical Geographyvol 16 no 4 pp 432ndash455 1992

[40] P Maheras I Patrikas T Karacostas and C Anagnosto-poulou ldquoAutomatic classification of circulation types in Greecemethodology description frequency variability and trendanalysisrdquo Theoretical and Applied Climatology vol 67 no 3-4pp 205ndash223 2000

[41] A J Cannon P H Whitfield and E R Lord ldquoSynoptic map-pattern classification using recursive partitioning and principalcomponent analysisrdquo Monthly Weather Review vol 130 no 5pp 1187ndash1206 2002

[42] S C Michaelides C S Pattichis and G Kleovoulou ldquoClassifi-cation of rainfall variability by using artificial neural networksrdquoInternational Journal of Climatology vol 21 no 11 pp 1401ndash14142001

[43] HM van denDool ldquoA new look at weather forecasting throughanaloguesrdquo Monthly Weather Review vol 117 no 10 pp 2230ndash2247 1989

[44] E N Lorenz ldquoAtmospheric predictability as revealed by natu-rally occurring analoguesrdquo Journal of the Atmospheric Sciencesvol 26 no 4 pp 636ndash646 1969

[45] A Philipp J Bartholy C Beck et al ldquoCost733catmdasha databaseof weather and circulation type classificationsrdquo Physics andChemistry of the Earth vol 35 no 9ndash12 pp 360ndash373 2010

Submit your manuscripts athttpwwwhindawicom

Forestry ResearchInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Environmental and Public Health

Journal of



EcosystemsJournal of


MeteorologyAdvances in

EcologyInternational Journal of


Marine BiologyJournal of


Hindawi Publishing Corporationhttpwwwhindawicom

Applied ampEnvironmentalSoil Science

Volume 2014

Advances in


Environmental Chemistry

Atmospheric SciencesInternational Journal of



Waste ManagementJournal of

Hindawi Publishing Corporation httpwwwhindawicom Volume 2014

International Journal of

Geophysics


Geological ResearchJournal of

EarthquakesJournal of


BiodiversityInternational Journal of


ScientificaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

OceanographyInternational Journal of


The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Journal of Computational Environmental SciencesHindawi Publishing Corporationhttpwwwhindawicom Volume 2014


ClimatologyJournal of


be severely affected in some circumstances as the abrasiveeffect of dust on aircraft hull and engines can have an accu-mulating adverse impact (see [20]) Dust events can lead to aconsiderable reduction of visibility at airports thus resultingin delays and cancellations (see [21]) Under severe condi-tions road transportation can also be affected by reducedvisibility [22]

A novel approach has recently been proposed contribut-ing to the diagnosis and prediction of dust events in theeasternMediterraneanThe proposedmethodology incorpo-rates the adoption of Artificial Neural Networks (ANN) inorder to diagnose and predict atmospheric pollutant levels inthe eastern Mediterranean resulting from the transportationof dust from the originating sources located in adjacentregions (see [23]) In that study synoptic circulation typesand surface measurements were employed along with neu-ral methodologies [24] and satellite measurements In thepresent paper the association between synoptic patterns thatwas established by using the ANN methodology and dustevents is presented Thus the first objective of the researchis to identify favorable synoptic situations that lead to dustevents over Cyprus in order to justify objectively the longexisting empirical knowledge of preference of occurrence ofdust events to synoptic types

Having performed this a second objective of this paperis to investigate whether the frequency of appearance ofsuch favorable or unfavorable synoptic patterns exhibits anytime tendencies which can subsequently be used to explainthe increase with time of dust episodes this increase hasbeen (qualitatively and subjectively) registered at least duringthe last decade at the weather observing station networkmaintained by the Meteorological Service of Cyprus

Section 2 summarizes the methodologies and data usedin this study also a brief account of the known relationshipbetween dust episodes and prevailing weather conditions isgiven The results section (Section 3) is separated into twoparts in the first the relationship between synoptic patternsand dust events is established (first objective above) and inthe second the trends with respect to time of the synoptic pat-terns are investigated (second objective above) Conclusionsare presented in Section 4

2 Methodologies and Data

Michaelides et al [23] employed digital meteorologicalcharts satellite data and surface measurements of dust inorder to determine which synoptic circulation types are mostlikely to produce conditions which favor dust transport epi-sodes over the island of Cyprus For that study a numberof integral factors were investigated and discussed namelythe atmospheric conditions leading to dust transportationand the meteorological conditions associated with the phe-nomenon along with a presentation of the surface measure-ments of PM10 (particles that are less than 10 120583m in aero-dynamic diameter) used for the definition of a dust eventAlso the methodology for using ANN in order to construct aclassification of synoptic patterns as well as the identificationof those selected types favoring dust transportation was dis-cussed Furthermore the exploitation of satellite technology

in estimating dust load in the atmosphere (using the Atmo-spheric Optical Thickness determined by the ModerateResolution Imaging Spectrometer (MODIS) sensor onboardthe Aqua-Terra satellites) was presented The application ofmultiple regression in combination with the synoptic clas-sification for the prediction of dust episodes was discussedas well as a neural network prediction methodology Finallyan integrated approach for the prediction of dust episodesthat makes use of either the multiple regression or the neuralapproaches was considered

In the following a brief outline of the ANNmethodologyadopted is given The reader is referred to Michaelides et al[23 25] for more details on this methodology the mathemat-ical formulations and the results obtained for the synopticclassification

The neural network architecture used in this research isKohonenrsquos Self Organizing Maps (SOM) [26 27] These net-works provide a way of representing multidimensional datainmuch lower dimensional spaces usually one or two dimen-sions An advantage of the SOM networks over other neuralnetwork classification techniques is that the Kohonen tech-nique creates a network that stores information in such away that any topological relationships within the training setare maintained for example even if the Kohonen networkassociates weather patterns with dust events inaccurately theerror obtained will not be of great amplitude since the resultwill be a class with similar characteristics For a recent reviewof the advantages in using SOM as a tool in synoptic clima-tology the reader is referred to Sheridan and Lee [28] Alsothe work by Tsakovski et al [29] stresses the importance andusefulness of Kohonenrsquos Self Organizing Maps in reachingsuch classification goals as well as their role in decision-making processes

Kohonenrsquos SOM algorithm in its unsupervised mode waschosen for building the neural network models becauseneither the number of output classes nor the desired output isknown a prioriThis is a typical example where unsupervisedlearning is more appropriate since the domain expert will begiven the chance to see the results and decide which modelgives the best resultsThe expertrsquos guidance can help to decidethe number of output classes that better represent the system[30]

In unsupervised learning there are no target values as inthe case of other methods of ANN Given a training set ofdata the objective is to discover significant features or regu-larities in the training data (input data) The neural networkattempts to map the input feature vectors onto an array ofneurons (usually one- or two-dimensional) thereby it com-presses information while preserving the most importanttopological andmetric relationships of the primary data itemson the display By doing so the input feature vectors can beclustered into 119899 clusters where 119899 is less or equal to the numberof neurons used Input vectors are presented sequentially intime without specifying the desired output (see [31 42]) Thetwo-dimensional rectangular grid architecture of KohonenrsquosSOM was adopted in the present research

The input vector is connected with each unit of the net-work through weights whose number is equal to the numberof grid points for the area in study The training procedure









3 Results





Janu

ary

Febr

uary

Mar

ch

April

May

June July

Augu

st

Sept

embe

r

Oct

ober

Nov

embe

r

Dec

embe

r0

5

10

15

Month

Num

ber o

f eve

nts


0 5 10 15 20 25 30 350

1

2

3

4

5

6

7

Class number

Num

ber o

f day

s (

)






Category Classes


1 31 11 15 20 23 6 7 1032 16 17 35 13 14 26 28 18 19 22 24 29 30


3 4 5 9 12 21 25 27and 28

0

2

4

6

8

10

12

Num

ber o

f eve

nts

0 5 10 15 20 25 30 35Class number








5840 5840

5840

5760

5760 5760

5760

5680

5680

5680

5600

5600

5600

5520

5520

5520

5440

5440

5440

5360

5360

5360

52805200

60∘N

50∘N

40∘N

30∘N

20∘N

0∘

20∘E

40∘E20

∘W

(a)

5920

58405840

58405760

5760

5760

5680

5680

5680

5600

5600

5600

5520

5520

5520

5440

60∘N

50∘N

40∘N

30∘N

20∘N

0∘

20∘E

40∘E20

∘W

(b)


0 5 10 15 20 25 30 350

50

100

150

200

250

300

350

400

450

500

Class number

Num

ber o

f day

s











1980 1985 1990 1995 2000 20050

5

10

15

20

25

30

35

Year

Num

ber o

f eve

nts

(a) Class 31

1980 1985 1990 1995 2000 20050

5

10

15

20

25

30

35

Year

Num

ber o

f eve

nts

(b) Class 1

1980 1985 1990 1995 2000 20050

5

10

15

20

25

Year

Num

ber o

f eve

nts

(c) Class 11






Acknowledgments







References




















































Journal of












Volume 2014

Advances in









Geophysics






















3 Results





Janu

ary

Febr

uary

Mar

ch

April

May

June July

Augu

st

Sept

embe

r

Oct

ober

Nov

embe

r

Dec

embe

r0

5

10

15

Month

Num

ber o

f eve

nts


0 5 10 15 20 25 30 350

1

2

3

4

5

6

7

Class number

Num

ber o

f day

s (

)






Category Classes


1 31 11 15 20 23 6 7 1032 16 17 35 13 14 26 28 18 19 22 24 29 30


3 4 5 9 12 21 25 27and 28

0

2

4

6

8

10

12

Num

ber o

f eve

nts

0 5 10 15 20 25 30 35Class number








5840 5840

5840

5760

5760 5760

5760

5680

5680

5680

5600

5600

5600

5520

5520

5520

5440

5440

5440

5360

5360

5360

52805200

60∘N

50∘N

40∘N

30∘N

20∘N

0∘

20∘E

40∘E20

∘W

(a)

5920

58405840

58405760

5760

5760

5680

5680

5680

5600

5600

5600

5520

5520

5520

5440

60∘N

50∘N

40∘N

30∘N

20∘N

0∘

20∘E

40∘E20

∘W

(b)


0 5 10 15 20 25 30 350

50

100

150

200

250

300

350

400

450

500

Class number

Num

ber o

f day

s











1980 1985 1990 1995 2000 20050

5

10

15

20

25

30

35

Year

Num

ber o

f eve

nts

(a) Class 31

1980 1985 1990 1995 2000 20050

5

10

15

20

25

30

35

Year

Num

ber o

f eve

nts

(b) Class 1

1980 1985 1990 1995 2000 20050

5

10

15

20

25

Year

Num

ber o

f eve

nts

(c) Class 11






Acknowledgments







References




















































Journal of












Volume 2014

Advances in









Geophysics















Janu

ary

Febr

uary

Mar

ch

April

May

June July

Augu

st

Sept

embe

r

Oct

ober

Nov

embe

r

Dec

embe

r0

5

10

15

Month

Num

ber o

f eve

nts


0 5 10 15 20 25 30 350

1

2

3

4

5

6

7

Class number

Num

ber o

f day

s (

)






Category Classes


1 31 11 15 20 23 6 7 1032 16 17 35 13 14 26 28 18 19 22 24 29 30


3 4 5 9 12 21 25 27and 28

0

2

4

6

8

10

12

Num

ber o

f eve

nts

0 5 10 15 20 25 30 35Class number








5840 5840

5840

5760

5760 5760

5760

5680

5680

5680

5600

5600

5600

5520

5520

5520

5440

5440

5440

5360

5360

5360

52805200

60∘N

50∘N

40∘N

30∘N

20∘N

0∘

20∘E

40∘E20

∘W

(a)

5920

58405840

58405760

5760

5760

5680

5680

5680

5600

5600

5600

5520

5520

5520

5440

60∘N

50∘N

40∘N

30∘N

20∘N

0∘

20∘E

40∘E20

∘W

(b)


0 5 10 15 20 25 30 350

50

100

150

200

250

300

350

400

450

500

Class number

Num

ber o

f day

s











1980 1985 1990 1995 2000 20050

5

10

15

20

25

30

35

Year

Num

ber o

f eve

nts

(a) Class 31

1980 1985 1990 1995 2000 20050

5

10

15

20

25

30

35

Year

Num

ber o

f eve

nts

(b) Class 1

1980 1985 1990 1995 2000 20050

5

10

15

20

25

Year

Num

ber o

f eve

nts

(c) Class 11






Acknowledgments







References




















































Journal of












Volume 2014

Advances in









Geophysics















5840 5840

5840

5760

5760 5760

5760

5680

5680

5680

5600

5600

5600

5520

5520

5520

5440

5440

5440

5360

5360

5360

52805200

60∘N

50∘N

40∘N

30∘N

20∘N

0∘

20∘E

40∘E20

∘W

(a)

5920

58405840

58405760

5760

5760

5680

5680

5680

5600

5600

5600

5520

5520

5520

5440

60∘N

50∘N

40∘N

30∘N

20∘N

0∘

20∘E

40∘E20

∘W

(b)


0 5 10 15 20 25 30 350

50

100

150

200

250

300

350

400

450

500

Class number

Num

ber o

f day

s











1980 1985 1990 1995 2000 20050

5

10

15

20

25

30

35

Year

Num

ber o

f eve

nts

(a) Class 31

1980 1985 1990 1995 2000 20050

5

10

15

20

25

30

35

Year

Num

ber o

f eve

nts

(b) Class 1

1980 1985 1990 1995 2000 20050

5

10

15

20

25

Year

Num

ber o

f eve

nts

(c) Class 11






Acknowledgments







References




















































Journal of












Volume 2014

Advances in









Geophysics















1980 1985 1990 1995 2000 20050

5

10

15

20

25

30

35

Year

Num

ber o

f eve

nts

(a) Class 31

1980 1985 1990 1995 2000 20050

5

10

15

20

25

30

35

Year

Num

ber o

f eve

nts

(b) Class 1

1980 1985 1990 1995 2000 20050

5

10

15

20

25

Year

Num

ber o

f eve

nts

(c) Class 11






Acknowledgments







References




















































Journal of












Volume 2014

Advances in









Geophysics



















References




















































Journal of












Volume 2014

Advances in









Geophysics


















































Journal of












Volume 2014

Advances in









Geophysics


















Journal of












Volume 2014

Advances in









Geophysics














Documents

Research Article Trends of Dust Transport Episodes in ...downloads.hindawi.com/archive/2013/280248.pdf · dust events to synoptic types. Having performed this, a second objective