objective classification of air mass types in hungary, with special interest to air pollution
TRANSCRIPT
Meteorol Atmos Phys 92, 115–137 (2006)DOI 10.1007/s00703-005-0143-x
1 Department of Climatology and Landscape Ecology, University of Szeged, Hungary2 Hungarian Meteorological Service, Budapest, Hungary3 Department of Physics, Laboratory of Meteorology, University of Ioannina, Ioannina, Greece
An objective classification system of air mass types for Szeged,Hungary, with special interest in air pollution levels
L. Makra1, J. Mika2, A. Bartzokas3, R. Beczi1, E. Borsos1, and Z. Sumeghy1
With 3 Figures
Received October 26, 2004; revised January 21, 2005; accepted March 18, 2005Published online: September 15, 2005 # Springer-Verlag 2005
Summary
This paper determines the characteristic air mass types overthe Carpathian Basin for the winter (December, January, andFebruary) and summer (June, July and August) monthsdependant on levels of the main air pollutants. Based on theECMWF data set, daily sea-level pressure fields analysed at00 UTC were prepared for each air mass type (cluster) in orderto relate sea-level pressure patterns with the level of air pol-lutants in Szeged. The data comprise daily values of twelvemeteorological and eight pollutant parameters for the period1997–2001. Objective definition of the characteristic air masstypes is achieved by using the methods of Factor Analysisand Cluster Analysis. According to the results, during thewinter months five air mass types (clusters) were detectedbased on higher concentrations of primary pollutants that oc-cur with high irradiance and low wind speed. This is the casewhen an anticyclone is found over the Carpathian Basin andover the region south of Hungary, influencing the weather ofthe country. Low levels of pollutants occur when zonal currentsexert influence over Hungary. During the summer monthsanticyclones and anticyclone ridge situations are found overthe Carpathian Basin. (During the prevalence of anticycloneridge situations, the Carpathian Basin is found at the edge ofa high pressure centre.) As a result of high irradiance andvery low NO levels, secondary pollutants are highly enriched.
1. Introduction
Air pollution has become a very important envi-ronmental problem, mostly in crowded cities.
Most human activities produce some kind ofpollutants, which are progressively accumulated.Air pollution has negative effects not only on thesurroundings of its source, but can also affectwider regions. Most air pollutants are released inrelation to processes involving combustion. Theemission sources include: transportation, fuelcombustion, industrial processes, solid waste dis-posal, amongst others. These harmful particles inthe air may damage human health, vegetation andthe global environment. In general, in many casesthey can form brown or hazy clouds at groundlevel with unpleasant smells. The damages tohuman health caused by air pollution generallyresult from repeated exposure to even low con-centrations for long periods. Metals tend to cor-rode faster in polluted environments. Paints donot last as long as in clean conditions; tires andother rubber materials also tend to fail due toozone cracking, unless they were produced viathe utilization of antioxidant additives. As withhuman health, the degree of damage caused large-ly depends on the concentration and the du-ration of exposure. Although most gaseous airpollutants are totally colourless, there are someexceptions such as NO2, which is brown. Mostvisible effects of air pollution are the outcome of
the interaction of light with suspended particles(De Nevers, 2000).
Air quality and the concentration of air pol-lutants are influenced not only by physical andchemical processes, but also by meteorological,geographical and social factors. Some weatherconditions, for example, moderate winds or calmair with temperature inversions as typically pre-vailing under anticyclonic conditions, can alsosignificantly influence the rise of extreme con-centration rates of pollutants in the air.
In Europe, many air pollution studies, es-pecially for Athens due to its long summerswith undisturbed irradiation and calm or weakbreezes, have appeared in the international lit-erature. This weather and the mountains, whichsurround the city to the north, favour extremeaccumulation of the air pollutants (Kambezidis
et al, 1995, 1996, 1998; Adamopoulos et al,2002).
According to P�eeczely (1959), based on his obser-vations made on air pollution rates in Budapest,air pollution tends to peak during extensive anti-cyclone events characterized by weak easterlybreezes prevailing over the city. Conversely, airpollution is relatively low during the prevalenceof cyclonic weather systems characterized bystrong and turbulent air currents prevailing overthe Carpathian Basin (Fig. 1a), especially whenHungary is in the rear part of the cyclone.
The major aim of the present study was todevelop an objective, reliable classification ofair mass types prevailing over the city of Szegedduring the summer and winter months via theapplication of multivariate statistical methods. Foreach air mass type, following characterization by
Fig. 1b. Location of Szeged in Csongrad county(centre); Csongrad county in Hungary (topright)
Fig. 1a. Location of the Carpathian Basin
116 L. Makra et al
homogenous temperature and humidity condi-tions, the concentration of the main air pollutantsis estimated. Additionally, in order to revealthe possible relationship between the prevailingweather conditions, the spatial distribution ofsea-level pressure fields and the concentration ofair pollutants in the area of Szeged, mean sea-level pressure fields have been calculated for thedifferent air mass types for the North Atlantic –European region. Unfortunately, very few studieswith such scope are known from the literature.The work of Sindosi et al (2003) could be men-tioned here as an example of a similar analysisapplied to Athens. Alternatively, stability classesare often used, e.g., in air quality modelling, toclassify whether the dispersion of air pollutants ishigh or low due to the prevailing meteorologicalconditions (determined empirically from windspeed, temperature gradient, cloud cover or solarradiation). Pasquill’s method (1962) for diffusionwas developed to estimate vertical and lateralplume widths from a continuous source at var-ious distances. These widths depend primarily onthe standard deviations of vertical and horizontalwind direction fluctuations, respectively. Thesequantities are functions of height, roughnesslength and the Monin-Obukhov length. Wheretemperature profile measurements are not avail-able, Pasquill suggested a set of six stabilityclasses specified in terms of wind speed and irra-diance only. Turner (1964) modified Pasquill’sscheme and defined seven stability categories.In this method, net irradiance is estimated froma knowledge of solar altitude and existing condi-tions of cloud cover and ceiling height. Both thePasquill and Turner classes are independent ofheight and surface roughness (Golder, 1972). Themethod used in this paper is an objective classi-fication compared to the subjectively determinedcategories of Pasquill and Turner. Furthermore,our method takes into account more meteorolog-ical parameters for classifying air mass typesand the efficiency of the classes received ingrouping pollutant concentrations is statisticallyevaluated.
The methodology used in the paper is not pro-posed as a substitute for other chemical transportmodelling systems but as a supplement to theexisting methodologies, attempting to contributeour efforts to forecast pollution levels and thusadopt appropriate measures when necessary.
This study represents an objective weather clas-sification method, which might also serve as astarting point for the creation of an air pollutionmonitoring=forecasting system, with the ultimategoal of facilitating debate regarding air pollutionin the city of Szeged.
2. Topography, climatology and air qualityof Szeged area
2.1 Topography
The city of Szeged is the largest town in SEHungary. The city (20�060 E; 46�150 N) is locatedat the confluence of the Tisza and Maros Riverscharacterized by an extensive flat landscape withan elevation of 79 m a.s.l. (Fig. 1b). The built-uparea covers a region of about 46 km2 with about155,000 inhabitants.
Szeged and its surroundings are not onlycharacterized by extensive lowlands but it alsohas the lowest elevation in Hungary and theCarpathian Basin as well. This results in a‘‘double basin’’ situation. Due to the position ofthe city in a basin (a smaller one within a largerone), temperature inversions form more easily inthe area (e.g., due to cool air flow from the basinslopes) and prevail longer than in flat terrain,leading to an enrichment of air pollutants withinthe inversion layer.
2.2 Climatology
According to the climatic classification system ofK€ooppen, the majority of the Hungarian territory,including Csongrad county and the agglomerationof Szeged, belongs to the Cf climate zone char-acterized by temperate-warm climates with analmost even distribution of precipitation or thatof Trewartha’s D.1 climate zone characterizedby continental climates with long warm seasons.
The more detailed, higher resolution climaticclassification of Hungary is based on the meantemperature values of the growth season (tVS) andthe aridity index (H) [where H¼ES=L �C (ES isthe annual mean radiation balance; L is the latentheat of vaporization and C is the total annualmean precipitation)]. Based on the climatologicalcharacteristics of the period 1901–1950, the cli-mate of Szeged can be considered as warm-drywith tVS>17.5 �C and H>1.15 (P�eeczely, 1979).
Classification system of air mass types for Szeged, Hungary 117
The ES energetic component of the aridityindex (H) changes only slightly in Hungary.Its mean annual value, considering the period1901–1950, is 1760 MJ m�2 yr�1 (P�eeczely, 1979).On the other hand, the total annual precipitationamounts may fluctuate quite substantially. Theextreme values for Szeged in this period werePmin¼ 203 mm; Pmax¼ 867 mm. On the basisof this information, the aridity index calculatedfor the most arid year was H¼ 3.47; while that ofthe most humid year was H¼ 0.81. The typicalvegetation assigned to the former value is desert,while that of the latter is woodland. Thoughpersistence of high H values is not observedall-year or all-summer round, the climate ofsome summers inclines towards semi-arid or aridconditions in the Szeged area. This is reflected inthe composition of the natural vegetation. In thesouthern part of the Great Hungarian Plains na-tive plants include semi-desert species such nee-dle grass (Stipa stenophylla) (Makra et al, 1985).
The climate of Szeged is characterized by hotsummers and moderately cold winters. The dis-tribution of rainfall is fairly uniform during theyear: with a share of 29 and 19% for the summer(JJA) and the winter (DJF) seasons, respectively.Mean daily summer temperatures are around22.4 �C, while the mean daily winter tempera-tures are 2.3 �C. The irradiance values also exhi-bit large-scale variances with an average of 20.2and 4.2 MJ m�2 in summer and winter days,respectively. The most frequent winds blowalong the NNW–SSE axis, with prevailing aircurrents arriving from NNW (42.3%) and SSW(24%) in the summer and from SSE (32.6%) andNNW (30.8%) during the winter. Thanks to itsunique geographical position, Szeged is charac-terized by relatively low wind speeds with aver-age daily summer and winter values of 2.8 and3.5 m s�1, respectively. The highest hourly windspeeds have been recorded during the spring witha rate of 5 m s�1 (P�eeczely, 1979).
The dry and warm climate of the region is espe-cially favourable for the Turanian (Aralo-caspian)semi-arid species. One group of plants preferextremely arid conditions; while another groupcontains species which indicate aridity, namely,these are xeroindicators (indicate long and dryperiods in the climate). The Ellenberg’s indicatornumber (Ellenberg et al, 1991) for the first groupis 1, while for the after group is 2. The lower the
Ellenberg indicator (min¼ 1) the more a speciesprefers a dry and warm climate. Similarly,the higher the number (max¼ 12) the more itfavours a humid environment. Hence, Ellenberg’sindicator number is a 12-degree scale, first ap-plied by Borhidi (1995) in Hungary, and wasintroduced into the Hungarian literature as aWB-value. The WB-value is a relative indicatornumber for ground water or soil humidity (WaterBorhidi). Borhidi also indicated the characteristicspecies for the Hungarian flora, which Ellenbergdealt with little on the scale.
The analysis is applied to the two extremeseasons of the year because the winter and thesummer show the most marked difference in at-mospheric circulation over the Carpathian Basin.In the summer, the subtropical (Azores) anticy-clone of the Atlantic is most frequent with 20–25% of all weather types. Northerly currents arealso characteristic because of the blocking an-ticyclones. In the winter, southerly currents aremost frequent, which are followed by westerlyflows. In both seasons, anticyclone centres andanticyclone ridge weather situations are most fre-quent over the Carpathian Basin, as a result ofthe basin nature of the region (P�eeczely, 1983).
2.3 Air quality
Air quality is modified and influenced by theprevailing atmospheric conditions, which arecontrolled by the prevailing meteorological pa-rameters, mainly the temperature profile close toground level. The recorded averages for the cityof Szeged are the following: annual mean tem-perature: 11.2 �C; mean January and July tem-peratures: �1.2 �C and 22.4 �C, respectively;mean annual relative humidity: 71%; mean an-nual precipitation total: 573 mm; mean annualsunshine duration: 2102 hours; mean annual windspeed: 3.2 m s�1.
The city structure is very simple and is charac-terized by an intertwined network of boulevards,avenues and streets sectioned by the River Tisza.However, this simplicity also largely contributesto the concentration of traffic as well as air pol-lution within urban areas.
The industrial area is mainly restricted to thenorth-western part of the city. Thus, the prevailingwesterly and northerly winds tend to carry thepollutants from this area towards the city centre.
118 L. Makra et al
Though Szeged is considered to be a centre oflight industry in the southern part of the GreatHungarian Plain (wooden- and paprika process-ing industry, mill industry, hemp processing,salami manufacturing) in general, some elementsof heavy industry can also be found in the ruralareas (rubber and paint industry, oil and gasmining). After the collapse of the socialist polit-ical system in 1989, the industrial potential ofthe city has undergone a heavy freefall. The tex-tile, cable, clothing and canning factories wereall closed down; furthermore, the Light-weightConstruction Building and Supplying (K�EESZ)Ltd. moved its headquarters to another city,Kecskem�eet.
The total urban spread extends well beyondthe city limits and includes the largest oil fieldin Hungary with several oil torches located justnorth of the town. This oil field is also a signifi-cant source of such air pollutants as NOx andsulphur dioxide. The power station, located inthe western part of the city, is also a major sourceof pollution. Exhaust fumes have also largelycontributed to the steady increase of nitrogenoxide and carbon monoxide in Szeged. Besides,as a result of the heavy traffic, deposited dust isoften suspended in the atmosphere.
In a detailed analysis, Szeged was ranked 32ndof 88 Hungarian cities, according to the qualityof the environment and the level of environmen-tal awareness. [The city ranked 1st was consid-ered to have the best environmental conditions(Makra et al, 2002)].
On the basis of the frequency of pollutant con-centrations exceeding air quality limits, mea-sured at the Regional Immission Examining(RIE) network of stations for Hungary in 2001,the air quality of Szeged, according to a three-category classifying system (satisfactory, moder-ately polluted, polluted), can be listed in the‘‘polluted’’ category (Mohl et al, 2002).
Nitrogen oxides (NOx), ozone (O3) and par-ticulate matter (PM10) concentrations tend toexceed air quality limit values established by EUstandards. [Daily (24-hour) concentration of par-ticulate matter is 11–19 times higher, while itsannual concentration is twice as much as the EUstandard proposed since January 1, 2005!] Theinformation that Szeged was ranked 32nd of 88Hungarian cities gives the impression that this isa city with rather moderate air quality. Therefore,
the information that Szeged is ranked ‘‘polluted’’according to the RIE database seems surprising.In the above-mentioned analysis, the cities wereranked according to seven different categories(of nineteen environmental indicators), whichare as follows: water consumption (1), energyconsumption (3), public utilities supply (4), traf-fic (1), waste management (3), settlement ame-nities factors (4) and air quality (with averageconcentration of particulates deposited, sulphur-dioxide and nitrogen-dioxide). As air quality isonly one of the seven categories considered andis only represented by three parameters (environ-mental indicators), it contributes little to the rankof the cities.
The high concentrations of particulate matterare closely connected with the development ofrespiratory diseases. The annual trend of the airpollutant levels follow a unimodal distribution.Concentrations of NO, NO2 and PM10 are char-acterized by winter maxima and summer minima.At the same time, ozone reaches its maximumin the summer, in accordance with the annualchange of irradiation (Makra and Horvath, 2001;Makra et al, 2001a, b; Mohl et al, 2002; Mayeret al, 2004).
About 50% of the particulate matter comesfrom the northwest region of Szeged, coveredby shifting sands, loess and sandy ridges. At thesame time, the industrial district is located in thenorthwest region of the town. Hence, the domi-nant northwesterlies transport particulate matteras well as pollutants of industrial origin overSzeged. The remaining PM10 originates fromtraffic. Particulates are produced partly by theengine of vehicles, and partly by air currents gen-erated by vehicles (Mohl et al, 2002).
The traffic system of Szeged is highly over-crowded. Among vehicles participating in thetraffic, the ratio of passenger cars is the highest:84%. In the year 2000, after the modernization ofvehicles, CO concentrations in the city werereduced to 36–40% of that measured in 1990.On the other hand, the traffic on the main roadsincreased to 3–70% during the same period.During a regular day (a 24-hour period) about70,000–90,000 vehicles, on average, pass throughthe city (Mohl et al, 2002).
A serious environmental health problem arisesduring late summer and early autumn caused bythe pollen of ragweed with mugwort leaves or
Classification system of air mass types for Szeged, Hungary 119
short ragweed (Ambrosia artemisiifolia¼Ambrosia elatior), considered to be the most dan-gerous of all pollens. Annual pollen counts ofragweed in Szeged are one or two orders of mag-nitude higher than those in other European cities(Makra et al, 2004a, b).
The ‘‘double basin’’ situation of Szegedfavours the longer persistence of anticyclonicweather types both in summer and winter, whichcontributes to the enrichment of not only MajorAir Pollutants (including sulphur dioxide, nitrogendioxide, carbon monoxide, particles, lead andozone) but also pollens as biological agents[belonging to Hazardous Air Pollutants (HAPs)].The role and efficiency of large-scale weathersituations established in grouping pollutant con-centrations was the main motive behind thepresent study.
3. Data collection
3.1 The air pollution data
The air pollution monitoring station, mentionedabove, is located in downtown Szeged in a cross-road with heavy traffic (Kossuth Avenue andDamjanich Street), about 10 m distance fromthe Kossuth Avenue. This is one of the busiestcrossroads of Szeged. The monitoring stationwas put into operation in September 1, 1996. Ata distance of 10 m from the station there is atwo-storey building which affects wind andirradiance parameters. Sensors, measuring con-centrations of the air pollutants, are placed 3 mabove the surface.
Concentrations of CO, NO, NO2, SO2, O3 andTSP at the monitoring station are measured by5 different analyzers. The concentration of COis measured by non-dispersive infrared absorp-tion (type of the instrument: CO11M-LCD). Themeasurement of NO and NO2 concentrations isbased on the principle of chemiluminescence;the concentrations of NOx are obtained so thatthe instrument adds up latest NO and NO2
values automatically (type of the instrument:TE 42C). The measurement method of SO2 isUV fluorescence emission (type of the instru-ment: FHAF21M-LCD). O3 concentration mea-surements are based on the UV absorption ofwavelength of 254 nm (type of the instrument:TE 49C). TSP concentrations are recorded by
absorption of �-radiation (type of the instru-ment: FH 62 I-N).
Gas analyzers are calibrated in two ways. Oneof these is the 0-point, adjustment which occursautomatically every 24 hours. The other calibra-tion point is adjusted by a verified sample everytwo weeks. Measurements of TSP are verifiedonce in every quarter year. The instruments arecontrolled and the gathered data are stored on apersonal computer. From the 10-second measure-ments one-minute averages are derived; then,from the latter data, 30-minute averages arecalculated.
The type of instruments measuring meteorolog-ical parameters at the station are as follows:temperature: THS-611, humidity: THS-611, irra-diation: RS 81-I and wind speed: WS-12 Hþ.Temperature and humidity values are measured3 m above the surface, while wind direction, windspeed and irradiation are recorded at a height of6 m above the ground level.
3.2 The meteorological data
The data consist of a 30-minute data set fromthe five-year period between 1997–2001 for thewinter (December, January and February) andsummer (June, July and August) months. Theelements considered are the average mass con-centrations of the main air pollutants [CO, NO,NO2, SO2, O3 and TSP (mg m�3)]; and the dailyvalues of the main climatic parameters (tempera-ture, humidity, atmospheric pressure, global irra-diance and wind speed).
The 12 meteorological parameters used are:mean temperature (Tmean,
�C), maximum tempera-ture (Tmax, �C), minimum temperature (Tmin,
�C),daily temperature range (�T¼Tmax�Tmin,
�C),wind speed (WS, m s�1), relative humidity(RH, %), irradiance (I, MJ m�2 day�1), saturationvapour pressure (E, hPa), water vapour pressure(VP, hPa), potential evaporation (PE, mm), dewpoint temperature (Td, �C) and atmospheric pres-sure (P, hPa).
The 8 pollution parameters considered are theaverage diurnal mass concentrations of the fol-lowing pollutants: CO (mg m�3); NO (mg m�3),NO2 (mg m�3), SO2 (mg m�3), O3 (mg m�3) andTSP (mg m�3) as well as the daily ratios ofNO2=NO and the daily maximum concentrationsof O3 (mg m�3).
120 L. Makra et al
Daily sea-level pressure fields measured at 00UTC (Universal Time Centre) were acquired fromthe ECMWF (European Centre for Medium-Range Weather Forecasts) Re-Analysis ERA 40project, in the frame of which daily data havebeen re-analyzed since September 1st, 1957.The procedure has been performed with a uni-form method from the data being available inthe investigated period. Data for the ECMWFRe-Analysis ERA 40 project are verified, dynam-ically correct, the pressure field is true evenover the Atlantic Ocean and there are no missingdata. When using this method, the measured falseinput data are omitted. On the other hand, if orig-inal station data are used, false data can fre-quently be encountered.
The investigated area is the North-Atlantic –European region between 30� N–70.5� N lati-tudes and 30� W–45� E longitudes. The grid net-work is selected with a density of 1.5��1.5�,which indicates 28�51¼ 1428 grid points acrossthe region.
4. Methods
4.1 Cartographical background
For each objective type, average daily isobarmaps on the basis of daily sea-level pressure datacalculated at each grid point of the investigatedregion were constructed by applying the Surfer7.00 GIS software. Isobars for an average day,i.e., for an average objective type, were drawnusing 28� 51¼ 1428 grid data on the basis ofthe standard Kriging method without increasingelement number of data and with maximumsmoothing. As a result of the procedure, thecurved surface on the Earth as a spherical trape-zoid with 40.5� difference of latitude and 75�
difference of longitude was transformed into aplane rectangle with equal spacing both horizon-tally and vertically. Isobar maps produced in thisway can only be fitted to those informative back-ground maps which are prepared with the sameprojection. For this reason, the background mapof the investigated region was produced in anequidistant cylindrical projection. The majoradvantage of this map is that it is free from longi-tudinal distortion along each meridian and,therefore, the determination of the points of thecompass is simple at any location of the map.
Namely, the north–south and east–west directionsare parallel with the vertical and horizontal sidesof the rectangle; hence, geographical co-ordi-nates of various locations and air pressure forma-tions can easily be determined with the help ofthe spaces indicated on the rectangle both hori-zontally and vertically (and, if required, by ap-plying linear interpolation). The only drawbackof its use is that the background map becomeslonger at higher latitudes in the east–west direc-tion. The grid denotes a horizontal distance ofabout 107.3 km at the latitude of 50� N.
4.2 �2-test, independence analysis
In order to decide whether or not the sea levelpressure fields examined differ significantly fromeach other, the �2-test independence analysis wasapplied. This method determines whether tworandom variables (� and �) are independent.According to the 0-hypothesis, � and � are notindependent.
4.3 Factor analysis
In order to reduce the dimensionality of theabove collected meteorological data sets and thusto explain the relations among the 12 meteorolog-ical variables, the multivariate statistical methodof factor analysis is used. The main object offactor analysis is to describe the initial variablesX1, X2, . . ., Xp in terms of m linearly independentindices (m< p), the so called factors, measuringdifferent ‘‘dimensions’’ of the initial data set.Each variable X can be expressed as a linearfunction of the m factors, which are the maincontributors to the climate of Szeged:
Xi ¼Xm
j¼1
�ijFj; ð1Þ
where �ij are constants termed factor loadings.The square of �ij represents the part of the var-iance of Xi that is accounted for by the factor Fj.
One important stage of this method is the deci-sion for the number (m) of the retained factors.On this matter, many criteria have been pro-posed. In some studies, the Guttmann criterionor Rule 1 is used, which determines to keep thefactors with eigenvalues >1 and neglect thosethat do not account for at least the variance of
Classification system of air mass types for Szeged, Hungary 121
one standardised variable Xi. Perhaps the mostcommon method is to specify a least percentage(80% in this paper) of the total variance in theoriginal variables that has to be achieved(Jolliffe, 1993; Sindosi et al, 2003). Extractionwas performed by Principal Component Analysis(kth eigenvalue is the variance of the kth prin-cipal component). There is an infinite numberof equations alternative to Eq. (1). In order toselect the best or the most desirable factors, theso-called ‘‘factor rotation’’ is applied, a process,which either maximises or minimizes factorloadings for a better interpretation of the results.In this study, the ‘‘varimax’’ or ‘‘orthogonal fac-tor rotation’’ is applied, which ensures that thefactors remain uncorrelated (Jolliffe, 1990, 1993;Bartzokas and Metaxas, 1993, 1995; Sindosi et al,2003).
Factor analysis was applied to the data setconsisting of 12 columns (12 meteorologicalvariables) and 450 rows (450 days) both for thewinter and summer months, in order to reducethe 12 interrelated meteorological parametersand reveal the main independent meteorologicalfactors which are responsible for the formation ofthe weather in Szeged.
4.4 Cluster analysis
Cluster analysis is applied to the factor scorestime series in order to group objectively dayswith similar weather conditions. The aim of thismethod is to maximize the homogeneity ofobjects within the clusters and also to maximizethe heterogeneity between the clusters. Eachobservation (day) corresponds to a point in them-dimensional space and each cluster consists ofthose observations, which are ‘‘close’’ to eachother in this space. The characterization of a dis-tance between two observations k and l as‘‘close’’ or ‘‘far’’ is determined by the squareof their Euclidean distance:
D2kl ¼
Xm
i¼1
ðxki � xliÞ2; ð2Þ
where xki is the value of the i-th factor for the k-thday and xli is the value of the i-th factor for thel-th day.
There are two main clustering techniques: hier-archical and non-hierarchical. Their basic differ-
ence is that in the non-hierarchical technique thenumber of clusters must be known a priori. Incontrast, in the hierarchical method the ultimatenumber of clusters is determined by a variety ofstatistical criteria. In this paper, the hierarchicaltechnique is applied because of the lack of anobjective classification of weather types for theSzeged region. The hierarchical technique canbe applied by using various methods. Here, the‘‘average linkage’’ method is used, since it doesnot depend on extreme values, besides it producesmore realistic groupings and properly combinesextreme weather days into distinct meteorologicalunits (Anderberg, 1973; Kalkstein et al, 1987;Hair et al, 1998; Sindosi et al, 2003).
Then, for each of the derived clusters of days,the mean value for every meteorological and pol-lution parameter is computed. In this way, therelationships between weather conditions andthe corresponding concentration levels of air pol-lutants are revealed. Finally, for each weathertype, the composite maps of the mean sea levelpressure distribution over the North-Atlantic –European region (00 UTC) are constructed. Theaim of these maps is to associate atmosphericcirculation patterns and pollution levels in theSzeged region. The classification of synoptic pat-terns into distinct groups enables us to describethe most important synoptic types consideringthe position of the Szeged region.
4.5 ANOVA and Tukey’s honestly significantdifference test
When determining the synoptic types, only mete-orological parameters are taken into account,excluding pollution data. Hence, the differencesof the mean pollution levels calculated for eachsynoptic type need further statistical evaluation.This is performed by the method of one-way Anal-ysis of Variance (ANOVA) for each pollutant. Byusing this method, significant differences in pol-lutant concentrations of different synoptic types(clusters) can be determined. Finally, the Tukey’shonestly significant difference test is applied inorder to quantitatively compare the mean air pol-lution levels between each pair of synoptic type(pairwise multiple comparisons) (McGregor andBamzelis, 1995; Sindosi et al, 2003).
All statistical computations were performedwith SPSS (version 9.0) software.
122 L. Makra et al
5. Results
5.1 Winter months
The application of factor analysis to the timeseries of the meteorological elements yielded 4factors explaining 86.51% of the total variance(Table 1). Table 2 displays the factor loadingsafter orthogonal rotation.
Factor 1 explains 50.86% of the total variance(Table 1) and includes the three main air tempera-ture variables (mean, maximum and minimumtemperatures) and three important humidity pa-rameters (saturation vapour pressure, water vapourpressure and dew point temperature). It can be
seen that the temperature variables are notdirectly related to irradiance, which during thewinter depends on the third factor. The reasonis that winter air temperature is mainly influ-enced by the synoptic-scale air mass affectingthe area and, to a lesser extent, by the local irra-diance. The high loadings of these temperatureand humidity parameters show their strong rela-tionship. Namely, the high loading of the watervapor pressure is due to the increase in vaporcapacity of the atmosphere as temperature rises.Dew point temperature covaries with the aboveparameters, since an increase (decrease) in watervapour pressure is explained by higher (lower)temperature at which air becomes saturated (dewpoint) (Table 2).Factor 2 (19.85% of the total variance; Table 1)includes only relative humidity with a negativesign and potential evaporation. The high loadingsof opposite sign indicate an inverse relationshipbetween the two variables. Namely, high (low) po-tential evaporation is associated with low (high)relative humidity (Table 2).Factor 3 explains 8.72% of the total variance(Table 1) and comprises the daily temperaturerange and irradiance. High irradiance values,which indicate low cloud cover, generally causehigh Tmax. Whenever clear skies persist also in thefollowing night, nocturnal long wave radiationleads to larger cooling of the surface and to lowerTmin; thus, Trange is generally larger under theseconditions (Table 2). (This is only true, if noadvection of, e.g., cooler air masses takes place.)
Table 1. Initial eigenvalues and cumulative variances,winter months (December, January and February)
Component Initial eigenvalues
Totalvariance
Relativevariance, %
Cumulativevariance, %
1 6.10 50.86 50.862 2.38 19.85 70.703 1.05 8.72 79.424 0.85 7.08 86.515 0.75 6.29 92.796 0.46 3.82 96.617 0.34 2.86 99.478 0.04 0.32 99.799 0.02 0.19 99.97
10 0.00 0.02 100.0011 0.00 0.00 100.0012 0.00 0.00 100.00
Table 2. Factor loadings for the winter months (December, January and February). Values higher than j0:60j are onlypresented
Meteorological parameters Factor 1 Factor 2 Factor 3 Factor 4
Mean temperature, Tmean 0.94Maximum temperature, Tmax 0.81Minimum temperature, Tmin 0.84Daily temperature range, �T¼Tmax�Tmin 0.87Wind speed, WSRelative humidity, RH �0.89Irradiance, I 0.74Saturation vapor pressure, E 0.93Water vapor pressure, VP 0.97Potential evaporation, PE 0.74Dew point temperature, Td 0.97Atmospheric pressure, P 0.96
Values higher than j0:10j are statistically significant at the 95% level; however, Table 2 shows only those exceeding j0:60j. Thismeans that at least 36% of the total variance of a parameter can be explained by a single factor
Classification system of air mass types for Szeged, Hungary 123
Factor 4 is slightly weaker than Factor 3 andexplains 7.08% of the total variance (Table 1).It comprises atmospheric pressure only (Table 2).
Next, cluster analysis was applied to the fourfactor score time series and, as a result of this,six homogenous clusters of days were revealed.However, one cluster included only 4 days(January 5, 10, 18, 19, all in year 2001; a mere0.89% of the total number of days). These dayswere related to extreme weather conditions withincreased pollutant concentrations, due to ananticyclone. This cluster was therefore omittedfrom further consideration, since our aim wasto analyze days with dominant air mass types.The main characteristics of the other five domi-nant clusters involving the prevailing air masstypes are shown in Table 3, which presents themean values of their meteorological parametersas well as the mean values of the correspondingpollution parameters.
Considering the basic statistical parametersof the pollutants, variation coefficients (standarddeviation expressed in the unit of the average)for NO and SO2 are twice as high as those
for other contaminants examined, which denotestheir higher variability. The difference ofjmedian � averagej remains within the so calledinterquartile half extent (the interval given by thelower quartile and the upper quartile) for eachpollutant. The greatest differences are detectedfor CO, NO, NO2=NO and O3.
The mean sea level pressure distribution be-longing to the clusters examined and the numberof days in each cluster (air mass type) are shownin Fig. 2.
Mean sea level pressure fields for the 5 clustersreceived in the winter were compared on thebasis of the used grid values. In order to decidewhether the mean sea level pressure fields of the5 clusters in the winter differ significantly fromeach other in the period examined, the �2-testwas applied. The 0-hypothesis means that thereis no significant difference between the mean sealevel pressure fields of the clusters compared. Onthe basis of our computations, probability of the0-hypothesis in the winter months between thesea level pressure fields of clusters 1–2 and 3–4is 1. Namely, in these cases the mean cluster
Table 3. Mean values of the meteorological and pollution parameters for the days belonging to the five dominant clusters of thewinter months (December, January and February)
Clusters 1 2 3 4 5
Number of cases (days) 56 134 73 94 90Freqency (%) 12.4 29.7 16.2 20.8 20.0
Tmean (�C) �4.9 0.24 5.5 2.4 6.0Tmax (�C) �1.0 2.1 9.6 7.6 8.1Tmin (�C) �8.1 �1.6 0.9 �1.4 2.5�T¼Tmax�Tmin, (�C) 7.0 3.7 8.7 9.0 5.6WS (m s�1) 0.3 0.6 1.0 0.4 0.9RH (%) 81.5 83.5 63.6 80.6 84.3I (MJ m�2) 5.4 2.7 6.2 5.4 2.7E (hPa) 5.9 8.5 12.8 10.1 12.8VP (hPa) 4.8 7.1 8.1 8.1 10.8PE (mm) 0.6 0.8 2.0 1.0 1.1Td (�C) �7.4 �2.0 �0.7 �0.4 3.8P (hPa) 1009.2 1004.0 986.2 1001.2 994.6
CO (mg m�3) 0.878 0.799 0.565 0.933 0.790NO (mg m�3) 24.3 20.9 24.1 44.0 28.5NO2 (mg m�3) 42.0 34.4 41.0 47.2 39.5NO2=NO 3.0 6.1 2.5 1.8 3.3O3 (mg m�3) 27.2 23.3 33.4 23.1 20.2O3 max (mg m�3) 50.7 39.2 59.5 46.5 39.0SO2 (mg m�3) 15.8 10.7 11.6 11.7 9.9TSP (mg m�3) 61.9 49.1 50.1 61.4 44.7
1
Fig. 2. Mean sea-level pressure fields for each air mass type (cluster), and monthly variation of the number of days, North-Atlantic – European region, winter months (December, January and February)
124 L. Makra et al: Classification system of air mass types for Szeged, Hungary
fields mentioned can not be considered indepen-dent. On the other hand, the probability of the0-hypothesis for all the other cluster pairs is 0;namely, the mean cluster fields compared areconsidered to be independent (Table 4).
The five air mass types and the correspondingpressure patterns with the associated pollutionlevels are described as follows.
Cluster 1: This can be named ‘‘anticyclone overthe Carpathian Basin’’. This pressure pattern ischaracterized by a high pressure system overCentral Europe. This air mass type accounts for12.5% of the total number of days and is asso-ciated with the following weather characteristicsin Szeged: high amounts of irradiance (meanvalue¼ 5.4 MJ m�2), the lowest values of tem-perature parameters (mean daily as well as max-imum and minimum temperatures), the lowestvalues of humidity parameters (water vapourpressure, saturation vapour pressure, potentialevaporation and dew point temperature) andvery low wind speed (0.3 m s�1). During suchweather conditions primary pollutants (CO,NO2, SO2 and TSP, except NO) are highly con-centrated in the city, as a consequence of poorventilation and temperature inversions formedduring the night (Horvath et al, 2002). Underthe prevalence of this air mass type, in accor-dance with the low cloudiness, concentration ofthe secondary pollutants (O3 and O3 max) is rela-tively high (Fig. 2; Table 3).
Cluster 2: This type is named ‘‘anticyclone overthe Mediterranean’’. This cluster contains 30.0%of the total number of days and represents themost frequent situation. This pressure patternforms an anticyclonic ridge over the CarpathianBasin with calm or weak breezes. Cloudy condi-tions, on average higher temperatures due to less
nocturnal cooling are characteristic during thisair mass type. The temperature parameters aresignificantly higher during this air mass type,compared to those in Cluster 1. The ozone levelsare lower because of the higher cloudiness. Thereason of the lower concentration in primary pol-lutants might be the higher wind speed (Fig. 2;Table 3).
Cluster 3: Anticyclone stretches out from theregion of the Azores. This situation is only char-acteristic in February. During prevalence of thistype an anticyclone can reach Central and evenEastern Europe resulting in calm, sunny weather.This situation causes high temperatures and highwinds. Lower concentrations of CO, SO2 andTSP in this cluster compared to those in Cluster 1can be explained with the highest average windspeed in Cluster 3. Since average concentrationsof NO in Clusters 1 and 3 are the same, higherenrichment of the ozone in Cluster 3 relating tothat in Cluster 1 can be explained with the lowercloudiness (This is only true if advection isneglected.) (Fig. 2; Table 3).
Cluster 4: Anticyclone over Southern Europe andNorth Africa. This cluster does not differ signifi-cantly from Cluster 3. Here, the high-pressuresystem from SW Europe extends over the East-ern part of the Mediterranean. Because of thevery low wind speeds, concentrations of primarypollutants (except SO2) are extremely high(CO¼ 0.93 mg m�3; NO¼ 44.0mg m�3; NO2¼47.2mg m�3; TSP¼ 61.4 mg m�3). At the sametime, irradiance is also high. However, concen-tration of ozone is not high due to the highestlevels of NO, implying the lowest levels ofNO2=NO, which inhibit the formation of O3 viathe destruction process: NOþO3�!NO2þO2
(Fig. 2; Table 3).
Cluster 5: Intense zonal current over Europe.This air mass type amounts to 20.1% of thetotal number of days and is mostly frequent inDecember. During this situation strong windsare observed at Szeged. The pressure pattern cor-responds to a zonal current over the CarpathianBasin, which involves fairly low levels of the pri-mary pollutants especially those of SO2 and TSPwith their lowest concentrations. On the otherhand, the highest cloudiness (I¼ 2.7 MJ m�2, asin Cluster 2) with a medium level of NO results
Table 4. �2-test, independence analysis of the average sea-level pressure fields for the five clusters derived for thewinter months (December, January and February), probabil-ity of the 0-hypothesis
Cluster 1 2 3 4 5
1 – 1.0000 0.0000 0.0000 0.00002 – 0.0000 0.0000 0.00003 – 1.0000 0.00004 – 0.00005 –
126 L. Makra et al
in the lowest concentration of the ozone param-eters (O3¼ 20.2 mg m�3; O3 max¼ 39.0mg m�3)(Fig. 2; Table 3).
In order to determine the influence of air masstype on pollutant levels, analyses of variance(ANOVA) were performed on the pollutantparameters. The results are shown in Table 5a.It can be observed that, with the exceptionof NO2=NO, all pollutants present significantinter – air mass type differences in mean con-centration values at the 99% probability level.Considering that differences are found amongthe mean pollutant levels, Tukey’s tests wereapplied in order to receive a pairwise multipleassessment of the differences. The statisticallysignificant differences are shown in Table 5bat the 95% and 99% probability levels, respec-
tively. It can be seen that the pairs of air masstypes (clusters) 3–4 differ significantly for fivepollutants, while the types 1–2, 1–5 and 2–3differ substantially for four pollutants. Gener-ally, clusters 3–4 can be considered to be themost different, since levels of most pollutantpairs show substantial difference between them.This can mainly be explained by the fact thatthese two types show nearly the highest differ-ence in wind speed. On the other hand, Cluster 2seems to be an intermediate cluster with respectto pollution, since it shows fewer pairwise dif-ferences than the others. An exception to this isNO2 with 3 out of 4 pairs of Cluster 2 beingdifferent. The pairwise multiple comparisons be-tween Clusters 2 and 5 did not detect significantdifference in any pollutant.
Table 5a. ANOVA statistics for inter – air mass comparison of pollutant concentrations in the winter months (December,January and February)
CO NO NO2 NO2=NO O3 O3 max SO2 TSP
Mean squarebetween groups
1516531.41 8183.16 2361.19 305.70 2057.77 6255.12 332.56 4971.82
Mean squarewithin groups
137957.12 585.10 257.83 212.15 186.97 464.40 65.23 534.98
F-Ratio 10.99 13.99 9.16 1.44 11.01 13.47 5.10 9.29Levelof significance, %
99 99 99 78 99 99 99 99
Table 5b. Air mass type – pollution difference matrix. Each matrix cell represents the comparison between two air mass types.Pollutants appearing in the matrix cells indicate significant inter – air mass difference in concentrations according to Tukey’ssignificant difference test (light-faced characters: 95% of significance, bold characters: 99% of significance), winter months(December, January and February)
Classification system of air mass types for Szeged, Hungary 127
5.2 Summer months
The application of factor analysis to the timeseries of the meteorological parameters resultedin 4 main factors explaining 84.36% of the totalvariance (Table 6). Table 7 shows factor loadingsof the summer months after orthogonal rotation.
Factor 1, with 47.35% of the total variance(Table 6), includes the same parameters as inthe winter case. These are temperature (mean,maximum and minimum temperatures) and hu-midity variables (saturation vapour pressure, watervapour pressure and dew point temperature).They have the same positive sign, which denotesthat higher (lower) temperature values involvehigher (lower) values of the humidity parameters.
While this factor was classified in the same wayas in the winter months, the remaining factorshave high loadings with completely differentmeteorological variables from their winter ver-sions (Table 7).Factor 2 (19.44% of the total variance; Table 6)comprises irradiance and potential evaporation,both with positive sign and relative humiditywith negative sign. Increasing irradiance involvesan increase in potential evaporation and a paralleldecrease of relative humidity (Table 7).Factor 3 (8.86% of the total variance; Table 6)consists of only atmospheric pressure (Table 7).Factor 4 (8.22% of the total variance; Table 6) isslightly weaker than Factor 3 and contains onlywind speed (Table 7).
In the following, cluster analysis was appliedto the four-factor score time series. The analysisrevealed ten clusters of days (air mass types),each with at least 3.7% of the total number ofdays. The summer season is characterized by agreater number of air mass types compared to thewinter period. During the whole summer only twomain pressure systems influence the CarpathianBasin: the Icelandic low from NW Europe and thesubtropical high from the region of the Azores.Hence, the differences among these dominanttypes, considering both the mean values of theparameters and the spatial pressure distribution,are rather small. The ten air mass types (clusters)with mean values of both meteorological and airpollution parameters are shown in Table 8.
Basic statistical parameters of the pollutantsare calculated for the summer months. Variation
Table 6. Initial eigenvalues and cumulative variances,summer months (June, July and August)
Component Initial eigenvalues
Totalvariance
Relativevariance, %
Cumulativevariance, %
1 5.68 47.35 47.352 2.39 19.94 67.293 1.06 8.86 76.154 0.99 8.22 84.365 0.89 7.38 91.746 0.59 4.92 96.657 0.37 3.07 99.728 0.02 0.17 99.899 0.01 0.10 99.99
10 0.00 0.01 100.0011 0.00 0.00 100.0012 0.00 0.00 100.00
Table 7. Factor loadings for the summer months (June, July and August). Values higher than j0:60j are only presented
Meteorological parameters Factor 1 Factor 2 Factor 3 Factor 4
Mean temperature, Tmean 0.88Maximum temperature, Tmax 0.75Minimum temperature, Tmin 0.82Daily temperature range, �T¼Tmax�Tmin
Wind speed, WS 0.98Relative humidity, RH �0.94Irradiance, I 0.71Saturation vapor pressure, E 0.87Water vapor pressure, VP 0.95Potential evaporation, PE 0.80Dew point temperature, Td 0.95Atmospheric pressure, P 0.92
Values higher than j0:10j are statistically significant at the 95% level; however, Table 7 shows only those exceeding j0:60j. Thismeans that at least 36% of the total variance of a parameter can be explained by a single factor
128 L. Makra et al
Table
8.
Mea
nval
ues
of
the
met
eoro
log
ical
and
po
llu
tio
np
aram
eter
sfo
rth
ed
ays
bel
on
gin
gto
the
ten
do
min
ant
clu
ster
so
fth
esu
mm
erm
on
ths
(Ju
ne,
July
and
Au
gu
st)
Clu
ster
s1
23
45
67
89
10
Nu
mb
ero
fca
ses
(day
s)2
84
05
87
61
74
67
24
75
12
5F
req
uen
cy(%
)6
.18
.71
2.6
16
.53
.71
0.0
15
.71
0.2
11
.10
.1
Tm
ean
(�C
)1
7.7
20
.21
9.7
22
.82
1.0
26
.12
5.9
20
.82
0.0
28
.1T
max
(�C
)2
3.9
26
.22
5.4
28
.32
9.6
32
.73
1.8
25
.22
5.1
34
.7T
min
(�C
)1
2.5
14
.91
6.9
17
.01
4.0
18
.62
0.1
16
.51
4.9
21
.2�
T¼
Tm
ax�
Tm
in,
(�C
)1
1.3
11
.38
.51
1.3
15
.61
4.1
11
.78
.71
0.2
13
.6W
S(m
s�1)
0.8
0.6
0.9
1.0
1.0
0.6
0.8
1.9
1.3
1.5
RH
(%)
59
.86
8.4
82
.26
7.8
26
6.1
53
.56
9.0
74
.06
4.9
55
.3I
(MJ
m�
2)
18
.91
8.6
15
.62
3.5
18
.42
4.5
23
.51
8.4
23
.42
6.6
E(h
Pa)
27
.73
2.4
31
.63
8.3
34
.84
6.9
46
.43
4.1
32
.35
2.7
VP
(hP
a)1
6.5
22
.02
6.0
25
.92
3.1
25
.13
1.9
24
.92
0.7
28
.9P
E(m
m)
4.4
4.4
2.8
5.2
4.9
8.1
6.1
4.0
4.7
8.9
Td
(�C
)1
0.0
14
.41
6.9
16
.91
4.7
16
.42
0.2
16
.21
3.5
18
.7P
(hP
a)1
01
8.0
10
16
.11
01
3.4
10
15
.01
01
5.7
10
17
.41
01
5.6
10
13
.71
01
3.8
10
15
.7
CO
(mg
m�
3)
0.2
39
0.3
61
0.3
96
0.3
48
0.3
23
0.4
40
0.4
43
0.2
49
0.2
41
0.4
84
NO
(mg
m�
3)
6.2
10
.07
.16
.48
.79
.96
.73
.55
.67
.9N
O2
(mg
m�
3)
20
.13
1.4
26
.62
5.4
28
.53
3.9
27
.71
7.5
22
.53
3.2
NO
2=
NO
6.1
7.4
9.8
9.0
8.0
10
.59
.61
4.4
27
.56
.4O
3(m
gm
�3)
54
.75
5.4
53
.85
6.1
58
.96
4.2
58
.95
7.4
60
.47
7.5
O3
max
(mg
m�
3)
89
.71
03
.49
9.8
98
.51
13
.71
14
.71
08
.29
2.7
98
.81
29
.8S
O2
(mg
m�
3)
5.0
4.8
2.9
4.3
5.6
5.3
4.6
3.6
4.3
5.0
TS
P(m
gm
�3)
25
.13
5.5
34
.23
8.2
38
.04
7.3
46
.73
1.3
32
.65
3.4
Classification system of air mass types for Szeged, Hungary 129
coefficients for O3 and O3 max decreased substan-tially compared to their values in the wintermonths. The difference of jmedian � averagej forNO2=NO is found beyond the interquartile halfextent in clusters 2, 3, 4, 5, 7 and 9. Furthermore,the averages for TSP and SO2 are also detectedbeyond this interval in Cluster 9. This indicatesthat distribution functions of the pollutant con-centrations in the clusters mentioned above aredistorted; namely, the means of the samples arenot representative of the data sets.
The mean sea-level pressure distribution overthe North Atlantic – European region and thevariation of the number of days within the sum-mer season are presented in Fig. 3a, b.
In order to decide whether the mean sea-levelpressure fields of the 10 clusters in the summerdiffer significantly from each other, the �2-testwas applied. According to our results, only themean sea-level pressure fields of clusters 5–9 canbe considered independent. For all the other clus-ter pairs independence is not realized (Table 9).
The pressure patterns and the correspondingpollution levels in Szeged are discussed below.
Cluster 1: Comprises 6.1% of the total number ofdays. It is characterized by a high pressure systemextending over Europe except Scandinavia andincludes the Carpathian Basin. At the same timethe thermal low of SW Asia is also developed.In this situation, air temperature is the lowest ofall the ten clusters. The reason is that most ofthese days belong to June. Therefore, the primary(CO, NO, NO2, NO2 vs NO and TSP except SO2)and the secondary pollutants (O3 and O3 max) havethe lowest concentrations (Fig. 3a; Table 8).
Cluster 2: Early summer. In this weather type(8.7% of the total number of days) the abovepressure pattern is less characteristic, since boththe high and the low pressures weaken. Windspeed is the lowest of the all clusters. The con-centration of pollutants increases, apart fromSO2, while the levels of NO reach highest values(Fig. 3a; Table 8).
Cluster 3: Typical summer, with 12.6% of thetotal number of days. Values of the meteorologi-cal elements represent a typical summer day in
Szeged. During this type the high pressure sys-tem from Azores slightly withdraws, while thethermal low of SW Asia develops comparing tothe pressure pattern in Cluster 2. During this airmass type, the level of CO increases, while theconcentration of SO2 falls (Fig. 3a; Table 8).
Cluster 4: This is the most frequent type with16.5% of the total number of days and is char-acteristic of each summer month. Its pressurepattern is very similar to that of Cluster 3. Theonly basic difference is that the extensive lowpressure system in Northern Europe is absent inthis cluster. The levels of CO decrease, while asubstantial decrease in cloudiness causes a slightincrease in O3 concentration. (NO levels prac-tically do not change compared to those inCluster 3.) (Fig. 3a; Table 8). The fact that asubstantial decrease in cloudiness causes only aslight increase in O3 concentration, might beexplained by the change of transport processes.Namely, a decrease in cloudiness might be theresult of change in circulation, which transportsless ozone over Szeged. On the other hand, thesmaller ozone concentration is only partly com-pensated by the photochemical processes acceler-ated by increased irradiance. Long-range transportcan also influence the local ozone concentrationand, in this way, the rate of local ozone formationdepending little on local irradiance.
Cluster 5: Typical early summer with the lowestvalue of the total number of days (3.7%). The highpressure system from Azores develops and extendsover Eastern Europe avoiding the CarpathianBasin and, at the same time, a low pressure centredeepens over the North Atlantic. The daily tem-perature range is high, with cloudy conditions andmoderate winds. There are not substantial differ-ences in the concentration of pollutants comparedto those of Cluster 4 (Fig. 3a; Table 8).
Cluster 6: Typical late summer (10.0% of thetotal number of days). The high pressure systemfrom Azores extends over Eastern Europe and in-cludes the Carpathian Basin. There are no weatherfronts in Northern Europe. Irradiance is very high,which involves high values of the temperaturevariables. On the other hand, wind speed is
1
Fig. 3a, b. Mean sea-level pressure fields for each air mass type (cluster), and monthly variation of the number of days,North-Atlantic – European region, summer months (June, July and August)
130 L. Makra et al: Classification system of air mass types for Szeged, Hungary
132 L. Makra et al
low. Hence, primary pollutants are highlyenriched. Though both irradiance and NO con-centration (having opposite effect on the levels ofO3 and O3 max) are higher than in Cluster 5,higher weight of irradiance is indicated by result-ing in a slight increase of the secondary pollu-tants (Fig. 3b; Table 8).
Cluster 7: This is the second most frequent typewith 15.6% of the total number of days. The highpressure centre from Azores is not present and, atthe same time, a low pressure pattern deepensover Northern Europe indicating a more charac-teristic pressure system than that of Cluster 6.However, both weather and pollutant levels prac-tically do not change when compared to the for-mer cluster (Fig. 3b; Table 8).
Cluster 8: This type occurs with the same fre-quency in each summer month (10.2% of the totalnumber of days). The high pressure centre fromAzores strengthens and extends over centralEurope, while the low pressure pattern in North-ern Europe is divided into two parts: the Icelandiclow and the Baltic low. The Carpathian Basin isunder the influence of the Baltic low. Hence,cloudiness increases which involves a decreasein the temperature parameters and wind speedreaches its maximum of all clusters. This is why
both the primary and the secondary pollutantlevels are so low (Fig. 3b; Table 8).
Cluster 9: Consists of 11.1% of the total numberof days. The location of the high pressure centrefrom Azores does not change, while Northernand Eastern Europe is covered by an extremelylarge and uniform low pressure pattern. TheCarpathian Basin lies at the edge of the highpressure system. As the weather situation betweenClusters 8 and 9 is extremely similar, there areonly minor differences in the meteorological pa-rameters. Hence, no significant differences occurin levels of the pollutants (Fig. 3b; Table 8).
Cluster 10: Typical and late summer with 5.4%of the total number of days. The high pressurecentre from the Azores weakens and graduallydisappears. On the other hand, the low pressurepattern over Ukraine and Romania, presented inCluster 9, disappears and a small high pressurecentre forms in its place, whilst an extended lowpressure centre develops over Northern Europe.The Carpathian Basin lies between the two highpressure centres ensuring undisturbed irradiancewith very high temperatures and fairly moderatewinds. This air mass type results in the highestlevels of both primary and secondary pollutants,except SO2 (Fig. 3b; Table 8).
Table 9. �2-test of the mean sea-level pressure fields for the ten clusters of the summer months (June, July and August),probability of the 0-hypothesis
Cluster 1 2 3 4 5 6 7 8 9 10
1 – 1.0000 1.0000 1.0000 1.0000 0.9998 1.0000 1.0000 1.0000 0.98062 – 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.00003 – 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.00004 – 1.0000 1.0000 1.0000 1.0000 1.0000 1.00005 – 1.0000 1.0000 1.0000 0.0113 1.00006 – 1.0000 1.0000 1.0000 1.00007 – 1.0000 1.0000 1.00008 – 1.0000 1.00009 – 1.0000
10 –
Table 10a. ANOVA statistics for inter – air mass comparison of pollutant concentrations in the summer months (June, July andAugust)
CO NO NO2 NO2=NO O3 O3 max SO2 TSP
Mean square between groups 332509.51 174.27 1178.53 1873.59 1465.91 4555.88 26.59 2732.57Mean square within groups 21776.86 37.17 125.86 942.20 253.81 694.29 11.63 134.77F-Ratio 15.27 4.69 9.36 1.99 5.78 6.56 2.28 20.28Level of significance, % 99 99 99 96 99 99 98 99
Classification system of air mass types for Szeged, Hungary 133
Similarly to the winter months, the signifi-cance of inter – air mass type differences in pol-lutant levels was determined by ANOVA. Theresults are shown in Table 10a. Mean concentra-tions of CO, NO, NO2, O3, O3 max and TSP pres-ent significant inter – air mass type differencesat the 99% probability level, while SO2 at the98% and NO2=NO at the 96% levels, respec-tively. Performing the pairwise comparisons(Tukey’s significant difference tests), the statisti-cally significant differences are shown in Table10b at the 95% and 99% probability levels,respectively. There are no two air mass typesfor which inter – air mass type differences inconcentrations of all the eight pollutants consid-ered are significant. The highest inter – air masstype difference is indicated by five pollutants forthe following comparisons: types 1–10, 6–8, 8–10 and 9–10. Air mass types 1–8, 1–9, 2–3, 2–4,
2–5, 3–4, 3–5, 4–5, 5–6, 5–7, 5–9, 6–7 and 8–9are the most similar, considering that no signifi-cant differences in levels of any pollutants can bedetected between them. The pairwise multiplecomparisons indicated significant differences inconcentrations of at least four pollutants for thefollowing cluster pairs: types 1–6 (CO, NO2,O3 max and TSP); types 1–10 (CO, NO2, O3,O3 max, and TSP); types 2–10 (CO, O3, O3 max
and TSP); types 3–6 (NO2, O3, SO2 and TSP);types 3–10 (CO, O3, O3 max and TSP); types 4–6(CO, NO2, O3 max and TSP); types 4–10 (CO, O3,O3 max and TSP); types 6–8 (CO, NO, NO2,O3 max, and TSP); types 6–9 (CO, NO, NO2 andTSP); types 8–10 (CO, NO2, O3, O3 max and TSP)and types 9–10 (CO, NO2, O3, O3 max and TSP).In general, air mass type 6 and type 10 differmostly from the others, since the pairwise multi-ple comparisons detected significant differences
Table 10b. Air mass type – pollution difference matrix. Each matrix cell represents the comparison between two air mass types.Pollutants appearing in the matrix cells indicate significant inter – air mass difference in concentrations according to Tukey’shonestly significant difference test (light-faced characters: 95% of significance, bold characters: 99% of significance), summermonths (June, July and August)
134 L. Makra et al
for them in levels of the most pollutants. Thereason for this might be the fact that these twotypes show a considerable difference in windspeed. At the same time, type 5 seems to be anintermediate cluster, since it shows fewer pair-wise differences than the others.
6. Discussion
In order to assess the effect of different air masstypes on air pollutant levels in Szeged, objectivemultivariate statistical methods were applied tometeorological and air pollution data. By defin-ing objective pressure patterns over the North-Atlantic – European region, air mass types dom-inating the Carpathian Basin were determined.
The procedure itself has been applied in theliterature (Sindosi et al, 2003); however, this isa new approach for classifying air mass typesin the region examined, since for Hungary onlythe subjective categorization of the prevailingpressure patterns made by P�eeczely (1957, 1983)is known. The basis of his classification is thesame as with the objective categorization: dailysea-level pressure fields measured at 00 UTC.P�eeczely determined 13 large-scale weather sit-uations for the Carpathian Basin. As regardsthe winter months, four groups of the P�eeczely’smacrotypes are characteristic over the CarpathianBasin: (1) types connected with southerly cur-rents, (2) an anticyclone extending from the west,(3) an anticyclone in the north from Hungary and(4) an anticyclone over the Carpathian Basin.These weather types account for more than 70%of the total number of days in this season. On theother hand, the five objective clusters detected inthis paper for winter are basically characterizedby zonal currents (87.5% of the total number ofdays). In more detail they are as follows: an an-ticyclone south of Hungary (Clusters 2 and 4), ananticyclone extending from the west (Cluster 3)and a zonal cyclonic type (Cluster 5). Thesetypes are completed with an anticyclonic clusterover the Carpathian Basin (Cluster 1) (12.5% ofthe total number of days). Concerning the sum-mer months, four P�eeczely types are emphasized:(1) Hungary lies in the rear part of an East-European cyclone, (2) an anticyclone extendingfrom the west, (3) an anticyclone north ofHungary and (4) an anticyclonic type over theCarpathian Basin. These air mass types comprise
a total of more than 60% of the total number ofdays. At the same time, the ten objective clustersfor summer are determined mainly by the follow-ing groups: an anticyclone extending from thewest (Clusters 2–5, 7–9) (zonal currents), an an-ticyclone over the Carpathian Basin (Clusters 1and 6) and an anticyclone east of Hungary (Clus-ter 10) (meridional currents). Predominance ofthe anticyclonic and anticyclonic ridge situationsin the summer is very clear both in the P�eeczelytypes and in the objective clusters, too.
The air mass types determined for the winterand summer months were also related to levels ofair pollutants in Szeged downtown. Relation ofthe objectively determined air mass types and airquality of Szeged detected that pollution levelscan be connected to different pressure patternsruling the region examined. Hence, with knowl-edge based on weather forecasts, expected pollu-tion levels can be indicated, which contributes tothe abatement of severe air pollution episodes.However, it has to be underlined that atmosphericcirculation is not the only factor controlling theair pollution in Szeged. The revealed pressurepatterns can only influence the concentration ofthe pollutants, which in their vast majority are ofhuman origin. Thus, for a precise air pollutionforecast, apart from a good weather forecast, agood knowledge of human activities is necessary.For example, days with traffic peaks, days ofvacation or holidays must also be consideredbefore certain restrictions on the emissions areimposed. Finally, another factor, which cannotbe ignored either, is weather persistence. It mustbe kept in mind that the presence of pressurepatterns favouring high concentrations of pollut-ants for a long period of time may cause evenworse air quality conditions.
7. Conclusions
The paper analyzed the levels of air pollutants inSzeged recorded under characteristic sea levelpressure patterns over the Carpathian Basin. Spe-cific air mass types given by the pressure patternsfor both the winter and summer months werefound to play a significant role in the concentra-tion of pollutants in downtown Szeged. Resultsfor the winter months revealed that primary pol-lutants appear in higher concentration when bothcloudiness and wind speed are low (air mass
Classification system of air mass types for Szeged, Hungary 135
types 1 and 4; Fig. 2, Table 3). This is the casewhen an anticyclone is found over the CarpathianBasin (Cluster 1) and when an anticyclone rulesthe region south of Hungary influencing theweather of the country (Cluster 4). Low concen-trations of primary pollutants are detected whenHungary lies under the influence of zonal cur-rents (wind speed is the highest) (Cluster 3, atransitional type and Cluster 5). Pressure patternsin the summer months are not as easily groupedinto clusters as those in the winter, concerningthe variability of the pressure fields and the mag-nitude of the gradients. This is mainly due to thepredominance of the anticyclonic conditions andanticyclonic ridge types. Due to low cloudinessand very low NO concentrations, rather highlevels of secondary pollutants are observed. Itis to be noted that O3 records exhibit about dou-ble concentrations than in the winter months.
Prediction of air mass types favours to preventthe development of extreme concentrations.
Acknowledgements
The authors thank the Department of Analysis and Method-ology, Hungarian Meteorological Service for providing thesea-level pressure data for the investigated period, GaborMotika (Environmental Protection Inspectorate of Lower-Tisza Region, Szeged, Hungary) for handling meteorologicaland air pollution data, Laszl�oo Haszpra and Laszl�oo Horvath(Hungarian Meteorological Service) for valuable advice onozone related processes and �AAron Deak J�oozsef for advice onplant ecology. This study was supported by the HungarianNational Foundation for Scientific Research (OTKA No.T 034765).
References
Adamopoulos AD, Kambezidis HD, Zevgolis D (2002) Casestudies of total NO2 column in the atmosphere of Athens,Greece: comparison between summer and winter. FresenEnviron Bull 11: 484–487. Sp. Iss. SI
Anderberg MR (1973) Cluster analysis for applications.New York: Academic Press, 353 pp
Bartzokas A, Metaxas DA (1993) Covariability and climaticchanges of the lower troposphere temperatures over theNorthern Hemisphere. Nouvo Cimento C 16C: 359–373
Bartzokas A, Metaxas DA (1995) Factor analysis of someclimatological elements in Athens, 1931: Covariabilityand climatic change. Theor Appl Climatol 52: 195–205
Borhidi A (1995) Social behaviour types, the naturalness andrelative ecological indicator values of the higher plants inthe Hungarian Flora. Acta Bot Hun 39: 97–181
De Nevers N (2000) Air pollution control engineering.Singapore: McGraw-Hill, 2nd ed., 586 pp
Ellenberg H, Weber HE, D€uull R, Wirth V, Werner W,Paulissen D (1991) Zeigerwerte von Pflanzen in Mittel-europa. Scripta Geobot 18: G€oottingen: Goltze-Verlag(in German)
Golder D (1972) Relations among stability parameters in thesurface layer. Bound Layer Meteor 3: 47–58
Hair JF, Anderson RE, Tatham RL, Black WC (1998)Multivariate data analysis, 5th ed. New Jersey: PrenticeHall, 730 pp
Horvath F, Dobolyi ZK, Morschhauser T, L€ook€oos L, Karas L,Szerdahelyi T (1995) Flora database 1.2, Vacrat�oot:Ecological and Botanical Institute of the Hungarian Acad-emy of Sciences (in Hungarian)
Horvath Sz, Makra L, Motika G (2002) An objective assess-ment of the connection between meteorological elementsand the concentrations of the main air pollutants atSzeged, Hungary. AMS 4th Symp. on the Urban Environ-ment and the 12th Joint Conf. on the Applications of AirPollution Meteorology with the Air and Waste Manage-ment Association, Norfolk, VA, USA, 20–24 May 2002.Proc. J4.3: J58–J59
Jolliffe IT (1990) Principal component analysis: a begin-ner’s guide – I. Introduction and application. Weather 45:375–382
Jolliffe IT (1993) Principal component analysis: a beginner’sguide – II. Pitfalls, myths and extensions. Weather 48:246–253
Kalkstein LS, Tan G, Skindlov JA (1987) An evaluation ofthree clustering procedures for use in synoptic classifica-tion. J Climate Appl Meteorol 26: 717–730
Kambezidis HD (1996) Daylight levels in Athensduring spring and autumn. Fresen Environ Bull 5:61–66
Kambezidis HD, Tulleken R, Amanatidis GT, Paliatsos AG,Asimakopoulos DN (1995) Statistical evaluation ofselected air-pollutants in Athens, Greece. Environmetrics6: 349–361
Kambezidis HD, Weidauer D, Melas D, Ulbricht M (1998)Air quality in the Athens basin during sea breeze and non-sea breeze days using laser-remote-sensing technique.Atmos Environ 32: 2173–2182
Makra L, Horvath Sz (2001) Assessment of air pollution inSzeged. L�eegk€oor 46: 14–18 (in Hungarian)
Makra L, Kiss �AA, Palotas J (1985) The spatial and temporalvariability of drought in the southern part of the greathungarian plain. Acta Clim Univ Szegediensis 18–20:65–85
Makra L, Horvath Sz, Zempl�eeni A, Csiszar V, R�oozsa K,Motika G (2001a) Air quality trends in southern hungary.EURASAP Newsletter 42: 2–13
Makra L, Horvath Sz, Zempl�eeni A, Csiszar V, Fodr�ee Zs,Bucsin�ee Kapocsi I, Motika G, S€uumeghy Z (2001b) Ana-lysis of air quality parameters in Csongrad county. ActaClim Chorol Univ Szegediensis 34–35: 23–44
Makra L, Horvath Sz, S€uumeghy Z (2002) An objectiveanalysis and ranking of cities on environmental and socialfactors. In: IGU 2002. Geographical Renaissance at theDawn of the Millennium. Durban, South-Africa, 2002.(Nkemdirim LC, ed), Climates in Transition MinutemanPress, pp 161–172
136 L. Makra et al
Makra L, Juhasz M, B�eeczi R, Borsos E (2004a) The historyand impacts of airborne Ambrosia (Asteraceae) pollen inHungary. Grana 43: 1–8
Makra L, Juhasz M, Borsos E, B�eeczi R (2004b) Meteoro-logical variables connected with airborne ragweedpollen in Southern Hungary. Int J Biometeorol 49:37–47
Mayer H, Makra L, Kalberlach F, Ahrens D, Reuter U(2004) Air stress and air quality indices. Meteorol Z 13:395–403
McGregor GR, Bamzelis D (1995) Synoptic typing and itsapplication to the investigation of weather – air pollutionrelationships, Birmingham, United Kingdom. Theor ApplClimatol 51: 223–236
Mohl M, Gask�oo B, Horvath Sz, Makra L, Szab�oo F (2002) 2ndEnvironmental Programme of Szeged, 2003–2007. Manu-script (in Hungarian) (Mayor’s Office, 6720 Szeged,Sz�eechenyi t�eer 10, Hungary)
Pasquill F (1962) Atmospheric diffusion. London: VanNostrand, 209 pp
P�eeczely G (1957) Grosswetterlagen in Ungarn. KleinereVer€ooffentlichungen der Zentralanstalt f€uur MeteorologieBudapest 30, 86 pp
P�eeczely G (1959) Air pollution of Budapest in different largescale weather situations. Id€oojaras 63: 19–27 (in Hungarian)
P�eeczely G (1979) Climatology. Budapest: Tank€oonyvkiad�oo,336 pp (in Hungarian)
P�eeczely G (1983) Catalogue of the macrosynoptic types forHungary (1881–1983). Budapest: Hungarian Meteorolog-ical Service 53: 116 pp (in Hungarian)
Sindosi OA, Katsoulis BD, Bartzokas A (2003) An objec-tive definition of air mass types affecting Athens,Greece; the corresponding atmospheric pressure pat-terns and air pollution levels. Environ Technol 24:947–962
Turner DB (1964) A diffusion model for an urban area.J Appl Meteor 3: 83–91
Authors’ addresses: L. Makra (E-mail: [email protected]), R. B�eeczi (E-mail: [email protected]),E. Borsos (E-mail: [email protected]), Z. S€uumeghy(E-mail: [email protected]), Department of Clima-tology and Landscape Ecology, University of Szeged,6701 Szeged, P.O. Box 653, Hungary; J. Mika (E-mail:[email protected]), Hungarian Meteorological Service, 1525Budapest, P.O. Box 38, Hungary; A. Bartzokas (E-mail:[email protected]), Department of Physics, Laboratory ofMeteorology, University of Ioannina, 45110 Ioannina,Greece
Classification system of air mass types for Szeged, Hungary 137