Scientists' Contributions  
   

Some characteristics of air quality parameters in Southern Hungary

Laszlo Makra1, Szilvia Horvath1, Andras Zempleni2, Villo Csiszar2, Katalin Rozsa2* and Gabor Motika3

Contact: Laszlo Makra, e-mail: makra@geo.u-szeged.hu

BACKGROUND

Air pollution is one of the most important environmental problems, which concentrates mostly in cities. Generally, human activities induce monotonous accumulation of pollutants. Possible reasons of worsening air quality are population growth in cities and, in connection with this, increasing built-in areas there. A considerable part of population growth is coming from the migration to the cities. The ever-increasing urban population, together with the growing industrialisation and energy consumption and the extensive transportation, raise air pollution, which becomes a more and more serious challenge for the interest of survival. The aim of the study, considering the above-mentioned classification, is complex: namely, to determine partly spatial and temporal characteristics, partly statistical interrelationships of air pollutants and to analyse their connection with meteorological elements.

The database of the study results, on the one hand, from an automatic environmental monitoring station at Szeged downtown, for the period between 1997-1999. This station is located in the downtown, at a busy traffic junction (in the intersection of Kossuth Lajos blvd. and Damjanich street - Terez street), and measures mass concentrations of CO, NO, NO2, SO2, O3 and TSP (total suspended particulate) (mg m-3). Calibration of gas analysers occurs at two points. One of them is the 0-point, which is set automatically, in every 24 hours. Whereas the other calibration point is set once in every second week by a verified sample. Calibration of the O3 instrument is performed by gas phase titration. Verification of TSP measurements occurs once in every quarter of a year. A personal computer serves to perform instrumental control and data storage. Data are produced primarily as one-minute averages from ten-second measurements. Then, thirty-minute averages are determined and stored. Another source of the database is the RIE (Regional Immission Examination) network for Szeged and Csongrad county. From this network, we used monthly mean concentrations of NO2 and SO2, as well as monthly totals of deposited dust (g/m22 / number of days of the month) for the period between 1985-1999.

 

EXTREME VALUE ANALYSIS

Estimating the joint occurrences of extreme high values for two pollutants is especially challenging. We carried out a step-by-step analysis of the data in order to model first the univariate and then the joint tail behaviour of the pollutants. Here analysis for the pair of CO-NO is presented (analogous results for some other pairs will be shown later). As most of the extreme value analysis methodology is suitable for independent and identically distributed observations, the linear trends and periodic parts from the observations (Fig. 1), (the latter ones were based on daily maxima), were removed as otherwise the high number of usually small values would outweigh the most important larger ones. The data show a natural annual cycle and trends in the residuals can also be observed. We have been able to model it suitably by a linear regression. We concentrate on the dependence of the extremes: i.e. on the observation of high values for both pollutants. Coles et al. (1999) give a review for dependence measures for extreme value analysis and we followed some of their methodology in our investigation. The joint distribution can be modelled by the logistic bivariate extreme value distribution, for which the probability of exceeding high values in both co-ordinates can be given as follows:

(1)
where 0 < r < 1 is the parameter, which was estimated by different methods.

 

A SPECIAL CASE OF THE TWO-SAMPLE T-TEST

A new statistical test is developed by Makra for determining if there is significant difference between expected values of non-independent time series (Makra et al., 2000a). The developed expression

(2)

is a random variable with N(0;1) distribution, where is the expected value of the whole sample with N elements, is that of the share sample with n elements (the latest sample is part of the earlier one) and is the joint standard deviation of the two samples considered (we suppose that standard deviations of the two samples are the same).

From the table of the distribution function of the standard normal distribution, that xp can be determined to a given p number (0 < p < 1), for which:

(3)

If the absolute value of the above random variable with N(0;1) distribution is higher than xp, then it is said that and differ significantly. The 0-hypothesis, according to which there is no difference between and , is realised not more than at the critical p probability. Significance-tests are carried out at p=0.01 probability level.

MODELLING EXTREME VALUES OF CO AND NO

Daily maxima of CO and NO are shown in Fig. 1.

Fig. 1. Daily maxima of CO- and NO concentrations, measured at the air quality monitoring station, Szeged downtown, for the period 01.01.1997-31.12.1999

The actual analysis was based on the weekly maxima of the detrended and deperiodised data. The motivation for considering this shorter time series was twofold. First, we could hope for a better fit to this weekly maxima by an extreme value distribution, and second, the dependence was also reduced this way. The dependence between neighbouring days is caused partly by the slow changes in pollutant levels, and partly by the sustained meteorological situations, effecting the pollutant levels. This dependence is reduced when we use weekly maxima only. We need our observations to be maxima or minima in order to apply the extreme value distributions. The longer periods we choose, the better fit can be expected. As the relative shortness has not allowed much more rarefication, we chose the time series of weekly maxima.

We estimated the so-called tail dependence coefficient c, which gives a natural description of dependence between extreme values as follows:

(4)

where , corresponds to the parameter r in (1) as =2-21/r. Our new method for fitting the logistic model is just based on this correspondance (due to the fact that it is easy to give a natural estimator for ). Before estimating these quantities, we transformed the marginals of the weekly maxima into uniform distributions (first plot of Fig. 2).

Fig. 2. Scatter plot of transformed (to uniform in both variables) data (first graph); the dependence of the estimated value of ron the threshold (the values between the vertical lines provided the actual value shown by the dotted line, second graph)

One can see the strong dependence between the co-ordinates. To estimate the parameter r or c, one has to define thresholds, where only the observations higher than the given threshold play a role in the estimator. The curve given in the second part of Fig. 2. shows the different estimators for r, depending on the threshold. As we are interested in the dependence for the extremes, we decided to take thresholds between 0.7 and 0.9 into consideration (see the vertical lines; we excluded the thresholds higher than 0.9 because of the high variation, which is due to the limited number of observations here). We accepted the average of the values in this area as our first estimator (r=0.5155; see the dotted horizontal line). We used two other methods for estimating r, too: maximum likelihood (ML) and a moment-type estimator. The results are given in Table 1.

Table 1. Comparison of different estimators for the logistic model
r
(CO; NO)
in mkg/m3 		Estimator based on 
0.5155
return period in years 		Maximum likelihood
0.4608
return period in years 		Moment-type
0.4951
return period in years
(6000; 500)
(8000; 600)
(14000; 900) 		3.934
9.333
105.061 		3.603
9.809
99.460 		3.802
10.128
102.740

One can summarise the results as follows: the difference among the estimated periods is less than 10%, with the largest value given by the estimator based on c, which is probably due to the fact that here only the higher values were taken into consideration, which seems to be appropriate in our case, where one cannot hope for perfect fit of the extreme value model (supposed for the ML method).

TEMPORAL VARIABILITY OF AIR POLLUTANTS

Fig. 3. shows the average annual cycles of NO, NO2, O3 and Ox at the air quality monitoring station Szeged, for the period between 1997-1999. Ox, according to its definition, is the sum of NO2 and O3. Ox concentration is suitable for estimation oxidating capacity of the atmosphere. Some wet chemical procedures (with absorbing solution), used earlier for measuring O3, actually provided Ox concentration data.

Fig. 3. Average annual cycles of NO, NO2, O3 and Ox at the air quality monitoring station, Szeged downtown, for the period 1997-1999.

The annual cycle of the primary air pollutant NO shows the greatest values in November, December and January, with its maximum at the end of January (Fig. 3.). As NO concentration depends not only on emission but on weather conditions as well, higher winter values refer to atmospheric stability with frequent inversions. The average annual cycle of NO has the lowest values in summer, (June and July). This can be explained by quicker oxidation of NO in summer, but dilution because of intensive vertical exchange in the atmosphere is more important. The annual cycle of NO2, which is secondary substance and produced mainly by chemical reactions, presents similar course to that of NO. Tropospheric ozone is produced by the action of short-wave radiation on substances emitted from anthropogenic sources. Consequently, the average annual cycle of O3, together with that of Ox, has the greatest values in summer (June and July). The annual cycles of these pollutants show similar picture to those in Stuttgart, Germany (Mayer, 1999).

The diurnal cycles of NO and NO2 (Fig. 4.) have the shape of a double wave, with bigger amplitudes for NO than for NO2. Due to the traffic density, the concentration of NO is relatively higher on weekdays, than on weekends. This effect can also be observed for the secondary substance NO2. The average diurnal variations on weekdays are greater for NO than for NO2, because NO2 has a longer lifespan than the more reactive NO. Generally, the NO concentrations are higher in the morning, then in the evening. This can be explained by the fact that in the morning the rush hour is shorter and the atmosphere near the surface is more stable than in the evening. The low NO concentrations early in the afternoon result mainly from the oxidation of NO by O3 and more intensive vertical exchange (dilution) of the atmosphere. The diurnal cycles of O3 show a clear daily course with one wave. A maximum takes place in the early afternoon caused by photochemical O3 formation, while a minimum occurs after midnight. On the basis of its definition, the diurnal cycle of Ox is similar to that of O3. On weekends, the average O3 maximum values are a little higher than on weekdays, but this is not valid for Ox (Fig. 4).

Fig. 4. Average weekly and diurnal cycles of NO, NO2, O3 and Ox at the air quality monitoring station, Szeged downtown, for the period 1997-1999.

As a comparison with the air pollutant characteristics in Stuttgart (Mayer, 1999), average weekly and diurnal cycles of the pollutants at Szeged show, on the one hand, less extremities, namely, less amplitudes of the values; on the other hand, the values themselves are lower at Szeged. At the same time, secondary extremes at Szeged can be shown only in the diurnal cycles of NO and NO2 (Fig. 4) and in their peak values (presenting only for NO) (Fig. 6). At Szeged, lower extremities and smaller concentrations can be explained by less traffic. Peak values for NO occur late in the evening, while secondary maxima are measured in the morning (Fig. 6). Diurnal course of mean NO2 concentrations shows a similar double wave (Fig. 4.), whilst that of mean NO-concentration has a characteristic primary maximum at the beginning of the week (Monday, Tuesday and Wednesday) (Fig. 4). Maximum values in the diurnal course for mean O3 concentrations (Fig. 4.) as well as its peak values (Fig. 5) can be found on weekends (on Saturdays and Sundays). Peak values of O3 and NO represent lower concentrations and less extremities (Fig. 5-6), furthermore, secondary extremities are less characteristic as those in Stuttgart (Mayer, 1999).

Fig. 5. Average weekly and diurnal cycle of percentile values of O3 at the air quality monitoring station, Szeged downtown, for the period between 1997-1999.

Fig. 6. Average weekly and diurnal cycle of percentile values of NO at the air quality monitoring station, Szeged downtown, for the period between 1997-1999.

SIGNIFICANT BURSTS IN SHARE SAMPLES OF DATA SETS

The Makra-test was applied to monthly mean concentration data sets of NO2 and SO2 as well as monthly totals of deposited dust (g/m2 / number of days of the month) for 10 RIE stations at Szeged. At 7 stations of the 10, means of all possible share periods with different length (n = 3, 4, ... , N; where n is the number of elements of the chosen share period and N is the number of elements in the whole data set), chosen from the whole data set of deposited dust (calculating step-by-step moving averages for n) are significantly lower towards the end of the whole data set, than that of the whole data set. Accordingly, the above-mentioned share periods show characteristic bursts in most of the examined deposited dust data sets. On the other hand, there are no such share periods in the data sets of NO2- and SO2-concentrations, means of which differ significantly either by positive or by negative signs from that of the whole data set. Namely, there are no share periods in the NO2- and SO2-data sets, showing significant bursts (Makra and Horvath, 1999; Makra et al., 2000b).

CONCLUSIONS

>> The extreme value analysis showed strong dependence between the components of the pair (CO,NO) (we have got similar patterns for most of the other pairs, as well) for the whole range of observations. Thus, the bivariate modelling allowed us to give much more exact return periods for given joint threshold than it would have been possible, based on univariate analysis only.

>> In Hungary, emissions of SO2 and total suspended particulate (TSP) are decreasing.

>> Average annual cycles of NO and NO2 (with minimum in the summer) are opposite to those of O3 and Ox (with minimum in the winter).

>> Average weekly and diurnal cycles of NO, NO2, O3 and Ox have lower values and less extremes in a middle-sized Central-European city (Szeged), compared to those of a big one (Stuttgart) in a developed country (Germany).

>> At Szeged, trend for NO2 presents an important decrease, while those for NO, O3 and Ox do not show a clear tendency (in the data sets of the monitoring station).

>> According to the Makra-test, share periods towards the end in most of the whole data sets of monthly totals of deposited dust show characteristic bursts. On the other hand, these bursts (indicated by significant difference between the average of the given share period and that of the whole data set) can not be shown for NO2 and SO2 (in the data sets of the RIE stations).

The statement of the earlier point, according to which trend for NO2 at Szeged shows characteristic decrease and the consequence of this point, according to which means of share periods of NO2 concentration data set do not differ significantly from that of the whole data set, does not indicate a contradiction. That is to say, if e.g. a significant trend can be presented in any share period of a given data set, it does not imply that the mean of the same share period differs significantly from that of the whole data set, as well. The significant decreasing trend in the time series of NO2 concentrations is realized not by bursts of means in share periods but by gradual decrease of the elements of the whole data set.

ACKNOWLEDGEMENT

The authors are indebted to L. Lukacsovics for fruitful discussions. The authors thank Z. Sumeghy for his contribution with digital maping. This research was supported by the grants from the Ministry of Education (FKFP-0429/1997 and FKFP-0001/2000) and OTKA (T-34765).

REFERENCES

       
  Scientists' Contributions  
   

[To Contents]    [To Next Topic]