A tool for determining urban emission characteristics to be used in exposure assessment

Environment International 36 (2010) 281–289

Contents lists available at ScienceDirect

Environment International j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / e n v i n t

A tool for determining urban emission characteristics to be used in exposure assessment P. Kassomenos a,⁎, S. Lykoudis b, A. Chaloulakou c a b c

University of Ioannina, Department of Physics, GR-45110, Ioannina, Greece National Observatory of Athens, Institute of Environmental Research and Sustainable Development, I. Metaxa & V. Pavlou, P. Pendeli, GR-15236, Athens, Greece National Technical University of Athens, School of Chemical Engineering, Heroon Polytechniou 9, GR-15780, Zografos, Athens, Greece

a r t i c l e

i n f o

Article history: Received 16 September 2009 Accepted 26 December 2009 Available online 25 January 2010 Keywords: Air pollution Probability distribution function Exposure assessment Emission reduction Exceedances

a b s t r a c t The exposure of citizens to elevated air pollution concentrations is one of the major factors leading to the deterioration of the quality of life and possibly to health problems in urban areas. The concentration of air pollutants depends largely on pollutant emission levels. If the statistical probability distribution function of the concentration of an air pollutant is known, it is possible to estimate how many times this concentration exceeds the air quality standards, or estimate changes in the emission levels in an area. It can be also used to estimate the long term exposure of population to certain pollutants. In this paper fifteen theoretical probability distribution functions, were used to fit the actual concentration frequency distributions of CO, NO2, O3, SO2, and Black Smoke (BS) in Athens, Greece for a 23-year period. The results showed that the theoretical distribution type best describing the distribution of the pollutants is Inverse Gaussian followed by the Extreme value distribution. The number of exceedances of air quality limits was used to validate the performance of the theoretical distributions that were best fitted to the observed ones. The temporal evolution of emission strength was estimated through the temporal evolution of the parameters of the probability distribution functions. Missing periods were accounted for by estimating the respective distribution functions through interpolation or extrapolation from the existing ones. The derived variation of emission levels consistently represents the emission reduction strategies enforced over the years, as well as the escalating growth of the passenger car fleet volume, and could be a useful tool for the design and assessment of emission control strategies. © 2009 Elsevier Ltd. All rights reserved.

1. Introduction Air quality is of major interest during the last years, both to the people and the relevant authorities. Policy makers set, and authorities apply air quality standards to protect human health and the environment. Relevant authorities are obliged to use these standards to establish emission control strategies especially in urban environments and manage the risk of exceeding air quality objectives (Longhurst et al., 2006). The measured concentrations of air pollutants are the means by which the air quality status of an area is assessed, but no monitoring activity is free of missing data, especially in the case of air quality networks that are in use for many years. Thus, a general understanding of the behavior of a pollutant could be better described by a probability distribution function, which provides all the necessary statistical indices, and accounts for any possible missing information from the monitoring activity.

⁎ Corresponding author. Tel.: +30 26510 08470; fax: +30 26510 08671. E-mail address: [email protected] (P. Kassomenos). 0160-4120/$ – see front matter © 2009 Elsevier Ltd. All rights reserved. doi:10.1016/j.envint.2009.12.009

When the probability distribution function (PDF) is correctly chosen, it can be used to provide general information on the behavior of the pollutant, and of course to predict the frequency of concentrations that exceed ambient air quality standards. However there is no universal frequency distribution that can represent the PDF of a given air pollutant. There are many types of theoretical PDFs that can be used to fit an air pollutant concentration data. Lognormal is the favorite of many researchers that have successfully applied it to describe the distribution of several air pollutants (Mage and Ott, 1984; Ott, 1990; Miles et al., 1991; Kao and Friedlander, 1995; Lu, 2002, among others). Georgopoulos and Seinfeld (1982) offer a summary of some possible PDFs, other than Lognormal, specifically proposing the Weibull distribution as an alternative. Also Jakerman et al. (1986) and Lu (2003) used the same distribution function for SO2 and PM10. Among the other distribution functions used for fitting air pollutant concentrations are the Gamma distribution (Berger et al., 1982) and the Type V Pearson distribution (Morel et al., 1999; Lu, 2002) etc. During the last few years the interest of using PDFs for air quality management purposes has been rekindled. Lu and Fang (2002) used three different PDFs (namely Lognormal, Weibull, and Type V Pearson)

282

P. Kassomenos et al. / Environment International 36 (2010) 281–289

to fit PM10 and SO2 concentration distributions in three air-monitoring stations of Taiwan. Lu (2004) predicts the exceedances of a critical concentration, fitting the higher portion of the data of each station with a two-parameter Exponential distribution and an asymptotic Extreme value distribution. Lu and Fang (2003) estimated the frequency distributions of PM10 and PM2.5 according to certain wind speed classes, at Sha-Lu in Taiwan. Gavriil et al. (2006) studied the theoretical probability distribution of PM10/PM2.5 time series in Athens. Also, Hadley and Toumi (2003) investigated whether there has been any change in the probability distribution of sulfur dioxide concentration at 10 United Kingdom monitoring sites over time periods of up to 40 years. In this study 15 theoretical PDFs were tested against 23 years of air quality data, namely CO, NO2, O3, SO2, and BS recorded in Athens, Greece. The data were arranged in 5-year periods in order to smooth out inter-annual variations that could be the result of the anomalous behavior of external factors, such as the meteorological conditions. We selected the most appropriate PDF per period, site and pollutant using well-known goodness of fit statistics. Based on these PDFs we estimated how many times a concentration exceeds the air quality limit value, set by local authorities and the European Union (EU). Furthermore, using simple interpolation on the parameters of the existing PDFs, we attempted to estimate the PDFs of the missing periods. This attempt required the selection of a single, uniform for all periods, PDF per station and pollutant. We also estimated the changes in emission levels along the years, as well as the emission reduction required, for each pollutant, in order to comply with the existing air quality standards. This information is useful for estimating the performance of emission reduction measures as well as the number of violations of the existing air quality standards, and developing a control strategy in the area. Moreover the statistical distributions of air pollutants could provide a comprehensive way to assess the long term exposure of the population to certain pollutants, harmful for the human health. 2. Monitoring sites and data used All data available from the 19 stations of the Hellenic Ministry for the Environment monitoring network (MinEnv) operated in Athens, Greece, were considered (MinEnv, 2007). The period considered, extends from 1983 to 2005 inclusive, with the exception of BS that covers the period 1983-2000. The original data — hourly values for CO, NO2, O3 and SO2, and daily values for BS — had to be processed in order to be compatible with EU limits. Existing limits refer to hourly values only for SO2 (350 μg/m3) and NO2 (200 μg/m3) (EU, 1999). CO (10 mg/m3) and O3 (120 μg/m3) limits refer to the maximum daily 8-hourly moving average values, considering the ending date of the 8-hour period as the date that the value refers to (EU, 2001, 2002) (Table 1). When calculating means at least 75% of the data should be available, that is 6 out of 8 hourly values for the 8-hourly running means, and 18 out of 24-hourly values for the 24-hourly average values.

All stations use automatic analyzers utilizing the measurement methods presented in Table 1, along with the respective, generally accepted, minimum detection limits (NARSTO, 2005; CHMI, 1999). Apparently, data values less than the respective minimum detection limit should be considered as dubious; therefore we have replaced all such cases with the half of the respective minimum detection limit (Lu, 2003). Checking the original data revealed that the limits proposed by NARSTO were more relevant to our dataset, so we used the lowest limits provided by NARSTO. For SO2, the only limit is proposed by CHMI (1999) yet it was too low for our data. A minimum detection limit value of 2 ppbv (5.2 μg/m3) was considered to be more suitable since values between 2 and 3 ppbv persisted for large periods in several stations. Air pollution levels over an area are determined by emission levels and meteorological conditions. The latter are known to present strong inter-annual variability modulated by the respective changes of larger scale atmospheric circulation. On the other hand emission levels might also change from year to year as a result of changes in the intensity of the emitting activities, socioeconomic reasons or administrative measures usually towards emission reduction. The combined effect of the meteorological and emission variability can produce contrasting results on the measured concentrations, so in order to smooth out the lesser of such effects yet retaining the most important of them we aggregated the data into 5-year periods. Using consecutive 5-year periods, possible unusual meteorological conditions are smoothed out. Furthermore, significant land use changes affecting the spatial distribution of pollution sources are rather slow, thus probably extending into more than one 5-year period. As a consequence, there is a good possibility that any changes in emission sources would be more or less smoothed out when examined on a consecutive 5-yearly aggregate basis. On the other hand this smoothing would also reduce the relative impact of significant changes in emissions e.g. as a result of a change in the fuel type used, or traffic regulation measures. Thus, breakpoints for the consecutive 5-year periods should be selected in a way to also separate periods according to such measures. In the case of Athens, Greece significant emission reduction measures have been adopted in several occasions. Specifically: in 1982 restrictions on traffic in an extended area around the city center; in 1992 a government subsidized withdrawal of vehicles without a catalytic converter; in 2000 the Athens Metro started its operation; in 2002–2003 the Attiki Odos highway, one of the major public works related to the 2004 Athens Olympics, becomes operational; and finally, the sulfur content of diesel for both vehicles and central heating has been reduced several times from 0.3% (1991) to 0.2% (1994), 0.05% (1996) and 0.035% (2000). Based on the above, significant changes in traffic related emissions and pollutant levels, therefore possible breakpoints as well, should be detected after 1982, 1992, and 2002. Changes in diesel related pollutants should be detectable in 1997 data, yet since these are also influenced by industrial emissions, we retained the breakpoint after 1997. The resulting 5-year periods start at 1983, 1988, 1993, 1998, and 2003.

Table 1 Air quality measurement methods, and detection limits according to EU Directives. Measurement Method

NO2 SO2 CO O3 BS

UV fluorescence Chemiluminescence NDIR photometry UV photometry Reflectometer

EU limits Period

Hourly Hourly Hourly Hourly Daily

Minimum detection limit (μg/m3) NARSTO

CHMI

– 9.4 140–560 3.9–9.8 5

1.5 1.0 125 2.0 3.5

Reference period

Value (μg/m3)

Exceedances allowed (per year)

Hourly Hourly Daily (max 8 h ma) Daily (max 8 h ma) Daily

200 350 10,000 120 250

18 24 1 25 7

P. Kassomenos et al. / Environment International 36 (2010) 281–289

Missing data are a serious problem, especially when calculating probabilities. Our 5-year periods were considered as valid for further investigation only when each season — extending according to the EU Directives from 1/10 to 31/3 and from 1/4 to 30/9 — had at least 70% valid data points (EU, 1999, 2001, 2002). The period 2003–2005 was included in the analysis despite the fact that it contains only 3 years. BS measurements stopped at 2000 so the respective last period is 1998–2000. 3. Methodology 3.1. Selecting the best-fit PDF For every station and period that met the completeness criteria, we estimated the PDF that was best fitted to the data, using BestFit v.4.5 software (Palisade Corp., 2004). The software offers several possible PDFs to be fitted on the data. From those we have selected a subset of 15 PDFs excluding the rest as not being suitable for the kind of data at hand. The PDFs tested were: Beta General, Chisquare, Erlang, Exponential, Extreme value, Gamma, Normal, Lognormal, Inverse Gaussian, Logistic, Loglogistic, Pareto, Type V Pearson, Rayleigh, and Weibull. The software used employs the maximum Likelihood Estimator (MLE) method. That is, for any PDF, f(x), with a set of parameters aj, j = 1,…,k and a data set of observations, it calculates the values of the aj parameters that maximize the probability of obtaining the given observational data set. Since the MLE method cannot always be applied, depending on the theoretical PDF, there is also a hybrid calculation method, which combines the standard MLE approach with a moment matching procedure. Goodness of fit statistics were calculated for each candidate PDF. We considered two widely used statistics, namely the Kolmogorov–Smirnov and the Anderson–Darling, in order to select the bestfitted theoretical PDF for each pollutant, period and station. A major advantage of these statistics is that neither of them requires binning of the data, thus avoiding the arbitrariness inherent in some other, equally widely used, goodness of fit statistics such as the Chi-squared. For each of these statistics the smaller the value is, the better the fit. The Kolmogorov–Smirnov (K–S) statistic, D, is better in detecting discrepancies around the middle of the range of the distribution values. It is defined as: D = supx ½jFn ðxÞ−FðxÞj

ð1Þ

where: n is the total number of data points, F(x) is the theoretical cumulative distribution function (CDF) and Fn(x) the observed one. Since the upper end of the distribution values, namely the elevated pollution levels, is of special interest to us, we also use the Anderson– Darling (A–D) statistic, A2, which highlights differences between the tails of the theoretical and the observed distributions. The statistic is defined as: 2

A = −n−

1 n ∑ ð2i−1Þ⋅fln½Fðxi Þ + ln½1−Fðxn−i + 1 Þg n i=1

ð2Þ

where: n is the total number of data points, and F(xi) is the theoretical CDF at data point xi. For each case (station, period and pollutant) the various PDFs were ranked according to the value of each statistic, thus producing two sets of scores (K–S and A–D) from 1 to 15. As mentioned above each of the statistical tests used, focuses on a different part of the data. K–S focuses on the middle part of the data range which covers the largest part of the time period represented by the data, hence is important for exposure assessment. On the other hand the A–D test focuses on the extreme values of the dataset hence it is important for the correct representation of limits' exceedances. As exposure assessment and limits' exceedances are considered equally important within the

283

scope of this work, combining the two scores into a single one by simple addition seemed reasonable. The selection procedure, outlined above, would give the overall best scoring theoretical PDF for each station, period and pollutant, and this could result in a wide variety of PDF types appearing as final selections, even for the same station. This might indicate that there is a change in the behavior of a specific pollutant among the examined periods that is reflected on a different type of PDF describing this pollutant, but it could also be a statistical artifact. In order to gain some insight on the actual differences among the various best-fitted PDFs for each station and pollutant, the three most commonly appearing PDFs were crossexamined with the 2-sample K–S test. As stated in the Introduction, one of the tasks set out in this work was to obtain PDFs for those cases with missing or incomplete data as well. This would be achieved by applying first or higher order, interpolation (or extrapolation) on the parameters of the theoretical PDFs fitted on the cases with valid data sets. This of course, required that the same PDF would be selected as the best-fit for all the cases that would be used as basis for the interpolation procedure. Thus we had to select a single type of PDF, henceforth called uniform PDF, for each pollutant and station. To do this we gathered the scores for all the available periods from a station and assigned each candidate PDF with the average ranking score. Only PDFs for which statistics were calculated for all the available periods were considered. The selection procedure was meant to provide a PDF that might not be the best for each subperiod separately but it would be as good as possible (using the scores of the K–S and A–D tests as criteria) for all the subperiods together. Furthermore, whenever the uniform PDF was of different type than the initial best-fit one, we examined their statistical differentiation using the 2-sample K–S test. 3.2. Calculation of exceedances Due to its importance, we evaluate the performance and usability of the derived PDFs in terms of accuracy of the exceedance predictions. The basis for the assessment is the actual frequency of exceedance. This can be found by simply counting the number of cases exceeding a set threshold value and dividing it by the number of available data, yet this may not be accurate since it does not account for missing values, the percentage of which can be significant, up to 30% in our data sets. Proportional correction of exceedance counts for missing data might provide better estimates that would be exact if the missing values were proportionally dispersed across the entire range of measured values. Expected exceedances were calculated using both the actual and uniform best-fit PDFs. 3.3. Estimation of emission levels and reduction The concentration of any air pollutant at a given point in time and space is determined by the emission sources the meteorological conditions and possible chemical reactions that the pollutant participates in. Transport and diffusion along with deposition are the mechanisms determining the concentration of a non-reactive pollutant measured at a certain location. Moving a source closer to the measurement location may result in higher concentrations being measured, even if the emissions were somewhat reduced as the diffusion and deposition mechanisms would not have time to act. Furthermore, a bias towards high or low winds, over a certain period, would result in enhanced or reduced transport and diffusion, leading to very different concentrations even if nothing had changed in the emission regime. Severe weather on the other hand might influence the emissions balance, favouring central heating related pollutants over those related to traffic. Assuming nonreactive species and that, the spatial distribution of the pollution sources remains unchanged and the meteorological conditions are not significantly different over the considered 5-year consecutive periods,

284

P. Kassomenos et al. / Environment International 36 (2010) 281–289

emissions are the sole determinant of the species concentration. In that case, the so-called rollback equation can provide a means of estimating a change in emission levels (Mage and Ott, 1984; Lu, 2002): R=

〈Cp 〉−〈Cs 〉 100% 〈Cp 〉−Cb ⋅

ð3Þ

where: Cp is the current (reference) mean (expected) value of a pollutant following a certain PDF, Cb is the respective background concentration, and Cs is the target mean (expected) value of that pollutant. If we wish to calculate changes in emissions between two periods, the reference and the target, then Cs is the mean pollutant concentration at the target period. On the other hand if we wish to calculate the required reduction in emission levels so that the air quality standards would not be violated, then Cs is the mean (expected) value of a PDF of the same type as the reference case, for which the probability of exceeding the limit value would not violate the adopted standards. In this work we use the uniform best-fit PDFs to estimate both the observed changes in emissions levels between the successive time periods, and to calculate the emissions reduction that was required at each period to achieve compliance with the EU air quality limits. 4. Results and discussion 4.1. PDF fitting results. Applying the methodology described above we found the best-fit PDF for each pollutant and station using those 5-year periods that fulfilled the completeness criterion of 70%. The interchangeability testing between the three most frequently high ranking best-fit PDFs for each case, revealed than on average less than 30% of the tested PDF pairs were found to be statistically different. Differences were more frequently observed for NO2 and SO2, yet not exceeding 40% of the cases, whereas for O3 differences were found for only about 21% of the cases. These results favour the notion that, for many of the cases, changes in the type of the best-fitted PDF across the examined period cannot be distinguished from being statistical artifacts. Uniform PDFs were of a different type than the actual best-fitted ones for about 50% of the cases, and from those, about 30% on average were also statistically different according to the K–S test. Overall the differences between uniform and actual best-fit PDFs are rather limited supporting the ability of the adopted selection procedure to provide theoretical PDFs that describe the observed ones in an appropriate manner. Ten out of the fifteen theoretical PDFs initially tested, appear at least once as best fitted to a data set. The most frequent is by far the Inverse Gaussian best fitted to 31% of the cases, followed by the Extreme value with 18% of the cases (Table 2). We also calculated the parameters of the PDFs for the cases with missing or incomplete data directly interpolating the parameters of the fitted PDFs. Linear interpolation was preferred over polynomial since the number of available points to interpolate between/extrapolate form was very small. The graphical representation of a PDF provides significant information on the represented parameter. The value corresponding to the maximum (the most probable value, mpv) is a good indicator of the normal air quality status for this area, while the

minimum of the distribution might point to the existing background concentration. Of course, when air pollution concentrations are examined, the tail of the distribution and its shape are the most important factors, since they represent the higher concentrations, and the way the concentrations are distributed. A long thin tail indicates the existence of concentrations much higher than the average but rather rare, while a shorter thick tail corresponds to concentrations not so high (with respect to the average) but more frequent. The distribution (scale and shape) parameters are significant quantities related to the emission characteristics of an area. More specifically, the scale parameter is positively correlated to the strength of the emission sources (Lu, 2004). On the other hand, the shape parameter of the PDF, although in fact describing the dispersion of the concentrations around their mean, does not seem to be affected by changes in the pollutant emission levels (Morel et al., 1999). Thus, it should be related to other factors affecting the pollution levels, namely the dispersion characteristics prevailing in the area. Since meteorology is a factor highly variable when examined in short time scales, but rather stable on a larger scale, the shape of a pollutant PDF seems to be primarily related to the distribution of the various emission sources, indicating that if the shape of the curve had not changed during the years, the spatial and temporal distribution of the emission sources should also have remained unchanged. Of course this is a coarse approximation of the reality and it depends greatly on the type of PDF, so it should be considered with caution. Since presenting the fitting results for all the stations would not be practical, we selected three representative stations: Patision (PAT) station located in the center of the city, operating since 1983, Peiraias1 (PE1) located in the southern part of Athens, near the harbor of the city, also operating since 1983, and Marousi (MAR) operating since 1988 and located in the north sector of Athens. All stations are characterized as Urban-traffic ones. Unfortunately stations characterized as suburban-industrial that could possibly present contrasting behavior, only had measurements for a few of the pollutants considered (Table 2). Fig. 1 presents the uniform best-fit PDFs per 5-year period for the selected cases. There is a marked migration of the CO mpv towards lower concentrations and with higher probabilities after the 1988–1992 period, indicating overall lower concentrations as time progresses (Fig. 1a). Also the contraction of the right hand tails suggests diminishing maximum values. These changes in the shape of the curves reflect the changes of the emission sources through the years. CO is a pollutant related to combustion processes, and since emissions by central heating and other diesel-fueled activities have been controlled since the early '90s, it is expected that CO emissions would have dropped significantly. Traffic emissions have decreased significantly, after 1992, due to the introduction of new, catalytic technology vehicles and the renewal of the vehicle fleet, but this was offset by the equally significant, and sometimes larger, increase in the traffic volume. Apparently the combination of both emission reduction strategies has resulted in the, more or less uniform, significant decrease of CO concentrations over Athens. The temporal evolution of NO2 probability curves is rather weak. The right hand tails are moving towards lower values, yet the mpv remains more or less stable indicating that even though maximal concentrations might have been reduced over the years, overall, the NO2 concentrations have not changed, except perhaps in PAT station (Fig. 1b). In this station, the mpv and the higher values decrease between 1983 and 1997 affected by the increased replacement, at that time, of old technology vehicles by catalytic ones. Afterwards, when the replacement procedure was saturated, the steadily increasing volume of the fleet and the relative increase of traffic in the city center overwhelmed the emission decrease. This is reflected by an increase of the mpv and the maximum values during 1998–2002. The decrease during 2003–2005 could be explained by a reduction of the traffic volume in the city center. Even though detailed traffic volume data for the major road axes of the Athens center are not available,

Table 2 Final selection of uniform best-fit PDF types per station and pollutant. Station

Pollutant

Name

Abbr.

Type

CO

NO2

O3

SO2

BS

Athinas Geoponiki Liossia Lykovrisi Marousi Patision Peiraias1 Peristeri N. Smirni Ag. Paraskevi Aristotelous Elefsis Galatsi Goudi Renti Thrakomakedones Zografou

ATH GEO LIO LYK MAR PAT PE1 PER SMI AGP ARI ELE GAL GOU REN THR ZOG

Urban-traffic Suburban-industrial Suburban-background Suburban Urban-traffic Urban-traffic Urban-traffic Urban-background Urban-background Suburban-background Urban-traffic Suburban-industrial Urban-traffic Urban-traffic Urban-industrial Suburban-background Suburban-background

Loglogistic Inv. Gaussian N/a Inv. Gaussian Inv. Gaussian Pearson V Inv. Gaussian Pearson V Lognormal N/a N/a N/a N/a N/a N/a N/a N/a

Loglogistic Extr. value Extr. value Extr. value Inv. Gaussian Pearson V Gamma Extr. value Inv. Gaussian Extr. value Loglogistic N/a Beta General N/a N/a Loglogistic Extr. value

Beta General Beta General Beta General Inv. Gaussian Inv. Gaussian Inv. Gaussian Pearson V Beta General Loglogistic Chi-square N/a Weibull Weibull N/a N/a Weibull Chi-square

Lognormal Extr. value Extr. value N/a Inv. Gaussian Inv. Gaussian Inv. Gaussian Extr. value Inv. Gaussian N/a Extr. value Inv. Gaussian Inv. Gaussian N/a N/a N/a Inv. Gaussian

Pearson V Exponential N/a N/a N/a Beta General Lognormal Gamma N/a N/a Pearson V N/a N/a N/a Inv. Gaussian N/a N/a

N/a not available.

P. Kassomenos et al. / Environment International 36 (2010) 281–289

285

Fig. 1. Uniform best-fit PDFs, originally fitted and interpolated, for CO, NO2, SO2, and BS at PAT station and O3, at PE1 and MAR stations.

indicative measurements at Patision street, a major road in the area of PAT station, show a decrease of traffic volume of about 20% between 2001 and 2005 (MinEnv, 2009). This could be the result of the introduction of the new metro lines and the new highways before the Athens Olympics in 2004. Actually, that infrastructure affected the entire Greater Athens Area since, despite the fact that the total number of vehicles circulating in Athens increased by about 20% during the period 2001–2005, the corresponding number of kilometers traveled decreased by about 3.5% (MinEnv, 2006). On the other hand the PDF curves for MAR have practically the same mpv and quite similar higher values distribution. The major differentiation between 1993–1997 and 2003–2005 is that the lower values of the distribution have moved to somewhat higher levels, possibly the result of the changing character of this area from residential to Urban-traffic due to the re-design of this part of the city. Finally, at PE1 station the right tails of the distributions have moved to lower values leaving the mpv unaffected. Ozone PDF evolution with time presents a graduation from south (PE1) to north (MAR). At the south, the high-end tail does not seem to change, yet the distribution moves to significantly lower values. At the northern station, on the other hand, the tail gets less extended but the mpv slightly increases (Fig. 1c, d). PAT station, in the center of the city, presents migration of both the mpv and the upper end tail to lower values. Since tropospheric O3 is a secondary pollutant, its concentration is controlled by the ozone precursors that also act as its consumers, namely NOx and hydrocarbons. In PE1 and PAT stations, where the local production of precursors is significant due to the high traffic volume, ozone concentrations seem to follow the slow reduction of NO2. In MAR station though, ozone transported, by the local circulations, from the city center to the northern suburbs builds up, assisted by the lack of adequate emissions of ozone consumers in that area (Kambezidis et al, 1998). As already mentioned, SO2 emission reduction measures were adopted since the early '90s targeting both central heating and the numerous small industries that are scattered in the Athens area. The near doubling of the mpv in PAT station between 1983– 1987 and 1988–1993 is representative of the magnitude of the problem, while the subsequent decrease of the mpv in all stations is characteristic of the impact of the emission control measures (Fig. 1e). Despite the fact that the mpvs of the SO2 concentration have become very low during the last years, there are still rather extensive tails in the PDFs, indicating the occurrence of, at least, a few very high values due to unfavorable meteorological conditions that sometimes prevail over Athens during winter, when the SO2 emissions are also elevated. BS has been recorded in several stations, yet not in MAR, for almost 30 years, and after 2000 it was replaced by PM10. The temporal evolution of BS PDFs is practically

opposite from that of CO and more notably for SO2. Specifically in both PE1 and PAT stations, there is a decrease of the mpv between the first two quinquennia, and a small but steady increase beyond 1992. Regarding emissions, the annual average smog content of both the industrial and central heating related exhaust gases presented a slight but steady downward trend since 1994 (MinEnv, 2002). On the other hand, even though only taxis and heavy duty vehicles are allowed to use diesel in Athens, so their number is limited, diesel-fueled vehicles are responsible for almost 4 times more kilometers traveled per year than gasoline fueled ones (MinEnv, 2002), indicating that traffic related emissions of particulate matter might be responsible for the observed increase.

4.2. Predicted exceedances of the critical concentration The probability of any pollutant concentration to exceed the limit values listed in Table 1 was calculated from the uniform best-fit PDFs, along with the observed frequencies of limit exceedance corrected for the amount of missing data, as described in Section 3.2. The overall relative root mean square error (RMSE) of the limit exceedance estimates using uniform best fit PDFs was 20% for SO2, 7% for NO2, 5% for O3, 3% for CO and 6% for BS. Table 3 presents the observed and uniform PDFs exceedance estimates for each 5-year period, station and pollutant for stations with more than one period available. Overall, the PDF-based results were found to be in good agreement with the observed values, while also reproducing successfully the observed temporal variation, except for a few cases. The squared correlation coefficients, R2, between predicted and observed numbers of exceedances were 0.90 for O3, 0.96 for NO2 and 0.99 for CO and BS, yet it was only 0.67 for SO2. Also, on average, the predicted exceedances overestimated the observed ones by 12% for NO2, 22% for O3, and 16% for BS, whereas CO exceedances were underestimated by about 5%. SO2 exceedance estimates were not very close to the observed ones, even though they presented similar trends. Severe cases of underestimation were identified for PAT and PE1 stations coinciding with the occurrence of the maximum SO2 values over Athens for the entire examined period. Apparently, the existence of very intense episodes led to very long tails of the observed PDFs that could not be reproduced by the fitting procedure applied in this work. Furthermore, the estimated NO2 exceedances for MAR station 1993–1997 and SMI station 1983–1987, significantly overestimated those observed, while the estimates for PE1 station consistently underestimated the observed exceedances.

286

P. Kassomenos et al. / Environment International 36 (2010) 281–289

Table 3 Estimated and observed exceedances of EU limits presented in Table 1, per 5-year period. ATH

GEO

Obs.

PDF

CO (days) 1983–1987 1988–1992 1993–1997 1998–2002 2003–2005

N/a 60 31 15 2

0 59 26 17 4

NO2 (hours) 1983–1987 1988–1992 1993–1997 1998–2002 2003–2005

N/a 149 88 53 18

O3 (days) 1983–1987 1988–1992 1993–1997 1998–2002 2003–2005

Obs.

LIO PDF

Obs.

1 5 2 0 0

3 3 2 1 0

N/a

410 158 78 51 23

M M M 18 9

76 47 25 11 5

9 M M M 9

N/a N/a 21 29 22

22 20 20 30 28

N/a 98 57 40 31

58 85 57 45 30

N/a 140 79 M 75

SO2 (hours) 1983–1987 1988–1992 1993–1997 1998–2002 2003–2005

N/a 9 9 0 M

4 6 11 0 0

M M 9 0 0

BS (days) 1983–1987 1988–1992 1993–1997 1998–2000

N/a 2 1 1

6 4 1 2

N/a

0 0 0 0 0

0 M 0 9 9

N/a

LYK PDF

0 0 0 1 2

114 147 83 52 80

0 0 0 0 0

MAR

PAT

PE1

PER

Obs.

PDF

Obs.

PDF

Obs.

PDF

Obs.

PDF

Obs.

N/a N/a M 0 0

66 37 M 1 0

N/a N/a M 1 0

88 48 17 2 0

188 183 68 30 1

180 173 70 32 2

N/a 22 6 0 M

82 25 7 0 0

N/a

N/a N/a M 9 9

155 64 17 2 0

N/a N/a 18 M 9

427 368 132 40 17

342 640 307 158 88

389 696 310 127 67

N/a 53 18 18 M

114 30 3 2 1

N/a N/a 35 9 9

M M 54 81 77

32 52 68 93 79

N/a N/a 89 83 40

67 74 82 92 44

N/a 6 1 0 0

19 10 5 3 2

M 34 35 M 36

32 34 34 25 45

N/a

N/a

N/a N/a 0 0 M

0 0 0 0 0

26 70 18 0 9

10 36 10 0 0

N/a 26 9 9 0

75 22 5 5 26

N/a N/a 9 M 0

N/a

N/a

68 18 6 5

68 13 6 5

2 0 0 N/a

2 0 0 0

N/a

SMI PDF

146 79 34 10 8

0 0 0 0 0

Obs.

PDF

M 0 1 5 0

0 0 0 13 1

18 44 26 26 9

71 26 35 33 28

N/a 101 88 51 106

89 104 84 25 89

M 9 9 9 M

0 4 21 0 0

N/a

M: missing. N/a not available.

Apparently, these discrepancies are the result of non optimal PDFs being used under the requirement for uniformity across the entire examined period. It was found that for NO2, the introduction of the catalytic technology in passenger cars after 1993 resulted in a significant reduction of the hours that the air quality limit was exceeded. The situation at the center of Athens (ATH, PAT) in the earlier years was dramatic. The limit of 200 μg/m3 was exceeded for more than 350 h/year in the years 1983–1987, and almost 700 h/year in the next 5-year period. The decrease after 1993 was sharp, yet the air quality of the Athens center and other areas with heavy traffic (PE1) still remains far from the acceptable levels, as a result of the increased traffic flow. On the other hand, stations located at residential areas like MAR, PER and SMI have reached acceptable air quality levels during the last few years. Similar results were also found for CO. During the first two 5-year periods the exceedances per year remained practically unchanged in PAT (∼180 h/year) but after that, they were progressively reduced down to 2 h/year during the last 5-year period (Table 3). A similar decreasing trend was also detected for the other stations following the introduction of the catalytic technology. However, CO levels are still relatively higher than the EU standards in the center of Athens (Economopoulou and Economopoulos, 2003). Ozone is a severe problem for the entire Athens area, except its center. Apart from PAT and ATH the rest of the stations still presented exceedances well over those allowed by the EU. In PAT the excessive traffic generated enough NOx to reduce the large amounts of photochemicaly generated O3 (Kassomenos et al., 2003) while significant amounts of O3 were also transported northwards. The predicted exceedances indicated a fluctuating variation with time, matching that of the observed ones, while the high values of ozone observed in the stations surrounding the city center indicated that O3 exceedances were mainly influenced by the local meteorology (Paschalidou and Kassomenos, 2004). This is the possible explanation for the increased number of exceedances during the period 1998–2002. SO2 exceedances were significant in the center of the city (PAT) and the harbor (PE1) till 1992, yet the restrictions on sulfur content of diesel, adopted in the early '90s, led to a sharp decrease. After 1997, SO2 exceedances have been reduced to values well below the EU limits at all stations. There seems to be a background number of 9 limit exceedances, consistent throughout the Athens area, probably attributable to the wintertime central heating emissions. Finally, BS exceedances, in general, follow a course similar to SO2 (Table 3). Again there is a stagnancy of the detected exceedances after their abrupt reduction during 1988–1992, but there is no uniform background as in SO2.

4.3. Estimation of emission levels and reduction Using 5-year aggregate data sets instead of annual aggregates, should improve the compliance of our data set to the assumptions inherent in the rollback equation, namely unchanged meteorological conditions and spatial distribution of the emissions sources. The rollback Eq. (3) was used to estimate the temporal evolution of emissions between the consecutive 5-year periods from 1983 to 2005. CO, SO2 and BS could be easily considered as non-reactive, whereas ozone could not, and was excluded from the analysis. Finally, NO2 even though also involved in the photochemical reaction, is included since it is a primary pollutant and this leaves some space for the rollback estimates to be meaningful, nevertheless treated with caution. Fig. 2 presents the rollback estimations of the percentage change of the emission levels between consecutive 5-year periods. Positive values indicate an emission reduction. For CO the rate of decrease increased with time especially in the northern stations of MAR and LYK, reaching 44% and 53% respectively between the 5-year periods 1998– 2002 and 2003–2005. In GEO there was an increase during the first periods that turned to decrease during the last two 5-year periods. Finally SMI station presented a steady increase of CO emission during the first two periods, followed by a steep increase during 1998–2002 and an equally large decrease during the last period (2003–2005). The emission changes for the last two periods were quite inconsistent. A closer examination of the original data for SMI station revealed that more than 60% of the maximal (upper 5%) values of that 5-year period were recorded in 2000, indicating a change in the emission sources of the area. Also, the station was moved to a nearby location at the end of 2000 and as a result CO levels after 2000 were almost half of their values before 2000 (MinEnv, 2003). These findings indicate a significant violation of the rollback equation assumptions therefore the last two periods of CO from SMI station were excluded from further discussion (Fig. 2a). NO2 emissions continuously decreased in the center of Athens (ATH, PAT) since 1993. Similarly, the stations of PER and LYK (urban-background and suburban respectively) presented, more or less steadily, decreasing emissions (by 4–5% and 13– 15% respectively) throughout the examined period. On the other hand, the urbanbackground station of SMI presented a significant increase, especially for the period 1987–1992 which was continued till 2002 and a small decrease in 2003–2005, possibly as a result of the station's relocation mentioned above. The reason of this increase is probably the construction of a new highway during that period which connected the

P. Kassomenos et al. / Environment International 36 (2010) 281–289

Fig. 2. Percentage of emission changes per 5-year period in Athens, for (a) CO, (b) NO2, (c) SO2, and (d) BS.

Fig. 3. Emission reduction required for compliance with EU limits, per 5-year period in Athens, for (a) CO, (b) NO2, (c) SO2, and (d) BS.

287

288

P. Kassomenos et al. / Environment International 36 (2010) 281–289

center of Athens with its southern suburbs. In MAR a decrease of about 14% of the NO2 emissions was observed between 1988–1992 and 1993–1997, followed by increasing emissions till 2005. The latter is due to the re-designing of the area (construction of new highways etc.) for the 2004 Olympics that changed the emissions regime in the neighbor of MAR station. PE1 station presented the same but weaker behavior. Finally the suburbanbackground station of LIO and, to a lesser extent, the suburban-industrial station of GEO, presented increasing emissions throughout the entire period yet this increase was declining over the years. Both stations are fairly close to major traffic arteries with increased share of heavy duty vehicles due to the industrial activities located in these or adjacent areas; especially LIO station is located in an area with massive development over the past 20 years. Therefore, the increasing emissions could be the result of the increasing traffic volume, and in the case of LIO station to increasing industrial activity (Fig. 2b). SO2 emissions seemed to decrease throughout the examined period, at an accelerating rate in PE1, GEO, ATH stations and more steadily in MAR, LIO and PER stations (about 25% between consecutive 5-year periods). PAT station had increasing SO2 emissions between the first two periods (by 45%) that were turned to decreasing later on. Only in SMI station there was an increase estimated till 1997, followed by an abrupt emission reduction during the two remaining periods (Fig. 2c). BS emissions were initially reduced between the 1983–1987 and 1988–1992 periods in all stations, especially in PAT and PE1. The rate of reduction was greatly reduced for the 1993– 1997 period, actually turning to increase in PE1 station, and in the last period (1998–2000) all stations presented increasing BS emissions. As already discussed, neither central heating nor industrial related BS emissions increased during that last period, whereas traffic related emissions were steadily rising throughout the period following the general increase of the number of vehicles in Athens (MinEnv, 2002) (Fig. 2d). Finally, for each 5-year period, we estimated the percentage of emissions reduction required to achieve compliance with the EU air quality limits listed in Table 1 (Fig. 3). These estimates should at least be considered as indicative and could assist in developing an emission control strategy. According to Fig. 3a, at the beginning, CO emissions had to be significantly reduced — ranging from 15% to 68% reduction — in all stations but SMI. The adopted emission control measures, after 1992, seem to have worked rather well since for the last period (2003– 2005) only the stations in the center of the city still needed to have their emissions reduced in order to meet the EU directives, ATH station by 34% and PAT station by just 6%. The situation was practically the same for NO2. Again there was a great need for emissions reduction during the periods 1983–1987 and 1988–1992 (63%, 36% and 28% for ATH, MAR and PAT respectively, and approximately 10–20% for the rest of the stations) that was successfully met by the adopted measures. Only in the center of the city and, surprisingly the urban–background station of SMI, there is still a need for a small reduction in emissions (4%–6%) (Fig. 3b). PE1 is the only station exceeding SO2 limits nowadays, after a very long period of compliance with the standards (Fig. 3c). This recent need for an SO2 emission reduction by 2% may indicate a differentiation of the spatial distribution of emission sources during that period that would violate the assumptions of our estimation procedure. Finally for BS only PAT station had to reduce its emissions in order to meet the EU limit and that was in the periods 1983–1987 and 1988–1992 (Fig. 3d). Emissions of various air pollutants should be further reduced in Athens. Restrictive measures such as toll-controlled traffic zone limitations, and prohibitions related to the ownership and usage of older technology vehicles should be imposed. On the other hand, incentives such as the subsidized renewal of the traffic fleet and the replacement of diesel fuelled central heating burners by newer technology, preferably gas fuelled, ones, must also be adopted so as to create a positive public opinion on the issue. Finally the series of actions that aimed to the development of the transportation infrastructure, especially public transport, should be continued and complemented by initiatives for re-defining land use and claiming more green space and space for public activities.

5. Conclusions In this study fifteen theoretical distributions were selected to fit the observed distributions of hourly values of SO2, NO2, daily maximum 8-hour moving average values of O3 and CO, and daily BS, arranged in 5-year periods, from the air quality monitoring network of Athens. Initially we have selected the most appropriate PDF per site and pollutant using well-known goodness of fit statistics. It was found that Inverse Gaussian and Extreme value were the most frequently selected PDFs, followed by Beta General and Type V Pearson. The temporal evolution of the shape of the PDF curves was successfully linked to the history of air pollution emissions in the center, the northern and southern parts of the city, and provided indications on the actual effect of the emission control strategies adopted along the years. Furthermore, using simple interpolation on the parameters of the existing PDFs, we attempted to estimate the PDFs of the missing periods. Based on these PDFs we estimated the number of times that a concentration exceeds the air quality limit values set by European Union

(EU). The relative RMSE of the exceedances predicted by the fitted PDFs was less than 10% for all pollutants except for SO2. The number of exceedances of the respective limit values has been significantly reduced, after 1993 with the introduction of the catalytic technology, for all traffic related pollutants, and earlier with the reduction of sulfur in diesel, for pollutants related to sources involving diesel combustion. Finally the rollback Eq. (3) was used to illustrate the change of the emissions between successive periods, as well as the emission reduction needed for certain pollutants to comply with the existing air quality standards, at any period. The results obtained by this procedure can be very useful in terms of emission control planning and assessment of abatement strategies, yet they should be regarded with caution. Their validity depends on the satisfaction of the assumptions inherent in the rollback equation methodology, namely the inert nature of the pollutant, the similar meteorological conditions and the unchanged spatial distribution of the emission sources. Since fulfilling these assumptions in a constantly changing urban environment is doubtful, an attempt was made to improve the compliance of the data and consequently the quality of the results, by using 5-year periods that smoothed out possible meteorological extremes and, at least to an extent, changes in the emissions regime. A common pattern was an abrupt drop of the required reduction followed by a gradual decrease till no further emission reduction was needed. This pattern reflects the combined effect of the introduction of emission reduction measures and the gradual intensification of certain sources, such as traffic. Overall, the results of this work provide useful information for air quality management and can also be used to develop an air pollution control strategy in order to protect human health. Utilizing theoretical PDFs fitted to the actual data compensates, at least partly, for the possible bias caused by missing data and can also provide a probable estimate of the pollution levels for periods with no data or too much missing data. Once the temporal evolution of the pollutant PDF has been established, it can be imported into the rollback Eq. (3) to either evaluate the performance of adopted emission related measures, or to investigate the effect of various emission scenarios on the pollution levels. One of the strong points of this methodology is that is provides information on the entire range of the possible values of a pollutant under a certain scenario. On the other hand the quality of the results depends on how well the selected theoretical PDF fits the actual data, and, as already mentioned, on the compliance to the assumption of the rollback equation. Acknowledgement Authors would like to thank Air pollution and Noise Control Department of the Greek Ministry of Environment that kindly offered the air quality data used in this study. References Berger A, Melice JL, Demuth CL. Statistical distributions of daily and high atmospheric SO2 concentrations. Atmos Environ 1982;16:2863–77. CHMI. Annual tabular overview. Retrieved 03 May 2005 from http://www.chmi.cz/ uoco/isko/tab_roc/1999_enh/ENG/kap_01/komentar_1_4.html. 1999. Economopoulou AA, Economopoulos AP. Air pollution in Athens basin and health risk assessment. Environ Monit Assess 2003;80(3):277–99. EU Council Directive 1999/30/EC of 22 April 1999 relating to limit values for sulfur dioxide, nitrogen dioxide and oxides of nitrogen, particulate matter and lead in ambient air. Off J Eur Comm 1999;L 163/41. EU Commission Directive of 17 October 2001 amending Annex V to Council Directive 1999/30/EC relating to limit values for sulfur dioxide, nitrogen dioxide and oxides of nitrogen, particulate matter and lead in ambient air. Off J Eur Comm 2001;L 278/35. EU Directive 2002/3/EC of the European Parliament and of Council of 12 February 2002 relating to ozone in ambient air. Off J Eur Comm 2002;L 67/14. Gavriil I, Grivas G, Kassomenos P, Chaloulakou A, Spyrellis N. An application of theoretical probability distributions to the study of PM10 and PM2.5 time series in Athens, Greece. Glob NEST J 2006;8:241–51. Georgopoulos PG, Seinfeld JH. Statistical distribution of air pollutant concentration. Environ Sci Technol 1982;16:401A–16A. Hadley A, Toumi R. Assessing changes to the probability distribution of sulfur dioxide in the UK using a lognormal model. Atmos Environ 2003;37:1461–74.

P. Kassomenos et al. / Environment International 36 (2010) 281–289 Jakerman AJ, Taylor JA, Simpson RW. Modeling distributions of Air pollutants concentrations — II. Estimation of one and two parameter statistical distributions. Atmos Environ 1986;20:2435–47. Kambezidis HD, Weidauer D, Melas D, Ulbricht M. Air quality in the Athens basin during sea breeze days using laser-remote-sensing technique. Atmos Environ 1998;32:2173–82. Kao AS, Friedlander SK. Frequency distributions of PM10 chemical components and their sources. Environ Sci Technol 1995;29:19–28. Kassomenos P, Sindosi O, Lolis C, Chaloulakou A. On the relation between seasonal synoptic circulation types and spatial air quality characteristics in Athens, Greece. J Air Waste Manag Assoc 2003;53:309–24. Longhurst JWs, Beattie CI, Chatterton TJ, Hayes ET, Leksmono NS, Woodfield NK. Local air quality management as a risk management process: assessing, managing and remediating the risk of exceeding an air quality objective in Great Britain. Environ Int 2006;32:934–47. Lu H-C. The statistical characters of PM10 concentration in Taiwan area. Atmospheric Environment 2002;36:491–502. Lu H-C. Comparisons of statistical characteristic of air pollutants in Taiwan by frequency distribution. J Air Waste Manag Assoc 2003;53:608–16. Lu H-C. Estimating the emission source reduction of PM10 in central Taiwan. Chemosphere 2004;54:805–14. Lu H-C, Fang G-C. Estimating the frequency distributions of PM10 and PM2.5 by the statistics of wind speed at Sha-Lu, Taiwan. Sci Total Environ 2002;298:119–30. Lu H-C, Fang G-C. Predicting the exceedances of a critical PM10 concentration — a case study in Taiwan. Atmos Environ 2003;37:3491–9. Mage DT, Ott WR. An evaluation of the methods of fractiles, moments and maximum likelihood for estimating parameters when sampling air quality data from a stationary lognormal distribution. Atmos Environ 1984;18:163–71.

289

Miles GH, Jakeman AJ, Bai J. A method for predicting the frequency distribution of air pollution from vehicle traffic, basic meteorology, and historical concentrations to assist urban planning. Environ Int 1991;17:575–80. MinEnv. Audits of moving and stationary emission sources — field results for 2005. Retrieved 20 November 2009 from http://www.minenv.gr/4/41/00/etos.2005.auto.pdf. 2006 (in Greek). MinEnv. Section for vehicles and external combustion — summary report for 1994– 2001. Retrieved 20 November 2009 from http://www.minenv.gr/ek_a/g000.html. 2002 (in Greek). MinEnv. Air quality report for Athens 2002 Part 1. Retrieved 20 November 2009 from http://www.minenv.gr/4/41/download/report-2002/report-2002-part1.doc. 2003 (in Greek). MinEnv. General Secretariat of Public Works, Center of Traffic Management, personal communication, 2009. Morel B, Yen S, Cifuentes L. Statistical distribution of air pollutant applied to the study of the particulate problem in Santiago. Atmos Environ 1999;33:2575–85. MinEnv. Air quality report for Athens 2006. Retrieved 01 August 2007 from http:// www.minenv.gr/4/41/000/atmosferiki.rypansi.athina-2006.PDF. 2007 (in Greek). NARSTO. Compendium of measurement methods. Retrieved 03 May 2005, from http:// cdiac.esd.ornl.gov/programs/NARSTO/compendium.htm. 2005. Ott WRA. A physical explanation of the lognormality of pollutant concentrations. J Air Waste Manag Assoc 1990;40:1378–83. Palisade Corp. Guide to using BestFit distribution fitting for Windows. NY, USA: Palisade Corp.; 2004. Retrieved 03 May 2005, from http://www.palisade-europe.com. Paschalidou A, Kassomenos P. Comparison of air pollutant concentrations between weekdays and weekends in Athens, Greece for various meteorological conditions. Environ Technol 2004;25(11):1241–55.