Jun 30, 1986 - distributions of air quality data were well reproduced by employing the generalized gamma distribution. By restricting the shape-parameter.
The Science of the Total Enuironment, 61 (1987) 91-105 Elsevier Science Publishers B.V., Amsterdam ~ Printed
91 in The
Netherlands
APPLICATIONS OF THE GENERALIZED GAMMA DISTRIBUTION TO AIR POLLUTION PROBLEMS: THE CASE OF VENICE
IRMA
LAVAGNINI
Environmental Venice (Zta1.y)
Science
DAR10
CAMUFFO
National
Research
(Received
May
Department,
Council 2nd,
University
CNR-ICTR,
1986; accepted
Corso June
30th,
of Venice,
Stati
Uniti
Calle
Larga
4, I-35020
S. Marta
Padua
2137,
l-30123
(Italy)
1986)
ABSTRACT Ground level concentrations of sulphur dioxide in the Venetian area of Italy we.re analyzed from a receptor-oriented point of view with reference to the seasonal evolution of the atmospheric stability over a coastal site. The observed frequency distributions of air quality data were well reproduced by employing the generalized gamma distribution. By restricting the shape-parameter in the adopted distribution function to integer values the main features of the considered region could be modelled in a simple and unified way. If the shape-parameter was allowed to assume both integer and half-integer values a more refined representation was achieved, in which the heterogeneous behaviour of the region was characterized mathematically. Economically sensitive problems such as land use planning may be possible applications of these results. INTRODUCTION
Venice is on the northern coast of the Adriatic Sea and is subjected to many dynamic effects due to the local topography. Firstly, the sea and land breezes are complicated by an interaction with the northern Alpine chain. Secondly, the orographic reliefs cause disturbances to the general circulations and particularly cause Bora inflows when the southern Scirocco passes over the Alps or when either a high or low pressure approaches the Alps from the west (Camuffo, 1981, 1982; Camuffo et al., 1979, 1980). Finally, the difference in temperature and roughness between sea water and mainland causes internal boundary layers to be formed over the industrial area or over Venice on the opposite side of the lagoon. The large buildings and the local heat sources in the industrial area cause additional thermal and mechanical turbulence. This scenario shows clearly the extremely complicated fate of emissions and how difficult it is to model the ground level concentrations and to evaluate the emission strategy on a theoretical basis. Therefore, since a deterministic model seems to be rather unrealistic, an empirical receptor-oriented approach seems to be the more convenient for adoption for this site. In the recent past some investigations have been carried out on sulphur
0048.9697/87/$03.50
(1: 1987 Elsevier
Science
Publishers
B.V.
92
dioxide dispersion in the atmosphere of the Venetian area. These studies have been mainly focused on transport models (Gaussian, Plume and K-models) aiming at meaningful correlations of sources and receptors (Runca et al., 1976, 1978; Zannetti et al., 1977; Marziano et al., 1979; Lavagnini et al., 1982). Among the results a noteworthy disagreement between simulated and experimental data, mainly due to the lack of thorough knowledge of source field conditions and of local meteorology of the sea-land transition area, has been shown. Time series analysis, Fourier analysis and statistical analysis have also been performed using, in some instances, only limited amounts of both meteorological and pollution data (Zannetti, 1976; Mattioli, 1977; Finzi et al., 1979). Over recent years statistical aspects of air quality data in the Venetian area have been widely investigated in order to select an appropriate probability model for the data collected by different monitoring stations in the region (Lavagnini et al., 1984; Buttazzoni et al., 1986; Marani et al., 1986). After testing several candidate models it has been shown that the generalized gamma distribution may be a useful tool in air quality data analysis. The aim of this paper is to furnish a practical statistical description of the Venetian region, which has potential for use with land use models, where long-term and large-scale information as well as detailed space-time small-scale analysis are required. We have been led to the present analysis by the fact that the observed frequency distribution of air quality data in the region of interest, namely the generalized gamma distribution (Marani et al., 1986), has also been obtained by Lienhard (1964) in the prediction of the dimensionless unit hydrograph. The fact that the form of the observed statistical distribution appears to closely approximate the molecular speed distribution of the Maxwell-Boltzmann theory suggested to Lienhard that the theoretical hydrograph distribution may be derived by considering a minimum of physical details of the watershed properties using Boltzmann statistics. In that description the generalized gamma distribution of rainfall run-off from a watershed may be characterized by a shape-parameter equal to 2 or 3, depending upon the shape of the particular watershed, and by a power-parameter equal to 2. The present inquiry is concerned with the hypothesis that the area considered here could be similarly characterized by generalized gamma distributions, which show a few integer or half-integer shape- and power-parameter values. The atmospheric pollution data used in this research are the 24 h averaged sulphur dioxide concentrations recorded between 1 January and 31 December 1982 at 21 stations covering the urban centers of Venice, Marghera and Mestre, and the surrounding industrial area of Porto Marghera (Zilio Grandi et al., 1977) (Fig. 1). The industrial area of Porto Marghera covers about 22 km2, and its main activities are concerned with oil-refining, petrochemicals production, production of electric energy, and metallurgy. In the industrial area there are sixteen stations, which we have numbered from 1 to 16. Stations 17 and 18 are located in the urban centers of Marghera and Mestre, respectively, which are situated on the mainland and close to the industrial zone. The urban area of Venice is in the middle of a lagoon and 5 km distant from the mainland. Sulphur dioxide pollution levels in Venice were recorded at Stations 19, 20 and 21.
93
Fig.
1. Industrial
and urban
REPRESENTATION VENETIAN REGION
areas
of Venice
OF AIR QUALITY AS A CASE STUDY
including DATA
the monitoring IN
network.
NON-HOMOGENEOUS
SITES:
THE
Although the log-normal distribution has generally been used to represent air pollution concentration frequency distributions (Benarie, 1980), other models, such as the exponential (Barry, 1971), gamma (Bencala et. al., 1976; 13erger et al., 1982) or Weibull model (Buttazzoni et al., 1986), have proved successful in reproducing the data of atmospheric pollution. For the Venetian area a comparison of normal, log-normal, gamma, Weibull and beta distributions indicated that the Weibull and gamma functions provide satisfactory fits of the observed sulphur dioxide concentration distributions, while the lognormal distribution does not (Lavagnini et al., 1984). In order to represent the considered area by means of a unifying statistical distribution the generalized gamma distribution has been investigated and appears to be a useful probability model for this purpose (Marani et al., 1986). The generalized gamma distribution, introduced by Stacy (1962), is a threeparameter distribution with probability density function of the form
where x > 0, fi > 0, and q > 0. This model includes the (x = [I = l), Weibull (x = /J), and gamma (fl = 1) distributions
exponential as special
94
cases. The graphic of f(x; CC,/?J,q) can take a variety of shapes by changing the value of the parameter 3. At x = 0, function (1) vanishes, is finite and non-zero, or is unbounded according to whether x exceeds, equals, or is less than unity. It is also worth noting that function (1) has a unique smooth maximum when a > 1. In order to determine the values of the parameters of the distribution so that the distribution fits the data in an optimal manner the well-known method of maximum likelihood can be employed. The method consists of evaluating the parameters X, 8, and q so as to maximize the likelihood function, defined as the joint density of the observations x1, J,,
The maximum likelihood estimates of the parameters are the solutions of the likelihood equations obtained by equating to zero the first partial derivatives of the likelihood function with respect to the parameters. Since these differentiation techniques fail to yield equations which can be solved algebraically the correct values of the parameters must be found by iterative methods. In the applications to air quality data collected in the area of Venice the parameter values of the generalized gamma distribution, computed by employing the method of maximum likelihood, often exhibit slight differences from one station to another. This holds, in particular, for the estimates of the shape-parameter r. The problem then naturally arises as to whether a suitable value for this parameter could be simultaneously adapted for some stations in order to provide a unified characterization for a group of monitoring stations. A first attempt toward this goal is reported in this paper, where each yearly sample is analyzed separately for two different values of r, which we have taken as 1 and 2. Starting from a fixed value of the shape-parameter CX,we allow the power-parameter [j to assume the values 0.5,1,1.5 and 2, and determine the scale parameter yeby means of the equation
which is obtained by equating to zero the partial derivative of the natural logarithm of function (2) with respect to the parameter q itself. The parameter q can thus be obtained in an explicit form, after substitution of the fixed values for c( and 8. For each given set of data we compare how closely the estimated generalized gamma distributions fit the data by computing the level of significance of the chi-square statistics
Here N is the total number of classes into which the data are grouped, PO, and Pe, represent, respectively, the observed and the expected proportion of data falling in the ith class, and IZ is the total number of observations in the
95
data set. The values of the parameters X, fi and II, for which the corresponding level of significance of the chi-square statistics P(x’) has a maximum, are chosen. Because of the above drastic simplification in the range of possible values for 2 and fi, the adopted approach gives a fairly homogeneous, albeit non-detailed, description of the 21 monitoring stations, thereby providing a global representation of the considered industrial zone, which may be useful in, for example, land-use planning assessments or in long-term forecasting. In order to test the effectiveness of control actions, which could be adopted to reduce the emissions, a more refined representation of the air quality data may be required. In particular, one would like to ascertain how the behavior of each station depends upon its location and the season. For this purpose we divided the data collected at each station into four seasonal groups, and we allowed both the shape and the power parameters to take both integer and half-integer values over a wider range. For the seasonal subdivision of air quality data for the Venetian area we chose the following periods: a winter period from December to February; a spring period from March to May; a summer period from June to August and an autumn period from September to November. The possible values for cxand /j were allowed to vary from 0.5 to 4.0 ‘I’ABLE
1
MLE parameters chi-square level Station number
6
9 10 11 12 13 14 15 16 17 18 19 20 21
of the generalized of significance P(x’)
gamma
distribution
for
each
station
and
thee associated
3L
Ir
‘1 (ppb)
W)
1.26 1.54 1.63 2.17 1.08 1.72 0.75 1.17 1.40 0.89 1.33 3.17 1.72 3.42 2.24 2.05 1.12 0.98 1.45 1.91 1.15
1.70 1.50 0.77 0.70 0.81 2.27 1.87 1.46 0.74 2.15 1.32 0.60 1.44 0.40 1.29 1.53 1.61 0.80 0.59 0.36 0.37
40.2 31.4 9.4 7.6 12.8 62.9 42.6 30.6 9.5 59.7 28.5 1.4 32.4 0.1 17.0 35.0 38.3 8.8 3.1 0.1 0.3
0.66 0.14 0.31 0.22 0.005 0.16 0.98 0.15 0.02 0.90 0.01 0.42 0.29 0.28 0.32 0.39 0.70 0.13 0.09 0.12 0.19
96
on the basis of previous estimations (see Table 1). By computing the scale parameter yeby means of (3), each seasonal sample may be characterized again by that set of parameters for which P(x”) is a maximum. DATA
ANALYSIS
AND
COMMENTS
Large-scale analysis In order to describe air quality data by means of function (1) the data collected at each station during 1982 have been analyzed. Three of these data samples (Stations 6, 11 and 19) have been referred to in a previous paper (Marani et al., 1986), where a preliminary test of the ability of this function to reproduce atmospheric pollution data was performed. The first estimates of the parameters Z, b and q have been obtained using the method of maximum likelihood for ungrouped data, where the NelderrMead (1965) simplex procedure was applied to perform the log-likelihood function maximization. The maximum likelihood estimates (MLE) of the parameters are listed in Table 1, together with the chi-square level of significance P(x”) for each station. One can observe that, with the exception of Station 5, which has TABLE
2
Best values of the parameters I and integer and half-integer Station number
CC,/II, and q and the level of significance values for [I are chosen
P(y’)
when
integer
values
for
P
‘I (mb)
w
1
1.5
42.5
0.13
2
1 0.5
14.5 1.4
0.27 0.35
0.5
2.2 19.9
0.29 0.01
1.5
42.0 30.9
0.14 0.64
34.8 24.8 45.4
0.20 0.09
9 4 5 6
1 .5 1.5
7 8 9
1
10 11 12
1 2
13 14 15 16 IT 18 19 20 21
1.5
2 2 2 2
1.5
1 1 0.5 1 1.5 1.5
39.1 12.6
0.91 0.14 0.27
16.8 1.9 12.3 34.9 39.0
0.20 0.39 0.03 0.70 0.5.5 0.04 0.24 0.35 0.33
1 1 1 1
0.5
12.5 18.0 2.0
1
0.5
1.7
1
)
97
a poor level of significance, P(,$) = 0.005, all the experimental data sets agree fairly well. From this table it also appears that for some stations the MLE parameters r and b are similar (see, for instance, Stations 1 and 17. or Stations It2 and 14). The values of the shape- and power-parameters fall in a narrow interval, from 0.89 to 3.42 and from 0.36 to 2.27, respectively; the scale parameter q., however, has a wide range of values. The stations can thus be roughly divided into two groups by approximating a to 1 and to 2. The stations corresponding to these values of z can be further classified according to the values of fl; more precisely, Table 1 shows that this parameter is widely spread, so that it cannot be expressed by a unique value. Therefore, we will consider for fl the discrete approximations 0.5, 1, 1.5 and 2. According to these considerations we have therefore modeled the distributions, assuming for each station the selected discrete values and determining the unknown parameter q as illustrated in the previous section. The values of c~,fl and q which give the best fit to the data for each station, as well as the corresponding level of significance, are reported in Table 2. Comparison between the last column of this table and the corresponding column in Table 1 indicates that, in some instances, the data fit the new distributions more closely. On the basis of the goodness-of-fit criterion adopted here, a distribution with some parameters fixed fits the data better than the same distribution with all parameters free. A possible explanation for this is that the maximum likelihood estimation of the parameters uses ungrouped data, whereas the level of significance of the chi-square statistics is computed using grouped data. It is also worth observing that c( assumes the values 1 both for stations located in urban areas (Stations 17, 18, 19, 20 and 21) and for stations in the industrial area (Stations 1, 5, 7, 8, 9, 10 and ll), whereas the value 2 only characterizes stations in the southern and western interior of the industrial area. This is consistent with the fact that when c( = 1 the distribution is decreasing when the concentration increases, whereas a unimodal ‘curve corresponds to 2 = 2. In other words, the inner group of monitoring stations is characterized by a low probability of negligible concentration values, and the peripheral stations by a higher probability of low concentration values. Small-scale analysis In order to gain a deeper insight into the physical behavior of the monitoring stations a refined analysis has been performed, as discussed in the previous section of this paper. How well the data fit the distributions obtained in this way is shown in Fig. 2, which shows the isopleths of the level of significance P(x’). This figure shows that the generalized gamma distributions fit the experimental data almost satisfactorily. In winter, good agreement is obtained over a large portion of the industrial area and in the city of Venice (Fig. 2a). In spring, the highly variable meteorological conditions cause poor fits for a large number of stations, which are
:a)
0
1
2km
0 I
1 I
2km I
b)
L
J
largely located in the interior of the industrial area (Fig. 2b). Figure 2c shows that, in summer, the area under consideration is divided into two zones ~ with a lower agreement of fits to the data -which are separated by a large, central strip, running from west to east. Finally, Fig. 2d refers to autumn and presents features similar to those in Fig. 2a.
0 I
1 I
2km I
(/ c Fig. 2(a)~ (d). Seasonal isopleths (winter, spring, summer and autumn) of the chi-square significance P(,y’), obtained in fitting the generalized gamma distributions to 24 h averaged dioxide concentrations, recorded during 1982 at the stations covering the region of Venice. 0.80.
level of sulphur (o)P(;(‘)
b)
0 I
1 I
2km I
The seasonal isopleths for the parameter 2 are shown in Fig. 3. Winter (Fig. 3a) is characterized by a broad island for the highest values of x in the middle of the industrial area. This island is encircled by wide lower-value areas in the southeastern as well as in the northern part of the zone, whereas, in the western part, values of r < 1 are rapidly reached. In passing from winter to
101
0
1
2km
c c--x”.5
(ill ,
Fig. 3(a) (d). Seasonal isopleths (winter, spring, summer and autumn) for the shape-parameters mu of the generalized gamma distribution which fits the 24 h averaged sulphur dioxide concentrations with the highest levels of significance P(x’). The reference data were collected in the Venice region during 1982.
102
spring (Fig. 3b) a completely different pattern of isopleths is found, since three small islands for higher values of x are observed. Lower values characterize both the surroundings of the industrial area in its northern and eastern sections, and the interior of the industrial area itself. In summer (Fig. 3c) the highest values for 2 appear in Venice and in the southwestern part of the industrial area, while the bulk of this area shows low values. Figure 3d shows that autumn is similar to spring, with three islands corresponding to the highest values of x, and lower values in the urban centres. The largest island is again in the southern part of the industrial zone, and is larger than in the spring period. A comparison between Figs. 2 and 3 shows that the highest values for P(x’) occur at those stations which are characterized by the lowest values for 2. This indicates that the generalized gamma distribution is more efficient in representing samples for which the lower concentration values are more frequent. The spatial variation of parameter a can be related to the meteorological conditions over the region. The most outstanding feature can be seen from inspection of the two extreme situations which occur in the winter and summer periods. In winter the atmosphere is more stable over the mainland, but unstable over the warmer water of the lagoon. The prevailing winds blow from the NW, NE and SW. The winds from the NW and SW transport the pollutants, in a generally stable atmosphere, towards Venice. During the passage over the warm lagoon, the cold air masses become unstable causing fumigation. The heat over the city is responsible for a higher concentration of pollutant in the urban environment. Domestic heating is not a local source of sulphur dioxide since such heating is by means of natural gas. The dispersion of pollutants over the mainland is thus severely regulated by the stable atmosphere so that the pollution episodes are usually due to low level emissions. Contributions from tall stacks occur more rarely and are mainly limited to frontal situations, when the dynamic instability causes the plumes to reach ground level, or may occur during fresh winds. In these cases high concentrations can be reached at ground level over the mainland. This pollution pattern is fairly well represented by the computed shape-parameter, which assumes high values in Venice and in the bulk of the industrial region, especially because of the convective activity generated by the heat sources, and low values in the peripheral areas (Fig. 3a). In summer the atmosphere is more unstable over the mainland and the sea-breeze is from the SE and veers during the day. During the night two nocturnal breezes, from the NE and NW, transport the pollutants; the former over the stable mainland, the latter over the unstable lagoon and Venice. Both these conditions influence pollutant dispersion, resulting in lower values for x in the inner part of the industrial area. Figure 3c shows the combined effect of the NW nocturnal breeze and the fumigation caused by the warm lagoon water and heat island over Venice. During spring and autumn the concentrations at Venice are strongly reduced by the frequent rainfall which occurs in these seasons, causing a washout of pollutants from the atmosphere.
103
0’
Fig. 4. Seasonal trend of the shape-parameter data collected by Station 1.
a of the generalized
gamma
distribution
fitting
the
The above-mentioned seasonal trend in atmospheric stability is responsible for the seasonal trend of the shape-parameter cx at each station. Two basic patterns are shown in Figs 4 and 5. The former shows a minimum value in the warm season, with a better dispersion of pollutants in that period (Stations 1, 2, 7, 8, 9 and 18); the latter is characterized by a minimum value in the cold season (Stations 13, 16 and 19). The other stations show a seasonal variation which may be considered as a superimposition of these two patterns. The seasonal shape is a consequence of the distribution of low level emissions and tall stacks in the vicinity of the monitoring stations with reference to the prevailing winds.
0
Fig. 5. Pattern of seasonal values of the shape-parameter which fits the sulphur dioxide concentrations recorded
r for the generalized at Station 16.
gamma
distribution
104 CONCLUSIONS
In an analogy with hydrological applications, this analysis shows that the data from all the monitoring stations for sulphur dioxide in the Venice area can be represented by means of the generalized gamma distribution with the shapeparameter having only a few discrete values (integer or half-integer). This simplified representation, although with poorer levels of significance, has many advantages, because the pollution effects of an industrial area may be homogeneously described over a wide region. This representation may be convenient when employed in land-use planning assessments. On the other hand, when one wishes to describe the ground pollution levels, or to model the consequences of emission policy, a more detailed knowledge of concentration distributions is required. The situation obtained by optimizing the chi-square level of significance, which thus yields the values of parameters X, fl and ?I, seems to be suitable for such a goal. It was shown that the shape-parameter values vary with the season and account for the complex dynamics of the atmospheric stability in the coastal zone considered. In particular, from the analysis of the distribution functions for the concentration frequencies which are seasonally observed at each station, two basic patterns can be observed. They correspond to stations with the minimum or maximum value for the parameter 2 in the warm season, all other stations being characterized by a seasonal trend which may be regarded as a superposition of the two basic patterns. This is a consequence of the distribution of low level emissions and emissions from tall stacks in the vicinity of the monitoring stations with reference to the prevailing winds. The parameter values of the two extreme distribution functions relate to the different physical features of the corresponding stations and represent the heterogeneous behaviour of the zone in a mathematical and quantitative way. ACKNOWLEDGEMENTS
The authors would like to thank Ente Zona Industriale of Porto Marghera for granting access to the data recorded by their monitoring network. This study was supported jointly by CNR (Progetto Finalizzato Energetica 2) and MPI. REFERENCES Barry, P.J., 1971. Use of Argon-41 to study the dispersion of stack effluents. Proceedings of the Symposium on Nuclear Techniques Environmental Pollution, Salzburg. IAEA. Austria ST1 Publ. 268, pp. 241l253. Benarie, M., 1980. Urban Air Pollution Modelling. Macmillan, London, 405 pp. Bencala, K.E. and J.H. Seinfeld, 1976. On frequency distributions of air pollutant concentrations. Atmos. Environ., 10: 941-950. Berger. A., J.L. Melice and C.L. Demuth, 1982. Statistical distributions of daily and high atmospheric SO,-concentrations. Atmos. Environ., 16: 286332877.
105
Buttazzoni, C., I. Lavagnini, A. Marani, F. Zilio Grandi and A. Del Turco, 1986. Probability models for atmospheric sulphur dioxide concentrations in the area of Venice. J. Air Pollut. Control Assoc., 36: 1028 1030. Camuffo, D., 1981. Fluctuations in wind direction at Venice, related to the origin of the air masses. Atmos. Environ., 15: 1543 1551. Camuffo, D., 1982. The sea breeze at Venice, as related to daily global radiation. Boundary-Layer Meteorol., 23: 175184. Camuffo, D. and L. Cavaleri, 1980. Oscillations in pollutant concentration occurring in cold offshore flow over Venice. Atmos. Environ., 14: 1225 1262. Camuffo, D.. F. Tampieri and G. Zambon, 1979. Local mesoscale circulation over Venice as a result of the mountain sea interaction. Boundary-Layer Meteorol., 16: 83- 92. Finzi, G., P. Zannetti, G. Fronza and S. Rinaldi, 19’79. Real time prediction of SO, concentration in the Venetian lagoon area. Atmos. Environ., 13: 12491255. Lavagnini, I., A. Marani, C. Buttazzoni, A. Del Turco and F. Zilio Grandi, 1982. 1Simulazione dell’inquinamento atmosferico nella zona di Marghera Venezia. Atti de1 X Convegno Nazionale Ambiente e Risorse, Bressanone, 6 11 Settembre. Universita’ di Padova: A6. Lavagnini, I.. C. Buttazzoni and A. Marani, 1984. Confront0 di modelli di distribuzione statistica dei dati di qualita’ dell’aria. Acqua-Aria, 6: 551 557. Lienhard, J.H., 1964. A statistical mechanical prediction of the dimensionless unit hydrograph. J. Geophys. Res., 69: 5231 5238. Marani, A., I. Lavagnini and C. Buttazzoni, 1986. Statistical study of air pollutant concentrations via generalized gamma distributions, J. Air Pollut. Control Assoc., in press. hlarziano, G.L., C.C. Shir, L.J. Shieh, A. Sutera, L. Gianolio and M. Ciprian, 1979. Study of the SO, distribution in Venice by means of an air quality simulation model. Atmos. Elnviron., 13: 477 487. Rlattioli, F., 1977. Spectral analysis of wind and SO, concentration in the Venice area. Atmos. Environ. 11: 113- 122. Nelder. J.A. and R. Mead, 1965. A simplex method for function minimization. Comput. J ., 7: 308 313. Runca, E., P. Melli and P. Zannetti, 1976. Computation of long-term average SO, concentration in the Venetian area. Appl. Math. Modelling, 1: 915. Runca, E., P. Zannetti and P. Melli, 1978. A computer-oriented emissions inventory procedure for urban and industrial sources. J. Air Pollut. Control Assoc., 28: 584 588. Stacy, E.W., 1962. A generalization of the gamma distribution. Ann. Math. Statist,, 33: 118771192. Zannetti, P., 1976. Analisi delle serie temporali di misure della qualita’ dell’aria in Venezia. Proc. of the “XIV Convegno Internazionale di Automazione e Strumentazione”, Milano, 23 24 Novembre 1976. FAST: 1~ 17. Zannetti. P., P. Melli and E. Runca, 1977. Meteorological factors affecting SO, pollution levels in Venice. Atmos. Environ., 11: 605616. Zilio Grandi, F. and R. Magri, 1977. La rete di monitoraggio dell’Ente Zona Industriale di Porte Marghera. Suppl. a Inquinamento, XIX, 6: 31-36.