Research Journal of Recent Sciences _________________________________________________ ISSN 2277-2502Vol. 2(4), 36-45, April (2013) Res. J. Recent Sci. International Science Congress Association 36 Estimation of Air Pollution in Urban Streets by Modeling of PM10, O and CO Pollutants according to Regression Method (Case study-Yadegar and Azadi streets intersection, Tehran, IRAN) Yousefi Kebria D., Darvishi G.2,* and Haghighi F.Department of Civil Engineering, Babol University of Technology, Babol, IRAN Graduate Faculty of Environment, Babol University of Technology, Babol, IRANAvailable online at: www.isca.in Received 5th January 2013, revised 14th January 2013, accepted 19th February 2013Abstract Nowadays, the growth of civilization rate, expansion in use of vehicles and development of economic activities all leads to increase of urban air pollution. Daily increase of urban traffic and emission of various pollutants has confronted the human to severe environmental problems, so that, it’s effects are evident in the areas of physical health, psychological and economical losses. So, in order to reach to sustainable development and having clean air and transportations, we need a means to investigate the projects which are related to transportations with sufficient accuracy and numerically. In this study, the analysis and modeling of PM10, O and CO pollutants concentrations was investigated in Tehran. Parameters affecting the concentration of pollutants was specified according to weather and traffic statistics and the relation between various variables such as transportation, precipitation, maximum and minimum temperature and also relation between humidity and air pollution was investigated and finally, a model was presented according to regression method which will be capable to estimate the concentration of PM10, O and CO pollutants in city streets with appropriate accuracy for future years. one of applications of this model is assessment and prediction of production factors of these pollutants for manage and control tools. Keywords:air pollution, urban streets, weather variables, traffic, modeling, regression. Introduction Air pollution is one of major problems of human and its importance in increasing daily. Unlike past, not only poor air quality resulted of industrial and household heating systems, but also vehicles movement is main reason of air pollution. Since, for design of a strategy for reducing air pollution, it’s necessary to determine the air pollution sources and the amount of its emissions, and then control and manage it by an appropriate method, so, urban air quality studies are required. In Iran, by regarding of land usage, topography, traffic manner and so on, this issue is under discussion. Specification of various contaminants and adapting them with standards and finally practical solutions for reduction of air pollution, are some key parameters for specification and resolve of this problem in urban streets2,3. Currently, to achieve environmental goals and clarify the environmental issues for managers, preparing of environmental effects computational studies in traffic management designs and road design in order to control the air pollution has been under attention 4,5. So, various studies in most parts of the world by using various methods have been done. in Denmark, a research under the title of “real estimations of automobile exhaust pollution by measuring of air quality and streets pollution models” has been done which is concentrated on benzene among the other present contaminants. Benzene is in fuel and Denmark could reduce the amount of benzene from 0.38 g/km to 0.11 g/km by reduction of benzene in fuel and increase of vehicles powered by catalyst. in Taiwan, another research has been done under the title of “measurement and modeling of air pollution emissions in narrow urban streets” that measure the concentrations of SO, NO and CO contaminants and applied RNG K-4 model for modeling . Also in Russia, for estimation of air pollution emission in narrow urban streets, the semi-empirical and mathematical model and urban model (USM) was used and in addition to weather and traffic variables, the arrangement and density of the buildings around the street has been studied and the influence of high density and arrangement of buildings in air pollution and concentration of contaminants has been mentioned. Generally, modeling of air pollution in urban areas require of research in various scientific fields such as emissions science, fluid mechanics, meteorology, mathematic science and other sciences 9,10. Along with well done researches and application of models such as CFD, USM, RNG K-4, semi-empirical-mathematical model and so on, the regression model is also used in this study. Regression analysis is a statistic technique for investigation and conversion to model of relationships between variables11. Another step in regression analysis is so called “appropriate relationship of the model” which satisfies the Research Journal of Recent Sciences ______________________________________________________________ ISSN 2277-2502Vol. 2(4), 36-45, April (2013) Res. J. Recent Sci. International Science Congress Association 37 suitability of under study model and quality of fit 12. The results of relationships of the model help us to say that the model is defensible or correct the main fit. Finally, the general aim is presentation of a model which able us to estimate the air pollution of urban streets and hereby pay to management and proper control of air pollution. Material and Methods In this study, modeling was done based on various variables such as traffic, precipitation, maximum and minimum of temperature, maximum and minimum of humidity and the amount of pollutants, and also the relationship between traffic and weather variables and air pollution was determined. In Tehran metropolis, the intersection of two main streets (Yadegare-Emam St. and Azadi St.) was determined as place of study. The reason for selection of this place is that at present according to statistics of Tehran Air Quality Control Company, this place is considered as one of important parts of pollution and traffic13. Gathering of required data were done by measuring devices and some data sources. Meteorological data were taken from the weather measurement station of Mehr-Abad in Tehran. By placing of selected place (intersection of Yadegar and Azadi) within this station, the data obtained from Mehr-Abad station such as at various hours were used. Also the BABUK-A measurement device which has very sensible sensors was used to obtain data about the amount of pollutants. Measurement devices, in peak hours of traffic, were placed at two places of the intersection to make possible the accurate and comprehensive measurement of the amount of pollutants. Generally, during the gathering of information for this study, measurement of climate and traffic variables and emission measurement were done at peak hours (7 to 8 A.M. and 5-6 P.M.) according to statistics of Traffic and Transportation Company of Tehran 14. Results and DiscussionIn this study, the amount of air pollution was investigated for CO, Oand PM10 pollutants according to the climate and traffic conditions and relationship between them. Also, relationship between each of these pollutants and it’s meaningfulness with traffic and weather data were separately investigated and finally, a specific model for each pollutant was presented. The variables used in the models include is as follow: : the amount of traffic, X: the amount of precipitation, X: minimum temperature, X: maximum temperature, X: minimum humidity, X: maximum humidity, Y: CO, Y: O, :PM10. The model related to CO: For determination of optimum model, at first, the relation between various variables and amount of CO is determined and then, the amount of correlation coefficient is calculated according to statistics studies15,16. (table-1)The results obtained from correlation coefficient, according to level of significance (Sig0.06) showing that there is a meaningfulness relation between the amount of CO and traffic, maximum and minimum of temperatures and amount of precipitation. Then, the scatter plots of CO pollutant related to various variables are showed in figure1. By using of above diagram, only the linear model can be reviewed. (table-2) As can be seen, the results obtained from variance analysis, shows the meaningfulness regression of the variables. After the variance analysis, now, fitting of the model parameters is done (table-3). The results obtained from variables determination, according to level of significance (Sig0.06), shows that none of coefficients except the coefficient related to the traffic amount, are non-zero with confidence of more than 0.95%. But, it is possible to make a linear model between the amount of CO pollutant and maximum temperature with confidence of 0.93%. Now, by correction of coefficients, model is fitting (table 4). Finally, by regarding of results obtained from correction of coefficients, for CO pollutant, the linear model is obtained as follow: = 2.32 + 0.00073 X + 0.046 X4 (1) As can be seen, the obtained model shows that with a unit increase of maximum temperature, the amount of pollution related to CO pollutant is increases 0.046 unit, and also, with a unit increase in traffic amount, the amount of CO pollution increases 0.00073 units. After presenting of the model, finally, correlated or uncorrelated error and the amount of Watson - Camera index is investigated (table 5). The amount of coefficient of determination (R2) in above table shows that about 81% of CO pollutant pollution is described by defined variables in model. Also, the amount of Watson Camera index shows that errors are uncorrelated. Watson - Camera index confirms the Regression model assumption that the value is between 1.5 to 2.5. Validation of the model: After presenting the model, its validity in 2012 should be assessed to competence the model for estimation of the pollution in the coming years. This model is based on data of year 2011 and is set to estimate the pollution in year 2012 based on data that measured in same year. These data selected randomly from 4 different days of the year 2012 (table 6) Validation of the model shows that the model works relatively good. Research Journal of Recent Sciences ______________________________________________________________ ISSN 2277-2502Vol. 2(4), 36-45, April (2013) Res. J. Recent Sci. International Science Congress Association 38 (a) (b) (c) (d) Figure-1 Scattering of CO pollutant related to variables of (a) minimum temperature, (b) maximum temperature, (c) precipitation, (d) volume of traffic The model related to O3: For determination of optimum model of O pollutant, first, the relation between various variables and the amount of O is determined and then, according to statistic studies, the amount of correlation coefficient is calculated17. (table-7)The results obtained from correlation coefficients, according to level of significance (Sig0.06), shows that there is not any relation between O pollutant emissions and the amount of precipitation, but there is a meaningfulness relation between the amount of Oand the amount of traffic and maximum and minimum of temperatures and maximum and minimum of humidity. Then, the scatter curve of O pollutant related to other various variables is plotted according to figure-2.By using of above diagrams, only the second model can be reviewed (table-8). As can be seen, the results obtained from variance analysis, according to level of significance (Sig0.06) showing the meaningfulness of regression on variables. The meaningfulness of regression on variables means that at least one of the coefficients is none-zero. Now, this issue is investigated that which of the coefficients are none-zero (table-9). The results obtained from analysis of coefficients, according to level of significance (Sig0.06) showing the coefficients of maximum and minimum humidity, amount of traffic, and second-order of maximum humidity are became meaningful. Now, by correcting the coefficients and estimation of them by meaningful variables, model fitting is done (table-10). Finally, by regarding the results obtained from correction of coefficients, a model has been presented for O pollutant as below: = 19.069 – 0.026 X – 0.196 X + 0.0002 X + 0.001 X2 (2) As can be seen, obtained model shows that with a unit increase in minimum humidity, the contamination level of O3 pollutant Research Journal of Recent Sciences ______________________________________________________________ ISSN 2277-2502Vol. 2(4), 36-45, April (2013) Res. J. Recent Sci. International Science Congress Association 39 is decreased 0.026 units and with a unit increase in maximum air humidity, the contamination level of O pollutant is decreased 0.196 units. Also, with a unit increase in the amount of traffic, the amount of O pollutant is increased 0.0002 units. Now after presentation of this model, correlated or uncorrelated error and the amount of Watson - Camera index is investigated (table-11). Table-1 Determination of correlation coefficient related to CO pollutant low.t up.t m.rain low.wet up.wet Traffic CO 0.540 0.000 0.506 0.000 0.280 0.06 -0.075 0.62 -0.247 0.098 0.68 0.06 Correlation Sig.(2-tailed) Table-2 Variance analysis of CO pollutantCO Sum of Squares df Mean Square F Sig. Regression Residual Total 5.576 2 2.788 7.710 0.001 15.548 43 0.362 - - 21.123 45 - - - Table-3 Determination of coefficients related to CO pollutant CO Non-standardized Coefficients Standardized Coefficients t Sig. B Std. Error Beta (Constant) low.t up.t m.rain low.wet up.wet traffic0.510 0.960 - 0.532 0.598 0.000 0.048 -0.002 -0.005 0.996 0.078 0.042 0.866 1.839 0.073 0.016 0.011 0.215 1.442 0.157 0.012 0.007 0.369 1.727 0.092 0.003 0.008 0.099 0.380 0.706 0.056 0.001 0.852 2.157 0.036 Table-4 Corrected coefficients CONon-standardized Coefficients Standardize Coefficients t Sig. B Std. Error Beta (Constant) up.t traffic 2.321 0.549 - 3.587 0.001 0.046 0.012 0.508 3.880 0.000 0.00073 0.010 0.089 3.092 0.000 Table-5 Coefficient of determination CO R R Square Adjusted R Square Std. Error of the Estimate Durbin-Watson 0.1058 0.816 0.782 0.59765 1.643 Table-6 Validation results of CO   CO NO. 5.48 5.62 1 4.30 4.19 2 4.35 3.71 3 3.5 2.70 4 Research Journal of Recent Sciences ______________________________________________________________ ISSN 2277-2502Vol. 2(4), 36-45, April (2013) Res. J. Recent Sci. International Science Congress Association 40 Table-7 Determination Correlation coefficient for O pollutant low.t up.t m.rain low.wet up.wet traffic 0.742 0.000 0.783 0.000 -0.24 0.109 -0.668 0.000 -0.75 0.000 -0.532 0.048 Correlation Sig.(2-tailed) Table-8 Variance analysis of O3 pollutant 3 Sum of Squares Df Mean Square F Sig. Regression Residual Total 84.896 10 8.490 12.371 0.000 24.018 35 0.686 - - 108.914 45 - - - Table-9 Determination of variations related to O pollutant Non-standardized Coefficients Standardized Coefficients t Sig. B Std. Error Beta (Constant) low.t up.t low.wet up.wet traffic low.t2 up..t2 low.wet2 up.wet2 traffic222.829 5.852 - 3.901 0.000 0.139 0.151 0.515 0.920 0.364 0.047 0.155 0.228 0.301 0.765 -0.061 0.036 -0.862 -1.684 0.01 -0.270 0.076 -3.698 -3.567 0.001 -0.072 0.005 -2.502 -2.362 0.040 -0.014 0.010 -0.923 -1.368 0.180 0.000 0.004 -0.085 -0.111 0.912 0.000 0.000 0.420 0.956 0.346 0.002 0.001 3.141 3.371 0.002 3.296E-7 0.000 0.310 0.223 0.825 Table-10 Corrected coefficients Non-standardized Coefficients Standardized Coefficients t Sig. B Std. Error Beta (Constant) low.wet up.wet traffic up.wet2 19.069 1.271 - 15.000 0.000 -0.026 0.010 -0.367 -2.518 0.016 -0.196 0.034 -2.679 -5.732 0.000 0.0002 0.000 -0.191 -2.374 0.022 0.001 0.000 2.251 4.581 0.000 The amount of coefficient of determination (R) in above table shows that about 73.6% of O pollutant pollution is well defined by mentioned variables in model. Also, the amount of Watson - Camera index shows the that errors are uncorrelated. Validation of the model: After presenting the model, its validity in 2012 should be assessed to competence the model for estimation of the pollution in the coming years. This model is based on data of year 2011 and is set to estimate the pollution in year 2012 based on data that measured in same year. These data selected randomly from 4 different days of the year 2012 (table 12) Validation of the model shows that the model works very well. The model related to PM10: For determination of optimum model for PM10 pollutant, initially the relationship between various variables and the amount of PM10 is determined and then, according to statistical studies, the amount of correlation coefficient can be calculated (13,14). (table 13) The results obtained from correlation coefficients, according to level of significance (Sig0.06), shows that the amount of PM10pollutant only has a meaningful relation with the maximum and minimum of humidity, the amount of precipitation and maximum temperature. But it must be mentioned that, the relation between the amount of traffic and PM10 pollutant is accepted with confidence of 0.93%. Then, the curve of PM10pollutant relative to various variables is drawled according to figure 3. Research Journal of Recent Sciences ______________________________________________________________ ISSN 2277-2502Vol. 2(4), 36-45, April (2013) Res. J. Recent Sci. International Science Congress Association 41 (a) )b( (c) (d) (e) Figure-2 Scattering of O pollutant related to variables of (a) minimum temperature, (b)maximum temperature, (c) minimum humidity, (d) maximum humidity and (e) volume of traffic Research Journal of Recent Sciences ______________________________________________________________ ISSN 2277-2502Vol. 2(4), 36-45, April (2013) Res. J. Recent Sci. International Science Congress Association 42 (a) (b) (c) (d) (e) Figure-3 Scattering of PM10 pollutant related to variables of (a) maximum temperature, (b) precipitation, (c) minimum humidity, (d) maximum humidity and (e) volume of traffic Research Journal of Recent Sciences ______________________________________________________________ ISSN 2277-2502Vol. 2(4), 36-45, April (2013) Res. J. Recent Sci. International Science Congress Association 43 By application of above diagrams, the linear model is investigated and variance analysis is done (table 14). As can be seen, the results obtained from variance analysis, showing at least one of the coefficients is none-zero. Now, by determining of coefficients, this issue is investigated that which of the coefficients are none-zero (table 15). The results obtained from correlation coefficients, according to level of significance (sig�0.05), shows that the correlation coefficients of maximum and minimum of humidity and amount of traffic and minimum temperature have became meaningful. Now, the model is fitted by correction of coefficients (table 16). Finally, by regarding the results obtained from correction of coefficients, a linear model has been obtained for PM10 pollutant as below: = 201.054 – 2.571 X – 0.636 X – 0.904 X + 0.001 X1 (3) Obtained model for PM10 shows that with a unit increase in minimum temperature, maximum and minimum of humidity, the contamination level of PM10 is decreased 2.571, 0.636 and 0.904 units, respectively and with a unit increase in traffic, the contamination level of PM10 is increased 0.001 units. now after presentation of the model, correlated or uncorrelated error and the amount of Watson - Camera index is investigated (table 17).The amount of coefficient of determination (R) in above table shows that about 60% of PM10 pollutant pollution is well defined by mentioned variables in model. Also, the amount of Watson - Camera index shows that errors are uncorrelated. Validation of the model: After presenting the model, its validity in 2012 should be assessed to competence the model for estimation of the pollution in the coming years. This model is based on data of year 2011 and is set to estimate the pollution in year 2012 based on data that measured in same year. These data selected randomly from 4 different days of the year 2012 (table 18) Validation of the model shows that the model works very well.Table-11 Coefficient of determination R R Square Adjusted R Square Std. Error of the Estimate Durbin-Watson .858 0.736 0.710 0.83708 1.602 Table-12 Validation results of O Table-13 Determining of correlation coefficient for PM10 pollutant low.t up.t m.rain low.wet up.wet Traffic PM10 0.196 0.197 0.347 0.019 -0.290 0.054 -0.572 0.000 -0.542 0.000 -0.564 0.074 Correlation Sig.(2-tailed) Table-14 Variance analysis of PM10 pollutant PM10 Sum of Squares df Mean Square F Sig. Regression Residual Total 24998.994 5 4999.799 5.934 0.000 32860.983 39 842.589 - - 57859.977 44 - - - Table-15 Determining of coefficient related to PM10 pollutantPM10 Non-standardized Coefficients Standardized Coefficients t Sig. B Std. Error Beta (Constant) low.t m.rain low.wet up.wet traffic203.610 40.741 - 4.998 0.000 -2.743 1.222 -0.439 -2.244 0.031 0.178 0.587 0.046 0.303 0.763 -0.666 0.359 -0.404 -1.855 0.050 -0.935 0.430 -0.545 -2.173 0.036 -0.0010 0.011 -0.105 -2.047 0.042    O3 NO. 13.52 12.34 1 8.20 8.15 2 9.19 9.34 3 10.69 10.81 4 Research Journal of Recent Sciences ______________________________________________________________ ISSN 2277-2502Vol. 2(4), 36-45, April (2013) Res. J. Recent Sci. International Science Congress Association 44 Table-16 Corrected coefficients PM10 Non-standardized Coefficients Standardized Coefficients t Sig. B Std. Error Beta (Constant) low.t low.wet up.wet traffic 201.054 39.405 - 5.102 0.000 -2.571 1.072 -0.412 -2.399 0.021 -0.636 0.341 -0.386 -1.865 0.070 -0.904 0.413 -0.527 -2.188 0.035 0.0010 0.011 -0.112 -2.421 0.043 Table-17 Coefficient of determination PM10 R R Square Adjusted R Square Std. Error of the Estimate Durbin-Watson 0.826 0.614 0.537 28.69606 1.681 Table-18 Validation results of PM 10   PM NO. 91.02 100.07 1 38.82 42 2 61.37 64.65 3 103.84 100.64 4 ConclusionBy regarding of modeling results and according to the issue that the results obtained for each pollutant (according to traffic and weather variables) is different, studies have been conducted separately for each variable. After presentation of the model and investigation of relationship between variables and the amount of air pollution, it has been cleared that with a unit increase of maximum temperature, the amount of pollution related to CO pollutant is increases 0.046 unit, and also, with a unit increase in traffic amount, the amount of CO pollution increases 0.00073 units. These results showing that among 6 available variables, maximum temperature and traffic variables have the greatest impact on CO. About the O pollutant, with a unit increase in minimum humidity, the contamination level of O pollutant is decreased 0.026 units and with a unit increase in maximum air humidity, the contamination level of O pollutant is decreased 0.196 units. Also, with a unit increase in the amount of traffic, the amount of O pollutant is increased 0.0002 units. These results showing that the amount of traffic, maximum humidity and minimum humidity are influential variables on O. Also, for PM10 pollutant, the obtained results from model indicate that with a unit increase in minimum temperature, maximum and minimum of humidity, the contamination level of PM10 is decreased 2.571, 0.636 and 0.904 units, respectively and with a unit increase in traffic, the contamination level of PM10 is increased 0.001 units. Finally, by presenting of these models in this study, the influences of various parameters in the desired location and city, on the concentrations of CO, O and PM10pollutants are obtained and these models are capable to estimate the air pollution resulted from traffic.Validation of the model shows that the model works very well and modeling by regression method is suitable and works well for the streets of Iranian cities. Finally, the overall conclusions of this study are as below: Regression analysis is a suitable method for investigation of relation between air pollution and variables. Regression analysis is a statistical method for investigation and modeling of relation between variables and one of the main objectives is the estimating the parameters of model. The main influential variable on the amount of various pollutants concentrations in urban streets is traffic volume. By prediction and estimating of pollutants in urban streets, management and better control of air pollution can be paid and presented model in this study can be used as a tool in this regard. Traffic educations, increase of public transportations systems and modification of the streets geometry and surrounding buildings can be considered as some ways for management and controlling of the air pollution in urban streets in order to reduce the amount of air contaminants. Abbreviation: T = Temp, Sig = Significant, F = Fisher, Df = Degree of freedom, Std = Standard, T = T- student. Research Journal of Recent Sciences ______________________________________________________________ ISSN 2277-2502Vol. 2(4), 36-45, April (2013) Res. J. Recent Sci. International Science Congress Association 45 Reference1.Chouhan A., Iqbal S., Maheshwari R.S. and Bafna A., Study of air pollution tolerance Index of plants growing in Pithampur Industrial area sector 1, 2 and 3, Res.J.Recent Sci., 1(ISC-2011), 172-177 (2012)2.Kavuri N.C. and Paul K.K., Chemical Characterization of Ambient Pm10 Aerosol in a Steel City, Rourkela, India,Res. J. Recent Sci., Vol. 2(1), 32-38, January (2013)3.Ministry of the Environment (Ontario), Air Quality Assessment Related to Traffic Congestion at Sarnia’s Blue Water Bridge Available from. http:// www. ene.gov.on.ca /envision /techdocs/5111e.pdf. (2005). 4.Tiwari S., Foliar Response of Two Species of Cassia to Heavy Air Pollution Load at Indore City, India, Res. J. Recent Sci., 1(ISC-2011), 329-332 (2012)5.Rathore Kanishka Raj, Dhawankar Aditi and Gungun, Environmental Impact Assessment (EIA) For Bus Based Rapid Transit System (BRTS) Bhopal, MP, India, Res. J. Recent Sci., 1(ISC-2011), 166-171 (2012) 6.Finn P. and Ruwim B., Actual car fleet emission estimated from urban air quality measurements and street pollution models, The Science of the Total Environment, 235, 101-109 (1999)7.Tasi M.Y. and Chen K.S., Measurement and three-dimensional modeling of air pollutant dispersion in an urban street canyonAtmospheric Environment, 38, 5911–5924 (2004)8.Olga V.T. and Sergey V.M., air pollutant dispersion within urban street canyons Atmospheric Environment 43, 245–252 (2008). 9.Taha H. and Sailor D., Evaluating the effects of radiative forcing feedback in modelling urban ozone air quality in Portland, Oregon: two-way coupled MM5- CMAQ numerical model simulations, Boundary-Layer Meteorology, 137), 291-305 (2010) 10.Harrison R.M., Jones A.M. and Barrowcliffe R., Field study of the influence of meteorological factors and traffic volumes upon suspend particle mass at urban roadside sites of differing geometries.Atmospheric Environment, 3837), 6361-6369 (2004)11.Agbo G.A., Ibeh G.F. and Ekpe J.E.,Estimation of Global Solar Radiation at Onitsha with Regression Analysis and Artificial Neural Network Models, Res.J.Recent Sci.,Vol. 1(6), 27-31 (2012)12.Comrie A.C., Comparing neural networks and regression models for ozone forecasting, Journal of the Air & Waste Management Association,47, 653–663, ( 1997)13.Tehran Air Quality Control Company, Air Pollution report, (2012)14.Traffic and Transportation Company of Tehran, Traffic statistics report (2012)15.Chaloulakou A., Grivas G. and Spyrellis N., Neural network and multiple regression models for PM10 prediction in Athens. A comparative assessment, Journal of the Air & Waste Management Association,53, 1183–1190 (2003)16.Hubbard M.C. and Cobourn W.G., Development of a regression model to forecast ground-level ozone concentrations, in Louisville, KY, Atmospheric Environment 32, 2637–2647 (1998)17.Chaloulakou A., Saisana M. and Spyrellis N., Comparative assessment of neural networks and regression models for forecasting summertime ozone in Athens, The Science of the Total Environment, 313, 1–13 (2003)`