Allocation of Routinely Monitored Mixing Ratios of

allocation of NOx mixing ratios to traffic and industrial sources in the Haifa ..... (13) Miller, M. S.; Friedlander, S. K.; Hidy, G. M. A chemical el...
1 downloads 0 Views 422KB Size
Download PDF
Environ. Sci. Technol. 2007, 41, 7215-7221

Allocation of Routinely Monitored Mixing Ratios of Nitrogen Oxides to Their Sources YUVAL,* YAEL DUBOWSKI, AND DAVID M. BRODAY Department of Civil and Environmental Engineering Technion, Israel Institute of Technology, Haifa 32000, Israel

In many urban areas throughout the world, the rising mean and peak levels of nitrogen oxides (NOx) are a concern. Road traffic and local industry are usually the major NOx sources in urban environments, but their relative contributions to the spatial distribution of the NOx volumetric mixing ratio is normally unknown. This missing piece of information is required for designing effective abatement measures to reduce ambient NOx levels. A new method for estimating the shares of which traffic and industry contribute to the mean ambient NOx mixing ratios observed in urban environments is proposed in this paper. The estimation is based on data obtained by routine air pollution monitoring, using a few assumptions about the temporal emission patterns of NOx and SO2 in the area of study. A set of equations is formulated for the unknown industry and traffic contributions to the NOx mixing ratios at each monitoring site. These equations are solved using the gradient projection optimization method. The bootstrap technique is used to estimate the errors in the process. Spatial maps of the separate shares of industry and traffic in the total ambient NOx levels can be obtained where a sufficient aerial coverage of stations is available. An application of the method to the allocation of NOx mixing ratios to traffic and industrial sources in the Haifa Bay area, Israel, is demonstrated. The results are expected to be useful for future planning of traffic thoroughfares and industrial development in the area.

1. Introduction The rising levels of nitrogen oxides (NOx) and their atmospheric transformation products have been a main air pollution worry in major metropolitan areas during the last few decades (1-3). NO2, the main component of NOx, is one of the five air pollutants identified as being of the greatest concern from health perspectives (4). The need to attain the NO2 limit values set by the European Commission Air Quality Framework Directive 1999/30/EC prompted many recent studies to facilitate NOx abatement programs (e.g., 5-7). The first step in setting forth a NOx abatement plan is studying the impact that the current emissions from the various sources has on its present temporal and spatial patterns. Road traffic is in most cases the largest NOx source; however, in many areas around the world industrial plants and/or power generation are sources of commensurate magnitude (8, 9). Estimating the contributions of known sources to the observed levels of air pollutants at specific points is usually * Corresponding author phone: +972-4-8292676; fax: +972-48292606; e-mail: [email protected]. 10.1021/es0702317 CCC: $37.00 Published on Web 10/04/2007

 2007 American Chemical Society

termed source apportionment. The most comprehensive approach to source apportionment is by a direct numerical simulation of the coupled meteorological dispersion and chemical reactions of the emissions (10). Simplified dispersion schemes are sometimes used instead of a full threedimensional dynamical and chemical model (e.g., 7, 6, 11). Running simulations with the different sources switched “on” and “off” can provide an estimation of the impact of different sources on the observed air pollution levels at points of interest in the study area. This is a very appealing approach, as it has the potential of finding the contribution of each identified source to the overall ambient levels of the pollutant at any given point in space and time. However, developing and running such a model is a challenging enterprise, requiring a thorough emissions inventory and much time for the construction and testing of the model. Moreover, the errors accrued in the process might be very large and hard to estimate (7). Fine-tuning the model to minimize its errors can be carried out through an inverse modeling scheme (e.g., 12) but the human and computer resources needed for such a task are not always readily available. Simple alternatives to a comprehensive direct source apportionment are based on analyses of air pollution measurements. When dealing with pollutants which have identifiable composition signatures that can be tracked to specific sources, the chemical species collected at monitoring sites can be associated with the potential sources through statistical algorithms such as those of the chemical mass balance (13, 14) or the positive matrix factorization (15, 16). The chemical mass balance method requires complete emission composition profile information. The positive matrix factorization method requires a large number of receptor samples and estimation of their errors. Both methods are mainly suitable and have been applied mostly for the apportionment of particulate matter or volatile organic compounds (ref 17 and references therein). Apportionment of ambient levels of gaseous reactive pollutants can be facilitated based on the ratios of the levels of observed gases. Duncan et al. (18) used NOy and SO2 source concentrations to estimate what percentage of the total ambient NOy observed at a monitoring station is attributable to a specific point source. Their method involves knowledge of all the source concentrations and a few assumptions about losses due to dry and wet deposition and gas-phase reactions, which may not always hold. Using a few regression models, Nirel and Dayan (19) studied the role of the SO2/NOx ratio, encountered at a specific monitoring station, in attributing air pollution to potential sources. Wind directions and information about the structure of the atmospheric boundary layer were also used. Application of their method requires a lot of preliminary data exploration and analysis, which might be difficult to do for a multiple-station data set. The present study proposes an alternative approach that bypasses all the above-mentioned hurdles, using routinely observed air pollution data (i.e., obviating the need for a complex modeling) in a relatively simple and computationally cheap scheme. The goal is limited to the allocation of observed ambient NOx mixing ratios to generic types of sources, namely, traffic and industry. However, the assessment of the traffic and industry NOx contributions is a scientifically and practically important issue, which is in the focus of recently published studies (e.g., 6, 8). Our approach is based on an a priori qualitative knowledge of the different temporal emission patterns of the industry and the traffic in the study area. Using this information, we set forth a few general assumptions and derive from them a set of equations VOL. 41, NO. 21, 2007 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

7215

FIGURE 1. Map of the Haifa Bay area. Shown are the Mediterranean shoreline, topographic contours every 100 m, main roads, and the locations of the major industrial plants. The monitoring stations whose data were used in the study are denoted by circles and their numbers. The coordinate tick marks are in kilometers on the local Israel system. associating the unknown industry and traffic NOx contributions to the ambient NOx with the corresponding observed NOx mixing ratios at each monitoring site. The system of equations is then solved, and the separate contributions of the source types to the ambient NOx levels as measured at the monitoring stations are recovered. Given data from a sufficiently dense monitoring network, the separate contributions at the stations can be interpolated to maps of their spatial patterns. These maps can provide a readily available and worthy insight into the roles that the major sources play in the spatial NOx distribution. The proposed methodology is applied at the Haifa Bay area, Israel, where the NOx levels are rising and a better understanding of the share of the major NOx sources in the area in the ambient NOx levels is required for devising optimal management strategies.

2. Methodology The general approach outlined in the Introduction is specified in this section. While the formulation of the method is made to befit the conditions in the study area of our example, its application to other regions may require only slight modifications that will be discussed later. The basic assumptions which we use are as follows: (a) Industry and traffic are the only sources of NOx in the study area. (b) The industry is the only source of SO2 in the study area. (c) At any point in the study area, the ratio between the mean (over the study period) number of molecules of NOx and SO2 that originate from industrial sources during weekdays is identical to the corresponding ratio during weekends and holidays (henceforth referred to as holidays). Note that no assumptions are made regarding the industrial emission rates and composition except that, on average, the ratio NOx/SO2 in the industrial emissions is similar in weekdays and holidays. In particular, no assumption is made regarding a possible long-term change in this ratio. Assumptions a and b are probably valid in most typical industrialized urban environments (18), especially where nonelectrical space heating and other domestic fuel burning is minimal. (See in the Supporting Information the emissions inventory supporting these assumptions for the 7216

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 41, NO. 21, 2007

case study of this work.) Possible non-negligible background levels of NOx and SO2 are not accounted for, but as will be discussed later, they can be easily incorporated. For assumption c to be valid, the air pollution dispersion conditions and industrial plants’ NOx and SO2 emissions per consumed fuel unit must be, on average, the same during the working days and during the holidays. This may certainly be incorrect for short periods of time due to variations in the working conditions of the industrial furnaces and due to varying dispersion conditions. However, averaged over a long period of time (or over a short period in which the emission characteristics and the meteorological conditions can be considered similar during working days and holidays), assumption c is very reasonable. The validity of assumption c for the demonstrated case study was ascertained by verifying with the major industrial plants that their weekdays and holidays emission characteristics do not differ in any systematic way (i.e., the fuel type and grade and the furnaces’ operating conditions do not depend on the day of the week). Meteorological conditions are certainly blind to the calendar, so it is obvious that averaged over a long enough period of time they are similar for weekdays and holidays. From assumption a, the following mass balance equations are derived for each monitoring station’s data set: w w Nw p + Nt ) N o

(1)

Nhp + Nht ) Nho

(2)

and

where Nw p is the mean (over the study period) ambient NOx mixing ratio observed at the station during working days and which originated from industrial emissions (point sources), Nw t is the corresponding traffic contribution, and Nw o is the corresponding observed mean ambient NOx mixing ratio. Nhp, Nht , and Nho are the corresponding quantities during holidays. The definitions of weekends, holidays,

and the choice of the hour on which the days change are country/region specific. For example, in Israel, the official weekly rest day is Saturday, and we chose to switch days at 18:00, the average time for day change in the Jewish calendar. These choices optimally capture the weekly and daily business and traffic cycles in the case study used for demonstrating the proposed method. The study period can be any portion of the period for which data records are available (e.g., data from certain seasons, hours of day, etc.) as long as that portion includes sufficient sampling for assumption c to be valid. Assumptions b and c lead, after a simple algebraic manipulation, to a third equation of h w h w h Nw p /Np ) Sp /Sp ) So /So ≡ F

(3)

h where Sw o and So are the mean (over the study period) SO2 mixing ratios observed at the monitoring station during working days and holidays, respectively, and F is defined by eq 3 as the ratio between the means of the observed SO2 levels during weekdays and holidays. In cases where the background NOx and/or SO2 levels are non-negligible, their measured mixing ratios can be incorporated as known constants in eqs 1-3. Equations 1-3 can be written in the matrix form

Ax ) b where

A)

(

(4)

1 1 0 0 0 0 1 1 -1 0 F 0

)

w h h t w h t x ) [Nw p Nt Np Nt ] and b ) [No No 0] . The underdetermined system in eq 4 can be solved for x by minimization of ||x||2 (the Euclidean norm of x) subject to Ax - b ) 0 and the additional physical constraints

{

w 0 e Nw p e No

0 e 0 e 0 e

Nw t Nhp Nht

e e e

Nw o Nho Nho

}

Thus, the solution x contains the study period means of the separate contributions of traffic and industry to the ambient NOx levels at a monitoring station, derived from the observed NOx and SO2 levels averaged over the study period and subject to the constraints set on the elements of x. To find the solution to the minimization problem, we employed the gradient projection optimization method (20). As in any other estimation process, an assessment of the possible errors introduced by the proposed method is a concern. Errors arising due to invalidity of the underlying assumptions are hard to assess but are expected to be small. That is because our use of data averaged over long periods results in mutual cancellation of random observation errors. Biases and systematic data errors should be excluded in the initial data processing, but their complete removal may not always be possible. Applying the method to synthetic data sets (i.e., where the true allocation was known), we noted that slight systematic errors during relatively short durations might sometimes be a source of relatively large allocation errors. Another cause of errors is the inaccuracies introduced by the process of solving eqs 1-3, e.g., the minimization of ||x||2 is not necessarily appropriate in all cases. To estimate the expected solution errors we chose to use the moving blocks bootstrap technique (21). Bootstrapping using moving blocks of data is required due to the temporal autocorrelation

FIGURE 2. Contributions of industry and traffic to the mean observed NOx levels in the monitoring stations. The results shown are for the summers and winters of 2002-2005. The error bars indicate 1 standard deviation of the corresponding bootstrap replicas. (a) The contributions during working days (total of 15 024 and 14 880 time points in summers and winters, respectively). (b) The contributions during Saturdays and holidays (2640 and 2448 time points in summers and winters, respectively). in the data. The moving blocks bootstrap involves a division of the data series into blocks. A random selection, with replacement, of the data blocks is then carried out multiple times, yielding many new data sets called bootstrap samples. Due to the random selection, each bootstrap sample is in general slightly different from the others and from the original set. We applied the numerical solution of eqs 1-3 to each ˜w of the bootstrap samples and obtained a set of values N ˜w p, N t , h h w N ˜ p, and N ˜ t , which are called the bootstrap replicas of Np , Nw t , Nhp, and Nht , respectively. The standard deviations of the replicas N ˜w ˜w ˜ hp, and N ˜ ht can serve as measures of the p, N t , N w h errors expected in our real data solutions for Nw p , Nt , Np, and Nht , respectively. An application of the allocation process in all the monitoring stations results in a set of values, associated with w h the station locations, for each of the variables Nw p , Nt , Np , and Nht . Each of these sets can be used to create a corresponding spatial interpolation map. We used for that purpose the ordinary kriging method (22, 23). The accuracy of the interpolation depends on the density of the data points. Where the monitoring stations’ coverage is not sufficiently dense, large interpolation errors may occur. In such a case, only the main features of the interpolation maps should be considered. A detailed discussion of this point, regarding interpolation of data from the same monitoring array used here, can be found in Yuval and Broday (24).

3. Case Study Information Figure 1 shows a general map of the Haifa Bay area. The area includes the city of Haifa and satellite suburbs and towns, totaling nine municipalities and a population of about VOL. 41, NO. 21, 2007 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

7217

FIGURE 3. Spatial distributions of industry and traffic contributions to the mean NOx mixing ratios (ppb) during summers and winters of 2002-2005 (total of 17 664 and 17 328 time points, respectively). (a) The contribution of industry during the summer. (b) The contribution of traffic during the summer. (c) The contribution of industry during the winter. (d) The contribution of traffic during the winter. 500 000. At the heart of the region are a few of Israel’s major industrial plants. An oil-fired power plant and a petroleum refinery together contribute about 80 and 83% of the total industrial NOx and SO2 emissions in the area, respectively. (See the Supporting Information for more details of the emission inventory.) The refinery and a few other large plants in its vicinity work year-round, 24 h a day, with no significant day/night or workday/holiday variations. However, the power plant’s load and emissions vary according to demand. The Haifa District Municipal Association for the Environment (HDMAE) estimated that the mean total industrial emission rates in the study period 2002-2005 were 0.80 tonne/hour NOx and 1.74 tonne/hour SO2 (Supporting Information, Table S1). The roads in the study area experience daily traffic of about 125 000 private cars, 30 000 lorries and light trucks, and a few hundred buses. There are no appreciable seasonal variations in the traffic volumes. However, the weekly rest day (Saturday) and statutory holidays experience a very large reduction of vehicular activity, with the diesel-fueled commercial traffic, responsible for about 75% of the vehicular NOx emissions, coming to an almost complete halt. The estimated mean traffic emission rates in the 2002-2005 study period were 1.02 tonne/hour NOx and 0.03 tonne/hour SO2 (Supporting Information, Table S1). The Persian trough, which dominates the synoptic meteorology of the coastal regions of Israel from mid-May to October, results in ample sunshine and dominant northwesterly winds during the warm-season days. Weak easterly land breezes may occur during the warm-season nights. Due to the Azorean high-pressure system aloft capping the Persian trough at the surface, the mixing height in the warm season is shallow (about 700 m) and stable (25). During the cold season (November to April), southeasterly flow dominates, except when subtropical storms pass through the area (mainly from December to February) and result in strong winds veering from southwesterlies to northwesterlies (Figure S2 in the Supporting Information). The cold season is characterized by a lot of mixing except during cold clear nights. 7218

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 41, NO. 21, 2007

Due to the topography of Mt. Carmel on which the city of Haifa is built, the wind directions on the crest and northeasterly slopes of the mountain change counterclockwise from the regional winds (Figure S2 in the Supporting Information). The air-quality monitoring network in the Haifa Bay area consists of 20 stations, but due to the proposed method’s requirements, only data from the 9 stations observing both SO2 and NOx could be used. The monitoring devices are located on top of buildings 3.5-16 m high. The monitoring locations were chosen so that they optimize monitoring of ambient air and not specific microenvironments. The data set includes half-hourly records from 2002 to 2005 (a total of 70 128 half-hourly time points). The typical mean NOx and SO2 volumetric mixing ratios are 8-25 ppb and 1-4 ppb, respectively. The background NOx mixing ratios are usually below 0.5 ppb (26) and were neglected in this study. Further details of the monitoring network and the initial processing and quality assurance of the data were given by Yuval and Broday (24).

4. Results The main goal of the proposed method is to estimate the relative shares of the industry and traffic in the mean observed ambient NOx mixing ratios. A way to assess the validity of the method’s results is to test it on subperiods with clear known trends of emissions, dispersion, and chemical transformation. Thus, before we present the final results for the whole study period, we compare the NOx allocation during the summers and winters and during the morning and night hours. Winters and summers obviously differ in their light intensity and length of days, which impact indirect photochemical conversion processes of NOx to NOy species (e.g., reactions of NO2 with OH and RO2 radicals). The analysis should also account for other factors such as the differences in the dominant wind directions and the height of the atmospheric mixing layer. The ambient conditions during the morning

and night hours also differ in many natural meteorological variables, but the largest impact on the ambient NOx levels is due to the morning and night rush hour traffic emissions. Qualitative verification of the proposed method can be established if the differences between the winter/summer, morning/night, and the weekend/holidays results reflect the known variations in the corresponding natural and anthropogenic variables. Figure 2 shows the calculated industry and traffic contributions to the mean observed NOx levels in working days and in holidays during winter (December-February) and summer (June-August) in each of the monitoring stations. In agreement with our qualitative knowledge of the traffic emission patterns, in both seasons the calculated traffic contribution to the ambient NOx mixing ratios is significantly higher during working days than in holidays (a station-wise h average ratio Nw t /Nt of 4.12 in the winter and 3.46 in the summer were found). The winter ratio is probably higher because of the later onset of the indirect photochemical NOx removal processes of the winter morning rush hour emissions in the weekdays. These weekdays morning emissions have a large impact on the mean NOx values. The NOx contribution of industry is also higher during working days but with a h much smaller average ratio (Nw p /Np equals 1.17 in the winter and 1.25 in the summer). The ratios between the mean industry and traffic contributions in winter and summer (weighted weekdays and holidays means) are similar (1.27 for traffic and 1.23 for industry). The close similarity in these ratios is expected as the differences between the winter and summer NOx mixing ratios is primarily due to the differences in the thickness of the atmospheric mixing layer (25) and in the rate of indirect photochemical NOx removal, both having similar effects on traffic and industrial emissions alike. Figure 3 shows the spatial patterns of the traffic and industry contributions to the mean ambient NOx mixing ratios during the summer and the winter. The traffic contribution pattern (parts b and d of Figure 3) show slightly higher mixing ratios than the pattern of the industry contributions (parts a and c of Figure 3). Higher mixing ratios due to both industry and traffic are found in the lower parts of the city of Haifa (in the vicinity of stations 2, 5, and 9, see Figure 1), which is geographically close to the industrial center and to the area where most of the commercial activity takes place. Figure 4 shows the industry and traffic contributions to the mean observed NOx levels in the morning hours of 06: 00-10:00 and the night hours of 24:00-04:00. The workday morning results (Figure 4a) show higher contributions from traffic compared with those of industry in all stations except station 8, which is far from any significant NOx sources. During workdays nights, the trend is reversed, with a small dominance of the industrial contributions in all of the stations. These results agree well with the common knowledge that the morning hours experience a large increase in traffic volumes but only a slight increase in the industrial output. In the night hours, the industrial activity is only slightly lower than its daily mean but the decrease in the traffic volume is substantial. The traffic emissions pattern in holidays is very different from than that of working days, as clearly manifested in the results shown in Figure 4b. Industrial contributions clearly dominate during the morning holiday hours, when private car traffic is low and diesel commercial traffic is practically nonexistent. In the night hours of weekends and holidays, the industry and traffic contributions are quite similar in stations located at quiet residential areas (stations 3, 5, and 6). In a few other stations, the traffic contributions are higher, probably due to elevated entertainment-related traffic on Saturday and holiday evenings. Figure 5 shows the mean spatial patterns of the traffic and the industry contributions to the ambient NOx mixing

FIGURE 4. Contributions of industry and traffic to the mean observed NOx levels in the monitoring stations. Results are shown for the night hours (24:00-04:00) and the mornings (06:00-10:00) in 20022005. The error bars indicate 1 standard deviation of the corresponding bootstrap replicas. (a) The contributions in working days (10 944 time points). (b) The contributions in Saturdays and holidays (2205 time points). ratios during the morning and night hours. The morning patterns of the industry and traffic contributions (parts a and b of Figure 5, respectively) are quite similar, except the conspicuous high traffic contributions around station 5 and the general trend of the main traffic contribution pattern toward the east. The onset of the daytime northwesterly wind in summer mornings is around 10:00. Thus, the mean morning industrial contributions pattern is more influenced by the dominating winter easterly winds and the early summer easterly land breeze, which brings industrial NOx to the area west of the industry center. The traffic contribution pattern is more influenced by the location of the morning rush hour traffic, which is heavy in the area around station 2 and along the highways leading to the bay of Haifa from the northeast and southeast (see Figure 1). The spatial patterns of the industry and traffic contributions during the night (parts c and d of Figure 5, respectively) are very similar, with the major feature extending from the industrial center toward the west. This reflects the relatively stagnant meteorological conditions in the nocturnal stable boundary layer in which the late evening traffic emissions and the nightly industrial emissions mix and remain in place or advect slowly toward the west with the land breeze. Figure 6a shows the absolute industry and traffic contributions to the mean observed NOx mixing ratios in each of the stations for the whole study period. Figure 6b shows the corresponding relative shares of the total observed NOx. Interestingly, the dominance of the traffic’s share during the day hours of the workdays is almost balanced in all the stations by the dominance of the industry’s share during the workdays nights and holidays. In most stations, the differences are smaller than their error estimates. The traffic’s relative shares (Figure 6b) are between 50 and 56% of the VOL. 41, NO. 21, 2007 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

7219

FIGURE 5. Spatial distributions of industry and traffic contributions to the mean NOx mixing ratios (ppb) during the night hours (24:00-04:00, total of 13 149 time points) and the mornings (06:00-10:00, total of 13 149 time points) of 2002-2005. (a) The contribution of industry during the morning. (b) The contribution of traffic during the morning. (c) The contribution of industry during the night. (d) The contribution of traffic during the night.

total observed NOx. The most updated emission estimations reported by the HDMAE (Supporting Information, Table S1 and Figure S1) attribute to traffic sources 45% of the emitted NOx in the region during 2006. The higher traffic shares that we found are expected as, per unit of emission, the surfacelevel traffic emissions are yielding a larger contribution to the ground-level ambient NOx than the emissions from the tall industrial stacks (6, 8). (Note that the 2002-2005 traffic’s share in the NOx emission are higher due to recently acknowledged errors in the HDMAE traffic emission estimations during those years.)

5. Discussion A new method for allocating ambient NOx mixing ratios, observed at monitoring stations, to traffic and industry source types was developed. The method uses readily available monitoring data that are routinely observed and archived throughout the world. Spatial maps of the separate shares of industry and traffic in the total ambient NOx levels can be obtained but only where a sufficient aerial coverage of stations is available. While this method does not aim to estimate the absolute emission from each source, it can estimate the contributions from the major source groups (traffic and industry) to the mean observed mixing ratios. The NOx allocation results pertain to a specific study period so that intraperiod variations in the emission characteristics and meteorology are averaged out. In cases of long time-scale variations in the emission characteristics (e.g., due to a decrease in the sulfur content in the fuels), the method can be applied separately to homogeneous subperiods of the whole study period to obtain their specific allocation results. The method presented in this study facilitates computation of NOx contributions during specific hours of the day. As demonstrated by the night and morning NOx allocation results (Figures 4 and 5), large diurnal variations in the traffic 7220

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 41, NO. 21, 2007

FIGURE 6. Contributions of industry and traffic to the observed NOx ratios in the monitoring stations during the whole study period 2002-2005 (70 128 time points). The error bars indicate 1 standard deviation of the corresponding bootstrap replicas. (a) The absolute contributions (ppb) of each source to the observed ambient NOx. (b) The corresponding relative shares of the total ambient NOx.

and industry shares may occur. Thus, for risk assessment studies as well as for NOx abatement programs, it is advisable to consider the patterns at the hours of the day when the population at risk tends to be exposed. The method was applied at the Haifa Bay area, and the results fit well the qualitative known spatial and temporal distribution of NOx emissions in the region and their dispersion and transformation by the local meteorological and atmospheric chemistry conditions. This study was motivated by a debate regarding the relative shares of the sources of NOx in the study area. With the use of the proposed methodology and the results presented in this paper, abatement efforts and future industry development and highway construction can be better focused and rationally devised.

Acknowledgments The data were kindly provided by the Haifa District Municipal Association for the Environment and by the Israel Electric Corporation. This work was supported by grants from the Israeli Ministry of Science and Technology, the Seniel Ostrow Research Fund, and the Technion Research and Development Foundation Landau/Ben-David. The authors would like to thank Professor Eldad Haber of Emory University for providing the gradient projection Matlab code.

Supporting Information Available (a) Table containing the 2002-2006 emission inventories for the Haifa Bay area; (b) figure with pie charts showing the relative shares of the various sources in the NOx and SO2 emissions; (c) winter and summer windroses of wind data from two characteristic stations. This material is available free of charge via the Internet at http://pubs.acs.org.

Literature Cited (1) Seinfeld, J. H.; Pandis, S. N. Atmospheric Chemistry and Physics, from Air Pollution to Climate Change; Wiley: New York, 1998. (2) Hao, J.; Tian, H.; Lu, Y. Emission inventories of NOx from commercial energy consumption in China, 1995-1998. Environ. Sci. Technol. 2002, 36, 552-560. (3) Zhang, B. N.; Kim Oanh, N. T. Photochemical smog pollution in the Bangkok Metropolitan Region of Thailand in relation to O3 precursor concentrations and meteorological conditions. Atmos. Environ. 2002, 36, 4211-4222. (4) Kyle, A. D.; Woodruff, T. J.; Buffler, P. A.; Davis, D. L. Use of an index to reflect the aggregate burden of long-term exposure to criteria air pollutants in the United States. Environ. Health Perspect. 2002, 110, 95-102. (5) Dixon, J.; Middleton, D. R.; Derwent, R. G. Sensitivity of nitrogen dioxide concentrations to oxides of nitrogen controls in the United Kingdom. Atmos. Environ. 2001, 35, 3715-3728. (6) Leksmono, N. S.; Longhurst, J. W. S.; Ling, K. A.; Chatterton, T. J.; Fisher, B. E. A.; Irwin, J. G. Assessment of the relationship between industrial and traffic sources contributing to air quality objective exceedance: a theoretical modelling exercise. Environ. Modell. Softw. 2006, 21, 494-500. (7) Farias, F.; ApSimon, H. Relative contributions from traffic and aircraft NOx emissions to exposure in West London. Environ. Modell. Softw. 2006, 21, 477-485.

(8) Bhanarkar, A. D.; Goyal, S. K.; Sivacoumar, R.; Chalapati Rao, C. V. Assessment of contribution of SO2 and NO2 from different sources in Jamshedpur region, India. Atmos. Environ. 2005, 39, 7745-7760. (9) Abdul-Wahab, S. A.; Azzi, M.; Johnson, G. M.; Bouhamra, W. S.; Ettouney, H. M.; Sowerby, B.; Crittenden, B. D. Using the integrated empirical rate-reactive plume model in assessment of the potential effects of Shuaiba industrial area NOx plumes on photochemical smog concentrations. Process Saf. Environ. 2003, 81, 363-374. (10) Seaman, N. Meteorological modelling for air-quality assessments. Atmos. Environ. 2000, 34, 2231-2259. (11) Hurley, P.; Manins, P.; Lee, S.; Boyle, R.; Ng, Y. L.; Dewundege, P. Year-long, high-resolution, urban airshed modelling: Verification of TAPM predictions of smog and particles in Melbourne, Australia. Atmos. Environ. 2003, 37, 1899-1910. (12) Schichtel, B. A.; Malm, W. C.; Gebhart, K. A.; Barna, M. G.; Knipping, E. M. A hybrid source apportionment model integrating measured data and air quality model results. J. Geophys. Res. 2006, 111, D07301. doi:10.1029/2005JD006238. (13) Miller, M. S.; Friedlander, S. K.; Hidy, G. M. A chemical element balance for the Pasadena aerosol. J. Colloid Interface Sci. 1972, 39, 65-176. (14) Watson, J. G.; Chow, J. C.; Pace, T. G. Chemical mass balance. In Receptor Modeling for Air Quality Management; Hopke, P. K., Ed.; Elsevier: Amsterdam, 1991; pp 83-116. (15) Paatero, P. Least squares formulation of robust, non-negative factor analysis. Chemom. Intell. Lab. Syst. 1997, 37, 23-35. (16) Paatero, P.; Tapper, U. Positive Matrix Factorization: a nonnegative factor model with optimal utilization of error estimates of data values. Environmetrix 1994, 5, 111-126. (17) Hopke, P. K. Receptor Modeling for Air Quality Management; Elsevier: Amsterdam, 1991. (18) Duncan, B. N.; Stelson, A. W.; Kiang, C. S. Estimated contribution of power plants to ambient nitrogen oxides measured in Atlanta, Georgia, in August 1992. Atmos. Environ. 1995, 29, 30433054. (19) Nirel, R.; Dayan, U. On the ratio of sulfur dioxide to nitrogen oxides as an indicator of air pollution sources. J. Appl. Meteorol. 2001, 40, 1209-1222. (20) Nocedal, J.; Wright, S. J. Numerical Optimization; Springer: New York, 1999. (21) Efron, B.; Tibshirani, R. J. An Introduction to the Bootstrap; Chapman and Hall: New York, 1993. (22) Isaaks, E. H.; Srivastava, R. M. An Introduction to Applied Geostatistics; Oxford University Press: New York, 1989. (23) Cressie, N. A. C. Statistics for Spatial Data, revised ed.; John Wiley & Sons: New York, 1993. (24) Yuval; Broday, D. M. High resolution spatial patterns of longterm mean concentrations of air pollutants in Haifa Bay area. Atmos. Environ. 2006, 40, 3653-3664. (25) Dayan, U.; Rodnizki, J. The temporal behavior of the atmospheric boundary layer in Israel. J. Appl. Meteorol. 1999, 38, 830836. (26) Alper-Siman Tov, D.; Peleg, M.; Matveev, V.; Mahrer, Y.; Seter, I.; Luria, M. Recirculation of polluted air masses over the East Mediterranean coast. Atmos. Environ. 1997, 31, 1441-1448.

Received for review January 30, 2007. Revised manuscript received August 19, 2007. Accepted August 22, 2007.

ES0702317

VOL. 41, NO. 21, 2007 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

7221