ARTICLE pubs.acs.org/est
Prediction Model of the Buildup of Volatile Organic Compounds on Urban Roads Parvez Mahbub,*,† Ashantha Goonetilleke,† and Godwin A. Ayoko‡ † ‡
School of Urban Development, Queensland University of Technology, GPO Box 2434, Brisbane 4001, Queensland, Australia Chemistry Discipline, Queensland University of Technology, GPO Box 2434, Brisbane 4001, Queensland, Australia
bS Supporting Information ABSTRACT: A model to predict the buildup of mainly trafficgenerated volatile organic compounds or VOCs (toluene, ethylbenzene, ortho-xylene, meta-xylene, and para-xylene) on urban road surfaces is presented. The model required three traffic parameters, namely average daily traffic (ADT), volume to capacity ratio (V/C), and surface texture depth (STD), and two chemical parameters, namely total suspended solid (TSS) and total organic carbon (TOC), as predictor variables. Principal component analysis and two phase factor analysis were performed to characterize the model calibration parameters. Traffic congestion was found to be the underlying cause of traffic-related VOC buildup on urban roads. The model calibration was optimized using orthogonal experimental design. Partial least squares regression was used for model prediction. It was found that a better optimized orthogonal design could be achieved by including the latent factors of the data matrix into the design. The model performed fairly accurately for three different land uses as well as five different particle size fractions. The relative prediction errors were 1040% for the different size fractions and 2840% for the different land uses while the coefficients of variation of the predicted intersite VOC concentrations were in the range of 2545% for the different size fractions. Considering the sizes of the data matrices, these coefficients of variation were within the acceptable interlaboratory range for analytes at ppb concentration levels.
’ INTRODUCTION Volatile organic compounds (VOCs) are among the major pollutants generated by transport activities in urban areas.13 Several studies have characterized the influence of traffic on the generation of VOCs in ambient urban atmosphere.46 Toluene, ethylbenzene, ortho-xylene, meta-xylene, and para-xylene are among the most common species of VOCs (boiling points ranging from 50 to 260 °C). These are termed as gasoline range organics.3,7,8 The International Agency for Research on Cancer has listed them as carcinogenic to humans.9 Urban stormwater runoff acts as the primary medium for the transport of pollutants from urban road surfaces.1013 Consequently, the risks to urban water resources from toxic and mutagenic pollutants generated by traffic activities have increased due to the rapid urbanization witnessed in most parts of the world. Though the investigation of pollutant buildup processes on urban roads has received significant attention in recent years, there has been very limited work undertaken with regard to the characterization of the buildup of VOCs on actual road surfaces. Olson et al.14 studied VOC concentrations in the near-road environment and found that the concentrations decreased exponentially with increasing distance from the roadway. However, they collected air samples from various locations near roadway. Limited information on the spatial variability of VOCs has led researchers to propose prediction methods for certain VOCs such as toluene, benzene, ethylbenzene, ortho xylene, meta xylene and para xylene in r 2011 American Chemical Society
the atmosphere (e.g., 15). Their approach was based on six GISbased ancillary variables which included distance to nearest border crossing, elevation, population density, distance to roads, traffic intensity, and distance to nearest petroleum facility. These ancillary variables may be irrelevant elsewhere mainly due to lack of indepth understanding of the VOC buildup characteristics on road surfaces as well as lack of knowledge on identifying the influencing parameters on such processes. This paper presents a robust model to predict VOCs (toluene, ethylbenzene, ortho-xylene, meta-xylene, and para-xylene) on urban road surfaces using three traffic parameters and two chemical parameters. This buildup model excludes the VOCs in the ambient air which may not necessarily get deposited on roads. To the best of our knowledge, there has not been any previous research which has been based on the physical collection of VOCs from the actual road surfaces. The closest is the work undertaken by Olson et al.14 where sample collection was performed at some distances from the roads. The selection of the model parameters was based on the characterization of the buildup of VOCs on urban roads. Different land uses, namely residential, industrial, and commercial, as well as different particulate and dissolved fractions are incorporated Received: January 25, 2011 Accepted: April 13, 2011 Published: April 22, 2011 4453
dx.doi.org/10.1021/es200307x | Environ. Sci. Technol. 2011, 45, 4453–4459
Environmental Science & Technology into the model. The model differentiates the VOC buildup on urban roads from that of ambient urban atmosphere. It reduces the degree of complexity involved in the prediction of trafficgenerated VOC buildup on urban road surfaces. This prediction model will be useful to environmental planners in terms of reliable prediction of VOCs on urban roads in changing traffic conditions and the use of the model structure as a guide to predict other traffic-generated pollutants on urban roads.
’ MATERIALS AND METHODS Site Selection. It is hypothesized in this study that vehicle congestions due to increased traffic volumes in urban areas would directly affect the pollutant buildup on urban roads. Hence, eleven road sites were selected as study sites for the buildup sample collection in three suburbs in the Gold Coast region in South-East Queensland, Australia. The sites selected for sample collection for VOC buildup covered three types of land uses including residential, commercial, and industrial which provided a crosssection of traffic activities on road surfaces within the Gold Coast region. The selected suburbs were also representative of the transport infrastructure developed in the Gold Coast region in the past decade. Average daily traffic (ADT), volume to capacity ratio (V/C), and surface texture depth (STD) were selected as the primary traffic parameters for this study. ADT and congestion as described by V/C are two important traffic parameters in terms of characterizing urban roads. Whereas ADT refers to the traffic count at a fixed point of the road section, V/C describes the congestion on the stretch of road during the peak hour.16 Additionally, the STD refers to the deviation of the surface profile of a pavement from a true planar surface often known as macrotexture of the pavement.17 Organic matter can act as binding agents between solids and other pollutants such as heavy metals in source, transport, and deposit phases.18 Hwang and Foster19 have reported 6897% enrichment of polycyclic aromatic hydrocarbons in particulate phase in urban runoff. In another study, Moilleron et al.20 have shown that 85% of the total aliphatic hydrocarbon content in urban runoff was attached to particulate phase. In this context, surface roughness becomes an issue because of the potential trapping of organics to the particulate matter. Smooth surfaces reduce trapping but provide a more efficient transport mechanism, and vice versa. While describing a physically based urban wash-off model, Shaw et al.21 have addressed the role of surface roughness of asphalt pavement in capturing and attenuating particulate matter. The road texture also affects the interactions between tires and driving surface. Kreider et al.22 have found that the characterization of roadway particles, tire wear particles, and tread particles are totally different during their build up and attributed friction between tire and the road surface as one of the principal causes of such characterization. Hence in this study, surface texture depth (STD) has been selected as a pavement parameter affecting the buildup of VOCs on the roads. A transport model called “ZENITH”,23 currently being used by the Gold Coast City Council for pavement infrastructure planning and design purposes, was used to predict the current ADT and V/C of the selected study sites while STD was directly measured in the field. The selected study sites with their traffic and pavement characteristics are given in the Figure S1 in the Supporting Information.
ARTICLE
Buildup Sample Collection. A method referred to as “wet and dry vacuuming”24 was employed for the collection of pollutant buildup samples. This consisted of the use of a domestic vacuum cleaner with a water filtration system for the collection of road dust from a 2 1.5 m2 plot area in 25-L plastic containers. Immediately thereafter deionized water was sprayed at 2 bar pressure on the plots and vacuuming was undertaken again to collect any remaining buildup matter. Samples were then transported to the laboratory for subsampling and further analyses. The pollutant buildup process has been characterized as four main functional forms such as linear, power, exponential, and MichaelisMenten.25 Among these, the nonlinear asymptotic form proposed by Sartor et al.26 has been most often cited and also used in several stormwater quality models including DR3QUAL, FHWA, and SWMM.25 Huber25 also suggested that buildup limits of paved surfaces influence the accumulation of solids especially due to the fact that surface materials on highways tend to remain in a state of suspension due to eddies created by traffic. In this context, Egodawatta27 noted that pollutant buildup on road surfaces asymptote to an almost constant value after a 7-day antecedent dry period. Hence, 7 dry days were allowed at each site prior to any sample collection in conformity with the findings of Egodawatta.27 The samples collected in the field were transferred into the splitting chamber of the churn splitter in the laboratory. The total composite sample volume was 8 L, of which 6 L was used to prepare 500-mL subsamples for particulate fraction analyses of the VOCs. The remaining sample in the churn splitter was used for the dissolved fraction analyses. The particle size distributions of the solids in the subsamples were determined using a Malvern Mastersizer S Particle Size Analyzer capable of analyzing particles between 0.05 and 900 μm diameters. The particle size distributions of the subsamples were compared with each other and used as a guide for homogeneity maintained in the subsampling process. The total particulate analytes were then fractioned into four size ranges, namely, >300, 150300, 75150, and 175 μm using wet sieving. The filtrate passing through a 1-μm membrane filter was considered as the potential total dissolved fraction. In each case, the prepared subsamples were stored in 500-mL amber glass bottles with PTFE seals, preserved at 4 °C in the laboratory, and analyzed within 14 days of collection. A total of 55 buildup samples were prepared for the 11 selected sites with each site having 5 samples for the 5 size fractions mentioned above. Seven replicate samples for each subsample at each fraction were also prepared, and means, standard deviations, and intersite percentage coefficient of variations have been reported in the Supporting Information. Sample Testing. The samples were tested for volatile organics such as toluene, ethylbenzene, ortho-xylene, meta-xylene, and paraxylene according to USEPA method 5035 and 8260B 7 using purge and trap extraction followed by gas chromatography/mass spectrometry (GC/MS). A brief description of the quality control measures is given in the Supporting Information. The total and dissolved organic carbon (TOC and DOC) concentrations and total suspended and dissolved solids (TSS and TDS) concentrations were also determined for the samples.28 Data Analysis. The data matrices consisted of eleven objects and nine variables for each of the five particle size fractions. The results for the different volatile organics, suspended solids, and organic carbons due to the traffic attributes, ADT, V/C, and STD were defined as “buildup values”. The eleven traffic objects were designated with object identifiers R1, R2, R3, I4, I5, C6, C7, R8, C9, C10, and R11 where the prefixes C, I, and R represented 4454
dx.doi.org/10.1021/es200307x |Environ. Sci. Technol. 2011, 45, 4453–4459
Environmental Science & Technology
ARTICLE
Table 1. Correlation Coefficients Matrix after VARIMAX Factor Rotation with Bold Coefficients Showing Variables Strongly Associated with Corresponding Factors correlation coefficients
measured variables
Figure 1. PCA biplot of total particulate volatile organic buildup on urban roads; objects are indicated by labels with the prefix I, C, or R for industrial, commercial, and residential sites, respectively.
commercial, industrial, and residential sites, respectively. After initial observation of the probability distribution of the objects and variables, standardization of each variable was undertaken as a pretreatment measure. The data analysis was designed to investigate the buildup process of volatile organics generated from traffic on urban roads and then to use the findings from the initial investigations to develop a prediction model for volatile organics buildup. Multivariate chemometrics methods such as principal component analysis (PCA), two phase factor analysis (FA), and partial least-squares regression (PLS) were employed for the data analysis. A brief description on these techniques with relevant references is outlined in the Supporting Information.
’ RESULTS AND DISCUSSION Exploratory PCA. In the six buildup data matrices (five for different size fractions described above and one as total particulate), the eleven traffic objects were considered having object attributes such as average daily traffic (ADT), volume to capacity ratio (V/C) and surface texture depth (STD) that were responsible for generating the VOCs, TSS, and TOC. PCA was performed on each of the pretreated buildup data matrices. Figure 1 shows the PCA biplot of the total particulate buildup data matrix. In Figure 1, VOCs form a group (A) consisting of toluene (TOL), ethylbenzene (ETB), meta and para-xylene (MPX), and ortho-xylene (OX) that are very strongly correlated with each other on both PCs. This suggests that they originated from similar or even the same sources. The negative loading of ADT and the strong positive correlation between V/C and the group A volatiles on PC1 suggest that a highly congested traffic lane of a road is more likely to cause the volatile organic buildup than the traffic count (ADT) on the road. The texture of the road surface could also affect the capture of particulate matters that might be associated with organics during the buildup phase. Patra et al.29 have illustrated that vehicle-induced turbulence caused faster resuspension of coarser particulate matter than the finer particles on road surfaces.
correlation coefficients
after rotation for total
after rotation for all
particulate fraction
fractions taken together
factor 1
factor 2
factor 1
factor 2
TSS
.954
0.253
0.252
.956
ADT
-.952
0.267
0.270
-.953
TOC STD
.889 .719
0.450 0.682
0.468 0.677
.880 .722
VC
0.051
.872
.853
0.059
OX
0.540
.811
.835
0.508
MPX
0.578
.803
.822
0.548
TOL
0.579
.799
.811
0.563
ETB
0.528
.792
.838
0.456
Mahbub et al.30 have shown that coarser particles from 75 to >300 μm exhibit rapid resuspension and redistribution due to traffic activities in commercial and industrial areas. Hence, in this study, the VOCs that were associated with coarser particulate matters from 75 to >300 μm were getting quickly redistributed and caused the inverse relationship with the free-flowing traffic. None of the objects except R1 had any noteworthy positive scores on PC1. Only four out of eleven objects (I5, R2, I4, and R8) had negative scores along with negative loadings from group A organics on PC2 which accounts for only 11.5% variance. This meant that the group A VOCs were slightly present in two industrial sites (I4 and I5) and three residential sites (R1, R2, and R8). However, these VOCs were not present in any of the commercial sites (C6, C7, C9, and C10) or in the two remaining residential sites (R3 and R11). This suggests that land use (commercial, residential, and industrial) did not have any clear-cut significant influence on the buildup of trafficgenerated VOCs, even though partial influence from some residential and industrial areas could be observed. Similar findings for the four particulate and the potential dissolved fractions strengthened the above conclusions relating to the measured variables. Modeled traffic data as well as small sample size might have contributed to such facts. Strong correlations between group A VOCs and TSS and TOC were observed in all size fractions. Organics’ association with solids during buildup could be attributed to such phenomenon. PCAs for each size fraction are provided in the Supporting Information. Wang and Zhao3 reported that the target VOCs were mostly found in the gaseous phase due to their high vapor pressure. This might have caused re-emission of the VOCs and hence the resulting weak correlations between land use and VOC concentrations in buildup. The deposition of VOCs on the pervious and impervious surfaces has been related to various meteorological conditions such as temporal fluctuations in wind activities31 as well as wind velocity and vertical diffusivity with the height above ground.32 Furthermore, Johnson and Belitz33 have reported statistically significant correlation between VOC concentrations with urban land uses while investigating the groundwater quality near water supply wells. In another study, Pitt et al.34 illustrated high detection frequencies of organic toxicants in particulate sample fractions originating 4455
dx.doi.org/10.1021/es200307x |Environ. Sci. Technol. 2011, 45, 4453–4459
Environmental Science & Technology
ARTICLE
the two levels were set for high and low.37 The design incorporated fifteen experiments altogether with twelve individual experiments and three replicate experiments at center. The calibration set for the experiment was chosen from the five different size fractions with each contributing three experiments. At each size fraction, the residential, commercial, and industrial sites contributed one experiment each. The central or replicate experiments were chosen in order to identify any curvature present on the response surface. However, these replicate experiments were chosen from the same site at different fractions in order to minimize the intrasite variability. The optimized calibration set is provided in the Supporting Information. The PCA biplot of the calibration matrix is presented in Figure 2. In Figure 2, ADT, TOL, ETB, MPX, OX, and VC have positive loadings and STD has negative loading on PC1. Only four experiments (E1, E4, E5, and E7) have either positive or negative scores on PC2 while none of the target volatile organics have any significant loadings on PC2. ADT, V/C, and the target VOCs are well presented in at least 6 of the experiments (E2, E3, E5, E6, E8, and E11), while TSS, TOC, and STD are presented in the rest (E1, E4, E7, E9, E10, and E12) on PC1. Moreover, the three central experiments (C1C3) are very close to the origin and have no significant scores on any of the PCs. This means that these central experiments are reruns of the same experiments and do not need to be included as part of the main experiments. The curvatures of the two factor response surfaces for toluene, ethylbenzene, meta and para xylene, and ortho-xylene were also checked by comparing their mean response values with the means of the three central experiments and no curvature was found for the target volatile organics in the calibration set. PLS Model Validation. In the PLS regression, the target volatile organics TOL, ETB, MPX, and OX were considered as dependent or measured variables and TSS, TOC, ADT, VC, and STD were considered as the predictor variables. A cross validation method,38 leaving one experiment out at a time from the calibration set was used to measure the standard error in cross validation (SECV). The following three criteria were used to decide the required number of PLS components for regression: • the SECV < 1 • 15% maximum difference between the percentage variance explained by the predictor and the measured variables • there is no dramatic increase in the percentage variance explained by the predictor with the inclusion of an additional PLS component. Table 2 shows the outcome of the PLS regression based on the above criteria. As a final step in model validation, data matrices were constructed from the eight remaining road sites. Each matrix consisted of the five predictor variables from the calibration set and
from vehicle parking and service areas. Therefore, this study incorporated the different land uses into the model calibration even though weak correlations existed between land use and VOC concentrations at the study sites. Factor Analysis. To extract the independent factors latent in the data structure, factor analysis (FA) was performed in two phases, namely, PCE for factor extraction followed by orthogonal VARIMAX rotation for the data matrices of all size fractions. The rotated component matrices for the total particulate fractions as well as all fractions including the dissolved fractions showed that two independent factors effectively separated the nine measured variables in Table 1. Table 1 shows that in both matrices TSS, ADT, TOC, and STD are associated with one factor while all the target volatile organics and V/C are associated with the other. These two independent factors were extracted based on the initial eigenvalue criteria g1. Out of eleven road sites, three were chosen to establish a calibration set and eight were chosen for the validation set for the PLS prediction model. Both the calibration and the validation sets consisted of a mix of residential, commercial, and industrial sites from both suburbs. Experimental Design. The calibration set for the PLS model was optimized using a two-level orthogonal experimental design.35,36 As the levels of the extracted factors were unknown,
Figure 2. PCA biplot of the calibration set shows the scores of fifteen experiments and the loadings of nine measured variables.
Table 2. PLS Regression Parameters for the Predictor Variables variance explained
variance explained by
coefficient of
measured
PLS
by predictor
measured
determination,
variables
components
variables, %
variables, %
r2
TOL ETB
1 2
65.42 68.28
67.84 53.26
0.7 0.6
regression coefficients for predictor variables SECV
TSS
TOC
ADT
VC
STD
0.66 0.99
0.00004 0.00001
0.00041 0.00025
0.00001 0.00000
0.09551 0.03352
0.16459 0.05634
MPX
1
65.51
59.97
0.69
0.73
0.00003
0.00034
0.00000
0.07334
0.12632
OX
1
65.64
71.14
0.71
0.60
0.00001
0.00024
0.00000
0.03529
0.06083
4456
dx.doi.org/10.1021/es200307x |Environ. Sci. Technol. 2011, 45, 4453–4459
Environmental Science & Technology
ARTICLE
Figure 3. Prediction results of (a) toluene; (b) and (c) ethylbenzene; (d) meta and para xylene; (e) ortho xylene at different size fractions showing most predictions were within 95% confidence limit.
eight objects. As the TSS and TOC were measured for the five different size fractions mentioned above, the PLS model was then applied to each of them in order to predict the target volatile organics in the five fractions. A 95% confidence limit was set to deal with the uncertainty in the prediction.39 The model was able to predict the target volatile organics for each of these fractions at most study sites with very few exceptions. Figure 3 presents the prediction results for each size fraction which includes at least one of the target volatile organics. In Figure 3a, at around 50% of the study sites, the predicted measurements of toluene are very close to their observed measurements for >300 μm. In Figure 3c, around 88% of the predicted measurements of ethylbenzene are very close to their observed measurements for 75150 μm. Acceptable results were achieved for other target volatile organics in Figure 3b, d, and e except for toluene at only one study site for >300 μm in Figure 3a. To achieve a better perception of the overall PLS model performance, five different error statistics were measured based on the relative prediction error.40 The relative prediction errors (RPE) for each of five size fractions were evaluated as
follows: 2
30:5
f
2
∑ ðXi, observed Xi, predicted Þ 7 6i¼1 7 RPEf ¼ 6 4 5 f 2 ∑ ðXi, observed Þ
ð1Þ
i¼1
and the total prediction error at the study sites was expressed as follows: 2
f
30:5
n
2
∑ ∑ ðXij, observed Xij, predicted Þ 6 6i¼1 j¼1 RPETotal ¼ 6 f n 4 ∑ ∑ ðXij, observed Þ2
7 7 7 5
ð2Þ
i¼1 j¼1
where n and f are number of sites and number of fractions respectively, and Xij denotes the concentration of the volatiles 4457
dx.doi.org/10.1021/es200307x |Environ. Sci. Technol. 2011, 45, 4453–4459
Environmental Science & Technology
ARTICLE
fractions as well as the total sample at three different land uses. Regardless of the test methods and the analytes, Horwitz42 suggested the acceptable interlaboratory coefficients of variation at the ppb level concentration to be (45%. Thus, in this study, the above-mentioned CV(%) suggest that the prediction model performed fairly accurately in three different land uses as well as for the five different size fractions.
’ ASSOCIATED CONTENT
bS Figure 4. Relative prediction error (RPE) at each size fractions and total RPE at all study sites.
Supporting Information. Detailed description of quality control measures in sample testing; the data analysis techniques used in this study; an additional figure (Figure S1) to present the study sites and the relevant traffic and pavement data. This information is available free of charge via the Internet at http:// pubs.acs.org.
’ AUTHOR INFORMATION Corresponding Author
*Tel: 61 7 3138 9945; fax: 61 7 3138 1170; e-mail: s.mahbub@ qut.edu.au.
Figure 5. Coefficients of variation (CV) at each size fraction and total sample at all study sites.
at i th fraction at the j th study site. Figure 4 gives the error statistics. In Figure 4, the RPE for the five size fractions ranged between 10% and 40% while the total RPE at all study sites ranged between 28% and 40% for the different target volatile organics using single component PLS prediction. Ni et al.41 reported up to 36% RPE for the prediction of nitrobenzene and nitro-substituted phenols using the single-component PLS method. The calibration set for their model could be prepared in the laboratory under stringent quality control measures, whereas in this study, the preparation of the calibration set was beyond laboratory control as measurements of some variables such as ADT, V/C, and STD could not be performed otherwise. Nevertheless, this study produced a robust PLS prediction model by exhibiting sufficient confidence interval (95%) in uncertainty of prediction. The main advantage of this model was the identification of a better calibration set for the orthogonal design by considering the independent factors latent in the original data matrix using the two phase factor analysis. This approach provided the calibration set underpinned by the factor optimization technique whereas the earlier study by Ni et al.41 used the number of measured variables as the factors in the optimization of the calibration matrix. Bias could be introduced into the calibration if all dependent variables are considered as factors. The intersite coefficients of variation (CV) for the VOC concentrations at ppb level at five different size fractions as well as the total sample were calculated and are shown in Figure 5. In Figure 5, the percentage CV of the predicted VOC concentration varied between 37% and 45% for toluene, 25% and 34% for ethylbenzene, 24% and 36% for meta and para xylene, and 36% and 45% for ortho xylene for the five different size
’ ACKNOWLEDGMENT This study was undertaken as a part of an Australian Research Council funded linkage project (LP0882637). P.M. gratefully acknowledges the postgraduate scholarship awarded by the Queensland University of Technology to conduct his doctoral research. Contributions from Vince Alberts at Queensland Health for VOC analysis and the support from Gold Coast City Council are also gratefully acknowledged. ’ REFERENCES (1) Buczynska, A. J.; Krata, A.; Stranger, M.; Godoi, A. F. L.; Kontozova-Deutsch, V.; Bencs, L.; Naveau, I.; Roekens, E.; Grieken, R. V. Atmospheric BTEX-concentrations in an area with intensive street traffic. Atmos. Environ. 2009, 43, 311–318. (2) Gilli, G.; Scursatone, E.; Bono, R.; Palin, L. Use of leaded gasoline and volatile halogenated hydrocarbon emission from automotive exhaust. Sci. Total Environ. 1989, 79, 281–286. (3) Wang, P.; Zhao, W. Assessment of ambient volatile organic compounds (VOCs) near major roads in urban Nanjing, China. Atmos. Res. 2008, 89, 289–297. (4) Mohamed, M. F.; Kang, D.; Aneja, V. P. Volatile organic compounds in some urban locations in United States. Chemosphere 2002, 47, 863–882. (5) Prev^ot, A. S. H.; Dommen, J.; B€aumle, M. Influence of road traffic on volatile organic compound concentrations in and above a deep Alpine valley. Atmos. Environ. 2000, 34, 4719–4726. (6) Staehelin, J.; Schl€apfer, K.; B€urgin, T.; Steinemann, U.; Schneider, S.; Brunner, D.; B€aumle, M.; Meier, M.; Zahner, C.; Keiser, S.; Stahel, W.; Keller, C. Emission factors from road traffic from a tunnel study (Gubrist tunnel, Switzerland). Part I: Concept and first results. Sci. Total Environ. 1995, 169, 141–147. (7) EPA. Test Methods for Evaluating Solid Waste, Physical/Chemical Methods; SW-846, 3rd ed.; U.S. Environmental Protection Agency: Washington, DC, 2008; Methods 5035 and 8260B. (8) WI DNR. Methods for Determining Gasoline Range Organics; PUBL-SW-140; Wisconsin DNR: Madison, WI, 1995. (9) IARC. IARC Monographs on the Evaluation of Carcinogenic Risks to Humans, 2009. http://monographs.iarc.fr/ENG/Classification/crthgr01.php. Accessed 3 Feb 2010. 4458
dx.doi.org/10.1021/es200307x |Environ. Sci. Technol. 2011, 45, 4453–4459
Environmental Science & Technology (10) Hunter, J. V.; Sabatino, T.; Gomperts, R.; MacKenzie, M. J. Contribution of urban runoff to hydrocarbon pollution. J. Water Pollut. Control Fed. 1979, 51, 2129–38. (11) Hoffman, E. J.; Latimer, J. S.; Mills, G. L.; Quinn, J. G. Petroleum hydrocarbons in urban runoff from a commercial land use area. J. Water Pollut. Control Fed. 1982, 54, 1517–1525. (12) Hoffman, E. J.; Mills, G. L.; Latimer, J. S.; Quinn, J. G. Urban runoff as a source of polycyclic aromatic hydrocarbons to coastal waters. Environ. Sci. Technol. 1984, 18, 580–587. (13) Herngren, L. Build-up and Wash-off Process Kinetics of PAHs and Heavy Metals on Paved Surfaces Using Simulated Rainfall. PhD Dissertation, Queensland University of Technology, Brisbane, 2005. (14) Olson, D. A.; Hammond, D. M.; Seila, R. L.; Burke, J. M.; Norris, G. A. Spatial gradients and source apportionment of volatile organic compounds near roadways. Atmos. Environ. 2009, 43, 5647–5653. (15) Smith, L.; Mukerjee, S.; Gonzales, M.; Stallings, C.; Neas, L.; € Norris, G.; Ozkaynak, H. Use of GIS and ancillary variables to predict volatile organic compound and nitrogen dioxide levels at unmonitored locations. Atmos. Environ. 2006, 40, 3773–3787. (16) Ogden, K. W.; Taylor, S. Y. Traffic Engineering and Management; Institute of Transport Studies, Monash University: Australia, 1999; pp 592594. (17) Federal Highway Administration. Technical Advisory Report T5040.36; U.S. Department of Transportation, 2005. http://www.fhwa. dot.gov/pavement/t504036.cfm. Accessed 10 December 2010. (18) Charlesworth, S. M.; Lees, J. A. Particulate associated heavy metals in the urban environment: Their transport from source to deposit, Coventry, UK. Chemosphere 1999, 39, 833–848. (19) Hwang, H. -M.; Foster, G. D. Characterization of polycyclic aromatic hydrocarbons in urban stormwater runoff flowing into the tidal Anacostia River, Washington, DC, USA. Environ. Pollut. 2006, 140, 416–426. (20) Moilleron, R.; Gonzalez, A.; Chebbo, G.; Thevenot, D. R. Determination of aliphatic hydrocarbons in urban runoff samples from the “Le Marais” experimental catchment in Paris centre. Water Res. 2002, 36, 1275–1285. (21) Shaw, S. B.; Parlange, J.-Y.; Lebowitz, M.; Walter, M. T. Accounting for surface roughness in a physically-based urban wash-off model. J. Hydrol. 2009, 367, 79–85. (22) Kreider, M. L.; Panko, J. M.; McAtee, B. L.; Sweet, L. I.; Finley, B. L. Physical and chemical characterization of tire-related particles: Comparison of particles generated using different methodologies. Sci. Total Environ. 2010, 408, 652–659. (23) GCCC. Gold Coast Priority Infrastructure Plan- Derivation of Trunk Road Infrastructure Network Charges: Final Report, Veitch Lister Consulting, 2006, 8 pp. http://www.goldcoast.qld.gov.au/ gcplanningscheme_0509/attachments/planning_scheme_documents/ part8_infrastructure/reference_documentation.pdf. Accessed 2 June 2010. (24) Mahbub, P.; Ayoko, G.; Egodawatta, P.; Yigitcanlar, T.; Goonetilleke, A. Traffic and climate change impacts on water quality: Measuring build-up and wash-off of heavy metals and petroleum hydrocarbons. In Rethinking Sustainable Development: Urban management, Engineering and Design; Yigitcanlar, T., Ed.; Engineering Science Reference: New York, 2010; pp 147167. (25) Huber, W. C. Deterministic modeling of urban runoff quality. In Urban Runoff Pollution; Torno, H. C., Desbordes, M., Marsalek, J., Eds.; NATO ASI series; Springer-Verlag: Berlin and New York; pp 167242. (26) Sartor, J. D.; Boyd, G. B.; Agardy, F. J. Water Pollution Aspects of Street Surface Contaminants. J. Water Pollut. Control Fed. 1974, 46, 458–467. (27) Egodawatta, P. K. Translation of small-plot scale pollutant build-up and wash-off measurements to urban catchment scale. PhD Dissertation, Queensland University of Technology, Brisbane, 2007. (28) Eaton, A. D., Clesceri, L. S., Rice, E. W., Greenberg, A. E., Eds. Standard Methods for the Examination of Water and Wastewater, 21st ed.; American Public Health Association: Washington, DC, 2005.
ARTICLE
(29) Patra, A.; Colvile, R.; Arnold, S.; Bowen, E.; Shallcross, D.; Martin, D.; Price, C.; Tate, J.; Apsimon, H.; Robbins, A. On street observations of particulate matter movement and dispersion due to traffic on an urban road. Atmos. Environ. 2008, 42, 3911–3926. (30) Mahbub, P.; Ayoko, G. A.; Goonetilleke, A.; Egodawatta, P.; Kokot, S. Impacts of traffic and rainfall characteristics on heavy metals build-up and wash-off from urban roads. Environ. Sci. Technol. 2010, 44, 8904–8910. (31) Filella, I.; Pe~ nuelas, J. Daily, weekly, and seasonal time courses of VOC concentrations in a semi-urban area near Barcelona. Atmos. Environ. 2006, 40, 7752–7769. (32) Chrysikopoulos, C. V.; Hildemann, L. M.; Roberts, P. V. Modeling the emission and dispersion of volatile organics from surface aeration wastewater treatment facilities. Water Res. 1992, 26, 1045–1052. (33) Johnson, T. D.; Belitz, K. Assigning land use to supply wells for the statistical characterization of regional groundwater quality: Correlating urban land use and VOC occurrence. J. Hydrol. 2009, 370, 100–108. (34) Pitt, R.; Field, R.; Lalor, M.; Brown, M. Urban stormwater toxic pollutants: assessment, sources, and treatability. Water Environ. Res. 1995, 67, 260–275. (35) Deming, S. N.; Morgan, S. L. Experimental Design: A Chemometric Approach, 2nd ed.; Elsevier Science Publishers: Amsterdam, 1993; pp 262264. (36) Lan, W. G.; Wong, M. K.; Chen, N.; Sin, Y. M. Orthogonal array design as a chemometric method for optimization of analytical procedures part 2.* Four-level design and its application in microwave dissolution of biological samples. Analyst 1994, 119, 1669–1675. (37) SIRIUS (2008). SIRIUS version 7.1 copyright 19872008. Pattern Recognition Systems AS, Help Topics. http://www.prs.no. (38) Wold, S. Cross-Validatory Estimation of the Number of Components in Factor and Principal Components Models. Technometrics 1978, 20, 397. (39) Watson, C. J.; Billingsley, P.; Croft, D. J.; Huntsberger, D. V. Statistics for Management and Economics, 5th ed.; Allyn and Bacon: Boston, 1993; pp 343348. (40) Otto, M.; Wegscheider, W. Spectrophotometric multicomponent analysis applied to trace metal determinations. Anal. Chem. 1985, 57, 63–69. (41) Ni, Y.; Wang, L.; Kokot, S. Simultaneous determination of nitrobenzene and nitro-substituted phenols by differential pulse voltammetry and chemometrics. Anal. Chim. Acta 2001, 431, 101–113. (42) Horwitz, W. Evaluation of Analytical Methods Used for Regulation of Foods and Drugs. Anal. Chem. 1982, 54, 67A–76A.
4459
dx.doi.org/10.1021/es200307x |Environ. Sci. Technol. 2011, 45, 4453–4459
