ARTICLE IN PRESS
Atmospheric Environment 39 (2005) 1323–1334 www.elsevier.com/locate/atmosenv
Source apportionment based on an atmospheric dispersion model and multiple linear regression analysis Akihiro Fushimia,, Hiroto Kawashimab, Hideo Kajiharac a
Traffic Pollution Control Research Team, PM2.5 and DEP Research Project, National Institute for Environmental Studies, 16-2 Onogawa, Tsukuba, Ibaraki 305-8506, Japan b Department of Management Science and Engineering, Faculty of Systems Science and Technology, Akita Prefectural University, 84-4 Ebinokuchi, Tsuchiya, Honjo City, Akita 015-0055, Japan c Graduate School of Science and Technology, Niigata University, 8050, Ikarashi-Ninocho, Niigata, Niigata 950-2181, Japan Received 22 March 2004; received in revised form 19 October 2004; accepted 10 November 2004
Abstract Understanding the contribution of each emission source of air pollutants to ambient concentrations is important to establish effective measures for risk reduction. We have developed a source apportionment method based on an atmospheric dispersion model and multiple linear regression analysis (MLR) in conjunction with ambient concentrations simultaneously measured at points in a grid network. We used a Gaussian plume dispersion model developed by the US Environmental Protection Agency called the Industrial Source Complex model (ISC) in the method. Our method does not require emission amounts or source profiles. The method was applied to the case of benzene in the vicinity of the Keiyo Central Coastal Industrial Complex (KCCIC), one of the biggest industrial complexes in Japan. Benzene concentrations were simultaneously measured from December 2001 to July 2002 at sites in a grid network established in the KCCIC and the surrounding residential area. The method was used to estimate benzene emissions from the factories in the KCCIC and from automobiles along a section of a road, and then the annual average contribution of the KCCIC to the ambient concentrations was estimated based on the estimated emissions. The estimated contributions of the KCCIC were 65% inside the complex, 49% at 0.5-km sites, 35% at 1.5km sites, 20% at 3.3-km sites, and 9% at a 5.6-km site. The estimated concentrations agreed well with the measured values. The estimated emissions from the factories and the road were slightly larger than those reported in the first Pollutant Release and Transfer Register (PRTR). These results support the reliability of our method. This method can be applied to other chemicals or regions to achieve reasonable source apportionments. r 2004 Elsevier Ltd. All rights reserved. Keywords: Benzene; Air pollutants; Industrial complex; Industrial source complex model (ISC); Receptor model
1. Introduction
Corresponding author. Tel.: +81 29 850 2752; fax: +81 29 850 2569. E-mail address:
[email protected] (A. Fushimi).
It is important to understand the contribution of each emission source of air pollutants to ambient concentrations to establish effective measures for risk reduction. For example, in 1997, automobiles in Japan contributed 66% of the nationwide total emissions of benzene
1352-2310/$ - see front matter r 2004 Elsevier Ltd. All rights reserved. doi:10.1016/j.atmosenv.2004.11.009
ARTICLE IN PRESS 1324
A. Fushimi et al. / Atmospheric Environment 39 (2005) 1323–1334
(Japan Ministry of Economy, Trade and Industry, 2000), a known human carcinogen (US Environmental Protection Agency (EPA), 1998). Although industrial complexes also apparently contribute to the ambient benzene concentrations in their surrounding area, judging from the ambient concentrations, their contributions have rarely been investigated quantitatively in Japan. In this paper, to obtain the contribution of each source of a pollutant to the ambient concentrations is called ‘‘source apportionment’’. Source contributions can be estimated by using available emission data and a simulation model. Although the Pollutant Release and Transfer Register (PRTR) system, which reports emissions and transfers of 354 toxic chemicals, was implemented in 2001 in Japan, the validity of the PRTR emission data is controversial. Fushimi et al. (2002) suggested that the benzene emissions reported in the PRTR, especially from automobiles, were underestimated. In the paper, it was shown that the nitrogen oxides (NOx) concentrations in Tokyo Metropolis and a city were calculated accurately by the Gaussian plume dispersion model called Industrial Source Complex Long-Term Model version 3 (ISCLT3, US Environmental Protection Agency and Office of Air Quality Planning and Standards (US EPA), 1995). In contrast, using the emission data reported in the 1998 PRTR Pilot Project (Japan Environment Agency, 1999), calculated benzene concentrations were much smaller than the measured values. When benzene emissions were re-estimated based on reported emission data other than the PRTR data, the re-estimated emission from automobiles increased 6.7-fold. Using the re-estimated emission data, the calculated benzene concentrations agreed well with the measured values. Later, in the first full implementation of the PRTR (2001 data, Japan Ministry of Environment and Japan Ministry of Economy Trade and Industry, 2003), the method of estimating emissions was modified; as a result, emission factor of benzene for gasoline passenger cars increased 5.6-fold, and that for all types of vehicles increased 1.7-fold, compared with the values used in the PRTR Pilot Project. Source contributions can be also estimated from ambient concentrations by using receptor models (Hopke, 1991), for example, by using the chemical mass balance (CMB) method (Watson et al., 2001) or principal component analysis (PCA). To adapt these receptor models in the vicinity of industrial complexes, source profiles, that is, emission amount ratios for multiple chemicals, are needed for specific factories. Source profiles are reported for typical sources (Watson et al., 2001), but are difficult to obtain for specific sources. Some attempts (Derwent et al., 1995; Palmgren et al., 1999) have been made to estimate the emission rates from automobiles along a specific road based on ambient concentrations measured at the roadsides and
simulation models. However, estimating emission rates from factories in industrial complexes is difficult because many emission sources are located in a small area. Thus, no practical method has been available for estimating the emission rates from factories in industrial complexes based on the ambient data. In this paper, we have developed a source apportionment method that can be applied in the vicinity of industrial complexes and does not require knowledge of emission amounts or source profiles. The method is based on an atmospheric dispersion model and multiple linear regression analysis (MLR) in conjunction with ambient concentrations simultaneously measured at sites in a grid pattern. We have tested the method by applying it to the case of benzene near an industrial complex.
2. Ambient measurements 2.1. Methods 2.1.1. Study area The coastal residential area of Ichihara City was chosen as the study area (Fig. 1). In this area, high ambient benzene concentrations have been observed for several years; annual average benzene concentrations in the area, measured by the local governments in 1997 (Japan Environment Agency, Air Quality Bureau, 1998), were 9.8 mg m3 at Iwasaki-nishi, 5.6 mg m3 at Kashi, and 4.2 mg m3 at Tsuiheiji Station. These values are much higher than the national average concentration of 3.0 mg m3. Ichihara City is within a 50-km radius of the Tokyo Metropolitan Area. It has an area of 368 km2 and a population of about 280,000 (Ichihara City, Department of Environment, 2000). In this area, a residential area lies inland (east) of Route 16. The Keiyo Central Coastal Industrial Complex (KCCIC), which is one of the biggest industrial complexes in Japan, lies more than 20 km along the seaside (west) of Route 16 (Fig. 1). Thus, the residential area is only a few kilometers from the KCCIC. The KCCIC comprises more than 20 factories related to the petrochemical and iron industries and produces about 1.8 million ton yr1 of ethylene (Petrochemical Press, 1998). Although the KCCIC as well as automobiles apparently contribute to the ambient benzene concentrations in this area, the source contributions have not been quantified. Our goal was to estimate quantitatively the relative contributions of the KCCIC and automobiles to the ambient benzene concentrations in the residential area, which includes Iwasaki-nishi, Kashi, and Tsuiheiji Station.
ARTICLE IN PRESS A. Fushimi et al. / Atmospheric Environment 39 (2005) 1323–1334
1325
Fig. 1. Study area, target sources and measurement sites. Iw: Iwasaki-nishi, Ks: Kashi, Ts: Tsuiheiji. 0-km sites: Sites 1–16, 0.5-km sites: Sites 17–20, 1.5-km sites: Sites 21–23 and Sites 30 and 31, 3.3-km sites: Sites 24 and 25 and Sites 27–29, and 5.6-km site: Site 26.
2.1.2. Sampling design Benzene concentrations were measured simultaneously at 31 sites of a grid established in the KCCIC and the surrounding residential area. The measurement sites, shown as small circles in Fig. 1, are collectively called the ‘‘grid network’’ in this paper. To discriminate the contribution of each factory and road, it was suggested that the concentrations should be measured between each factory. Therefore, the grid network was established with intervals of about 1–2 km between measurement sites. To avoid a large effect from nearby roads, each measurement site was at least about 100 m from the closest arterial road. Measurements were carried out six times: on 7 (Fri), 20 (Thu), and 25 (Tue) December in 2001, and on 14 (Fri) and 24 (Mon) June and 12 (Fri) July in 2002 (Table 1). Sites 1–16 were within the KCCIC, and Sites 17–31 were in the residential area.
2.1.3. Sampling procedures Active sampling was carried out with adsorption tubes and Pocket Pumps (210-1002, SKC Inc., Eighty Four, PA, US). The sampling time for each sample was about 3 h. The time of day that each sample was collected differed by about 3 h because only one car was used for setup and recovery of all samples, but this difference was not taken into account in the analysis. Two adsorbents were used: Carbopack B (60/80 mesh)+Carboxen 1000 (60/80 mesh) (‘‘Carbo’’ hereafter) was used in December 2001, and Tenax TA (60/80 mesh, 170 mg, ‘‘Tenax’’ hereafter) was used from June to July 2002. Glass tubes (1/4f 3.5 in, Supelco, Bellefonte, PA, US) were used with both adsorbents. For the Carbo sampling, the sampling flow rate was 40 mL min1, and the sampling volume was 7.2 L. For the Tenax sampling, the sampling flow rate varied from 17.4 to 32.0 mL min1, and the sampling volume varied from 3.6 to 8.0 L.
ARTICLE IN PRESS A. Fushimi et al. / Atmospheric Environment 39 (2005) 1323–1334 2.8 (n ¼ 129) (0.1–37.0) 4.1 (n ¼ 29) (1.0–33.5)
Breakthrough of benzene did not occur during either the Carbo or Tenax samplings, given the breakthrough values reported by PerkinElmer Japan (2000) and Scientific Instrument Services, Inc. (2002).
n is the number of the sampling sites.
9.1 (n ¼ 7) (0.8–37.0) Avg Min–Max Avg
1.6 (n ¼ 20) (0.1–5.5)
2.0 (n ¼ 18) (0.5–5.3)
2.2 (n ¼ 24) (0.7–3.7)
2.0 (n ¼ 31) (0.7–11.1)
1.5 (n ¼ 50) (0.2–4.2) 1.6 (n ¼ 13) (1.0–3.1) — — Avg Min–Max Residential area
1.2 (n ¼ 7) (0.2–2.8)
1.1 (n ¼ 5) (0.5–2.6)
1.9 (n ¼ 10) (0.7–2.6)
1.3 (n ¼ 15) (0.7–4.2)
3.7 (n ¼ 79) (0.1–37.0) 6.2 (n ¼ 16) (1.4–33.5) 9.1 (n ¼ 7) (0.8–37.0) Avg Min–Max Inside the KCCIC
1.9 (n ¼ 13) (0.1–5.5)
2.3 (n ¼ 13) (0.8–5.3)
2.5 (n ¼ 14) (1.7–3.7)
2.6 (n ¼ 16) (0.7–11.1)
Avg 25 December 2001 NNE 20 December 2001 NW 7 December 2001 N Date Wind direction
Table 1 Summary of the benzene concentrations (mg m3) simultaneously measured at 31 sites in the grid network
14 June 2002 SE
24 June 2002 SSE
12 July 2002 ESE
1326
2.1.4. Chemical analyses After each sampling, the adsorption tubes were sealed with a Teflon-ferruled brass end-cap and stored in a sealed aluminum foil bag until analysis, and were then analyzed within about 7 days. Samples were thermally desorbed and analyzed by gas chromatograph/mass spectrometer (GC/MS) (HP 6890A GC, 5973N MS, Agilent Technologies, Palo Alto, CA, US); a TurboMatrixATD (PerkinElmer, Wellesley, MA, US) was used for thermodesorption. The analytes were firstly thermodesorbed from Carbo or Tenax, and concentrated by a second trap, and then thermodesorbed for GC/MS. For Carbo, the first desorption was carried out at 320 1C for 10 min in a flow of 70 mL min1 helium. The analytes were trapped by a second thermodesorption tube (Air Monitoring, PerkinElmer) at 30 1C in a flow of 40 mL min1 helium. The second thermodesorption was carried out at 320 1C for 10 min in a flow of 11 mL min1 helium. For Tenax tubes, the first desorption was carried out at 240 1C, and the second desorption, in a tube filled with Tenax TA (60/80), was carried out at 240 1C. The analytes were detected by MS in SCAN mode with m/z of 35–200. A capillary column, type HP-5MS, 30 m, 0.25 mm i.d., 0.25 mm df (Agilent Technologies), was used in a flow of 1 mL min1 helium, employing the following temperature program: for Carbo; 5 min at 25 1C, from 25 to 250 1C in steps of 10 1C min1, from 250 to 325 1C in steps of 20 1C min1, and 10 min at 325 1C; the same program was used for Tenax except that the initial temperature was 30 1C. The Japanese Indoor Air Standard Mix (100 mg mL1, 52 chemicals, methanol:water ¼ 95:5, Supelco) was used for calibration. The adsorption tubes were spiked with three levels of the standard solution in a flow of nitrogen and analyzed by GC/MS under the same conditions as were used for the ambient samples. Using the three levels of the analyzed data and the origin, calibration curves were obtained. Determination coefficients of the calibration curves for benzene were greater than 0.98 for every measurement. Seven standard samples with the lowest level of the calibration curves were analyzed, and the detection limits (DL) and the quantification limits (QL) were calculated as 3 times and 10 times the standard deviations of the analyzed values, respectively. In July 2002, for example, DL and QL of benzene concentrations were 0.18 and 0.59 mg m3, respectively.
ARTICLE IN PRESS A. Fushimi et al. / Atmospheric Environment 39 (2005) 1323–1334
2.2. Results Measured benzene concentrations at the 31 sites on the 6 days are summarized in Table 1. In December 2001, a northerly (onshore) wind was dominant. In contrast, in June and July 2002, a southerly or easterly (offshore) wind was dominant. On all of the measurement days, benzene concentrations in the inland residential area were relatively low; most were lower than the Japanese environmental standard of 3 mg m3. Inside the KCCIC, benzene concentrations were higher than those in the residential area, and they varied widely among the sites within the complex. The maximum observed concentration was 37.0 mg m3. At sites at the outer ends of the piers, benzene concentrations were consistently lower than about 1 mg m3 when the wind was onshore. In contrast, concentrations higher than 10 mg m3 were observed at those sites when the wind was offshore. Benzene concentrations tended to decrease with the distance from the KCCIC. This trend was recognized under both offshore and onshore wind conditions; no significant difference in concentrations between offshore and onshore wind conditions was found. The benzene concentrations measured on the six days were averaged at each of the 31 sites (Fig. 2). In the residential area, benzene concentrations were 0.8–2.7 mg m3, that is, relatively low. Inside the KCCIC, the concentrations were 1.7–10.0 mg m3, that is, higher than those in the residential area. Benzene
1327
concentrations decreased with distance from the KCCIC. Measured benzene concentrations were averaged according to their distance from the KCCIC. Thus, the 31 measurement sites were classified into 0-km sites (inside the KCCIC, n ¼ 16), 0.5-km sites (0.4–0.6 km, n ¼ 4), 1.5-km sites (1.4–1.6 km, n ¼ 5), 3.3-km sites (2.8–3.9 km, n ¼ 5), and a 5.6-km site (n ¼ 1), where the distance from the KCCIC was the distance from the freight-train tracks, which parallel Route 16 at a distance of 10 m on its seaward side. Average benzene concentrations were 3.7 mg m3 inside the KCCIC, 1.8 mg m3 at 0.5-km sites, 1.2 mg m3 at 1.5-km sites, 1.3 mg m3 at 3.3-km sites, and 1.0 mg m3 at the 5.6-km site.
3. Source apportionment 3.1. Framework We developed a source apportionment method based on an atmospheric dispersion model and MLR in conjunction with the ambient concentrations measured simultaneously in the grid network. The method assumes that the benzene concentration measured at a specific site can be described by the sum of the concentrations contributed by the surrounding major sources of benzene and the background concentration.
Fig. 2. Benzene concentrations simultaneously measured at 31 sites in the grid network (6-day average). The values in parentheses are the numbers of the monitoring data.
ARTICLE IN PRESS A. Fushimi et al. / Atmospheric Environment 39 (2005) 1323–1334
1328
Step 1: Measure the ambient concentrations simultaneously at sites in a grid network. Step 2: Calculate concentrations contributed by each source based on the standardized common emission rate by a dispersion model. Step 3: Perform multiple linear regression analysis on a several-day average basis. Emission rate of each source is obtained. Step 4: Calculate annual average concentrations contributed by each source using estimated emissions Step 5: Calculate contributions of each source on an annual average basis Fig. 3. Flow of this study.
This relationship is described as a multiple linear regression equation
→ Meas. sites →
→ Sources →
c1 c2
x1.1 x1.2 x2.1 x2.2
x1.n x2.n
a1 a2
=
cm Meas. conc
b b +
xm .1 xm .2 Calc. conc.
xm .n
(1)
an
b
Reg. coef.
Const.
where ci (mg m3) is the measured concentration at site i ( ¼ 1, 2, y, m), xij (mg m3) is the concentration at site i contributed by source j ( ¼ 1, 2, y, n) obtained by assuming a standardized common emission rate, aj is the regression coefficient for source j; and b (mg m3) is a constant. Although the benzene emission from each source may vary from day to day, this variation was not considered. The study was divided into five steps as shown in Fig. 3. Based on the measured and calculated concentrations obtained during Step 1 and Step 2, respectively, severalday averages of benzene emissions from each source were estimated in Step 3. The several-day average contribution from each source can be obtained from the results of Step 3, but Steps 4 and 5 were performed to obtain annual average contributions. The method and the results of Step 1 are described above. Those of the remaining steps are described below. 3.2. Estimation of emissions on a several-day average basis In Step 2, the benzene concentration xij at each measurement site contributed by major sources was calculated by assuming that each source had the same emission rate per unit area. Seventeen major sources were designated: 11 factories and five piers (each pier
consisted of multiple factories), representing the KCCIC, and a section of Route 16 (Fig. 1). The daily traffic count for Route 16 is about 50,000 vehicles. The target section of Route 16, most of the section shown in Fig. 1, was 24.2 km long with an average width of 39 m. These 17 sources used in the source apportionment are called ‘‘target sources’’ hereafter. Four factories in the KCCIC whose benzene emissions as reported in their environmental reports were considered to be zero were eliminated in advance from consideration as target sources. Moreover, two factories and a pier that were part of a petroleum refinery were eliminated from the analysis, because the total benzene emission from these factories was much smaller than that from the chemical and iron industries in this area (Japan Ministry of Economy, Trade and Industry, 2001). For the calculation, the ISCST3 (US EPA, 1995) was used. The ISCST3 and the ISCLT3 (US EPA, 1995), recommended by the US EPA, have been used worldwide (Angius et al., 1995; Hao et al., 2000; Lorber et al., 2000; Evans et al., 2002). Calculation conditions for the ISCST3 and assumptions used in the analysis were as follows: Each factory was represented by one or more rectangles (Fig. 1), and a uniform emission rate in the area of each factory (area source) was assumed. A standardized common emission rate of 31.5 ton km2 yr1 was used for all sources. The emission altitude for all factories was assumed to be 10 m. The effect of buoyancy was not considered because benzene is expected to be emitted from storage tanks, drying facilities, and other non-buoyant sources. Route 16 was represented by multiple narrow rectangles, the emission rate from the target section was assumed to be constant, and the emission altitude used was 1.5 m. The altitude used for the calculation sites, that is, corresponding to the measurement sites, was 1.0 m. A comparison of wind data among several coastal stations suggested that the wind directions obtained at the Tokyo Electric Power Company (altitude 173 m) were representative of the wind directions in the coastal area. Therefore, wind direction and wind speed, converted to the surface speed by an empirical formula, measured at the Tokyo Electric Power Company were used. Temperature and insolation measured at Tsuiheiji were used, because these parameters were not measured at the Tokyo Electric Power Company. Atmospheric stability classes, based on the Pasquill stability classes, were calculated every hour from surface wind speed, insolation, and cloudiness. Mixing heights were 600, 500, 500, and 350 m for stability classes A, B, C, and D, respectively (Higashino et al., 2000). Briggs urban dispersion parameters were used. Benzene concentrations were calculated for each measurement hour, and then all of the calculated concentrations for each site were averaged. Then, the 6-day average concentration xij for a 31 17-matrix basis was calculated.
ARTICLE IN PRESS A. Fushimi et al. / Atmospheric Environment 39 (2005) 1323–1334
Fig. 4 shows the relationship between 6-day averages of measured concentrations and those estimated by MLR. Each point represents the data from one measurement site. The measured concentrations and the estimated concentrations agreed well (R2 ¼ 0:79). Thus, the seven sources retained as positive sources sufficiently explained the measured concentrations. There was no clear bias between the measured concentrations and the errors between the measured and the estimated concentrations. Estimated benzene emissions from the sources retained at the end of the MLR are shown in Table 2. Here, emissions obtained as 6-day averages were converted to annual average values. Estimated benzene
The correlation matrix among the target sources was obtained from the calculated concentration vectors xi : Four pairs of the sources that were highly correlated (R40:70) with each other in the correlation matrix were merged and treated as single sources: Piers A and B (R ¼ 0:96), Factories B and C (R ¼ 0:82), Factories E and F (R ¼ 0:81), and Piers C and D (R ¼ 0:77). This procedure decreased the number of the target sources from 17 to 13. In Step 3, MLR was performed based on 6-day averages of the measured concentrations ci and the calculated concentrations xij for the 31 measurement sites. The regression coefficients aj and the constant b were obtained by MLR. After the first MLR iteration, negative sources, that is, those with negative regression coefficients, were eliminated one by one, starting with the negative coefficient with the largest absolute value. MLR was performed until no negative source remained. The benzene emission rate of each source was obtained by multiplying aj by the standardized common emission rate, because xij values were calculated based on the standardized common emission rate, and concentrations contributed by a specific source were proportional to the emission rate of that source. Results of the MLR are summarized in Table 2. The seven sources (Factories A, B+C, E+F, I, J, and K and Route 16) that were retained as positive sources are shown in Table 2. Six sources (Factories D, G, and H, and Piers A+B, C+D, and E) were eliminated during the MLR as negative sources. The p-value of each source was 0.41 or lower. The calculated constant, 1.2 mg m3, was close to the averaged 6-day concentrations (1.3 mg m3) measured at the 3.3–5.6 km sites. This constant value seems reasonable for the sum of the concentrations contributed by wide-area sources, excluding the target sources used in the analysis, such as roads other than Route 16.
1329
Fig. 4. Measured benzene concentrations and the concentrations estimated by MLR (6-day average).
Table 2 Results of MLR (6-day average) Source
Area (km2)
Regression coefficient
p-value
Estimated benzene emission (ton yr1) Mode
95% lower limit
95% upper limit
17 28 7.1 14 6.5 8.3 28
34 61 98 49 44 20 66
0.54a
1.8a
Factory A Factory E+F Factory I Factory K Factory B+C Factory J Route 16
0.10 0.87 0.67 0.62 1.90 0.62 0.94
8.7 1.6 2.5 0.91 0.32 0.31 0.64
o0.01 o0.01 0.03 0.26 0.14 0.39 0.41
26 45 52 18 19 6.0 19
Constant
—
1.2
o0.01
1.2
Total (factories) Total (all sources) a
Unit: mg m3.
166 184
a
ARTICLE IN PRESS A. Fushimi et al. / Atmospheric Environment 39 (2005) 1323–1334
1330
emissions from each factory were 6.0–52 ton yr1. The total emission from the six factories that represented the KCCIC was 166 ton yr1. The estimated emission from Route 16 was 19 ton yr1. 3.3. Source apportionment on an annual average basis In Step 4, annual average concentrations x0ij contributed by the sources retained at the end of the MLR (Table 2) were calculated by using the emission rates estimated in Step 3 and the annual meteorological data. For this calculation, the ISCLT3 was used. The spatial distribution of the annual average concentrations contributed by the KCCIC was also obtained. Benzene concentrations contributed by the KCCIC to the residential area were 1–4 mg m3 near Route 16, and more than 4 mg m3 near Iwasaki-nishi Station. The estimated concentrations, which consisted of the sum of those concentrations contributed by the KCCIC and Route 16 and the constant 1.2 mg m3, were averaged for each distance from the KCCIC (Fig. 5). The estimated benzene concentrations contributed by all sources and by the KCCIC declined sharply with distance from the KCCIC. In Step 5, annual average contributions of the KCCIC to the surrounding ambient benzene concentrations were estimated. The annual average contributions of the KCCIC were calculated as the ratio of Saj xij ðfactoriesÞ=fSaj xij ðfactories and roadÞ þ bg: The spatial distribution of the annual average contribution of the KCCIC was also obtained. Here, the constant b obtained by the MLR in Step 3 was used. Fig. 6 shows the spatial distribution of the estimated annual average contributions of the KCCIC. Large contributions of KCCIC were found inside the KCCIC. Contributions of the KCCIC at the 31 measurement
6
Benzene conc. ( gm-3)
5
80
4 60 3 40 2 20
1
0
Contribution of the KCCIC (%)
100 Estimated conc.. by alll sources Estimated conc. by the KCCIC Estimated contribution of the KCCIC
0 0
1
2 3 4 Distance from the KCCIC (km)
5
6
Fig. 5. Decline curve of the estimated benzene concentrations and the estimated relative contributions of the KCCIC (annual average).
sites were averaged according to their distance from the KCCIC (Fig. 5). The estimated relative annual average contributions of the KCCIC was 64% inside the KCCIC, 49% at 0.5-km sites, 35% at 1.5-km sites, 20% at 3.3-km sites, and 9.2% at the 5.6-km site; the relative contribution declined monotonously with distance. The contributions of the KCCIC were also estimated at the local governments’ measurement sites (Table 3). The estimated relative contribution of the KCCIC was 72% at Iwasaki-nishi, 41% at Kashi, and 28% at Tsuiheiji. Thus, at the three sites, the contributions of the KCCIC were large, whereas the contribution from Route 16 was 5% or less, and the contribution from the background concentration, that is, contribution from other than the KCCIC or Route 16, was also relatively large (Table 3). The estimated annual average concentrations at the three sites were consistent within an error of 20% with those measured by the local governments (Table 3). Our measurements and the local governments’ measurements were done independently, and concentrations at Iwasaki-nishi and Kashi were not measured in this study. These results support the reliability of our method. The contribution of the industrial complex to the surrounding ambient concentrations has become clear by the application of our source apportionment method.
4. Discussion As described above, the 6-day average measured concentrations and those estimated by MLR for the 31 measurement sites agreed well (Fig. 4). Moreover, the estimated annual average concentrations at the local governments’ measurement sites were consistent with those measured by the local governments (Table 3). The constant of the multiple linear regression equation also seemed reasonable. Some aspects of the source apportionment method are discussed below. 4.1. Comparison of the estimated emissions with the PRTR data The benzene emissions of the factories in the KCCIC estimated in Step 3 were compared with the 2001 PRTR data, which was reported after our analyses were completed (Table 4). Those factories in the study area that emitted more than 0.1 ton yr1 are listed in the table. In general, the estimated emissions were slightly higher than those reported in the PRTR. The total emission reported in the PRTR for the six factories retained as positive sources by the analysis was 84.3 ton yr1, or about half of the of 166 ton yr1 estimated in this study. Although no emission from
ARTICLE IN PRESS A. Fushimi et al. / Atmospheric Environment 39 (2005) 1323–1334
1331
Fig. 6. Spatial distribution of the estimated relative contributions of the KCCIC to the ambient benzene concentrations (annual average).
Factory A was reported in the PRTR, Factory A might emit benzene because it uses benzene to produce phenol. The benzene emissions from Piers D and E reported in the PRTR were relatively large, but these piers were eliminated during this analysis. Factory K might have been incorrectly interpreted in the analysis as a benzene source in place of Piers D and E because Factory K does not use benzene, but it is in the same direction as Piers D and E from the measurement sites. However, the effect of the misunderstanding of emission sources on the relative contribution of KCCIC in the surrounding residential area was estimated to be less than 10%, from the source apportionment results. The sum of the estimated emissions of the factories in the KCCIC was 1.5 times larger than that reported in the PRTR. However, the estimated benzene emissions from the factories may not have been overestimated; instead, the PRTR data may be underestimated. The estimated spatial distribution of the benzene sources was in good agreement with that shown by the PRTR data. The benzene emission from Route 16 estimated in this study was compared with the emissions calculated by multiplying the average traffic volume on Route 16 by emission factors. Total traffic volumes counted at four points along Route 16 in Ichihara City in 1999 were 33,000–68,000 vehicles day1 (Japan Ministry of Land, Infrastructure and Transport, 2002). Assuming that
50,000 vehicles travel the target section (24.2 km-long) of Route 16 in a day, the total distance traveled was calculated to be 121 million vehicle km1 day1. Two emission factors for all types of vehicles, 10 mg km1, reported in the 2001 PRTR, and 52 mg km1, estimated previously by us (Fushimi et al., 2002), were used to calculate the emissions from Route 16. Using the PRTRbased emission factor and our estimated value, benzene emissions from the target section of Route 16 were estimated as 4.6 and 23 ton yr1, respectively. The value of 19 ton yr1 estimated in this study was between these two estimated values.
4.2. Treatment of negative sources When many sources are introduced in a MLR such as that used in Step 3, multicollinearity between sources often occurs, causing some regression coefficients, emissions in our analysis, to have negative values. When this happens, sources whose contributions to the ambient concentrations are not large should be eliminated or merged with other sources before or during the analysis. However, some subjectivity is introduced in the process of determining which sources should be eliminated, which can affect the results. In this paper, only the result that showed the most consistency between the
ARTICLE IN PRESS A. Fushimi et al. / Atmospheric Environment 39 (2005) 1323–1334
FY 2001 data (Japan Ministry of Environment, 2002), annual average concentrations obtained from 12 measurements conducted over a 24 h period. Average of two measurements. b
a
4.5 (0.59–11) 2.7 (0.25–10) 1.6 (0.17–3.4)
— — 0.80b
5.1 2.2 1.7
1.13 0.81 1.07
72 41 28
5.3 4.6 2.3
23 54 70
Table 4 Benzene emissions from the factories in the KCCIC estimated in this study and reported in the PRTR
Iwasaki-nishi Kashi Tsuiheiji
Route 16 KCCIC
Estimated contribution (%) Ratio (est./meas. by the local governments) Est. conc. (mg m3) Meas. conc. in this study (mg m3) Meas. conc. by the local governments (mg m3) a
Table 3 Measured and estimated benzene concentrations and the relative contributions from the KCCIC at the local governments’ measurement sites (annual average)
Background
1332
Factories
This study (ton yr1)
PRTRa (ton yr1)
I E+F A B+C K J
52 45 26 19 18 6.0
17.0 42.9 — 8.4 — 16.0
Pier D Pier E Pier B Others
— — — —
Total
166
b b b c
20.0 3.9 2.3 3.1 113.6
a
Japan Ministry of Environment and Japan Ministry of Economy Trade and Industry (2003). b Eliminated during the analysis. c Eliminated before the analysis.
measured and estimated concentrations was described. However, with the other approaches, total emissions and spatial distributions of the emissions were almost the same as described in this paper, although the sources retained at the end of the analysis were somewhat different.
4.3. Averaging the measured and calculated concentrations used in MLR In order to estimate the contributions of factories located close together within a small area with our method, it is desirable that concentrations measured under most wind directions be used. Otherwise, results may be biased, or may not be obtained. Although MLR can be performed six times using the measured data for each day, results could not be obtained in some cases, suggesting that when the wind is offshore, the measured concentrations at inland sites cannot be sufficiently explained by the target sources. To overcome this problem, MLR was performed using 6-day average values in this paper. A measurement determined over a period of one or several days might be enough if during that period most wind directions occurred. However, it is more difficult to maintain power to the pumps and measurement sites during longer measurement periods. Continuous measurements were also conducted at three sites that enclose the KCCIC in this study. However, only simultaneous data measured in the grid network were used in the analysis because the continuously measured data did not appear to be efficient enough in our method. It was found that simultaneous measurements at multiple sites in the grid network were
ARTICLE IN PRESS A. Fushimi et al. / Atmospheric Environment 39 (2005) 1323–1334
effective for quantitative source apportionment among nearby multiple sources. 4.4. MLR by wind direction In order to determine the sensitivity of the measured concentrations used in the analysis, MLR was performed using averages of measurements obtained when the wind was onshore (7, 20, and 25 December 2001) and of those obtained when the wind was offshore (14 and 24 June and 12 July 2002). The total estimated emissions were 190 and 187 ton yr1, respectively, for each case. These values are consistent with the 6-day average estimate of 184 ton yr1. Spatial distributions of the emissions were also consistent among the three cases. However, the results were slightly different in detail. When the wind was onshore, the estimated emission from Route 16 was zero, and the constant had a smaller value, 0.5 mg m3. When the wind was offshore, the estimated emission from Route 16 was 16 ton yr1, and the constant was larger, 1.5 mg m3. These results can be explained as follows: When the wind was onshore, emissions from the factories contributed greatly to benzene levels in the inland area, so the contributions estimated from Route 16 and non-target sources were relatively smaller. In contrast, when the wind was offshore, the constant was larger because inland benzene concentrations could not be sufficiently explained only by the contributions from the factories and Route 16.
1333
zene emissions from the factories in the KCCIC and a section of Route 16 were estimated by MLR based on 6day average measured and calculated concentrations. The emissions from the KCCIC and Route 16 were estimated to be 166 and 19 ton yr1, respectively. The annual average contributions of the KCCIC to the ambient concentrations were estimated based on the estimated emissions. The estimated relative contributions of the KCCIC were 65% inside the complex, 49% at 0.5-km sites, 35% at 1.5-km sites, 20% at 3.3-km sites, and 9% at the 5.6-km site. Large contributions from the KCCIC have been found at Iwasaki-nishi (72%), Kashi (41%), and Tsuiheiji (28%), where high benzene concentrations have been observed for several years. At these three sites, contributions from Route 16 were 5% or less, and the background concentration, that is, concentration contributed by sources other than the KCCIC and Route 16, was relatively large. Benzene concentrations estimated by the method agreed well with the concentrations measured at the 31 sites and at the local governments’ measurement sites. The estimated emissions of the factories and the road were slightly larger than those reported in the PRTR. These results support the reliability of our method. By applying our source apportionment method, the contributions of the industrial complex to the surrounding ambient concentrations have become clear. It has been found that simultaneous measurement at sites in a grid network is effective for quantitative source apportionment near multiple sources. The method can be applied with other chemicals or in other regions to achieve reasonable source apportionment.
5. Conclusions We developed a source apportionment method based on an atmospheric dispersion model and multiple linear regression analysis (MLR) in conjunction with ambient concentrations measured simultaneously at sampling sites in a grid network. A Gaussian plume dispersion model called the Industrial Source Complex Model (ISC) developed by the US Environmental Protection Agency was used in the method. The method does not require emission amounts or source profiles. The method was applied to the case of benzene in the vicinity of the Keiyo Central Coastal Industrial Complex (KCCIC), one of the biggest industrial complexes in Japan. Benzene concentrations were measured simultaneously in a grid network consisting of 31 sites in the KCCIC and the surrounding residential area from December 2001 to July 2002. This measurement grid network was established in the study area with intervals of about 1–2 km between grid lines. The benzene concentration at each site contributed by each source was calculated by using the dispersion model and assuming a standardized common emission rate. Ben-
Acknowledgements This paper was written based on the results obtained at the Graduate School of Environment and Information Sciences of Yokohama National University. We are grateful to the Environmental Management Division of Ichihara City for their help with the ambient sampling. Part of this study was supported by the junior fellowship program in FY 2001 (No. 0905002) from the New Energy and Industrial Technology Development Organization (NEDO) of Japan.
References Angius, S.P., Angelino, E., Castrofino, G., Gianelle, V., Tamponi, M., Tebaldi, G., 1995. Evaluation of the effects of traffic and heating reduction measures on urban air quality. Atmospheric Environment 29 (23), 3477–3487. Derwent, R.G., Middleton, D.R., Field, R.A., Goldstone, M.E., Lester, J.N., Perry, R., 1995. Analysis and interpretation of air quality data from an urban roadside
ARTICLE IN PRESS 1334
A. Fushimi et al. / Atmospheric Environment 39 (2005) 1323–1334
location in central London over the period from July 1991 to July 1992. Atmospheric Environment 29 (8), 923–946. Evans, J.S., Wolff, S.K., Phonboon, K., Levy, J.I., Smith, K.R., 2002. Exposure efficiency: an idea whose time has come? Chemosphere 49, 1075–1091. Fushimi, A., Kajihara, H., Yoshida, K., Nakanishi, J., 2002. Estimation of atmospheric concentrations and evaluation of the PRTR data using a dispersion model: a case study of benzene. Environmental Science 15 (1), 35–47. Hao, J., He, D., Wu, Y., Fu, L., He, K., 2000. A study of the emission and concentration distribution of vehicular pollutants in the urban area of Beijing. Atmospheric Environment 34, 453–465. Higashino, H., Kitabayashi, K., Yokoyama, O., Takatsuki, M., Yonezawa, Y., 2000. Development of the fate model for chemical substances—development of the model to estimate long-term average atmospheric concentrations. Journal of Japan Society for Atmospheric Environment 35 (4), 215–228. Hopke, P.K., 1991. Receptor Modeling for Air Quality Management. Data Handling in Science and Technology, vol. 7. Elsevier, Amsterdam, The Netherlands 344pp. Ichihara City, Department of Environment, 2000. Environment of Ichihara. Japan Environment Agency, Air Quality Bureau, 1998. The results of hazardous air pollutants monitoring by local government in fiscal year 1997. Japan Environment Agency, 1999. Report on the 1998 PRTR Pilot Project. Japan Ministry of Economy, Trade and Industry, 2000. About the progress of the self management plan for hazardous air pollutants. http://www.meti.go.jp/kohosys/press/0000892/0/ taiki1.htm Japan Ministry of Economy, Trade and Industry, 2001. About the progress of the self management plan for hazardous air pollutants. http://www.meti.go.jp/kohosys/press/0001961/0/ 011012yuugaitaiki.pdf
Japan Ministry of Environment, 2002. The results of hazardous air pollutants monitoring by local government in fiscal year 2001. http://www.env.go.jp/air/osen/mon_h13/index.html. Japan Ministry of Environment and Japan Ministry of Economy, Trade and Industry, 2003. Summary of the FY 2001 PRTR data Results of chemical emissions and transfers. Japan Ministry of Land, Infrastructure and Transport, Chiba National Highway Work Office, 2002. Chiba National Highway Investor Relations (IR) site. http://www.ktr.mlit. go.jp/chiba/ir/index.htm. Lorber, M., Eschenroeder, A., Robinson, R., 2000. Testing the USA EPA’s ISCST-Version 3 model on dioxins: a comparison of predicted and observed air and soil concentrations. Atmospheric Environment 34, 3995–4010. Palmgren, F., Berkowicz, R., Ziv, A., Hertel, O., 1999. Actual car fleet emissions estimated from urban air quality measurements and street pollution models. The Science of the Total Environment 235, 101–109. PerkinElmer Japan Co., Ltd., 2000. Analyses of indoor-air organic compounds by adsorption tube sampling-thermodesorption-GC/MS. PerkinElmer Japan Co., Ltd. Petrochemical Press Co., Ltd., 1998. Petrochemical industry year book 1998. Petrochemical Press Co., Ltd. Scientific Instrument Services, 2002. Adsorbent Resins. http:// www.sisweb.com/index/referenc/resins.htm. US Environmental Protection Agency, Office of Air Quality Planning and Standards (US EPA), 1995. User’s Guide for the Industrial Source Complex (ISC3) Dispersion Models, vols. 1 and 2. USEPA. EPA-454/B-93-003a and b. US Environmental Protection Agency, Office of Research and Development (US EPA), 1998. Carcinogenic Effects of Benzene: An Update. EPA/600/P-97/001F. Watson, J.G., Chow, J.C., Fujita, E.M., 2001. Review of volatile organic compound source apportionment by chemical mass balance. Atmospheric Environment 35, 1567–1584.