Environ. Sci. Technol. 2003, 37, 2958-2965
Application of EPA CMB8.2 Model for Source Apportionment of Sediment PAHs in Lake Calumet, Chicago AN LI,* JAE-KIL JANG, AND PETER A. SCHEFF School of Public Health, University of Illinois at Chicago, 2121 West Taylor Street, MC-922, Chicago, Illinois 60612-7260
A chemical mass balance model developed by the U.S. EPA, CMB8.2, was used to apportion the major sources of PAHs found in the sediments of Lake Calumet and surrounding wetlands in southeast Chicago. The results indicate the feasibility of applying CMB8.2 to pollutants found in aquatic sediments. To establish the fingerprints of PAH sources, 28 source profiles were collected from the literature. Some of the source profiles were modified based on the gas/particle partitioning of individual PAHs. The profiles under the same source category were averaged, and the fingerprints of six sources were established, including coke oven, residential coal burning, coal combustion in power generation, gasoline engine exhaust, diesel engine exhaust, and traffic tunnel air. Nine model operations with a total of 422 runs were made, differing in the choice of fitting species and the sources involved. Modeling results indicate that coke ovens and traffic are the two major sources of PAHs in the area. For traffic sources, either traffic tunnel alone or both diesel and gasoline engine exhausts were entered into the model. These two groups of model operations produced comparable results with regard to the PAH contributions from road traffic. Although the steel industries have shrunk in recent years, closed and still-active coke plants continue to contribute significantly to the PAH loadings. Overall, the average contribution from coke oven emissions calculated by different operations ranges from 21% to 53% of all sources, and that from traffic ranges from 27% to 63%. The pattern of source contributions shows spatial and temporal variations.
Introduction Polycyclic aromatic hydrocarbons (PAHs) are the byproducts of the incomplete combustion of virtually all organic matter. It is well-known that PAHs originate predominantly from anthropogenic processes. Major sources of PAHs include transportation vehicles, coke ovens, steel and iron furnaces, metal smelters, road paving using asphalt, manufactured gas plants, petroleum cracking, coal tar pitch, power generation using fossil fuels, incineration of municipal and industrial wastes, and domestic heating. Among these sources, traffic and coal related sources have been considered to be the two most important source categories in many metropolitan areas in the 20th century (1-3). * Corresponding author phone: (312)996-9597; fax: (312)413-9898; e-mail:
[email protected]. 2958
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 37, NO. 13, 2003
Chicago has been identified as a source of PAHs and other persistent organic pollutants in the Great Lakes region (4). Although numerous studies have focused on the chronology of PAH accumulation in the Great Lakes sediments, a historical profile of PAH contamination in Chicago urban area over the last century has not been established. Recently, we have completed an investigation of the distribution and temporal trend of PAH fluxes in Lake Calumet and its surrounding wetlands in urban Chicago (5). The data are highly valuable to the identification of contributing sources and the development of pollution reduction strategies for the entire region under the influence of the sources. For the identification and apportionment of pollutant sources, two basic approaches have been used. Sourceoriented models rely on the knowledge of emissions and subsequent transport processes to predict the concentrations of pollutant at a specific receptor site. For PAHs, because it is difficult if not impossible to establish emission inventories, source-oriented approaches have not been successful. In contrast to source based models, various methods have been developed based on the pollutant profiles observed at sampling sites, or “receptors”. These include the use of isomer ratios (6-8), alkyl homologue distribution patterns (8), or stable isotope signatures (9). Many of these methods involve just a portion of PAH data produced by analytical laboratories and/or provide only qualitative source contribution information, thus the results often need additional and independent confirmation. Receptor models apportion the contributions from all major sources from the concentration profiles of pollutants found at receptors. Fundamentals of the receptor models are described by Hopke (10, 11), Gordon (12), Henry et al. (13), and others. Principal component analysis (PCA) and multiple linear regression (MLR) are common types of receptor models and have been used for PAHs with certain degrees of success (6, 14). The chemical mass balance (CMB) model is another type of receptor model. It has been applied to air resources management for decades. In 2000, the U.S. EPA released its Windows-based CMB8.2 computer software, which substantially facilitates the estimation of source contributions to ambient air pollutants such as particulate matter and volatile organic compounds. However, the software was labeled as an air quality model and has not been tested for source apportionment of pollutant chemicals found in aquatic environments. The objective of this study is to quantitatively apportion the major sources of PAHs in the sediment of a lake near downtown Chicago by applying the EPA’s CMB8.2. To our knowledge, this is the first study to apply the CMB8.2 to a group of pollutants measured in aquatic sediments.
CMB Model As a tool for source apportionment, the chemical mass balance model was first proposed for aerosols in the early 1970s (15, 16). The model assumes that the profile of marker chemical species measured at a specific receptor site is a linear combination of concentration profiles of the chemical species emitted from independent contributing sources (17). The general equation is
C ) AS + E where C is a n × 1 vector of concentrations of chemical species i measured at a receptor site with 1 e i e n; A is n × m source composition matrix of n compounds for each of the m sources modeled; S is a m × 1 vector of the source contribution factor; 10.1021/es026309v CCC: $25.00
2003 American Chemical Society Published on Web 05/31/2003
and E is a n × 1 error vector. The number of sources (m) must be less than or equal to the number of chemical species (n). This equation is a set of n mass balance equations, which are solved simultaneously by minimizing the sum of squares for error to obtain the source contribution factor Sj for each of the m sources. To obtain quantitative source contributions at a receptor, several assumptions should be satisfied: (1) composition of source emissions is consistent over the period of ambient and source sampling; (2) chemical species do not react with each other, i.e., they add linearly; (3) all sources with a potential for significantly contributing to the receptor have been identified and have had their “fingerprints” determined; (4) the compositions of different sources are linearly independent of each other; and (5) measurement uncertainties are random, uncorrelated, and normally distributed. If significant losses of some chemical species occur during their transport from the sources to the receptor site, assumption (1) would be violated. In an attempt to account for such losses due to various possible transformation processes, a coefficient of fraction, Rij, has been proposed to modify the values of source composition matrix Aij (18-20). Collearity among sources will violate assumption (4) above. In CMB8.2, collinearity is identified by the eligible sources display, which is in turn defined by the maximum source uncertainty defined by the user. The eligible space is spanned by eigenvectors with inverse singular values less than or equal to the maximum source uncertainty. Contribution from a source is “estimable” if its projection into the eligible space is at least the minimum source projection defined by the user. Because ambient (sediment, in this work) data uncertainties and relative levels of source contributions vary from sample to sample, it is impossible to decide a priori that a set of source profiles is collinear or not. The decision is made for each set of profiles combined with each set of the ambient data (17). Several mathematical approaches can be used to solve for the source contribution vector S. The method adopted by the EPA’s CMB8.2 is the effective variance weighted solution, which achieves the most probable values of Sj by minimizing χ2, the sum of squares of the differences between the measured and calculated values of Ci, weighted by the analytical uncertainty σ2, of both the receptor concentration profile and the source fingerprint (14, 17): m
n
2
χ )
∑ i)1
(Ci -
∑A S ) ij j
j)1 m
σci2 +
∑σ j
2
2
AijSj
The CMB output includes a number of performance measures including χ2 (defined above), R2 (fraction of the variance in the measured concentrations that is explained by the variance in the calculated species concentration), % mass (percent mass explained by the model), standard error (single standard deviation), and t-statistic (ratio of estimated source contribution to its standard error) for each of the source contribution estimates as well as species-specific information about the uncertainty. An early application of a CMB model to PAHs was presented for New York City aerosols in the late 1970s (21). Atmospheric PAHs have also been apportioned for their sources in Los Angeles (20), Paris (22), and Chicago (23, 24). In addition, a number of CMB applications for ambient and indoor air included PAHs as tracers for combustion sources (19, 25). The CMB approach has also been used for PAHs in aquatic sediment (20, 26-31).
FIGURE 1. Sampling locations in the Lake Calumet area.
Methodology Study Area. Lake Calumet is located 15 miles south of metropolitan Chicago. The only surface water inflow is Pullman Creek, a drainage ditch at the northwest corner of the lake. The outlet is the Calumet River which flows, under the control of O’Brien Lock and Dam, either north to Lake Michigan or south to the Illinois River of the Mississippi River Basin (32). Rapid industrial development in southeast Chicago began in the 1860s with the laying of railroad tracks. The size of Lake Calumet has been significantly reduced since then, due mainly to the landfill of refuse from nearby industries, especially the steel industry, as well as municipal wastes including the ash and cinders from coal combustion for home heating and cooking. Other industries around the lake include petroleum refineries, chemical plants, building material companies, and grain processing as well as incinerators, landfills, and illegal dumping sites. Most of the east lakeshore has been turned into cargo ship loading docks. Interstate I-94 is adjacent to the west shore of the lake, with increasing traffic flows since its opening in 1962. A north-south rail corridor runs across the wetlands to the west of the lake, carrying industrial cargo. A number of residential communities are within a few miles of the lake, with some houses being in close vicinity to industries. Based on historical and present land uses, and a few previous studies on PAHs in the area (14, 24, 32, 33), major sources of PAHs in the area include diesel and gasoline powered vehicles, coke oven emissions, and domestic cooking and heating. These were the major targets of source apportionment in this study. Coal-fueled power generation was also included, because a power plant is located about 2 miles northeast (downwind most of the time) of the study area on the shore of Lake Michigan. Other potential sources are wood burning for residential heating, municipal waste incineration, ship traffic, and petroleum refineries. Sampling, Sediment Characterization, and Chemical Analysis. Detailed description of procedures is to be published elsewhere (5); below is a brief summary. Sediment cores were collected in June 1997 at nine locations in the Lake Calumet area. Sampling locations are illustrated in Figure 1. Sampling points D and E were at the VOL. 37, NO. 13, 2003 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
2959
northwest corner of the lake, near the Pullman Creek and the freeway I-94. Sites J and K were two ponds surrounded by wetlands east of Lake Calumet. Site I was in a ditch by a rail track, and site G was in the turning basin of the Calumet River. All cores were taken by a gravity corer except Core I, which was collected by a push corer. Length of the five long cores (D, E, I, J, and K) ranged from 42 to 51 cm. Each core was sectioned into 10-12 segments. The short cores (B, C, F, and G), with lengths being 8-20 cm, were cut into 4-6 segments. Each segment of the long cores (D, E, I, J, and K) was dated using 210Pb. The PAHs analyzed were naphthalene (Nap), acenaphthylene (AcNP), acenaphthene (AcN), fluorene (Fl), phenanthrene (PhA), anthracene (An), fluoranthene (FlA), pyrene (Py), benz[a]anthracene (BaA), chrysene (Chy), benzo[b]fluoranthene (BbFlA), benzo[k]fluoranthene (BkFlA), benzo[e]pyrene (BeP), benzo[a]pyrene (BaP), indeno[123-cd]pyrene (IP), dibenz[a,h]anthracene (dBahA), and benzo[ghi]perylene (BghiP). The analytical procedure included Soxhlet-extraction using 1:1 acetone:hexane for 24 h, volume reduction using a Kuderna-Danish (K-D) evaporator, silica gel chromatographic cleanup (34), and instrumental analysis using a HP Model 5890-II gas chromatograph equipped with a DB-5 capillary column (30 m × 0.25 mm id, 0.25 µm film) and a flame ionization detector. The identities of PAHs in selected samples were confirmed using a HP Model 6890+/5973 GC/ MS. Method blanks, laboratory method control samples, and matrix duplicated samples were analyzed as quality control measures. The recovery of the surrogate ranged from 48% to 90% with an arithmetic mean of 72.8%. The analytical procedure was further validated by analyzing NIST Standard Reference Materials 1649a (Urban dust). The measured concentrations ranged from 64% to 112% of the certified values. Two segments in each long core and one in each short core were analyzed in duplicates. The average relative standard deviations (RSD) ranged from 10.3% to 25.6% among the 16 PAHs, with an average of 16.8%. The precision for each reported PAH concentration, which is required by the CMB model, was derived from the product of each measured concentration and the segment- and PAH-specific RSD. Source Fingerprints. It should be noted that a definitive signature of a combustion process may not exist due to the complexity of the combustion process. Emissions of PAHs depend on numerous factors which may vary significantly even during a single combustion process. In addition, sampling methods differ, introducing additional differences among published source signatures. The key to a successful CMB application is to obtain a set of source fingerprints which are consistent with the measurements at the receptor locations. For the purpose of this work, 11 and 17 published PAH source profiles were collected from the literature for coal- and traffic-related sources, respectively (see Supporting Information for references). Large differences were observed between source profiles established by collecting both gaseous- and particulate-phase PAHs with the use of sorbents or gas traps and those that contain only particulate PAHs using only filters. Such differences are especially large for low molecular mass PAHs. For example, the concentration ratio of pyrene to benzo[e]pyrene is 2.4-6.0 for the particulate samples of coal combustion, while it ranges from 11.9 to 16.1 for combined gaseous and particulate samples. Recognizing that the PAHs in sediments are mostly those originally associated with airborne particles, percent particulate to total airborne PAHs (P%) were estimated from reported gas-particle partitioning reported in more than a dozen literature sources (see Supporting Information). The geometric means of P% were used to convert the literature source profiles containing both gaseous and particulate PAHs to particulate-only profiles. 2960
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 37, NO. 13, 2003
This conversion greatly reduced the discrepancies between source profiles and allowed us to select model-fitting species without the concern of gas-phase losses. The source profiles containing only particulate PAHs were not modified. After P% corrections, the concentrations of PAHs were normalized to that of benzo[e]pyrene. This normalization is necessary because the numbers of PAHs included in source profiles vary. BeP was absent in 4 out of 28 original fingerprints. In these cases, its concentration was assumed to be the same as that of BaP. This is reasonable because the reported BaP/BeP ratios are close to unity in almost all sources. The BeP normalized concentrations within the same source category were averaged, and the results are presented in Table S1 of the Supporting Information. Precision of the source PAH fractions was calculated based on the source- and PAH-specific percent standard deviation from the original publications. The average relative standard deviations of all sources ranged from 87% to 117% for lighter PAHs (up to pyrene) and from 60% to 75% for heavier PAHs. Such high uncertainties resulted in many “inestimable” source contributions during our preliminary runs. To avoid this problem, a constant relative uncertainty of 40%, which is twice the uncertainties (about 20%) often associated with the laboratory measurement for PAHs in receptor sediment samples, was used for all nonzero values in the source fingerprints. When a fingerprint value for a particular PAH in a source was not available, it was considered as zero, and an uncertainty of 0.1 was assigned. Chemical, photochemical, and biological degradation of PAHs before they were buried in the sediment may also affect the true profiles of PAHs entering sediments, because susceptibility to degradation differs among individual PAHs. For example, benzo[a]pyrene and anthracene are more susceptible to photodegradation than their respective isomers benzo[e]pyrene and phenanthrene. It is also known that gaseous molecules in the air and dissolved ones in the water are more vulnerable to degradation and long-range transport than the particle-bound molecules. The receptor sites involved in this study are in close proximity to the sources. Therefore, the stability of PAHs was not considered. Partitioning between water and settling particles in the lake may also affect the relative abundance of PAH species when they reach the bottom sediments. In general, however, due to the strong affinity of PAHs to particulate matter, their loss to the lake water may not be significant. Also not considered in this study is the degradation of PAHs after they are buried in the sediment. CMB Modeling. The inputs required by the EPA CMB Version 8.2 include the ambient data and source profiles. In this work, the “ambient” data are the experimentally measured concentrations, in µg kg-1, of PAHs in the sediment. Only the sediment segments that recorded a total PAH concentration greater than 2 µg g-1 were used. A total of 49 individual sediment segments were apportioned for the traffic and coal-related sources. The precision of each measurement was determined from duplicate laboratory analyses, as discussed previously. Naphthalene is not included in the modeling, because of the high uncertainties in its source fingerprints (RSD > 100% for all source categories, see Table S1) and possible evaporative losses during chemical analysis of the sediment samples. Difficulties were often associated with the GC separation of benzo[b]fluoranthene (BbFlA) and benzo[k]fluoranthene (BkFlA). For this reason, the sum of these two isomers was used with an abbreviation Bb+kFlA. Due to the lack of sufficient data for dibenz[a,h]anthracene (dBahA) in many source profiles, this compound was not selected as fitting species in most model runs. Nine sets of runs, referred to as operations #1 through #9 hereafter, were carried out with a total of 441 attempted CMB
TABLE 1. Summary of CMB Modeling Results operation # d.f.a
1
2
3
4
5
6
7
8
9
3-5
2-5
10-13
4-6
4-5
12-14
5-6
5-6
13-14
PAH Fitting Species AcNP AcN Fl PhA An FlA Py BaA Chy Bb+kFlA BeP BaP IP DBahA BghiP
x
x
49
47
Results Statistics 32 49
0.95 0.81 0.99
0.93 0.80 0.97
0.87 0.51 0.97
1.04 0.24 3.15
1.29 0.39 3.29
137 87 216
96 79 125
0.04 3.85 2.84
Source Contribution, mg kg-1 (All Run Average) 0.17 0.29 0.18 1.03 3.06 0.76 1.12 5.70 5.91 2.56 4.16 2.66
1.76 1.58
3.09 1.30
x
coal-power plant coal-residential coal-coke oven coal average traffic-gasoline engine traffic-diesel engine traffic-tunnel air traffic average
x x x
x x x x x x
x
x
x
x x x x x x x x x x x x x x x
x x x x x x
x
x
x
x x x x x x x x x x x x x x x
49
49
49
49
49
0.98 0.86 1.00
0.95 0.90 0.99
0.87 0.57 0.98
0.95 0.81 1.00
0.89 0.68 0.99
0.78 0.38 0.91
2.01 0.47 9.01
0.34 0.01 1.40
0.88 0.12 2.16
1.62 0.22 9.50
0.51 0.04 1.55
1.50 0.22 4.65
2.76 0.63 15.68
101 80 119
113 82 186
93 76 104
99 78 114
121 66 181
103 77 123
111 94 150
3.71
5.62
3.47
6.35
4.44
6.60
0.38
0.52
0.37
0.62
0.48
0.63
x x
x x x
5.14
x
x x x
6.28
0.00 0.39 0.22
Source Contributions, Fractions (All Run Average) 0.02 0.08 0.05 0.11 0.38 0.15 0.15 0.51 0.53 0.21 0.37 0.27
0.16 0.22
0.26 0.16
0.16 0.11 0.41
b
x
2.09 0.82 4.44
coal-power plant coal-residential coal-coke oven coal average traffic-gasoline engine traffic-diesel engine traffic-tunnel air traffic average d.f. ) degrees of freedom.
x
x
N R2 ave min max χ2 ave min max %Mb ave min max
a
x x x x x x
x x x x x x x x x x x x x x x
0.48
0.58
%M ) percent of mass explained by the model.
calculation runs. The nine operations differ in the selection of fitting species and the sources involved (see Table 1). A maximum of 49 runs are included in each operation. Operations #1 to #3 investigate five sources including coal fired power plants, residential coal burning, coke oven, gasoline engine exhaust, and diesel engine exhaust. For operations #4 to #6, power plant was dropped based on the results of operations #1 to #3, and traffic tunnel air was used as traffic source tracer. Operations #7 to #9 involve only two source category averages for coal and traffic sources. To compare the modeling results from using different sets of fitting species, operations #1, #4, and #7 used only heavier PAHs, while #3, #6, and #9 used all the 15 PAHs. In operations #2, #5, and #8, seven PAHs were selected, covering all molecular mass groups. The use of benzo[a]pyrene and anthracene was avoided because of their relatively high reactivity.
In using the model, the maximum number of iterations was set at 50. The maximum source uncertainty defines the eligible space within which sources may be estimated with an uncertainty less than the defined maximum (17). A constant of 50% is used throughout all the modeling runs. Note that this value is higher than the default value of 20% suggested for source apportionment of air pollutants, because higher source uncertainties were expected for PAHs in sediment due to the higher degree of complexity. The minimum source projection was set at 0.95, below which the source is considered by CMB8.2 as “inestimable”. In all runs, the optional “source elimination” was implemented, which eliminated physically impossible negative source contributions.
Results and Discussion Detailed discussions on PAH concentrations and fluxes in the area are given elsewhere (5). Briefly, the total concentrations of the 17 PAHs in the surface sediments ranged from VOL. 37, NO. 13, 2003 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
2961
FIGURE 2. Source fingerprints. 4.9 to 20 mg kg-1. The most abundant PAHs were fluoranthene and pyrene, followed by chrysene, benzo[a]pyrene, benzofluoranthenes, indeno[123-cd]pyrene, and phenanthrene. The PAH fluxes to the sediment ranged from 50% from residential coal consumption throughout the entire period up to the sampling date of 1997. This is unreasonably high, especially for the decades after 1950s. In view of the abundance of light PAHs in the residential coal burning fingerprint (see Figure 2), the results from operations #2 and
FIGURE 3. Shift of source contributions over time at four locations. Left: Operation #2; Right: Operation #5. #5 are considered more realistic than those of operations #1 and #4 regarding the contributions of domestic coal burning to the total PAHs. This suggests that the choice of PAH fitting species should ensure that different molecular mass groups are included, after the effects of gas-particle partitioning (as well as transformation, if necessary) are taken into account. By fitting the model with all 15 PAH species, operations #3, #6, and #9 have higher degrees of freedom than others. However, the performance measures R2 and χ2 are apparently poorer than the operations using fewer PAHs, indicating lower quality of fit to the measured receptor PAH profile. This is most likely due to the inconsistencies between the source profile and receptor concentration data sets. Using the same seven fitting PAHs but different sets of sources, operations #2, #5, and #8 produced consistent overall source apportionment results (see Table 1). The contributions from coal related sources estimated by operations #2, #5, and #8 are 58%, 52%, and 52%, respectively, while the corresponding contributions from traffic related sources are 42%, 48%, and 48%. As two major source categories, traffic and coal related sources contribute comparably to the total PAHs in the sediment of Lake Calumet. Detailed discussions given below are based primarily on the results of operations #2 and #5. Spatial Variations and Temporal Trend. Figure 3 illustrates the time trend of PAH source shifting at four locations in the area. As expected, the relative contributions from different sources show spatial variations. The results from operation #2 indicate that over the time periods covered by the lengths of individual cores, locations C, D, and E had received 24-45% of their PAHs from coke oven, while coke oven contributions at locations I, J, and K as well as G, range from 56% to 70%. On the other hand, traffic sources contributed 46-57% at locations C, D, and E, which are higher than the 30-38% found for locations I, J, K, and G. These results are in agreement with the known PAH sources at these locations. That is, the PAH input at the west side of the
lake is largely influenced by the nearby freeway traffic, while the wetlands to the east side of the lake have received significant amount of PAHs from the nearby coke manufacturers. Operation #5 predicted less distinct relative contributions from the two major sources on the west and east sides of the lake. Roadway traffic is one of the top contributing PAH sources in the region. In this study, three independent source profiles were investigated. They are the gasoline engine exhaust and diesel engine exhaust included in modeling operation #2 and traffic tunnel air in operation #5. According to the results of operation #2, PAHs contributed from gasoline powered vehicles outweighs those from diesel vehicles in 38 out of 47 runs. The fingerprint of traffic tunnel air appears to have a more even distribution of PAHs with different molecular masses (see Figure 2). It was used in operation #5 to compare with the sum of contributions from gasoline and diesel engines used in operation #2. The overall average of traffic contribution was estimated as 42% and 48% by operations #2 and #5, respectively. On the basis of individual runs, the differences in the source contribution estimates between operations #2 and #5 range from 1% to 41%. The discrepancies are higher for sediment cores J and K than for other cores. For location D, both operations #2 and #5 indicate a peak of traffic contribution of 12-15 mg kg-1 in the late 1980s and a declining trend after then. The recent decrease most likely attributes to the reduced emissions from gasoline engines, as the results of operation #2 show that sediment PAHs originated from diesel engines had increased from 3.0 mg kg-1 in 1982 to 4.3 mg kg-1 in 1997. Coal-related sources include coke ovens, residential coal consumption, and coal burning power plants. Coke ovens in the area were built to serve the steel industries, which started in 1875 (40). One of the coke plants, which was active at the time of sampling, is located less than 2 miles north of sampling location J. In this study, the results of CMB model confirm that coke ovens have contributed significant percentage of VOL. 37, NO. 13, 2003 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
2963
PAHs at almost all locations in the study area. The highest PAH input from coke oven to the sediment, 18.5 mg kg-1 (78% of the total), was found at location J in the sediment deposited around 1978 (not shown in Figure 3). However, little PAH input was recorded in the sediment at sites J and K before 1960 or even later. The reason is not clear. Overall, sediment at individual sampling locations have been receiving a relatively steady input of PAHs in an amount ranging from 2 to 7 mg kg-1 originated from coke ovens since the 1950s. Before then, little PAHs were received from coke ovens by the sediment at locations D and E. In fact, no known steel mill has been operated on the west shore of the lake. However, enormous amount of steel industries wastes, mainly slag, were used as landfill material since the early 1900s until the late 1960s to form the foundation for roadways, bridges, railroads, and industrial sites (41). Therefore, it is difficult to determine the percentage of the PAHs from current emissions versus those from the coke production in the past. In addition, the Lake Calumet is surrounded by metal smelters, paint and coating companies, rail car factory, incinerators, wastewater treatment facilities, and other industries (40, 42). Combined, they have produced sustaining input of PAHs into the environment of Lake Calumet. Location I has seen relatively high PAH input from coal combustion sources since the 1930s. The sediment core collected at this location is from a ditch next to the cargo railroad which was built perhaps after 1920. Coal-burning steam engines were operated on the rails until about the late 1950s when diesel replaced coal as the fuel. Coke oven is identified as the major PAH source at this location until recent years, although contributions from traffic sources show a dramatic increase after 1993. In addition, creosote in railroad ties may have also contributed to the PAHs found in the sediments near the rails at location I (43). Results of all operations #2, #3, #5, and #6 indicated a relative small overall percentage (
