Multimedia Fate Model for Hexachlorocyclohexane in Tianjin, China

Feb 19, 2004 - feasible to apply a steady-state multimedia model for investigating the fate of γ-HCH in Tianjin. In addition to illustrating the leve...
3 downloads 0 Views 201KB Size
Environ. Sci. Technol. 2004, 38, 2126-2132

Multimedia Fate Model for Hexachlorocyclohexane in Tianjin, China H O N G Y I N G C A O , † S H U T A O , * ,†,‡ FULIU XU,† RAYMOND M. COVENEY, JR.,‡ JUN CAO,† BENGANG LI,† WENXIN LIU,† XUEJUN WANG,† JIANYING HU,† WEIRAN SHEN,§ BAOPING QIN,§ AND REN SUN§ Department of Urban and Environmental Sciences, Laboratory for Earth Surface Processes, Peking University, Beijing 100871, China, Department of Geosciences, University of Missouri, Kansas City, Missouri 64110-2499, and Tianjin Environmental Monitoring Center, Tianjin 300191, China

A level III fugacity model was applied to characterize the fate of γ-HCH in Tianjin, China, before the 1990s when the contamination reached its maximum at steady state. Geometric means were used as model inputs. The concentrations of γ-HCH in air, surface water, soil, sediment, crops, and fish as well as transfer fluxes across the interface between the compartments were derived under the assumption of steady state. The calculated concentrations were validated by independent data collected from the literature. There was generally good agreement between the estimated and the observed concentrations, and the differences were all less than 0.6 log units for air, water, soil, sediment, and fish and approximately 1 order of magnitude for crops. Around 97% of γ-HCH accumulated in soil and sediment. Wastewater irrigation was not an important pathway for delivering γ-HCH to soil as compared to the dominant source of agricultural application. Degradation and advective airflow carried much γ-HCH out of the system. Sensitivities of the model estimates to input parameters were tested, and a coefficient of variation normalized sensitivity coefficient was defined for the test. The most influential parameters were degradation rates in sediment and soil, application rates, concentrations in wastewater, and adsorption coefficients. Monte Carlo simulation was conducted for model uncertainty analysis. The model was run 20 000 times using randomly generated data from predefined log-normal distribution density functions. All calculated concentrations and fluxes were lognormally distributed. The dispersions of the calculated and observed concentrations were compared in terms of coefficients of variation to distinguish between true variability and model uncertainty.

Introduction Hexachlorocyclohexanes (HCHs) are insecticides that were once widely used in China, resulting in widespread dispersal * Corresponding author tel/fax: +8610-6275 1938; e-mail: taos@ urban.pku.edu.cn. † Peking University. ‡ University of Missouri. § Tianjin Environmental Monitoring Center. 2126

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 38, NO. 7, 2004

in the environment (1, 2). Since 1980, HCHs have been banned in many countries (3), which is reflected in declining levels in environmental media and human tissues, especially in developed countries (4). Still, because of their persistency and potential of bioaccumulation and long-range atmospheric transport, HCHs continue to be of local, regional, and global concern today (4-7). Large amounts of HCHs (1) were applied over a period of nearly four decades in Tianjin, which became one of the most severely HCHs contaminated areas in China (2, 8, 9). Extensive agricultural application began in 1953 using locally produced technical HCHs with 12-13% of γ-HCH. The use of HCHs was banned in China for large-scale agricultural usage and was replaced with lindane (99.9% γ-HCH) in the early 1980s. Lindane was used for another decade until it was officially prohibited in 1992 together with DDT, nemafume, and galecron (10). Although the ban was not fully implemented in the area and there were occasional reports of illegal use of so-called “pest killing powder” (a mixture of waste slag from HCHs production and host material) smuggled either from leftover storage or local chemical companies where HCHs was produced as midbody of sodium pentachlorophenate, the extensive application over the area was indeed gradually phased out. In fact, according to the result of an extensive survey of more than 300 topsoil and crop samples, the levels of HCHs in surface soils and crops from Tianjin except a few isolated locations next to major chemical companies have decreased substantially since the 1980s. The mean concentration of HCHs in topsoil dropped from 186.8 ng/g in 1980s to the present level of 29.1 ng/g (8). In addition to agricultural application, wastewater discharged from two large HCH producers contributed a large quantity of the pesticides to the local loadings (11-13). Over the years, annual production of HCHs by the two largest chemical companies alone was around 3000 ton while their wastewater was not well treated. The facility for technical HCHs production in Tianjin Chemical Company, Tanggu, was shut down in 1983. The Dagu Chemical Company, Hangu, produced technical HCHs from1953 to 1985 and lindane from 1986 to 2000 when production was completely discontinued. The result of a preliminary dynamic modeling using a similar model structure as this study for that period indicated that concentrations of γ-HCH in the bulk media of air, water, soil, and sediment in Tianjin reached constant levels within a period of less than 10 yr (Supporting Information). It was concluded that a steady state was reached long before the early 1990s when γ-HCH was totally banned for agricultural application. It is desirable to reconstruct a picture of γ-HCH fate in Tianjin when the levels of γ-HCH reached their steadystate values in the hope of providing quantitative information for a better understanding of γ-HCH contamination in Tianjin as well as in the rest of China. Many models have been proposed for quantitative depiction of the fate of chemicals in the environment (14, 15). Among these, the fugacity model has been successively applied to predict the fate of organic chemicals on various scales and was proven useful to regulators (16-20). It is feasible to apply a steady-state multimedia model for investigating the fate of γ-HCH in Tianjin. In addition to illustrating the levels and transport pathways of γ-HCH before the 1990s, the model can also provide boundary conditions for a detailed dynamic modeling in future. The specific objective of this study was to model the fluxes of γ-HCH across various compartments and its concentrations in various media including air, water, soil, sediment, crops, and fish in the Tianjin area before the 1990s using a 10.1021/es0305860 CCC: $27.50

 2004 American Chemical Society Published on Web 02/19/2004

level III fugacity model. The reliability of the model estimates was evaluated by various means including concentration validation, sensitivity analysis, and uncertainty analysis.

Methodology Study Area. Tianjin lies on Haihe River alluvial plain formed by alluvial and marine sedimentation with Baohai Bay to its east. Bordered by Beijing on the northwest and by Hebei on the north, southwest, and south, most of the plain has an elevation of 3-30 m, except a small area in north with hills and mountains. The annual mean temperature is 12.3 °C with prevailing south wind. The annual precipitation is around 600 mm, a large portion of which concentrated in the summer. The plain is criss-crossed with a network of rivers and waterways, most of them man-made, that are largely used as open sewers (21). Tianjin is one of the most important industrial areas in northern China, and the population was around 8 million in 1980s (12). The total area is approximately 12 000 km2 with a large portion being used for agriculture (21). The study area is shown in the Supporting Information (Figure S1). Multimedia Modeling. A level III fugacity model was applied to describe the partitioning and transfer of γ-HCH in Tianjin based on an approach of Mackay and Paterson (22). Four bulk compartments including air (air, particulates, and crops), water (water, suspended solids, and fish), soil (air, water, and solids), and sediment (water and solids) were included. The processes taken into consideration are summarized in the Supporting Information (Table S1). The concentrations of γ-HCH in the compartments and the transfer fluxes between adjacent compartments were modeled. Observed concentrations from the literature were used for model validation. Modeling was performed using Matlab v.6.5 (23). SPSS v.10.0 and MS Excel were employed for statistical analysis and data manipulation. Details of the model framework and computation can be found in the Supporting Information (Figure S3). Parameter Identification. Parameters characterizing dimension and property of the compartments and subcompartments, input and output rates and concentrations, thermodynamics, and kinetics of both diffusive and nondiffusive processes were required for modeling the fate of γ-HCH. The parameters used were either collected from the literatures or measured in our laboratory. Temperature corrections were conducted for Henry’s law constant and vapor pressure according to Paasivirta et al. (24). Since there are usually a wide range of values for most of the parameters, reflecting both variability and error in measurements, as much data as possible were collected for model calculation and uncertainty analysis. Multiple values were collected for each of the 44 (39 input parameters plus 5 measured concentrations) parameters out of 67 (61 input parameters and 6 measured concentrations) in total. As a result, geometric means and standard deviations could be derived. For the rest of the 23 parameters (22 input parameters and 1 measured concentration) each with only a single value available, coefficients of variation were artificially assigned and used to derive standard deviations. Statistics including geometric means, geometric standard deviations, sample size, and a description of how coefficients of variation were artificially derived are provided in the Supporting Information (Figure S4). Sensitivity Analysis. The influences of individual parameters on the outcome of the multimedia model were addressed using sensitivity analysis. The sensitivity of the model results to variation of the input parameters was tested by running the model with an individual parameter multiplied by factors ranging from 0.1 to 10.0. Evaluation was conducted by calculating sensitivity coefficients that are the relative changes of major output estimates over the changes of input

parameters. A coefficient of variation adjusted sensitivity coefficient was proposed to normalize the dispersion of the input parameters on the influence. Uncertainty Analysis. Both concentrations and fluxes estimated by the multimedia model are inherently variable. So are the input parameters. In addition to the inherent variability, there are also uncertainties in the parameters and estimates (25). For assessment of the overall uncertainty and variability in predictions, Monte Carlo simulation was used to propagate collective variance of the inputs through the model (26). Each input parameter was represented as a probability density function that defined both the range of values and the likelihood of the parameter having that value. Based on evaluation of probability distribution of input parameters, log-normal distribution was adopted for all input parameters under consideration. For a given distribution, the simulation was undertaken repeatedly 20 000 times, with new values randomly selected for all parameters from within their probability distributions. A built-in function of “randn” in Matlab was used to select the values randomly for each parameter (23). Distributions of environmental media concentrations and transfer fluxes representative of the combined effects of uncertainty and variability were constructed and used for uncertainty evaluation. For media concentrations, the generated distributions were compared with the distributions of the observed values in an attempt to distinguish between the model uncertainty and variability.

Results and Discussion Statistical Distribution of the Model Parameters. For each of 39 input parameters (out of 61 total), there was more than one value collected. Examples of the distribution of several representative parameters are shown in the Supporting Information (Figure S5) as histograms, and a tendency for shifting toward normal distribution pattern after logtransformation can always be seen. In fact, most input parameters fall into two categories similar to either Cv and O33 (log-normally distributed) or Ps and Ul (right-skewed but not log-normally distributed) shown in the Supporting Information (Figure S5), and the deviation of these parameters from normal distribution is summarized in Figure 1 as coefficients of skewness and kurtosis before and after log transformation. Log-normal distribution of many parameters has been reported in the literatures. For instance, it has been repeatedly demonstrated that concentrations of micropollutants including pesticides in various media are log-normally distributed (27-29). Examples of log-normal distribution reported in the literatures include level of airborne particulates, wind speed, diffusion parameter, runoff frequency, soil organic matter, Henry’s constant, vapor pressure, and degradation rate constant (25, 26, 30-32). Therefore, geometric means and standard deviations were computed for all input parameters (Table S3, Supporting Information) and used in model calculation and uncertainty analysis. Concentrations of γ-HCH in Various Media and Model Validation. The concentrations of γ-HCH in various compartments including air, water, soil, sediment, crops, and fish were calculated using geometric means as inputs. The results are shown in Figure 2 with independently observed data also as geometric means. In addition to geometric means, geometric standard deviations were also shown in Figure 2 for the observed concentrations except air for which only a single measurement was available. The calculated air concentration (8.89 × 10-11 mol/m3) was 1 order of magnitude higher than that of the background (advective air flow, 7.11 × 10-12 mol/m3) due to extensive local application, while the difference of water concentration between the calculated (1.45 × 10-4 mol/m3) and background (upstream, 3.68 × VOL. 38, NO. 7, 2004 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

2127

FIGURE 1. Coefficients of skewness and kurtosis of selected input parameters used for the fugacity modeling before and after logtransformation. The parameters from left to right are as follows: C3, soil concentration; Cv, vegetable concentration; O33, soil organic carbon; Q02h, wastewater discharge rate; kp, dry deposition velocity; Gf, fish density; X23, solid fraction of water; Ps, vapor pressure; BCF, bioconcentration factor; Km4, degradation rate in sediment; X13, solid fraction in air; Q02t, advective water flow in; Ul, runoff rate of dissolved component.

FIGURE 2. Comparison between the calculated and the measured γ-HCH concentrations in air, water, soil, sediment, crops, and fish in Tianjin. For the measured concentrations, standard deviations are also provided as little bars except air for which only a single item was available. 10-7 mol/m3) was more than 2 orders of magnitude mainly because of wastewater discharge from the local HCH producers. For air, water, soil, sediment, crops, and fish, differences between calculated and measured geometric means were 0.11, 0.35, 0.60, 0.35, 1.06, and 0.52, respectively. Although most differences fall within an acceptable range (factors of 3-5, equivalent to a log-unit range of 0.5-0.7; 33), it is noticed that the calculated results were overestimated for all media. With seven compartments in the model, such overestimation was unlikely to occur by pure chance (less than 1% probability). Several reasons might contribute to the overestimation: (i) The application rate could be overestimated for the 1980s when technical HCHs were replaced with lindane. (ii) The degradation rate constants in major sinks could be underestimated. (iii) The observed concentrations of various media could be underestimated due to different technique used for sample analysis in the 1980s. The largest residuals were found for crops followed by soil. Later in this paper, it is demonstrated that the calculated soil concentration was very sensitive to three parameters, including application rate on soil (T03h), thickness of soil (h3), and degradation rate constant in soil (km3). Since h3 was carefully determined based on the measurement of several vertical profiles of γ-HCH in the area (Table S3, Supporting Information), the overestimation of soil concentration may be induced by error embedded in T03h and/or km3, both of which were from the literatures with large variation. Although the sensitivity of crop concentration was not directly tested in the study, an order of magnitude difference between the measured and the calculated γ-HCH concentrations in crops could be caused by inaccuracy of the air/vegetable partitioning coefficient (BCFv) and/or application rate to air (T01h), which are logically the most influential input parameters to concentration of crops either directly or indirectly. The total amount of γ-HCH accumulated in all media in Tianjin when the contamination reached its maximum was 2128

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 38, NO. 7, 2004

estimated as 102.4 t, equivalent to 1.3 times the quantity annually applied by agriculture in the area. If the system approximately reached steady state in 10 yr, the total accumulation was 13% of the amount applied during that period. Around 87% of the applied pesticide left the area mainly through degradation and advective airflow. There were 0.8, 1.6, 77.1, and 22.9 t of γ-HCH in air, water, soil, and sediment, accounting for 1, 2, 75, and 22% of the total, respectively. It appears that 97% of the pesticide accumulated in either soil or sediment because of its high affinity to organic matter. Transfer Fluxes of γ-HCH. The calculated transport fluxes are shown in the Supporting Information (Figure S3). The fluxes in and out of the area as well as each compartment were well-balanced. For instance, the relative error of the total flux in (27.001 mol/h) and out (26.999 mol/h) of Tianjin was 0.0099%. Figure 3 is a schematic diagram illustrating the transport processes in the area with arrows indicating the transport directions and arrow size (area) proportional to the calculated fluxes. The diagram was simplified by combining the processes in the same direction. For instance, the flux from air to soil includes diffusion, wet deposition, and dry deposition, which were calculated separately in the model. The Tianjin Chemical Company and the Dagu Chemical Company were among the largest producers of HCHs in China. They discharged a large volume of wastewater with γ-HCH concentrations as high as 0.078 mg/L directly into local rivers (11, 12). Still, the wastewater contributed less than 2% of total γ-HCH loading, while the primary source in Tianjin was agricultural application. Typically around 40% of applied γ-HCH is lost directly to atmosphere through evaporation (11, 34). Although a portion of the contaminant entering the air returned to the ground through precipitation and diffusion within the area (Figure 3), most of them were carried outside of the area by advective airflow. This is similar to the result reported by Takeoka et al. (35), who examined the flux of HCH during agricultural application in India a decade ago and found that almost all HCH applied was lost to the atmosphere. Consequentially, the advective airflow out of the area (6.19 mol/h) was more than 12 times that of the advective air flow in (0.50 mol/h). Regionally and globally speaking, therefore, Tianjin was a source of γ-HCH to other areas. The overall flux through surface water was small as compared to air. The area is well-known for its severe water shortage. Most waterways were gated for water conservation, and the results had a dual effect: (i) the water was extremely stagnant and (ii) the water in the channels was severely contaminated, both of which favored an active interchange between water and sediment (Figure 3). The calculated quantity of γ-HCH removed from Tianjin by advective water flow (0.18 mol/h) was slightly higher than that brought in (0.11 mol/h). However, the output was probably under-

FIGURE 3. Transport fluxes of γ-HCH in and out of Tianjin area and between the adjacent compartments. The arrows indicate the major transport and degradation pathways with their areas proportional to the calculated fluxes in geometric means.

FIGURE 4. Plots of the calculated C2 (water concentration) and F21 (water to air flux) against the factor by which h2 (water thickness) and X23 (solid fraction in water) were multiplied, respectively. estimated because the major wastewater discharge outlets of γ-HCH were located downstream of Jiyun River, very close to Bohai Bay (see locations of two chemical companies in Figure S1, Supporting Information). It is unlikely that the discharged wastewater could be well mixed with other water bodies in the area, as assumed in the multimedia model. Without spatial resolution, however, the current model failed to catch the effect of point source locations. There are many evidences that wastewater irrigation is one of the most important sources for heavy metal contamination of local soils and agricultural products in Tianjin (12, 36). However, as can be seen in Figure 3, not much γ-HCH was brought to farmland via wastewater irrigation. On the basis of an extensive survey in the area, Gong et al. have also found no significant difference in HCHs levels between the agricultural lands with and without wastewater irrigation (8). On the other hand, surface runoff was the second most important source of γ-HCH to water, next to air-water precipitation. Among other across-interface transport, all diffusive processes were slow. This is similar to the fate of benzo[a]pyrene in this area (37). As a typical persistent chemical, γ-HCH degrades very slowly in natural environment. Still, degradation was the primary process for disappearance of γ-HCH in the area (Figure 3), simply because it already accumulated to a very high level and the transfer processes reached a steady state after more than a decade of application. The total degradation flux of all media (20.6 mol/h) was more than three times that of the advective airflow out of the area (6.19 mol/h). With the highest concentration of γ-HCH, degradation in soil was the most important cause of removal of γ-HCH (Figure 3). Sensitivities of the Estimates to Input Parameters. For parameter sensitivity analysis, each parameter was multiplied by factors from 0.1 to 1.1 with an increment of 0.1 and from 2 to 10 with an increment of 1. Either linear or nonlinear responses of the calculated concentrations and fluxes to perturbations of input parameters were exhibited, and two examples are shown in Figure 4.

To compare the influence among parameters, a sensitivity coefficient (CS) similar to the one used by Mongan and Henrion was adopted (38). The coefficient approximately equals the tangent slope of the response curve at factor ) 1 point (the ratio of the relative variation of the estimated concentration to that of input parameter):

CS ) Abs((CF0.9 - CF1.1)/(0.2 × CF1.0))

(1)

where CF0.9, CF1.0, and CF1.1 represent the estimated concentrations when the tested parameter was multiplied by a factor of 0.9, 1.0, or 1.1, respectively. The absolute values were taken for an easy comparison of the extent of the influence. For the parameters with curvilinear sensitivity response (influence of h2 on C2 in Figure 4 for instance), the CS thus defined depends on the range used, and (0.1 in this study was small enough so that the effect caused by nonlinearity was neglected. Practically, the influence of a parameter on the estimates depends not only on the CS value but also on their actual perturbation in the real world. It is anticipated that a parameter with larger variability is more influential than a parameter with lower variability with equal CS values. To include the influence of the parameter variability in the analysis, Lohman et al. (39) classified all parameters into four categories of those showing large, medium, and local perturbations and those that are constant and evaluated the sensitivities of the parameters of each category separately. Since the variabilities of most parameters were quantified in this study, a CV (coefficient of variation) normalized coefficient of sensitivity (Cn) was proposed as

Cn ) CS × CV

(2)

Thus, sensitivities of model estimates on various input parameters could be evaluated in terms of combined influence of true model sensitivity and parameter variability. Both CS (bottom) and Cn (top) for the calculated concentraVOL. 38, NO. 7, 2004 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

2129

FIGURE 5. Coefficients of sensitivity of the calculated concentrations of the four major compartments to the input parameters with (Cn) and without (CS) CV normalization. The parameters are arranged in a descending order of the sum of four Cn values. The parameters from left to right are as follows: km4, degradation rate in sediment; T01h, application rate to air; C02h, wastewater concentration; T03h, application rate to soil; h3, soil thickness; km3, degradation rate in soil; Koc, adsorption coefficient; h4, sediment thickness; O43, organic carbon in sediment; O23, organic carbon of solid in water; Ps, vapor pressure; km2, degradation rate in water; B4, diffusivity in sediment; O33, organic carbon in soil; Q01t, advective air flow in; k12, mass transfer coefficient at air side over water; ... A2, area of water; A3, area of soil.

FIGURE 6. Coefficients of sensitivity of the calculated fluxes of the four major transfer and degradation processes to the input parameters with (Cn) and without (CS) CV normalization. The parameters are arranged in a descending order of the sum of four Cn values. The parameters from left to right are as follows: B4, diffusivity in sediment; T01h, application rate to air; C02h, wastewater concentration; L4, diffusion path length in sediment; km4, degradation rate in sediment; T03h, application rate to soil; Koc, adsorption coefficient; O43, organic carbon in sediment; O23, organic carbon of solid in water; Ps, vapor pressure; km2, degradation rate in water; O33, organic carbon in soil; Q01t, advective air flow in; k12, mass transfer coefficient at air side over water; h3, soil thickness; km3, degradation rate in soil; ... A2, area of water; A3, area of soil. tions of the four major compartments and for the fluxes of four important processes are shown in Figures 5 and 6 in descending order of Cn. On the basis of CV normalized sensitivity, the most significant parameters with respect to their overall influence on the model outputs are Km4 > T01h > C02h > T03h > h3 > Km3 > Koc for the estimated concentrations and B4 > T01h > C02h > L4 > Km4 > T03h > Koc for the predicted major fluxes. The later depend very much on the specific processes selected. For several parameters including A2, A3, Q01t, and Ps, large CS values for both concentration and flux were offset by their relatively small CVs, resulting low Cn values. They are therefore not considered among the most influential parameters in the model. In summary, the most significant parameters for modeling the fate of γ-HCH are degradation and source terms. This is similar to those reported by Liu et al. (26). In addition, the estimates were also very sensitive to thickness of soil (h3) and organic carbon-normalized adsorption coefficient (Koc). Uncertainty of the Model Estimates. As result of 20 000 Monte Carlo simulations, the predicted γ-HCH concentrations in four bulk media and two biological compartments are presented in Figure 7 as log-scaled distributions. For comparison, the observed concentration distributions are also illustrated except air for which only a single measurement was available. As can be seen in Figure 7, the calculated concentrations varied widely covering a range of several orders of magnitude, 2130

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 38, NO. 7, 2004

especially for soil and sediment. The dispersions of the calculated concentrations are significantly larger than those of the measured concentrations in all media. The latter are due to both variability (spatial and temporal) and measurement uncertainty. For uncertainty analysis, it is always hoped that the relative contributions of the variability and true uncertainty of the model estimates can be distinguished (25). In this study, the true uncertainty may be expressed to a certain extent by the difference between the dispersions of the two distributions for each compartment shown in Figure 7. To quantify the differences, coefficients of variation were calculated based on log-transformed data, and the results are presented in Figure 8 as stacked bars. If the random errors in the concentration measurements are neglected, the bottom parts of the stacked bars exhibit the concentration variability (both spatial and temporal), and the entire heights of the bars indicate the perturbations of the calculated concentrations. The differences between the two (the upper parts of the stacked bars) represent the true uncertainties for each compartment. It appears that the uncertainty of the modeled water concentration is significantly smaller than those of other media, although the variabilities of them are relatively similar to each other. Although fish were treated as a subcompartment of water in the model framework, the uncertainty of the calculated concentration in fish was much larger than that of water, simply due to relatively large variability of BCFf (CV was among the largest in all input parameters). According to the

FIGURE 7. Comparison of the log-transformed distributions of the observed and the estimated concentrations in various media. The observed distributions were generated based on geometric means and standard deviations of the measured concentrations, while the estimated distributions were from Monte Carlo simulation. For air, only a single observed data was available (shown as a broken line).

FIGURE 8. Comparison between the observed variability and model uncertainty. The CVs of the calculated concentrations are illustrated by the total heights of the stacked bars while the CVs of the observed concentrations are reflected by the bottom part of the bars. results shown in Figure 8, the variability of γ-HCH concentrations in all media studied was in a descending order of fish > sediment > soil > water > crops, while the uncertainties were in a descending order of soil > fish ≈ sediment ≈ crops > water. Without duplicated measurements available, air is not included in the lists. It was often reported that sensitive model input parameters usually account for the largest variance in model prediction (40). The large uncertainties of the calculated soil and sediment concentrations are related to the most influential parameters identified in the sensitivity analysis (km4, T03h, h3, km3, etc.). It should be pointed out that the accuracy of the true uncertainties addressed above depends very much on the accuracies of the distributions of both observed and modeled concentrations. Because of remaining uncertainties over input parameters and relatively small sample sizes for most media, the results cannot be assumed to be precisely model details of the fate of pollutants in Tianjin. With no observed results available, the variability and uncertainty of the modeled transfer fluxes are presented in Figure 9 as a Box-Whisker diagram including the 5th, 25th, 50th (median), 75th, and 95th percentiles as well as geometric means.

FIGURE 9. Variability and uncertainty of the estimated fluxes among various media as medians (P50), geometric means (MGEO), and the 5th (P5), 25th (P25), 75th (P75), and 95th (P95) percentiles. The results were generated by performing the Monte Carlo simulation using log-normally distributed input parameters. The process of diffusion, dry precipitation, wet precipitation, sedimentation, erosion as solid, and erosion as liquid were designated by d, p, w, s, e, and l, respectively. Semi-interquartile ranges (SIR, difference between the 25th and the 75th percentiles in mol/m3) of various processes were calculated for comparison. Most SIR calculated for fluxes were around 0.5 orders of magnitude, with the largest of 1.3 for diffusion from sediment to water (T42d), followed by diffusion from soil to air (T31d) and from water to air (T21d). It appears that all diffusion processes have relatively large errors due to large variations of molecular diffusion coefficients and transfer rate constants (Supporting Information, Table S3).

Acknowledgments Funding was provided by The National Scientific Foundation of China (Grant 40332015), National Basic Research Program of China (Grant 2003CB415004), and The National Scientific Foundation of China (Grant 40021101). VOL. 38, NO. 7, 2004 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

2131

Supporting Information Available Detailed information on study area, model structure, parameter identification, and results of uncertainty analysis and also modeled processes and input parameters. This material is available free of charge via the Internet at http:// pubs.acs.org.

Literature Cited (1) Chen, S. P.; Chen, Z. W. Atlas of Environmental Quality; Tianjin Scientific Press: Beijing, 1986. (2) Zhao, L.; Ma, Y. J. Agric. Environ. Dev. 2001, 67, 37-39. (3) Iwata, H.; Tanabe, S.; Sakai, N.; Tatsukawa, R. Environ. Sci. Technol. 1993, 27, 1080-1098. (4) Loganathan, B. G.; Kannan, K. Ambio 1994, 23, 187-191. (5) Wania, F.; Mackay, D. Environ. Sci. Technol. 1996, 30, 390A396A. (6) Gong, Z. M. Organochlorine Pesticide Residues in Soils and Crops from Tianjin. Ph.D. Thesis, Peking University, Beijing, 2003. (7) Wania, F. Environ. Sci. Technol. 2003, 37, 1344-1351. (8) Gong, Z. M.; Xu, F. L.; Dawson, R.; Cao, J.; Liu, W. X.; Li, B. G.; Shen, W. R.; Zhang, W. J.; Qin, B. P.; Sun, R.; Tao, S. Arch. Environ. Contam. Toxicol. (in press). (9) He, X. C. Organochlorine Pesticides in Water, Suspended Solids, and Bottom Sediment from Rivers in Tianjin. Masters Thesis, Peking University, Beijing, 2003. (10) Ministry of Chemical Industry. On Implementation of the Notification from the Office of the State Department on Management of Pesticides and Animal Medicines; MCI[1992]77; Beijing, 1992. (11) Li, Q. S. Environ. Sci. Ser. 1986, 6, 5-14. (12) Tianjin Environmental Protection Bureau. Environ. Qual. Statement 1986-1990; Tianjin Environmental Protection Bureau: Tianjin, 1991. (13) Hua, X. M.; Shan, Z. J. Adv. Environ. Sci. 1996, 4, 33-45. (14) Schnoor, J. L.; McAvoy, D. C. J. Environ. Eng. Div. ASCE. 1981, 107, 1229-1246. (15) Yoshida, K.; Shigeoka, T.; Yamauchi, F. Toxicol. Environ. Chem. 1987, 15, 159-183. (16) Mackay, D. Environ. Sci. Technol. 1979, 13, 1218-1223. (17) Mackay, D.; Paterson, S.; Shiu, W. Y. Chemosphere 1992, 24, 695-717. (18) Ling, H.; Diamond, M.; Mackay, D. J. Great Lakes Res. 1993, 19, 582-602. (19) Ares, J.; Miglierina, A. M.; Sanchez, R.; Rosell, R. Environ. Model. Assess. 1998, 3, 95-105. (20) Wania, F.; Mackay, D. Environ. Toxicol. Chem. 1999, 18, 14001407. (21) Fu, S. X.; Zhang, C. H.; Cao, G. F. Atlas of Eco-environment of Beijing-Tianjin Area; Science Press: Beijing, 1990; pp74-75.

2132

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 38, NO. 7, 2004

(22) Mackay, D.; Paterson, S. Environ. Sci. Technol. 1991, 25, 427436. (23) MathWorks. Using MATLAB version 6; The MathWorks, Inc.: Natick, MA, 2002. (24) Paasivirta, J.; Sinkkonen, S.; Mikkelson, P.; Rantio, T.; Wania, F. Chemosphere 1999, 39, 811-832. (25) McKone, T. E. Reliab. Eng. Syst. Saf. 1996, 54, 165-181. (26) Liu, C.; Bennett, D. H.; Kastenberg, W. E.; McKone, T. E.; Browne, D. Reliab. Eng. Syst. Saf. 1999, 63, 169-184. (27) Aigner, E. J.; Leone, A. D.; Falconer, R. L. Environ. Sci. Technol. 1998, 32, 1162-1168. (28) Gordon, S. M.; Callahan, P. J.; Nishioka, M. G.; Brinkman, M. C.; O’Rourke, M. K.; Lebowitz, M. D.; Moschandreas, M. J. J. Exposure Anal. Environ. Epidemiol. 1999, 9, 456-470. (29) Kim, S. K.; Oh, J. R.; Shim, W. J.; Lee, D. H.; Yim, U. H.; Hong, S. H.; Shin, Y. B.; Lee, D. S. Mar. Pollut. Bull. 2002, 45, 268-279. (30) James, S. C.; Chrysikopoulos, C. V. Chem. Eng. Sci. 2001, 56, 6535-6543. (31) Ali, S.; Singh, R. Appl. Eng. Agric. 2001, 17, 285-291. (32) Lu, H. C.; Fang, G. C. Sci. Total Environ. 2002, 298, 119-130. (33) Cowan, C. E.; Mackay, D.; Feijtel, T. C. J.; van de Meent, D.; Di Guardo, A.; Davies, J.; Mackay, N. The Multi-media Fate Model: a Vital Tool for Predicting the Fate of Chemicals. In Proceedings of a Workshop Organized by the Society of Environmental Toxicologists and Chemists (SETAC); SETAC Press: Pensacola, FL, 1995. (34) Chen, W. X. Protection of Agricultural Environment; Agricultural Press: Beijing, 1993. (35) Takeoka, H.; Ramesh, A.; Iwata, H.; Tanabe, S.; Subramanian, A. N.; Mohan, D.; Magendran, A.; Tatsukawa, R. Mar. Pollut. Bull. 1991, 22, 290-297. (36) Xu, S. P.; Tao, S.; Cao, J.; Li, B. G.; Shen, W. R. Chin. Environ. Sci. 1999, 19, 226-229. (37) Wang, X. L.; Tao, S.; Dawson, R.; Li, B. G.; Cao, J. J. Envioron. Qual. 2002, 31, 896-903. (38) Mongan, M. G.; Henrion, M. Uncertainty: A Guide to Dealing with Uncertainty in Quantitative Risk and Policy Analysis; Cambridge University Press: Cambridge, 1990. (39) Lohman, K.; Pai, P.; Seigneur, C.; Levin, L. Sci. Total Environ. 2000, 259, 3-11. (40) Wolt, J.; Singh, P.; Cryer, S.; Lin, J. Environ. Toxicol. Chem. 2002, 21, 1558-1565.

Received for review August 8, 2003. Revised manuscript received December 11, 2003. Accepted January 6, 2004. ES0305860