Multimedia Fate and Human Intake Modeling: Spatial versus

Multimedia Fate and Human Intake Modeling: Spatial versus Nonspatial Insights for ...... International Lake Environment Committee Survey of the State ...
1 downloads 0 Views 412KB Size
Environ. Sci. Technol. 2005, 39, 1119-1128

Multimedia Fate and Human Intake Modeling: Spatial versus Nonspatial Insights for Chemical Emissions in Western Europe DAVID W. PENNINGTON,* MANUELE MARGNI, CHRISTOPH AMMANN, AND OLIVIER JOLLIET Industrial Ecology - Life Cycle Systems, Institute of Environmental Science and Technology, Ecole Polytechnique Fe´de´rale de Lausanne, EPFL-GECOS, section 2, CH-1015 Lausanne EPFL, Switzerland

Multimedia fate and multipathway human exposure models are widely adopted in assessments of toxicological risks of chemical emissions at the regional scale. This paper addresses the question of how much spatial detail is necessary in such models when estimating the intake by the entire population in large, heterogeneous regions such as Europe. The paper presents a spatially resolved multimedia fate and multipathway exposure model for Western Europe, available as IMPACT 2002. This model accounts for relationships between the location of food production and drinking water extraction as well as where population cohorts live relative to where chemical emissions occur. The model facilitates estimation of environmental concentration distributions, related levels of contaminants in foods, and the fraction of a chemical release that will be taken in by the entire human population (the intake fraction) at the regional scale. To evaluate the necessary spatial resolution, the paper compares estimates of environmental concentrations and the intake fraction from the spatially resolved model with the results of a consistent clone without spatial resolution. An evaluation for disperse emissions of PeCDF (2,3,4,7,8-pentachlorodibenzofuran, CAS# 5120731-4) suggests reasonable agreement with monitoring data for most impact pathways with both versions of the model, but that the generic vegetation models for estimating contaminant concentrations in agricultural produce require improvement. A broader comparison for a range of organic chemicals demonstrates that the nonspatial models are likely to be appropriate in general for assessing dispersed sources of emissions. However, it is necessary to include generic compartments in such nonspatial models to account separately for emissions that enter lakes with long residence times versus rivers that feed directly into seas. For assessing an emission source in a specific location, using models that are not spatially resolved can result in underestimation, or overestimation, of the

* Corresponding author phone: +39 0332 785880; fax: +39 0332 785601; e-mail: [email protected]. Current address: TP 460, Soil & Waste Unit, Institute of Environment & Sustainability, Joint Research Centre, European Commission, Ispra (Va) 21020, Italy. 10.1021/es034598x CCC: $30.25 Published on Web 01/15/2005

 2005 American Chemical Society

population’s intake by at least 3 orders of magnitude for some chemicals.

Introduction It is timely to consider when nonspatial multimedia fate and multipathway exposure models remain sufficient for assessing chemicals and when spatial resolution is necessary. Estimating the fraction of an emission that will be taken in by the entire human population and the corresponding necessity of considering spatial variations over large regions have been addressed extensively, for example, for radionuclides and atmospheric pollution. Historically, this has not been the case in the context of multimedia/multipathway modeling. Multimedia/multipathway models have been developed to support various scientific, regulatory, and educational purposes (1). Such models facilitate assessment of the fate of chemicals in the environment at a regional scale, the associated potential for human exposure, as well as the risks of toxicological effects. For example, the multimedia/ multipathway model in EUSES provides a well-established, straightforward, consensus basis for screening chemical emissions in the context of unacceptable toxicological risks at a regional scale in the European Union (2). Straightforward multimedia/multipathway models have long provided estimates of risks for hypothetical individuals that dwell in a region of, for example, 200×200 km2 or larger, where an emission occurs (1). Spatial variations are not explicitly taken into account, within environmental media or when estimating human exposure. Individuals are modeled assuming exposure to average contaminant concentrations in the region. This provides a useful basis for screening in terms of individual risk in the region of the emission. Population-level estimates of human intake are necessary in other applications of growing importance in a policy context. Estimating the effects at the population level is necessary, for example, in cost-benefit analyses (3, 4) and life cycle assessments (LCA) (5-7). Intake by the entire population is characterized by the intake fraction, the fraction of the mass of a chemical released into the environment that will be taken in by the human population via food consumption, inhalation, and dermal exposure (7). Estimating the intake fraction for many chemicals and emission scenarios is likely to require modeling large regions, such as all of Europe. Assuming that concentrations of contaminants in each medium, crop production levels, and the population density are also distributed at average levels over such large regions may not provide a robust basis for estimating the intake fraction in all cases. Klepper et al. (8) compared the results of a nonspatial multimedia model with those of spatially resolved single medium models for air, soil, and water. Calculated maximum concentrations in the spatial models were a factor of 1000 higher than those of the nonspatial model for some chemicals, particularly for emissions to water. Spatially resolved multimedia models are now also becoming available through the growing exploitation of geographic information system (GIS) data and tools (9-14). Prevedourous et al. (13), for example, developed a spatially resolved multimedia model for Europe, adapting the BETR North American model (11, 12), and demonstrated likely variations of 2 orders of magnitude in the concentrations of γ-HCH (the pesticide lindane). Predictions were generally within an order of magnitude of monitoring data for most sites. Suzuki et al. VOL. 39, NO. 4, 2005 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

1119

(14) similarly presented results for Japan for four organic chemicals with air concentrations typically within an order of magnitude of measured values, although with greater discrepancies for other media. The results from a nonspatial multimedia model for the four chemicals generally corresponded with the average of the concentrations from the spatially resolved model, suggesting nonspatial models may be appropriate at such scales when an average estimate is sufficient. Reported differences in environmental concentrations could be further compounded when considering human intake by, for example, the spatial distribution of where food is produced or water is extracted relative to the location of an emission source and how a contaminant distributes in the environment. To evaluate the extent to which spatial distinction is necessary when estimating the intake fraction, this paper presents a spatially resolved multimedia/multipathway model for Western Europe, IMPACT 2002 (see Supporting Information). Intake fractions are calculated from the contaminant concentration in food produced at each location, the water extracted to serve a given population at each location, as well as the population distribution when considering inhalation. The results are compared with those of a clone of the same model that is not spatially resolved: (1) together against monitoring data for a disperse emission of the dioxin congener, PeCDF (2,3,4,7,8-pentachlorodibenzofuran; CAS# 51207-31-4) and (2) for a set of organic chemicals to broadly assess when spatial resolution is needed to estimate the human intake fraction versus when nonspatial models are likely to be sufficient for disperse emission sources and for emissions in specific locations.

Model Description Population Intake and the Intake Fraction. The overall intake of a chemical from an emission by the human population is characterized by the intake fraction (7). The intake fraction (iF) is the fraction of the mass of a chemical released into the environment that will be taken in by the entire human population via food consumption, inhalation, and dermal exposure (dermal exposure is not addressed in this paper). A high value, such as iF ) 0.001, reflects that the population will take in 1 part in 1000 of the emitted quantity. To estimate an intake fraction at the regional scale requires the combination of (1) a mass balance to describe how a chemical distributes after emission within as well as between each medium of the environment and (2) a model of the fraction of the contaminants at each location that will be taken in by the population. The environment is modeled here as a number of discrete compartments. A compartment can be in any environmental medium, a medium being divided into several compartments. The distribution of a contaminant among environmental compartments is presented in the form of a vector of contaminant masses, M B [kg]. A vector of intake rate coefficients, B E [day-1], describes the rate at which the population takes in the contaminant from each compartment through food, drinking water, and inhalation. Adopting the common assumption of steady state or timeaveraged distributions (dM/dt ) 0), the intake fraction is given by

iF )

B E‚M B S

(1)

where B E‚M B is the intake rate of the contaminant by the entire population [kg/day] and S is the emission rate [kg/day]. The next sections describe the calculations in the model for the mass distribution (M B ) and the intake rate coefficients (E B). 1120

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 39, NO. 4, 2005

Chemical Mass Distribution. To estimate the distribution of contaminants in the environment, M B in eq 1, two versions of a multimedia fate model were developed: (1) A model of Western Europe with spatial differentiation in each environmental medium. Each medium is subdivided into a number of homogeneous compartments, analogous to approaches in other spatially resolved multimedia models9-14 and as described in the following sections. (2) A nonspatial clone. Spatial resolution is not represented within each environmental medium (air, water, soil, sediment, vegetation). This nonspatial clone is based on the same data and algorithms as the spatial version. Nonspatial model parameters are presented in the Supporting Information and were calculated by summing or averaging spatial data, as appropriate. In both cases, a typical nested approach (1) was adopted to account for any intake that may occur as a result of contaminant advected outside of the modeled Western European region. The Western Europe model was nested in a nonspatial global model. Estimating the mass distribution in the spatially resolved multimedia model with n compartments required a mass balance solution for n simultaneous differential equations. Chemical transport can be between compartments in the same medium, or between compartments in different media. Transport can be by any mechanism (advection, diffusion, etc.). Typical of multimedia modeling (1), these rates were all described using first-order rate coefficients that reflect typical conditions and chemical transport at the regional scale (see next section). Matrix algebra provided the most straightforward and transparent solution to readily visualize and to solve the n simultaneous differential equations for this n-compartment model (see Supporting Information). For steady state, the vector of masses describing the contaminant’s distribution in the environment (M B ) is given by inversion of the matrix of transport rate coefficients

B M B ) - kh -1‚S

(2)

where B S is the vector of the emission rates [kg/day] and kh is the matrix of bulk transport rate coefficients [day-1]. Figure 1 illustrates the rate coefficient matrix (kh ) for a hypothetical model with four compartments in two media (2 in air and 2 in water). Rate coefficients off the diagonal describe transport between compartments. For example, kair(B)-water(B) is the rate coefficient for transport from the air compartment B to the water compartment B. Rate coefficients on the diagonal are the sum of all the removal rate coefficients for the given compartment. These overall coefficients account for the total rate of loss from a compartment by degradation, removal by intercompartment transport, or by any other losses modeled from the system. A diagonal element such as kair is therefore equal to the inverse of the overall half-life of the chemical in that air compartment (k ) ln(2)/τ1/2). Equation 1 can also be solved to provide an exact dynamic solution to address the importance of temporal variations. The implications of temporal variations and the dynamic solution are not considered in this paper. Intermedia Rate Coefficients. Rate coefficients describe the overall intermedia transport of chemicals from one compartment to another at a regional scale in multimedia models. These rate coefficients are commonly estimated for organic chemicals (1, 15). The coefficients considered in this model are listed in the Supporting Information. The algorithms adopted to estimate the coefficients were for transport from air (diffusion, wet and dry deposition, dissolution with corrections to account for intermittent rainfall events in a steady-state model) (16-18), from surface waters (diffusion,

FIGURE 1. Illustrative mass balance calculation using matrix algebra for a four-compartment system, consisting of the two air and two water compartments in the diagram. deposition) (19), from soils (infiltration, diffusion, and runoff - including accounting for impermeable surfaces and vertical chemical profiles) (20), from sediments (resuspension and burial) (16), from generic agricultural vegetation (roots, leaves, stems, and surface layer compartments with transport by diffusion, resuspension from foliage surfaces, root and stem uptake) (21, 22), and for oceanic water (advective transport across the thermocline/euphotic layer, diffusion, sedimentation) (16). Location specific parameters used in the calculations for each coefficient and each compartment in the model were obtained with the help of Geographic Information System (GIS) databases and tools, as described in the next sections. For vegetation, the focus here was on agricultural crops. Sensitivity analyses using CalTOX (23) and this model, with and without vegetation compartments, suggested that the influence of vegetation is generally negligible on the concentration of contaminants in other media. Estimates for contaminant concentrations in crops remain of vital relevance for the calculation of human exposure for many chemicals and agricultural vegetation was therefore retained. Exclusion of other vegetation may only be applicable given the assumption of steady-state, for nondissociating organic chemicals, when rates of degradation in vegetation are low relative to transport (typically unknown), and when it is not desirable to know the concentrations in other types of vegetation. Intramedium Rate Coefficients. Rate coefficients are also necessary for the advective transport of a chemical between compartments within air and water. These coefficients are essentially dominated by the physical flow rates and patterns of air and water. The basis for the calculation of these advective flows is described in the next sections. Chemical diffusion is not considered relevant at the regional scale addressed here. Location Specific Fate Model Parameters. Table 1 summarizes the parameters that were spatially resolved in

the multimedia fate part of the model. Generic values from the nonspatial version of the model were retained for other parameters. The nonspatial model parameters are listed in the Supporting Information. Figure 2 illustrates the geographic scope of the model and the spatial delimitations adopted. The delimitations help define the size of each compartment, assign spatially dependent parameters, and describe the average advective flow rates of air and water between compartments at the regional scale. The following subsections describe the choice of these delimitations. Watershed Delimitation. Watersheds define water flow patterns at the regional scale and therefore provided a convenient basis for delimitation of terrestrial compartments in the model, see Figure 2. Four compartments described the media in each watershed: soils, surface waters, sediments, and agricultural vegetation (see also Supporting Information). No spatial distinction was made within each compartment in a watershed, typical of regional scale multimedia models (1). Separate delimitations for soils were not considered here. This choice of watershed resolution was based on where there are large differences in population density and on the location of major lakes. Lakes have a key influence on spatial sensitivity of the model results due to their larger hydraulic retention time when compared to rivers. The Rhine watershed, for example, has a surface area of more than 160 000 km2 and represents the entire catchment of the river Rhine. It was therefore necessary to select smaller watershed resolutions in this case. Air Grid Delimitation and Vertical Mixing Height. The air was subdivided into compartments according to a typical 2 × 2.5 degree grid (around 200 × 250 km) (33), see Figure 2. Rate coefficients for intermedia transport took into account the cross-sectional area that was in common between a compartment in a watershed and an atmospheric grid cell. These fractions were calculated using standard GIS techniques (24). Air compartments could be defined using the VOL. 39, NO. 4, 2005 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

1121

TABLE 1: Spatially Resolved Parameters in the Multimedia Fate Modela parameter average annual air flow rates watershed boundaries

annual mean runoff from each watershed average annual surface water flow rates from watersheds rainfall rate in each watershed average water depth in each watershed

land cover in each watershed soil urban water

range in model

description and source

4×1011-5×1012 m3/h Calculated using the underlying wind velocity data of the model GEOS-CHEM (25) and the perpendicular cross-sectional areas, with air subdivided according to a grid (see the following sections) 3400-95500 km2 ERICA (European Rivers and Catchments) data set from EEA (26), and for regions not covered (Ireland, Eastern Europe, parts of Italy, etc.), from the model-basedU.S. Geological Survey’s (USGS) HYDRO1k data set (27). Fringe of North Africa based on grid cells 37-1135 mm/year Mean annual runoff (rainfall - evapo-transpiration) from 0.5 × 0.5 degree grid data from the Global Run Off Data Centre (GRDC) (28) 2000-9×106 m3/h Flow rate leaving each watershed estimated from the mean annual runoff multiplied by the surface area, adding upstream inflows. (See Supporting Information for water flow rate details and evaluation with measured data) 130-2400 mm/year Calculated from the sum of evapo-transpiration and the run off Evapo-transpiration data from the hydrologic Science Branch of the NASA data on 0.5 × 0.5 degree raster (29) 0.1-167 m International Lake Environment Committee (ILEC) lake database (30) and Bundesamt fu¨ r Wasser und Geologie (BWG) for Swiss lakes (31) (see Supporting Information for lake details). River depth was approximated in each watershed from the main channel length, the estimated water surface area from the land cover data, the flow rate, and average river velocity of 2 m/s (expert judgment) Coordination of Information on the Environment (CORINE) land cover data set of the EEA in grid form with resolution of 250m (32) (mean adopted in 77-99% absence of data for the UK) 0.5-22% 0.04-8%

a Straightforward GIS (Geographic Information System) manipulations provided the mean values for each compartment in cases where a finer resolution of data existed (24).

FIGURE 2. Spatial delimitations in the modelled Western Europe region of 6 400 000 km2 (U.S.A.: 9 629 091 km2). Delimitations based on 135 watersheds for soils and surface waters and on 2 × 2.5 degree grid cells for air and oceanic waters. The population in this region is approximately 420 million (U.S.A.: 280 million) distributed across 25 nations (10 only partially represented). same horizontal delimitations as the watersheds, as this would simplify these intermedia transport calculations. However, defending the relevance of watershed boundaries in the context of modeling atmospheric transport is problematic (see Supporting Information) (34). 1122

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 39, NO. 4, 2005

Advective transport rates were calculated from wind velocity data, accounting for the perpendicular crosssectional area of each air compartment (see Table 1) (24, 34). The time-averaged flows of air in each direction in each compartment were taken into account using the approach

TABLE 2: Spatially Resolved Parameters in the Intake Modeling, Distributed According to Grid Cells and Watershedsa parameter population density

range in model

description and source

20-825 inhababitants/ UNEP Grid Centre, 1990 data, global grid with a resolution of 1×1 degree (36) km2 EUROSTAT Annual national agricultural production data for 1999 (37) Food and Agriculture Organization (FAO) national statistics database for 1999 for non-EU countries (38)

annual food production fruits and vegetables meat (beef, pork,...) milk eggs freshwater fish

0.5-0.008 kg/m2 soil 0.1-4×10-4 kg/m2 0.3-0.001 kg/m2 0.02-1×10-4 kg/m2 0-0.003 kg/m2 water

sea fish

0.004-0.0004 kg/m2

drinking water extraction

1-90% from surface water

Agricultural, livestock, and dairy annual usable production (2) data were allocated to the respective watersheds and air cells in proportion to the fraction of national agricultural land coverage, based on CORINE data (32). The data include the fraction consumed by humans as well as by animals as well as the quantities exported from Europe. When information was available, the fractional loss from market to table was taken into account - although this typically did not strongly influence the results Fisheries Yearbook 2001 provided saltwater catch data for 7 regions (39). Catch in the oceanic compartments was calculated as a function of the fishing area within the Mediterranean Sea and the North East Atlantic regions of the yearbook. These data included a correction factor to account for the ratio between the live and landed weights Water extraction data (groundwater and surface water fraction) from the EEA Water Abstraction by Source data set (40). Population served was based on population density, assuming average drinking water ingestion rates of 2 L person-1 day-1 (41). Groundwater contaminant concentrations were assumed equivalent to concentrations at the bottom of the unsaturated vadose layer in the soil, which proved a negligible contribution to intake fraction in this model. Detailed modeling of the fate of contaminants that pass into the groundwater or underground sources were not considered further

a Default values for other parameters and the mean values adopted in nonspatial model are listed in the Supporting Information. Straightforward GIS manipulations provided the mean values for each compartment in cases where a finer resolution of data was available (24). Expert judgment was used where a sufficiently fine spatial resolution of data was not available, e.g. to divide fish catch rates among oceanic grid cells in proportion to water surface area and food production rates among watersheds in proportion to agricultural land use in each nation.

of Woodfine et al. (12) Each air compartment had an assumed height of 800 m to represent a typical boundary layer mixing height. A sensitivity analysis demonstrated that inclusion of more than one atmospheric layer for modeling Western Europe in the context of human intake fractions, with separate horizontal and vertical wind velocity data for each layer, did not have an important influence in this steady-state model (34). Mixing within the 800 m layer is likely to be rapid and further vertical subdivision or accounting for transport in higher layers was not necessary here. Oceanic Delimitation. Each oceanic compartment has the same delimitation as the corresponding air compartment (grid cell). Oceans were modeled as two compartments of seawater and a sediment compartment in each cell. The two water compartments are above and below the annual average thermocline/euphotic layer, assumed to be at 500 m. Average seawater flow rates between the different compartments in the surface layer (above 500 m) were assumed, as a starting premise, to be uniform in all directions and with a velocity of 1 knot (3704 m/h). No advective flows were assumed in deeper, less turbulent, layers. Intake at the population level was not sensitive to these assumptions when varied within plausible ranges of annual averages in this model. Seawater fish consumption had little influence on intake at a population level when considering land-based emission sources. This will not always be the case when considering individual intakes or where emissions to seawater and impacts on marine ecosystems are of interest. Multipathway Human Intake. Having estimated the distribution of contaminants from an emission in the environment, M B , it is necessary to estimate the vector of intake rate coefficients, B E [day-1] (see eq 1). This vector describes the rate at which the population takes in a contaminant from each compartment in the model through food, drinking water, and inhalation. Population intake is calculated from the amount of contaminants in the food produced in a specific location, the water extracted at a specific location to serve a given

population, and the population that dwells in a location when considering inhalation. This is termed here a productionbased exposure scenario, reflecting primarily that the contaminant levels in food and drinking water are associated with where food is produced and not necessarily the location of where the population lives. This differs from a subsistence scenario, which is more often adopted in chemical screening and reflects an individual who eats, drinks, and lives within the region of an emission. Each intake rate coefficient in the vector (E B) in eq 1 is calculated for direct and indirect exposure

Ei ) Ei,direct + Ei,indirect

(3)

Ei,direct [1/day] is the rate coefficient for direct exposure to contaminants in compartment i, through consumption of local water or inhalation by the population (direct human consumption of soil was not considered). As an example, eq 4 is the calculation for inhalation.

Ei,direct (inhalation) )

populationi‚breathing rate [m3/day] volumei [m3]

(4)

The inverse of this rate coefficient for inhalation is the residence time, reflecting the average time required for the population in compartment i to inhale the volume of air of that compartment. The second rate coefficient in eq 3 for indirect exposure, Ei,indirect, requires estimation of how much contaminant will be in an intermediate substrate (vegetation, livestock, fish) multiplied by the amount of the substrate that will be ultimately consumed by the human population. Substrates include fish, meats, dairy produce, and vegetables. Equation 5 outlines the calculations. Table 2 summarizes the locationspecific data considered for food production, water extraction, and the distribution of the population. The Supporting Information section presents the additional steps for accounting for contaminants in animal feed and the intake by the animals prior to intake by humans. VOL. 39, NO. 4, 2005 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

1123

∑BAF Ei,indirect )

e,i‚PRe,i

e

Fi‚Vi

(5)

Equation 5 estimates the rate that a contaminant in substrate such as beef (e) will be taken in by the population relative to the mass in an associated environmental compartment (i). BAFe,i [kgi/kge] is the bioaccumulation factor, the mass fraction of a contaminant in the substrate relative to the mass fraction of contaminant in the environmental compartment. The mass of the compartment i is given by the bulk density, Fi [kgi/m3], multiplied by bulk volume, Vi [m3]. PRe,i [kg/day] is the rate of substrate produced associated with environmental compartment i and destined for human consumption. Contaminant concentrations in crops were estimated using the generic vegetation model described in the previous sections. BAFe,i values were generally estimated for organic chemicals for milk and meats using biotransfer factors and animal intake rates of roughage, air, and water (see Supporting Information for calculation details and data). In the absence of measured data, BAFe,i values for freshwater and saltwater fish were calculated using the correlations of Meylan et al. (35).

Results Evaluation Using Monitoring Data. To provide preliminary evaluation insights, the intake fraction estimates from both the spatially resolved and nonspatial versions of the Western Europe model were evaluated using monitoring data for the dioxin congener PeCDF (2,3,4,7,8-pentachlorodibenzofuran; CAS# 51207-31-4). Emissions profiles and monitoring data generally remain unavailable in the public domain for evaluations of a broader range of organic chemicals, as addressed in the next sections. The evaluation insights are therefore limited here to chemicals with a behavior similar to PeCDF as well as to disperse emission sources. Margni et al. (42) summarized the similar insights for other congeners. The emissions profile of PeCDF was based on the spatially resolved dioxin emission data in Toxic Equivalents (TEQs) of Pacyna (43) and the average congener contributions to the TEQ emission profile across Europe of Vulykh and Shatalov (44) (see Supporting Information for emission input profile and physical-chemical properties). The emissions of PeCDF are widely dispersed to air across Europe (43), and the intake associated with any one specific source is not considered in this evaluation. Monitored concentrations of PeCDF in environmental media (soil, air, vegetation, and sediments) and exposure substrates (meat, milk, agricultural produce, fish, etc.) were similarly estimated from TEQs using the data for average congener contributions (44-46). Intake Fraction. The intake fraction for Western European emissions of PeCDF is high, between 0.002 and 0.05 based on monitored concentrations (45, 46) and using food production statistics (37, 38). The average, close to 0.01, suggests the human population will take in about 10 g for every 1 kg of PeCDF emitted. Figure 3 illustrates the breakdown of these intake fraction estimates, which are generally corroborated by the two versions of the model. The intake fraction estimates using the spatial and the nonspatial versions of the model are approximately 0.03 and 0.01, respectively. These are within the range estimated using the monitoring data. The overall influence of spatial differentiation in the model for estimating the intake fraction for these disperse emissions of PeCDF is negligible. Intake associated with vegetables, fruits, cereals, meats, and milk all contribute significantly, as suggested using the monitoring data. 1124

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 39, NO. 4, 2005

Detailed insights are provided in the following sections and in the Supporting Information for discrepancies in the contaminant concentrations in the substrates and in the environmental compartments. Concentrations in Air, Soils, Sediments, and Vegetation. Figure 4 compares the model estimates and monitored concentrations for contaminants in cereals, fruits, and vegetables as well as in air, soils, and sediments. The concentrations of contaminants in the stem in the generic vegetation model could be assumed representative of those in unexposed produce, such as root crops and produce protected from ambient air by a protective, removable skin such as citrus fruits and cereals. The concentration in the foliage represents exposed produce such as vegetables and fruits that are not protected. Figure 4 suggests, however, that the foliage in generic vegetation models is also more representative for unexposed produce such as cereals. Concentrations in the foliage were therefore used to calculate the intake fractions for cereals in Figure 3. While not influencing the intake fraction for PeCDF, concentrations in the root soil in Figure 4 were underestimated by about 2 orders of magnitude compared to the monitored values. Two loss processes dominate the overall removal rate from soils in the model: volatilization and degradation contribute 71% and 25% to removal, respectively. The magnitude of such processes at the regional scale is difficult to estimate and could be underestimated. Additionally, the depth at which soil data are sampled may be important in such evaluations. The model predicted the concentration of PeCDF at the surface is about 1000 times higher than at a 30 cm depth in soils. Figure 4 presents the average value for this 30 cm depth, which may not reflect the basis of the monitoring data. However, the model did not account for biological or mechanical mixing, which can reduce the importance of such differences and the necessity of accounting for vertical profiles in soils. Exposure Concentrations in Fish, Meat, and Dairy Produce. Figure 5 presents a comparison of the model estimates with the concentrations of PeCDF monitored in the food substrates milk, meat, eggs, and fish. With the exception of sea fish, the magnitude and the variations of the monitored concentrations and the spatially resolved model estimates (vertical bars) are in reasonable agreement. The nonspatial model results are not always conservative but are similar in magnitude to the average monitoring data and to the median concentration estimates from the spatially resolved version of the model. Concentrations in sea fish are 2 to 3 orders of magnitude lower than in the monitoring data for fish. Factors that explain these discrepancies include that the origin of the fish in the monitoring data is not clear, and the source of the contaminants in these fish may not be associated with the modeled Western European emissions. The Supporting Information outlines other more general reasons for model versus monitoring data discrepancies. Analysis of Spatial versus Nonspatial Model Estimates. The need for spatial differentiation was analyzed for a broader set of organic chemicals by comparing the results of the spatial versus the nonspatially resolved versions of the model. The analysis first considers diffuse emissions, as in the evaluation for PeCDF, and then emissions at specific locations. For these analyses, a set of representative organic, nondissociating chemicals reflect plausible differences in partitioning behavior, dominant human exposure pathways, overall environmental persistence, and long-range transport characteristics. Degradation rate coefficients, Henry’s Law constants, and other chemical properties were taken from databases using hierarchical selection guidelines (see Supporting Information for a list of the chemicals, data sources, and data adopted).

FIGURE 3. Comparison of intake fraction and associated contributions per substrate for PeCDF using the two versions of the model (spatial and nonspatial) and values based on monitoring data (the sum of the monitored concentrations in food substrates multiplied by statistics for European production, divided by the total emission quantity). Variations in the fractions estimated using monitored data reflect the plausible ranges using the maximum and the minimum contaminant concentrations. Intake by inhalation and drinking water were negligible for PeCDF in the model, hence not shown. Chemical properties were assumed to reflect variations under average conditions for a broad range of chemicals. The same values were used in both versions of the model to ensure consistency in the analyses. Spatial and temporal variations of the properties were not taken into account. Disperse Emissions. Figure 6 demonstrates that estimates of the intake fraction with the nonspatial version of the model are typically lower than those of the spatial version for a uniformly distributed emission, similar to the results for PeCDF in the previous sections. The difference is less than a factor of 2 for such emissions to air, a factor of 10 for releases to soils, but as high as 3 orders of magnitude underestimation for some releases to surface waters. This discrepancy for emissions to surface waters is primarily associated with the single average residence time adopted for all surface waters in the nonspatial model. The other differences are explained in the sensitivity analyses in the Supporting Information. Location Specific Emissions. For a selected subset of four of the organic chemicals considered, Figure 7 illustrates the variation of intake fraction associated with emissions in specific locations using the spatially resolved model. The median of these location-specific estimates typically corresponds closely with both the estimate for a uniform emission scenario as well as the estimate from the nonspatial version of the model. Discrepancies with the nonspatial model estimates for location-specific emissions are again most significant for chemicals released to surface waters (e.g. propachlor and dicofol emission to water in Figure 7).

Discussion This paper presents a spatially resolved multimedia chemical fate and multipathway exposure model for Western Europe, available as IMPACT 2002. The model facilitates estimation of concentration profiles of dispersed contaminants and human intake at the population level. The results are presented in the form of intake fractions, the fraction of an emission that will be taken in by the entire population. Results were compared with those of a nonspatial clone of the model as well as with monitoring data for disperse atmospheric emissions of the dioxin congener, PeCDF (2,3,4,7,8-pentachlorodibenzofuran; CAS# 51207-31-4). Disperse Emissions. Accounting for likely uncertainties, for example of 1 to 2 orders of magnitude in the estimates, the intake fraction predictions from spatial and nonspatial versions of the presented model will not be statistically distinct for most disperse emission sources. For disperse atmospheric emissions of PeCDF, inclusion of spatial distinction in such models did not improve the reliability of the intake fraction estimate when compared against monitoring data. The analysis presented for a broader range of organic chemicals suggests this may be the case for all emissions to air and for releases/applications onto soils when considering disperse emission sources. The disperse nature of the sources in these scenarios essentially cancels out the importance of spatial variability when assessing human intake at the population level. Models without spatial resolution that VOL. 39, NO. 4, 2005 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

1125

FIGURE 6. Comparison of human intake fractions for uniformly distributed emissions (in proportion to surface area) in the spatially resolved model versus the nonspatial version for representative organic, nondissociating chemicals.

FIGURE 4. Comparison of model estimates of concentrations with monitoring data compiled by Buckley-Golder (45) and Quass (46) for PeCDF. Vertical bars represent the range of monitored values as well as the spatially resolved model estimates. Horizontal bars are the median monitored data, the mean estimates using the spatial model (total mass/total volume per media), and the estimates from the nonspatial version of the model.

FIGURE 7. Illustration of the variation in intake fraction as a function of the emission compartment in the spatially resolved version of the model compared to estimates for uniformly distributed emissions as well as results of the nonspatial version of the model. Emissions were modeled separately for each compartment. This subset of four chemicals and emission scenarios reflects the main trends: toluene emitted to air with principle human exposure via inhalation, dicofol to water with principle exposure via fish, propachlor to water with principle exposure via drinking water, and DDT to air with principle exposure via crops. FIGURE 5. Comparison of model results with monitoring data of chemical concentration ranges in meat and dairy produce for PeCDF (45, 46). Horizontal bars are the median values for monitoring data and the average concentrations weighted by production quantity for the estimates of the spatially resolved model (total EU intake/EU production rate associated with each food type).

provide average insights are therefore likely to provide a suitable level of complexity for estimating intake fractions for most disperse emission sources. The presented nonspatial model was not suitable for assessing disperse emissions of some chemicals to water. This model resulted in underestimations of the intake fraction by up to 3 orders of magnitude (Figure 6). This was primarily related to the large differences in residence time between European water bodies. The surface water residence time in 1126

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 39, NO. 4, 2005

the nonspatial model was the sum of the total Western European water volumes divided by the total water flow rate out to sea or out of the modeled region. This residence time does not adequately account for the large variations across Western Europe. Some watersheds feed into large lakes with residence times of about 10 years, while others are associated with rivers that have residence times such as 48 h prior to discharge into seawater (see Supporting Information for lake details). As straightforward nonspatial multimedia models will remain essential for screening chemicals in terms of both individual risk as well as risk at the population level for disperse emission sources, it is recommended to distinguish between the fraction of emissions to lakes with large residence times versus to rivers with shorter residence times. This will eliminate the large underestimations for disperse emissions to surface waters of some chemicals.

For the assessment of many organic chemical emissions, including atmospheric emissions of persistent substances such as DDT or PeCDF, multimedia/multipathway models are essential as the food pathway dominates human exposure. Nevertheless, the evaluation for PeCDF suggests estimation of the concentrations of contaminants in agricultural produce remains problematic using generic vegetation models. Estimates of concentrations in agricultural produce are essential for estimating human intake and this will therefore be a significant source of uncertainty in many human health assessments. Further development, adoption, and evaluation of models for agricultural produce is necessary in multimedia/ multipathway assessments. Location Specific Emissions. In many applications of growing interest in policy support and decision-making, models are also needed for estimating the intake at a population level for emissions at specific locations. In general, the intake fraction estimates from the nonspatial model corresponded to the median of the location-specific estimates using the spatially resolved version. However, while reflecting the median, the nonspatial model does not provide insights into the likely variations. The variation can be at least 5 orders of magnitude for some chemicals depending on the emission location (Figure 7). The intake fraction may therefore be under, or over, estimated by 2 to 3 orders of magnitude using a nonspatial model for certain emissions, and this may impede decision-making in some cases. Variation in the intake fraction associated with emission location is particularly important for chemicals that are not subject to significant disperse transport at a regional scale. For example, variation in intake fraction will be high for chemicals that are unlikely to volatilize significantly if released to surface water. This is the case for chemicals with a low Henry’s Law constant (typically H 100 Pa m3/mol), intake was predominantly associated with inhalation. The intake fraction estimates for these chemicals varied in the model as a function of release location by up to a factor of 100 (maximum/ minimum). Differences in air concentrations due to advective flow patterns and population density were primarily responsible for these variations. Intake fractions were predominantly associated with contaminants in agricultural vegetation for emissions to air and to soil for most chemicals with a low Henry’s law constant, such as DDT in Figure 7 (especially if H < 1 Pa m3/mol). The intake fraction for these chemicals varied by a factor up to 250 (maximum/minimum). This variation was mainly attributable to spatial differences in food production quantities. Intake associated with contaminants in livestock was important for such chemicals if bioaccumulation was significant. Depending on the magnitude of other uncertainties and what is acceptable when making a decision, the use of a nonspatial model can be unacceptable for estimating the intake fraction for an emission in a specific location. From the presented insights, this can result in at least 2 to 3 orders of magnitude uncertainty for some chemical emissions. Insights using models with finer resolutions and taking into account other parameters that were not spatially varied in this research, including chemical and soil properties, may result in higher variations for emissions at specific locations. These variations may be further compounded by temporal variations. Caution is particularly recommended for emis-

sions that are not expected to disperse significantly at a regional scale, where local variations and related modeling can be more important for assessing individual as well as population-level intake.

Acknowledgments We thank Yasunari Matsuno, Norihiro Itsubo, Kikuo Yoshida, Haruyuki Higashino, Koh Harada, and Hiroaki Kondo of NIRE (now AIST) for inputs to the previous prototypes for the Kanto region, Japan; Matt MacLeod (U. of Trent, Canada) for general advice; Till Bachmann (U. of Stuttgart) as well as Thierry Pelichet, Isabelle Bey, and Bryan Duncan (EPFL) for air modeling support and data; Myriam Saade (EPFL) for exposure modeling data; and Raphael Charles (EPFL and RAC) and Christophe Bonnet (EPFL) for vegetation modeling inputs.

Supporting Information Available Further details related to property data, environmental parameters, the atmospheric delimitation choice, sensitivity analyses for identifying key parameters, and mass balance calculations using a matrix approach as well as the rate coefficients for fate and for human intake. This material is available free of charge via the Internet at http://pubs.acs.org.

Literature Cited (1) The multi-media fate model. A vital tool for predicting the fate of chemicals; Cowan, C. E., Mackay, D., Feijtel, T. C. J., van de Meent, D., Di Guardo, A., Davies, J., Mackay, N., Eds.; SETAC Press: Pensacola, Florida, U.S.A., 1994. (2) European Commission EUSES, the European Union System for the Evaluation of Substances, National Institute of Public Health and the Environment (RIVM), 1996. (3) European Council Directive 97/11/EC amending Directive 85/337/EEC on the assessment of the effects of certain public and private projects on the environment, 1997. (4) European Council Directive 2001/42/EC on the assessment of the effects of certain plans and programmes on the environment, 2001. (5) Udo de Haes, H.; Jolliet, O.; Finnveden, G.; Goedkoop, M.; Hauschild, M.; Hertwich, E.; Hofstetter, P.; Klo¨pffer, W.; Krewitt, W.; Lindeijer, E.; Mueller-Wenk, R.; Olson, S.; Pennington, D.; Potting, J.; Steen, B. Striving towards best available practice in life cycle impact assessment; SETAC Press: Pensacola, Florida, U.S.A., 2002. (6) Pennington, D. W.; Potting, J.; Finnveden, G.; Lindeijer, E.; Jolliet, O.; Rydberg, T.; Rebitzer, G. Life cycle assessment Part 2: Current impact assessment practice. Environ. Intl. 2004, 30, 721-739. (7) Bennett, D. H.; Margni, M.; McKone, T. E.; Jolliet, O. Intake fraction for multimedia pollutants: A tool for life cycle analysis and comparative risk assessment. Risk Analysis 2002, 22, 903916. (8) Klepper, O.; den Hollander, H. A. A comparison of spatially explicit and box models for the fate of chemicals in water, air and soil in Europe. Ecological Modeling 1999, 116, 183-202. (9) USEPA Total Risk Integrated Methodology. Status Report, United States Environmental Protection Agency, Office of Air Quality, Planning and Standards, 1999. (10) Wania, F.; Persson, J.; Di Guardo, A.; McLachlan, M. S. The POPCYCLING-Baltic Model. A non-steady-state multicompartment mass balance model of the fate of persistent organic pollutants in the Baltic Sea environment, NILU Norwegian Institute for Air Research, 2000. (11) MacLeod, M.; Woodfine, D. G.; Mackay, D.; McKone, T. E.; Bennett, D. H.; Maddalena, R. BETR North America: A regionally segmented multimedia contaminant fate model for North America. Environ. Sci. Pollut. 2001, 8, 156-163. (12) Woodfine, D.; MacLeod, M.; Mackay, D.; Brimacombe, J. R. Development of continental scale multimedia contaminant fate models: Integrating GIS. Environ. Sci. Pollut. 2001, 8, 164-172. (13) Prevedouros, K.; MacLeod, M.; Jones, K. C.; Sweetman, A. J. Modelling the fate of persistent organic pollutants in Europe: Parametrization of a grided distribution model. Environ. Pollut. 2004, 128, 251-261. (14) Suzuki, N.; Murasawa, K.; Sakurai, T.; Nansai, K.; Maysuhashi, K.; Moriguchi, Y.; Tanabe, K.; Nakasugi, O.; Morita, M. Georeferenced multimedia environmental fate model (G-CIEMS): VOL. 39, NO. 4, 2005 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

1127

(15) (16)

(17)

(18) (19) (20) (21) (22) (23) (24) (25)

(26) (27) (28) (29) (30) (31) (32)

(33)

Model formulation and comparison to the generic model and monitoring approaches. Environ. Sci. Technol. 2005, in press. Mackay, D. Multimedia environmental models, The fugacity approach; 2nd ed.; Lewis Publishers: Boca Raton, 2001. Mackay, D.; Di Guardo, A.; Patterson, S.; Kicsi, G.; Cowan, C. E. Evaluating the environmental fate of a variety of types of chemicals using the EQC model. Environ. Toxicol. Chem. 1996, 15, 1627-1637. Finizio, A.; Mackay, D.; Bidleman, T.; Harner, T. Octanol-air partition coefficient as a predictor of partitioning of semi-volatile organic chemicals to aerosols. Atmos. Environ. 1997, 31, 22892296. Jolliet, O.; Hauschild, M. Modelling the influence of intermittent rain events on fate and transport of organic air pollutants. Environ. Sci. Technol. 2004, resubmitted. Schwarzenbach, R. P.; Gschwend, P. M.; Imboden, D. M. Environmental Organic Chemistry; Wiley: New York, 1993. McKone, T. E.; Bennett, D. H. Chemical-specific representation of air-soil exchange and soil penetration in regional multimedia models. Environ. Sci. Technol. 2003, 37, 3123-3132. Cousins, I. T.; Mackay, D. Strategies for including vegetation compartments in multimedia models. Chemosphere 2001, 44, 643-654. Charles, R. Modelling pesticides residues, Doctoral Thesis, Swiss Federal Institute of Technology, Lausanne (EPFL), 2004. McKone, T.; Bennett, D.; Maddalana, R. CalTOX 4.0 Technical Support Document, Vol. 1, Lawrence Berkeley National Laboratory, 2001. Ammann, C. Development of a spatial multimedia model of chemical fate and human exposure for Europe, Diplome Thesis, Swiss Federal Institute of Technology, Lausanne (EPFL), 2002. Bey, I.; Jacob, D. J.; Yantosca, R. M.; Logan, J. A.; Field, B.; Fiore, A. M.; Li, Q.; Liu, H.; Mickley, L. J.; Schultz, M. Global modeling of tropospheric chemistry with assimilated meteorology: Model description and evaluation. J. Geophys. Res. 2001, 106, 2307323096. European Environment Agency ERICA (European Rivers and Catchments), 1998. U.S. Geological Survey HYDRO1k, 2002 http://edcdaac.usgs.gov/ gtopo30/hydro/europe.html 2002. Global Runoff Data Centre World Runoff Data, 2002 www. grdc.sr.unh.edu/2002. Hydrologic Science Branch of NASA http://hsb.gsfc.nasa.gov/ data_distribution.html 2001. International Lake Environment Committee Survey of the State of World Lakes, 2002 http://www.ilec.or.jp/2002. Bundesamt fu ¨ r Wasser und Geologie Naturseen der Schweitz (Natural lakes of Switzerland), 1983. European Topic Centre on Terrestrial Environment; European Environment Agency Coordination of Information on the Environment (CORINE), 250 m grid, 2000 http:// terrestrial.eionet.eu.int/. Simpson, D.; Fagerli, H.; Jonson, J. E.; Tsyro, S.; Wind, P.; Tuovinen, J. P. Transboundary acidification, eutrophication,

1128

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 39, NO. 4, 2005

(34)

(35)

(36) (37)

(38) (39) (40)

(41)

(42)

(43)

(44)

(45)

(46)

and ground level ozone in Europe. Part 1. Unified EMEP model description, Norwegian Metrological Institute, 2003. Pelichet, T. Development of a spatial multimedia chemical fate and human exposure model for Europe, with a focus on air advection, Diplome Thesis, Me´moire No. 82 Universite´ de Gene`ve, Lausanne, 2003. Meylan, W. M.; Howard, P. H.; Boethling, R. S.; Aronson, D.; Printup, H.; Gouchie, S. Improved Method for Estimating Bioconcentration/Bioaccumulation factor from Octanol/Water partition coefficient. Environ. Toxicol. Chem. 1999, 18, 664672. UNEP (United Nations Environment Programme) Grid Centre 1990 http://www.grid.unep.ch/data/grid/gnv198.php. EUROSTAT Agriculture in the European Union, Statistical and Economic Information 2001, 2002 http://europa.eu.int/comm/ agriculture/2002. Food and Agriculture Organization of the United Nations FAO Statistical Databases, 2002 http://www.fao.org 2002. EUROSTAT Fisheries Yearbook, European Commission, 2001. Scheidleder, A.; Grath, J.; Winkler, G.; Staerk, U.; Koereimann, C.; Gmeiner, C.; Gravesen, P.; Nixon, S.; Casillas, J.; Lack T. J.; Thyssen, N. Groundwater quality and quantity in Europe: Data and basic information, European Environment Agency, 1999. European Commission Technical guidance document in support of Commission Directive 93/67/EEC on risk assessment for new notified substances and Commission Regulation (EC) No 1488/ 94 on risk assessment for existing substances, European Commission, 1996. Margni, M.; Pennington, D. W.; Amman, C.; Jolliet, O. Evaluating multimedia/multipathway model intake fraction estimates using POP emission and monitoring data. Environ. Pollut. 2003, 28, 263-277. Pacyna, J. M. Appendix 1 to executive final summary report. Environmental cycling of selected persistent organic pollutants (POPs) in the Baltic region (Popcycling-Baltic project). 1990 http://www.msceast.org/pops/emission.html. Vulykh, N.; Shatalov, V. Investigation of dioxin/furan composition in emissions and in environmental media. Selection of congeners for modelling, Meteorological Synthesizing Centre, 2001. Buckley-Golder, D.; King, K.; Brown, K. Compilation of EU dioxin exposure and health data - Summary report, Produced for DG ENV by UK Department of the Environment, Transport and the Regions (DETR), 1999. Quass, U.; Fermann, M.; Gro¨ker, G. The European dioxin emission inventory, Stage II. Volume 1, Produced for DG ENV by the North Rhine Westphalia State Environment Agency, 2000.

Received for review June 13, 2003. Revised manuscript received November 3, 2004. Accepted November 23, 2004. ES034598X