III.5.2.Sensitivity Analysis

Sensitivity analysis was undertaken in order to evaluate the impact to exposure media concentration estimations with changes in fate and transport/transfer model parameters. Figure III-5 shows an example of sensitivity analysis conducted.

This figure describes the impact of key factors for the stack emission source category for determining biota impacts. The x-axis contains the names of the parameters evaluated. The key below the figure gives the definition of the parameters and the values selected for the demonstration scenarios.

The y-axis shows the numerical change to the key model result, in this case, vegetable and beef concentrations, to the changes made in the parameter. These changes are noted above and below the bars. For example, vegetable concentration is about 3 times higher at 200 ft from the stack emission source than it is at 500 meters from the source, the distance used in the demonstration scenario.

Some of the observations made for this test, typical for the type of observations which were made for sensitivity testing, include:

1) Mixing depth, described by the parameter dnot, has very little impact on final beef concentrations.
2) Nearer to and further from the stack had different impacts for above and below vegetable concentrations as compared to beef concentrations. The farm was assumed to ...

table Figure III-6 Results of sensitivity analysis of algorithms estimating above and below ground vegetation, and beef fat concentrations resulting from stack emissions.

... be 500 meters from the stack. Nearer to the stack at 200 meters, ambient air concentrations and dry deposition amounts were lower, but wet deposition was at its maximum. One effect of this was that vegetable concentrations increased. Below ground vegetables increased by about a factor of 4, due to the same increase in soil concentration as a result of much higher wet deposition.

Above ground vegetation increased by about 50%. Particle depositions dominated above ground vegetable/fruit concentrations. Therefore, an increase in overall particle depositions due to an increase in wet depositions led to increased above ground vegetable/fruit concentrations.

However, the trend was not the same for beef and milk fat. ...

expand table Figure VX X-X

... The reason for this was that grass and other cattle feeds were dominated by vapor contributions, not particle depositions, as were above ground vegetables. For bulky above ground vegetables, vapor phase impacts were empirically reduced considering the difference in bulk for these vegetables compared to the leafy grass and azalea leaf for which the air-to-leaf vapor transfer factor was developed. Therefore, a drop in ambient air vapor phase concentrations at 200 meters as compared to 500 meters dominated the result, and the net impact was to reduce beef fat concentrations. Further from the stack at 5000 meters, all biota concentrations were lower. Vapor phase air concentrations were roughly halved, and dry and wet deposition were lower by 60 and 80% respectively. This led to substantial reductions in vegetable concentrations.

Interestingly, beef concentrations were lower at 5000 meters than at 200 meters, but not by much. This is because vapor phase concentrations at 5000 meters were, in fact, greater than they were at 200 m. The net results, according to the modeled depositions and air concentrations, is that beef and milk fat impacts are ironically fairly similar at 200 and 5000 meters.
3) Changing the vapor/particle partitioning assumption also had inverse effects for above and below ground vegetables as compared to beef. The baseline vapor/particle partitioning for 2,3,7,8-TCDD was 55% vapor/45% particle.

When decreasing the vapor to 10% and increasing the particle to 90%, both vegetations increased. Below ground vegetables increased because below ground vegetables were not a function of vapor phase concentrations, only of soil concentrations, which were a function of particle depositions.

Above ground vegetable concentrations increased as well, as they are dominated by particle depositions. As noted above, however, cattle vegetations are driven by vapor transfers. Therefore, increasing the vapor portion tended to increase these vegetations and hence beef concentrations.

Following are key overall observations from the sensitivity analysis:

1) Source terms are the most critical for exposure media impacts.
Source terms include soil concentrations, stack emission rates, and effluent discharge rates. In all cases, the impact to exposure media is linear with changes to source terms. Proximity to the source term can be important as well, as demonstrated with differences in distance from the stack emission source.

2) Chemical-specific parameters, particularly the bioconcentration/biotransfer parameters, are the second most critical model inputs.
Some of these have lesser impacts within the range tested, such as the organic carbon partition coefficient, Koc, for surface water impacts. Generally, at least an order of magnitude in range in possible media concentrations is noted with the range of chemical-specific parameter ranges tested.

The impact of changes to bioconcentration/biotransfer parameters is mostly linear. This is because these transfer factors estimate media concentrations as a linear transfer from one media to another. For example, fish lipid concentrations are a linear function of the organic carbon normalized concentration of contaminants in sediments.

These transfer parameters are also identified as uncertain parameters. Tested ranges sometimes spanned over an order of magnitude for 2,3,7,8-TCDD.

3) All other parameters had less of an impact as compared to source strength and chemical specific parameters; nearly all impacts were within an order of magnitude for the range of tested values.
Part of the reason for this trend is that there is a reasonably narrow range for many of the non-chemical specific or source term parameters - soil properties, wind speeds, vegetation yields, and others.

4) The sensitivity analysis exercises unearthed a dichotomy in model performance between the soil source category and the stack emission source category.
The on-site soil source category was demonstrated with a 1 ppt soil concentration of 2,3,7,8-TCDD, a concentration similar to measured concentrations of 2,3,7,8-TCDD in rural settings.

Air concentrations are esimtated to be 4*10-5 pg/m3 (vapor+particle phases summed). Atmospheric transport modeling in the demonstation of the stack emission source category resulted in an exposure site air concentration (vapor+particle phases summed also) at 500 meters from the stack to be 1*10-5 pg/m3.

With similar air concentrations predicted to occur at the exposure site for the demonstration of the soil and stack emission categories, one might hypothesize that all subsequent impacts would be similar.

That was not the case. The stack emission source algorithms deposited particulates onto soil to estimate a soil concentration that was in the 10-3 ppt range for the 1-cm untilled depth and the 10-5 range for the 20-cm tilled depth.
This compares to the 1 ppt concentration for the on-site soil source category demonstration. With similar air concentrations but a 3+ order of magnitude difference in soil concentrations in the demonstration of the soil and the stack emission sources, the following trends were noted:

• Below ground vegetables had much higher concentrations for the soil source demonstration scenario.
• Soil-related exposures (dermal contact and soil ingestion) were much higher for the soil source demonstration scenario.
• Soil was significantly more critical in predicting beef and milk fat concentrations in the soil source category. The following shows the relative impact of soil versus vegetations (grass and cattle feed) for the on-site soil demonstration and the stack emission demonstration:

Diagram III-01

Subsequently, beef and milk concentrations were almost two orders of magnitude higher for the soil source category as compared to the stack emission source category.

• Because above ground vegetations are driven by air concentrations, above ground vegetables/fruit and grass/cattle feed concentrations were similar for both demonstrations.

Further examination of the results and other model testing did suggest that the air-to-soil algorithm may be underestimating soil concentrations, and the soil-to-air algorithms may be underestimating air concentrations. If both these observations are correct, and model or parameter adjustments corrected these underestimations, then model performance would be more similar for the two source categories.

The evidence for the air-to-soil underestimation came in an air-to-beef food chain model exercise (see Table III-5 below on model testing for a summary of this test). An air-borne reservoir of dioxin-like compounds was crafted to be typical of rural environments. Depositing this reservoir onto soil resulted in a predicted concentration about an order of magnitude lower than observed concentrations in rural settings. Speculated causes include:

1) the 10-year half-life for dioxin-like compounds may not be long enough,
2) vapor-phase transfers to soils were not modeled, and
3) detritus input to soils was not considered. Empirical evidence for the possible underestimation of air concentrations over soils came in two forms.

One, plant:soil ratios modeled in the soil source demonstration scenario appeared lower than experimentally determined plant:soil ratios by about an order of magnitude. This could be due to an underestimation of air concentrations.

Two, air concentrations of 2,3,7,8-TCDD predicted to occur over a 1 ppt soil concentration was lower than by an order of magnitude for concentrations found in a "remote" area of Sweden, and about two orders of magnitude lower than crafted to be typical of rural setting in the United States.

While the soil-to-air algorithm may be underestimating air concentrations, it is also possible that they are not underestimating these concentrations. The expectation that releases of dioxins from soils in background settings should result in air concentrations typical of background settings may not be a realistic expectation.The argument was developed in Section II.3.3. Conclusions for Mechanisms of Impact to Food Chain earlier in this Executive Summary that the food chain is impacted via atmospheric depositions, and that industrial emissions followed by long range transport ultimately explain media concentrations in background settings.

What is not known is, what portion of air concentrations in rural settings can be attributed to long range transport and what portion attributed to suspension of reservoir sources (soil and other reservoir sources).

What is really needed to test the soil to air algorithm are measured concentrations over soils not known to be otherwise impacted by dioxin like compounds. Such information could not be found in the literature.

III.5.3.Mass Balance Considerations for Soil Contamination

The purpose of this exercise is to evaluate whether a principal of mass balance will be violated with the models and parameters used for the demonstration of the off-site soil source category - that principal being that dioxin releases from a site cannot exceed the original amount at the site (assuming no replenishment).
A simplifying assumption for the off-site soil source category was that the soil concentration remained constant over the period of exposure - there was not a systematic depletion of the reservoir over time due to modeled dissipation processes.

First, an estimate of the "reservoir" of 2,3,7,8-TCDD that is implied with the demonstration parameters was made. Then, an estimate of the rate at which this reservoir dissipated using the solution algorithms for dissipation: volatilization and wind erosion flux from soils, and soil erosion, was made.

Other routes of dissipation that were examined are the soil ingestion by cattle and children, losses in runoff and leaching, the loss via dermal contact, and the removal via harvest of below ground vegetation. These were shown to be minuscule in comparison to air and soil erosion. The premise examined was that, if it takes substantially more time than the exposure period to dissipate the reservoir, then it may be fair to conclude that the assumption of a constant soil concentration may be suitable for purposes of exposure assessments.

On the other hand, complete dissipation within a time period less than or even near to the period of exposure would mean that exposures and risks are being overestimated. This analysis led to a conclusion that the reservoir modeled in the exercise above would take more than 90 years to dissipate.
This was not a definitive exercise, by any means, but it does lend some confidence that a principal of mass balance may not have been violated for the soil source categories, and for the assumption of 20 years exposure duration.


Some discussion of the issues commonly lumped into the term "uncertainty" is needed at the outset. The following questions capture the range of issues typically involved in uncertainty evaluations:

1. How certain are site specific exposure predictions that can be made with the methods?
2. How variable are the levels of exposure among different members of an exposed local population?
3. How variable are exposures associated with different sources of contamination?

The emphasis in Volume III is in providing the technical tools needed to perform site-specific exposure assessments.

For the assessor focusing on a particular site, question (1) will be of preeminent importance. Therefore the emphasis of the uncertainty evaluation is to elucidate those uncertainties inherent to the exposure assessment tools presented. This chapter examines the capabilities and uncertainties associated with estimating exposure media concentrations of the dioxin-like compounds using the fate, transport, and transfer algorithms, and also identifies and discusses uncertain parameters associated with with human exposure patterns (contact rates and fractions, exposure durations, etc.).

A site specific assessment will also need to address the variability of risks among different members of the exposed population, the second key question above. The level of detail with which this can be done depends on the assessors knowledge about the actual or likely activities of the exposed population. In this document, one approach to evaluating this variability is demonstrated. Separate "central" and "high end" scenario calculations are presented to reflect different patterns of human activities within a hypothetical rural population.

A key issue with regard to intra-population variability is that it is best (if not only) addressed within the context of a specifically identified population. If such information is available, a powerful tool that can be used to evaluate the variability within a population is Monte Carlo Analysis. Three recent Monte Carlo studies which have been done for exposure to 2,3,7,8-TCDD were reviewed. Assumptions on distributions of exposure patterns and fate and transport parameter distributions are described, as are the results of their analyses. Monte Carlo procedures require distributions for the input parameters used in the assessment. Such distributions have not been established by the Agency.

Decisions on the use and definition of such distributions affect assessments of all chemicals and cut across all Agency programs. Thus, it is not appropriate to establish such polices in this document. The Agency does have efforts underway to evaluate these generic issues. For example, the Office of Health and Environmental Assessment (OHEA) is in the process of revising the Exposure Factors Handbook and held public review meetings in 1993. In addition, OHEA is developing a guidance document on generating exposure scenarios. Several offices have projects specific to Monte Carlo:

• Office of Health and Environmental Assessment - A Workshop on approaches to evaluating uncertainty (including the use of Monte Carlo) was held in 1992.
• Office of Policy, Planning and Evaluation - A workshop on using Monte Carlo methods was held in 1993.
• Office of Pollution Prevention and Toxics - A handbook on the use of Monte Carlo is being developed for publication at a later date.

With regard to question (3), this document does not present a detailed evaluation of how exposure levels will vary between different sources of release of dioxin-like compounds into the environment. While Volume III does demonstrate the methodologies developed for sources of release of dioxin-like compounds into the environment with source strengths and environments crafted to be plausible and meaningful, there is still a great deal of variability on both the source strengths and on the environments into which the releases occur.

For example, the frequency with which farms and rural residences are near stack emissions of dioxin-like compounds is not addressed. Comprehensive comparisons and rankings of different sources and exposure patterns are generally not available, although pieces of the puzzle are beginning to come together. Volume II of this assessment does estimate national releases of dioxin-like compounds from several sources.

References to EPA and other assessments on dioxin-like compounds have been made throughout Volumes II and III of this assessment, such as those related to soil exposures (Paustenbach, et al., 1992), exposures to contaminated fish (EPA, 1991), and exposures resulting from land disposal of sludges from pulp and paper mills (EPA, 1990b).

There was a concerted effort to evaluate the capabilities of the fate, transport, and transfer algorithms by comparing key outputs from these models - predictions of concentrations and ratios of media to media concentrations - with literature reports. A summary of key comparative tests is given in Table III-5.