Page 3 of 6

6.3.3. Results

The results of the sensitivity analysis are principally described in a series of 14 bar graphs. The Y-axis is on a log scale and shows changes in media concentration estimation when the high and low parameter substitutions are made. The Y=1 line is the value of the media concentration with all baseline parameter selections; the precise value of that media concentration is noted on each graph. Other y-axis values are arrived at as the ratio of the pertinent media concentration estimated with the altered parameter over the baseline concentration; a y-axis value of 0.1, for example, means that the concentration with the parameter substitution was one-tenth the concentration under baseline conditions. Also noted on each graph is the pertinent source strength term - for air concentration sensitivities, soil concentrations are noted, and so on. The parameters tested are named on the x-axis, and these names correspond to the names in Table 6.1. The definition and baseline value of these key parameters are noted below each graph. The high and low values tested are appropriately placed either above (when the concentration increases with the parameter change) or below the bar graphs. These parameters are the only ones which impact the tested media concentration. Of course, the soil concentration also impacts the media concentration, but as noted in the previous section, soil concentrations have a direct and linear impact in all cases, and so are not displayed on the figures. Observations from each figure now follow. Estimation of off-site air concentrations in the vapor phase

Results for this test are shown in Figure 6-1. No single change resulted in estimations over an order of magnitude different from that made with baseline parameters. The model is insensitive to porosity and particle bulk density parameters, Eslp and Psoil. The results are also reasonably insensitive to ranges for organic carbon content of soil, OCsl, and windspeed, Um. For all other parameters, there appears to be roughly an order of magnitude spread over the range of parameters tested. Increasing the exposure duration to 70 years would decrease prediction by roughly one-half and decreasing the duration to 1 year would roughly double concentrations. As discussed earlier in Section 6.2., the volatilization algorithm assumes that contamination begins at the soil surface at time zero, and residues available for volatilization originate from deeper in the profile over time. The result of this assumption is that the flux decreases as time increases. This is the only algorithm of this assessment where an assumption of a decreasing source strength over time is made. The impact does not appear too critical, however, as estimated air concentrations only increase by a factor of 3 when assuming a 1 year duration or decrease by 45% when assuming 70 years duration. Estimation of off-site air concentrations in the particulate phase

Results from this test are shown in Figure 6-2. The y-axis in this test spans two orders of magnitude since changes in the parameters describing the inherent wind erodibility of the soil, Ut and F(x), results in over an order of magnitude higher and lower than concentration estimations as compared to estimations using the selected values of Ut and F(x). The assumption of bare soil conditions at the site of contamination led to a value of 0.0 for V, the vegetative cover parameter. If the contaminated site had a reasonably dense vegetative cover leading to a V of 0.9, air concentrations at the nearby site of exposure would be about an order of magnitude less. The impact of area (ASC), distance (DLe), and frequency (FREQ) on exposure site concentrations mirror those for vapor-phase air concentrations. That is because these three are used in the same far-field dispersion algorithm. Another parameter used for the far-field dispersion algorithm is windspeed, Um.

table Figure 6-1 Results of sensitivity analysis of algorithms estimating exposure site vapor phase air concentrations resulting from off-site soil contamination. table Figure 6-2 Results of sensitivity analysis of algorithms estimating exposure site particle phase air concentrations resulting from off-site soil contamination.
expand table Figure V3 6-1 expand table Figure V3 6-2

However, interestingly, the impact of that parameter is reversed between the vapor and particulate phase algorithms. For the particulate phase, the windspeed has more of an impact in increasing wind erosion and hence the reservoir of airborne contaminant - increasing windspeed increases air concentrations. For the vapor phase, windspeed does not playa role in estimating volatilization flux, but only a role in the far-field dispersion model. In that role, increasing wind speed increases dispersion and decreases concentrations. Noteworthy for the particle phase algorithm is that estimated concentrations are independent of any chemical-specific parameters; wind erosion suspending the particles is only a function of climate, ground cover, and soil erodibility. Also noteworthy is that the baseline air concentration of contaminants on particles is over an order of magnitude lower than the baseline air concentration of contaminants in the vapor phase. Besides having implications for particle phase and vapor phase inhalation exposures, this difference also has implications for impacts to vegetation concentrations and subsequently to beef and milk concentrations. Estimation of soil erosion impacts to nearby sites of exposure

Results from this test are shown in Figure 6-3. This model shows little sensitivity to two parameters, the bulk density of soil at the site of exposure, Bsoil, and the amount of "clean" soil (that which is between the contaminated and exposure site) which erodes onto the exposure site, SLec, along with the contaminated soil. These will not be discussed further. In contrast to SLec, the model has a direct linear impact with the amount of soil eroding from the contaminated site, SLs. Decreasing that amount by a factor of 10 decreases exposure site soil concentrations by the same amount, and doubling contaminated site erosion also doubles exposure site soil concentrations.

The model appears to show insensitivity to the distance between the exposure and contaminated site, DLe. However, this result should be viewed cautiously. The sediment delivery ratio equation, described in Chapter 4, Section 4.3.1., was developed to estimate sediment loads from construction sites to nearby surface water bodies, and from distances up to 250 m. Its application to distances beyond that are questionable, and applications from one land area to another land area rather than from one land area to surface water, should also be questioned. At the model baseline distance of 150 m, the SDs (sediment delivery ratio) is 0.26. At 1000 m, it is 0.17, which is a marginal dropoff for what appears to be a significant increase in distance. The distance becomes increasingly important when there are obstructions between the contaminated and the exposure site such as ditches, roads, and so on.

When using this methodology, one should consider not relying on the sediment delivery ratio equation for:
1) transport of soils beyond 250 meters,
2) when the exposure site is upgradient from the site of contamination (in its development for construction sites, the assumption that a water body is downgradient from the site seems reasonable), and
3) when there are obvious land features which would retard erosion.

At the model's baseline value of 150 meters to a nearby site of exposure and to a nearby water body, use of the sediment delivery equation is expected to yield reasonable results.

table Figure 6-3 Results of sensitivity analysis of algorithms estimating exposure site soil concentrations resulting from erosion from off-site soil contamination.
One noteworthy trend from Figure 6-3 is that exposure site soil concentrations are not a direct linear function of areas of the exposure and contaminated sites, which they shouldn't be. It is assumed in the erosion algorithm that contaminated soil delivered to the exposure site mixes evenly throughout the exposure site to a depth defined by the tillage depth.

The algorithm corrected the amount of contaminated soil delivered to an exposure site when the exposure site was smaller than the contaminated site; it reduced the amount of contaminated soil delivered according to an appropriate size ratio. The baseline scenario had the contaminated and exposure site the same size at 40,000 m2.
expand table Figure V3 6-3

For this situation, all contaminated eroded soil from the site reaches and mixes with soil at the exposure site. When the contaminated site increased an order of magnitude to 400,000 m2, the soil concentration at the exposure site only doubled; it did not increase by an order of magnitude. It is unreasonable to assume that all the eroded soil would crowd into the smaller exposure site. When the contaminated site decreased an order of magnitude to 4,000 m2, the exposure site soil concentration likewise decreased by an order of magnitude. In this case, like the case when the contaminated and exposure site were of the same size, all the contaminated soil eroding in the direction of the exposure site mixes into exposure site soil, so the resulting average soil concentration at the exposure site is linearly related to the concentration at the contaminated site. A similar trend is noted with changes in the exposure site area term.

The impact to changes in depth of tillage is nearly, but not quite, linear. Decreasing the no till depth of mixing, dnot, from 0.05 m to 0.01 m increased soil concentrations by a factor of 3.5 roughly, while increasing dnot to 0.10 decreased concentrations by 45%. A similar, nearly linear, impact is noted with the changes tested for tillage depth, dt. For figure clarity, these results were left off Figure 6-3, but decreasing the depth from an initial 0.20 m to 0.10 m increased concentrations by just under a factor 2, and decreasing it to 0.30 m decreased concentrations by just under 33%.

The model has a linear impact with the enrichment ratio, ER. Its baseline value is 3.0; decreasing it to 1.0 decreases exposure site soil concentrations by 67%, and so on. The most interesting result from this sensitivity analysis exercise, however, was the test with the dissipation rate constant for eroding contaminants, k. This parameter was introduced in the equation since it was hypothesized that dioxin-like contaminants at and near the surface might be subject to dissipation processes such as photolysis or volatilization. Dioxin-like contaminants would reside near the soil surface for eroding or depositing (the stack emission source category) contaminants. This contrasts the assumption that soil concentrations at a site of soil contamination are steady over time - no dissipation or degradation constant is applied. The baseline value for k is 0.0693 yr-1, which is equivalent to a ten-year half-life. With that and all other baseline parameters, exposure site non-tilled soil concentrations are 0.279 ppb, or 28% of the contaminated site concentration of 1.00 ppb. When the dissipation half-life is increased to 100 years, essentially no dissipation, the exposure site soil concentration increases to over 1.00 ppb, or over that of the contaminated site. This is an improbable, if not impossible, result. Of all the parameters which have an incorrect value to have led to this result, the most likely one is the enrichment ratio. This multiplier increases the concentration on eroded soil from that which was on the site. It has a direct linear impact on exposure site soils - increasing it from its baseline of 3.0 to 5.0 increased exposure site soils by 67%. It is reasonable to assume that soil which erodes from a site is finer and more rich in organic matter in comparison to in-situ soils on the average, and that the eroded soils are "enriched" in comparison to in-situ soils for organic contaminants whose binding to soils is a function of organic carbon content of soils. This concept of enrichment has been used and demonstrated with field data, although not with field data of dioxin-like compounds. This exercise may have demonstrated that the assumed enrichment ratio of 3.00 is too high. Further, it may be reasonable to conclude that since dioxin-like compounds are so tightly sorbed to soils in general that the enrichment effect is not as pronounced as a 3.00 enrichment ratio implies.

A final issue evaluated for the soil erosion algorithm is the assumption of a steady state. As discussed in Section 5.4.1, Chapter 5, the solution did have a time term, which at time t = 0, erosion from a contaminated to an exposure site begins. Assuming long periods of time results in this exponential term approaching 1.0. The term was, therefore, dropped, to arrive at the steady state algorithm for soil erosion impacts. To test the impact of this assumption, tests were run including the exponential term with t equal to 1, 2, 3, 4, 5, 10, and 15 yrs. Expressing results in terms of the percent of steady state concentration reached at the end of each year, the results for these time intervals are 28, 49, 63, 74, 81, 96, and 99%, respectively. As seen, 81% of steady state is reached after 5 years, and 96% is reached after 10 years. Therefore, if off-site contamination has existed for 5-10 years or more, estimates of exposure site concentration (and related exposures) with a steady state assumption should be reasonable. However, if an assessment is to be done for a site to be newly impacted, such as a planned landfill, than the steady state approach would lead to some overprediction of concentrations and exposures. For the first five years, concentrations would average about 60% of steady state, and for the first 10 years, concentrations would average about 75% of steady state. This would be of most concern for a childhood pattern of soil ingestion, which would be 40% lower for the first five years of a landfill operation as compared to a steady state assumption. Otherwise, it is seen that the steady state assumption does not greatly impact exposure site concentrations. Estimation of soil erosion impacts to nearby surface water bodies

Results from this test are shown in Figure 6-4. One immediate point to make about this bar graph is that the same magnitude and direction of change is noted for both water concentrations and bottom sediment concentrations. This is actually true for all but two of the parameters in Figure 6-4. These two are the organic carbon content parameters, OCssed and OCsed, and the organic carbon partition coefficient. First, the direction of the change is not the same. Increasing the sorption of dioxin-like compounds onto sediments increases the concentration on sediments (of course), but decreases the concentration in water. For the "low organics" test, water concentration increases by a factor of 2.7 rather than slightly decreases as in Figure 6-4, which for this case, displays only the impact to bottom sediments. For the "high organics" test, water concentrations decrease to 0.60 of what they were in baseline conditions. The high Koc decreases water concentrations to 0.1, and the low Koc increases water concentrations 7 times. Both these trends are distinctly different than the sediment trends; they were left out of the graph in order not to crowd the graph (or require another one be drafted), and also because water concentrations in the sub-ppq range are of minimal concern for exposure.

The comments in the above section concerning the enrichment ratio are pertinent for this algorithm. The comments in the above section concerning the sediment delivery ratio equation, which for this algorithm is used to determine a value for SDw, also pertains to this algorithm. However, if a site is near a surface water body, it seems that the origins of the sediment delivery ratio equation - developed from data on construction sites near surface water bodies - are more appropriate. The key source strength terms tested, the area of contamination, ASC, and the soil loss rate from the site of contamination, SLs, both have a direct linear impact on the both sediment and surface water concentrations. The other soil loss term, the erosion rate for the watershed, SLw, also has a direct linear impact. What also appears to be critical in this algorithm is the size of the watershed. The reduction of impacts seen with the "large" watershed necessitated a y-axis spanning two orders of magnitude.

The algorithm seemed fairly insensitive to the remaining four parameters tested. The average watershed concentration, initialized at 0.0 in order to just show the incremental impact from the contaminated site, was increased to 1 ppt. This was also the concentration used to demonstrate the on-site source category, and was identified as a possible background concentration of dioxin-like compounds.
table Figure 6-4 Results of sensitivity analysis of algorithms estimating surface water and bottom sediment concentrations resulting from a site of soil contamination.
If this is a reasonable selection for a background soil concentration, it is seen in Figure 6-4 that background soils have a marginal impact on a water body which is impacted from a site of elevated soil concentrations. The impact of the organic carbon partition coefficient, Koc, on bottom sediments appears small despite the fact that the Koc range spans two orders of magnitude. This is an indication that it is so high for the dioxin-like compounds, that (at least in the algorithm of this assessment), its assignment is not critical for sediment concentration estimations. The same lack of impact appears to be the case for the organic carbon content of water body sediments, and the level of suspended solids in the water column.
expand table Figure V3 6-4