Risk-Based Waste Classification in California (1999)

After consideration of the scenarios and models chosen by the Department of Toxic Substances Control (DTSC) of the California Environmental Protection Agency in Chapters 2 and 3, the NRC committee examined the quality of the data used in the models for classification of wastes. This chapter examines (1) detailed model parameters for exposure pathways such as those leading to dietary intake; (2) selection of parameters, such as dilution attenuation factors, for specific models; (3) analytical methods, especially the use of either the waste extraction test (WET) or the toxic characteristic leaching procedure (TCLP) extraction methods; and (4) human and ecological toxicity tests.

After the selection of scenarios and models to be used in the risk assessment, the models must then be implemented with the correct parameter values. It is incumbent on DTSC to review its modeling to ensure correct selection of parameter values to correspond to the scenario in which the model is used. The committee examined certain parts of the documentation and spreadsheets to evaluate whether a suitable quality-control process had been applied to DTSC's modeling (see Chapter 3). In its review of the DTSC report, the committee found numerous errors and inconsistencies in the selection of the component models and the model

parameters. The following list indicates some of the types of errors and problems that were found for input parameters. Given the nature of the task and the time allotted, the committee identified as many specific problems as it could and provides examples of these in this chapter. However, the committee did not prioritize these problems and notes that not all of them are equally serious or have the same impact on the outcome of the risk assessments. This list should be taken not as a complete set of problems that needs to be corrected, but as an illustration of the type of problems and errors that a complete quality-control program should be designed to locate and correct.

The problems and errors can be classified into several types, with some problems and errors occurring simultaneously:

• Transcription errors: The values have been incorrectly transcribed from an original reference.
• Mistaken identity: The values are correctly derived from measurements, but the measurements are of the wrong physical quantity in the context of the particular scenario and model.
• Mistaken derivation: The values are derived from measurements of the correct physical parameter, but the derivation is incorrect in the context of the particular scenario and model.
• Incorrect extrapolation: The values are derived from physical measurements by using an extrapolation that is inapplicable.
• Impossible: The values used are physically impossible.

For completeness, the following types of problems that should be corrected by an adequate quality-control process are also discussed in this chapter:

• Inappropriate model errors: The model used does not correspond to the physical processes occurring in the scenario
• Structural errors: Errors or ambiguities in the structure of the models that lead to errors in calculations.
• Documentation errors: The description of the model differs from the model that was intended to be adopted.
• Implementation errors: The implementation of the model differs from the mathematical model adopted.
• Calculation errors: Something has been incorrectly calculated, but it is not possible to determine what went wrong.

Of course, documentation errors and implementation errors often occur together, and either or both can occur at any stage in the translation from physical description to simplified physical description to mathematical model to simplified mathematical model to implementation of the model.

The most basic level in scenario development is the selection of the specific parameters needed to implement the models in the context of the scenario. Such parameters include food intakes, quantities of soil eaten, dust-deposition rates, bioconcentration factors, soil-mixing depths, vapor pressures, soil porosities, inhalation rates, and solubilities. Below is an analysis of the types of problems encountered when DTSC's choices of parameter values were examined.

The food intake values used in the scenarios are based on data that may not be directly relevant to the citizens of California. It may reasonably be expected that the scenarios outlined would result in a small number of individuals incurring a large, albeit unknown, risk. However, the number of such individuals relative to the population of California, and the risk incurred by these individuals, is not knowable without completing model runs using the exposure scenarios developed by DTSC. The committee is, therefore, unable to estimate the effect the food intake and population assumptions have on total risk. Some of the committee's concerns with the application of food intake data are described below.

The specific dietary intake parameter values need to be realistic. Those used by DTSC do not appear to have been selected for real-life conditions and draw upon data that are 10 to 30 years old (an example of the mistaken identity error). Focusing on the adjacent resident scenario, current data need to be gathered on the types of residents near facilities. What fraction are farm households? If they are not farm households, do these households produce and consume their own meat, eggs, and dairy products?

If DTSC selects farms as the basis for its adjacent resident scenario, it should collect data on the number of small, family-owned farms in

California because residents of these farms are most likely to use home-grown crops as principal sources of food. Are they located near waste sites? How many? How far? Are there California demographic data to support those given in the U.S. Environmental Protection Agency's (EPA) Exposure Factors Handbook (EFH) (EPA 1990a)? As with the discussion of the scenarios for the population subject to the food intake assumptions mentioned above, these questions also raise the problem of estimating the changes in population living near hazardous waste sites and producing their own food. With the changes in farming from small, family-owned farms to agribusiness, the risk to a small number of individuals may be reduced in time. However, the large, agribusiness farm subjected to contamination of crops by a nearby hazardous waste site could increase the risk (by a smaller amount) to a larger segment of the population through the sale of contaminated crops. Therefore, DTSC might also, in the definition of its scenarios, wish to take account of time trends in agriculture, perhaps resulting in fewer small farms near waste sites, and perhaps resulting in wider dispersion of contaminated produce from larger farms.

Some of the difficulty in the exposure assessments for the adjacent resident scenario can be traced to estimates of dietary intake, most specifically for home-grown foods. These estimates are presented in the CalTOX parameter values section of the DTSC report (1998a; pp. 611 ff). The primary reference for home-grown food intake is from the first revised EFH (EPA 1990a). Table 4-1 reports the fraction of various foods that are assumed home-grown.

Taking even the mean values for the fraction of foods consumed would require residential conditions that would be illegal under various ordinances in most communities, and are unlikely to be observed for any individual. The problem with using these food intake estimates stems from using data collected in a specific survey and attempting to apply it to a more general or inappropriate situation. The following sections examine the problems with using the data on home-grown food intake by specific food type.

The primary EFH reference for fruits and vegetables reports a decrease in the average size of the garden from 600 ft² in 1982 to 325 ft² in 1986. Extrapolation to 1999 would suggest that even smaller garden sizes are the current norm. Furthermore, the gardens are found to produce approximately 0.9 lbs of produce per square foot annually, or about 300 lbs of produce. Given the U.S. Department of Agriculture's results cited in the EFH indicating consumption of 201 g/day of vegetables, a 325 ft² garden would fully support two individuals on this intake of vegetables. However, this is an average yield and does not take into account the differential yield for different vegetables, for example, pumpkins versus spinach. The EFH further reports that the largest numbers of such gardens are in the Midwest and South and that more individuals in rural settings tend such gardens compared with those living in cities and suburban areas. Neither EFH nor the DTSC report provides specific data on the number of gardens in California. It is reasonable to assume that some state-specific data for consumption of home-grown fruits and vegetables are available for California, a major agricultural state, yet only data from the EFH are used.

Table 2-10 in the EFH shows that the percentage of home-grown fruits and vegetables consumed ranges from 4.2% for lettuce to 75% for lima beans; these values are used in the DTSC report. The EFH data cited were gathered for a specific survey and no evidence is given by DTSC to support the applicability of such data to conditions in California. The EFH specifically cautions the reader on the representativeness of these data, which were drawn from a small number of days and quantified by recall only. DTSC uses an average value taken from Table 2-10 and presumes that consumption of all fruits and vegetables matches

this average value from home gardens; however, no rationale is provided for this presumption.

Furthermore, these consumption values for home-grown fruits and vegetables seem excessively large for the California population as a whole. DTSC does not specify who the target group is or who will be protected. It would appear that DTSC is looking at a maximally exposed individual. Although these values might be accurate for home gardeners in 1986, their validity for a population that consists predominantly of urban and suburban dwellers (another mistaken identity error) is questionable. DTSC has not demonstrated that the population assumed to grow and consume these foods exists. DTSC provides no support for the size of the garden versus food consumption, nor do they provide information about subpopulations who might be vegetarians, low income and subsistence farmers, specific ethnic groups, or children. Whether explicit account needs to be taken of any such subpopulations depends on the scenarios under evaluation to meet specific policy goals. The public comments indicate that it is possible to ascertain the number of hazardous-waste sites in the state and the distances of the nearest residences. With such information, it should be possible to adequately characterize home gardeners living near hazardous-waste sites, including the average distance from their residences to those sites.

For the grain consumption pathway, DTSC makes use of data exclusively on corn, because corn is the only "grain" product mentioned in the EFH. However, the committee suspects that corn grown in home gardens is used as a vegetable, not a grain (a mistaken identity error). It is not aware of any data on grinding corn meal from corn grown in a home garden. DTSC further compounds this poor data analysis in that other grains, presumably wheat and similar products, are assumed to be identical to corn. Lacking any data supporting the use of wheat as a vegetable, its use can safely be assumed to be as a grain to make flour and other products. It is extremely unlikely that the typical home garden produces 12% of the wheat flour used in the residence.

The DTSC report states that, in farm households, the annual fraction of

home-grown beef consumed is 44% with a coefficient of variation (CV) of 0.5 (DTSC 1998a, p. 612). Similarly, for dairy products and eggs (by direct assumption of the equivalence between dairy products and eggs), the values are 40%, with a CV of 0.7. These values might be biased because they were based, according to EFH (EPA 1990a), on a survey of 900 rural farm households published in 1966, and they only apply to farm households. It is highly unlikely that such numbers can apply to suburban and urban settings, where keeping livestock is usually against local ordinances (a mistaken identity error). Again, the central issue is what population is being protected? Clearly, this component of the scenario uses a value based on a maximally exposed individual, not on the broader population. With the changes in U.S. agricultural practices since the 1960's from family farms to agribusiness, the application of these data to any residents of California must be justified by DTSC.

The consumption rate of fish for recreational or subsistence anglers and the fraction of fish eaten from local sources are also subject to controversy. The consumption rates in the EFH are based on data collected in 1973–1974 (EPA 1990a), and might no longer be valid, particularly given the number of no fishing advisories in effect for many California waters. DTSC also assumes that the shape of the distribution of intake is triangular, however, the shape of the triangle is not indicated and the basis for the assumption is unsubstantiated.

The committee urges DTSC to incorporate more recent exposure factors (e.g., those given in EFH published by EPA in 1997) as well as data that are representative of California urban, suburban, and rural populations.

This section highlights some of the committee's concerns regarding the use of models for soluble or extractable regulatory thresholds (SERTs) and some specific parameter problems for the preliminary endangerment assessment (PEA), LeadSpread, and CalTOX exposure models used to develop the toxicity threshold limit concentrations (TTLCs).

There are various structural problems with the DTSC's implementation process for SERTs in general:

• The use of a single dilution attenuation factor ensures that variability in the population's exposure to groundwater is not taken into account. As DTSC acknowledged during the second public meeting (DTSC, personal commun., November 20, 1998), the SERT scenario was modified for some sort of worst-case exposure, not for exposure at a 90th percentile of the population as documented.
• There is a logical disconnection between some toxicity indicators (surface-water-quality criteria, maximum contaminant level) and the calculated values with which they are compared (groundwater concentration). Logically, if such a comparison is meaningful, it should also be meaningful at all downgradient distances, not just those corresponding to a dilution attenuation factor of 100. This is probably connected to the previous problem, and its solution requires explicit specification and acknowledgment of the policy objective.

There are also problems with the SERT definition.

• The toxicity values used for the SERT calculations are particularly puzzling. These toxicity values include the ambient water quality criteria for aquatic life and maximum contaminant levels. Ambient water quality criteria apply to surface-water bodies, so that an extra dilution has to be taken into account, that is, where the groundwater runs into the surface water body. In particular cases, such as water bodies that are fed only by contaminated groundwater, the dilution factor might be greater than unity (e.g., if the water body evaporates and the contaminant is nonvolatile). If DTSC is attempting a probabilistic approach, then some distribution for this further dilution is required. If DTSC is attempting a worst-case analysis, then the worst-case would have to be applied. This type of error could be either a mistaken identity error if ambient water quality criteria were assumed to be relevant to groundwater, or an extrapolation error if it was assumed that groundwater concentrations correspond to surface-water concentrations.
• The maximum contaminant level is used as an indicator level applicable to groundwater. Given that health-based levels are separately

• derived, it appears that DTSC is using the maximum contaminant level as an enforceable standard for all California groundwater. Thus, the committee questions the use of an maximum contaminant level for a risk-based approach. This appears to be a mistaken identity error.
• DTSC's approach to the use of a liner protection factor to take better account of modern landfills is also misguided. DTSC has attempted to estimate the liner protection factor by comparing a lined landfill with an unlined landfill. However, the parameter values used for the unlined landfill appear to correspond to a fairly tight landfill with clay liner. But these parameter values do not correspond to the parameter values used in the original EPA modeling. Thus, the liner protection factor is calculated from an incorrect base value. This could be classified as a mistaken identity error for all the parameter values for the unlined landfill.
• The SERT scenario is so poorly defined that the committee cannot comment on its applicability—it can simply point out where the implementation does not agree with the documentation. The following paragraphs identify some of the specific problems that were encountered in the review of the lower (nonhazardous) and upper (hazardous) SERT calculations.

The DTSC spreadsheet for SERTs has a 100% correlation between the distributional calculations for risk and the hazard index. Although this does not affect the results of the current calculations, it is possible that in a more complex analysis such a correlation would be incorrect. (In fact, in the subsequent calculation of upper SERTs using the DTSC spreadsheet, this correlation is essential to get correct results, because the minimum function is applied at an intermediate stage of calculation). This is a potential structural error, although it does not affect the current calculations.

The logic in the spreadsheet for SERT calculations does not correspond to the description given in the DTSC documentation (DTSC 1998a, pp. 43–45). The minimum value of the health-based level, maximum contaminant level, or ambient water quality criteria is applied before the statistical lower 10th percentile is calculated for health-based level and the lower 10th percentile of this minimum has been found. In principle,

this should make no difference to the final results, although one can expect problems in labeling some intermediate results (see below). This is a documentation error, or possibly a structural error (although it does not affect the final result).

Possibly as a result of the preceding calculation, the values given in the column labeled ''Health-based level × 100'' (DTSC 1998a, p. 46) do not correspond to the health-based level × 100, where the health-based level is computed as the 10th percentile as described in the text. In fact, for each of six chemicals (aldrin, kepone, arsenic, beryllium, thallium, and vanadium), the value given in the table is correct to one significant figure, that is, it is indeed the health-based level × 100, where the health-based level is the lower 10th percentile value. For all but four of the remaining chemicals in the table, the value given can be obtained (to one significant figure) from the same calculation, but by using the mean values of each parameter in the calculation, not the lower 10th percentile of the distribution resulting from using the parameter value distributions. So the value given does not correspond to the text description. For the remaining four chemicals (chlordane, methoxychlor, chromium VI, and molybdenum), it is not clear how the values given in the table were derived, because they do not correspond to either calculation or to anything in the spreadsheet.

The spreadsheet apparently used a Monte Carlo approach to evaluate the 10th percentile of the lognormal distribution required for calculating the health-based level for the lower SERT. Although the spreadsheet entries show correct values (within 0.7%) for the 10th percentile in most cases, in five cases (cobalt, fluoride, molybdenum, thallium, vanadium) the entries are more than 15% in error. This appears to be a calculation error. The calculation for the health-based level involves a single lognormal distribution (for the hazard index) or a multiplicative combination of three lognormals (for risk), which is also lognormal. Therefore, the calculation of the lower SERT is analytically straightforward.

The difference between the calculation of the lower and upper SERTs is the liner protection factor. The upper SERTs are calculated by multiplying the lowest of the health-based level, the maximum contaminant level

or the ambient water quality criteria by a dilution attenuation factor of 100 and a liner protection factor. The DTSC documentation (DTSC 1998a, p. 47) specifies that an liner protection factor was entered as a "custom distribution" but provides no indication of how the values were derived. The custom distribution function in the Crystal Ball software by Decisioneering, Inc., allows various options for defining distributions (combinations of point values with assigned relative weights together with piecewise linear densities), but the DTSC documentation does not specify what options were used. During the first public meeting, DTSC stated that six values (two sites, three conditions, using the HELP model) were used as a custom distribution (DTSC, personal commun., September 10, 1998).

The SERT spreadsheet contains a list of six values for liner protection factors. They are (in the order listed): 36, 190, 1600, 22, 118, 970. The Crystal Ball custom distribution add-in, however, lists a different set of six values, entered as point values with equal relative weights. These values are (in increasing sequence): 18, 99, 118, 191, 970, 986. The DTSC documentation gives three generic values for liner protection factor—18, 99, 986, using an approximate model that takes into account leakage through the liner versus leakage through clay only (DTSC 1998a, p. 1,487).

The documentation later cites the HELP model as giving three values each for two precipitation and evapotranspiration regimes—Los Angeles and Eureka (DTSC 1998a, p. 1,488). The values are: 36, 190, 1,600 for Los Angeles and 22, 118, 970 for Eureka. These six values are identical to those listed in the spreadsheet, but not in the custom distribution in Crystal Ball. The documentation then appears to include a printout of a spreadsheet (source not provided by DTSC) that provides yet another set of values for all three cases: 36, 190, 1,600 for Los Angeles; 22, 120, 970 for Eureka; and 18, 99, 990 for a generic case (DTSC 1998a, p. 1,492). Thus, the documentation is not clear on what values are used to derive the upper SERTs—the resulting difference between documentation and implementation could be a documentation error or a transcription error.

The spreadsheet calculations for upper SERT results appear to correspond to the values for liner protection factors present in the Crystal Ball custom distribution (which do not correspond to any documented set of values). DTSC indicated that the spreadsheets contained the most valid calculations, so that the liner-protection-factor values in the Crystal

Ball custom distribution must presumably be taken as those to be used. Although adding in the liner-protection-factor custom distribution makes the distributions to be evaluated a sum of six lognormals, it is still relatively easy to compute analytic results—all that is required is the solution of one nonlinear equation each for the risk case and the hazard index case.

Examination of the table containing the liner protection factors (DTSC 1998a, p. 48) shows just four values: 18, 19, 24, and 63, although it is not clear how these values were derived. They are not calculated within the SERT spreadsheet. It is possible that the erroneous 19 and 24 arose from the same error that produced the >> 15% error in the health-based levels for the lower SERTs because the same five materials are affected (cobalt, fluoride, molybdenum, thallium, vanadium). The exact values using the liner-protection-factor set present in the Crystal Ball custom distribution should be: 18, 30.1, and 63.7. These apply when the upper SERT is based on maximum contaminant level-ambient water quality criteria, oral reference dose, and cancer potency, respectively. The value obtained when the upper SERT is based on maximum contaminant level-ambient water quality criteria is necessarily the lowest of the six-point estimates for the liner protection factor, because the liner-protection-factor distribution is the only distribution involved in this case. Each point in the liner-protection-factor distribution corresponds to one-sixth weight, so that all percentiles from 0 to 16.66666 are assigned this lowest value. It is not clear if this was DTSC's intent. For the six liner-protection-factor values specified for Los Angeles and Eureka, the three corresponding SERT values would be 22, 34.4, and 54.5. The results given by DTSC have to be considered a calculation error.

A major concern for the preliminary endangerment assessment (PEA) model is that DTSC fails to provide a thorough description of the scenarios to be modeled. A review of the component models of the PEA model indicates several areas where the component models are incorrect, as indicated below. In addition, many of the model values are not referenced, or differ from the documentation cited. Furthermore, the uncertainty distributions are not always documented. Some incorrect pathway models include evaporation of chemicals from a landfill and vapor

dispersion above a landfill, for which DTSC fails to use the available (roughly correct) pathway model in CalTOX.

For the adjacent resident and the resident on converted land scenarios, DTSC used two spreadsheets, one for hazard and one for risk. This might present a fundamental difficulty related to the definition of the level of protection desired, if the requirement is to protect each individual against particular levels of both noncancer and cancer risks at some percentile of the distribution across individuals. With the two spreadsheets, the requirements are imposed completely separately. Thus, even though 90% of individuals are protected at the selected levels against cancer risk, and 90% against noncancer risk, less than 90% (in principle, as few as 80%) could theoretically be protected against both cancer risk and noncancer risk simultaneously. This is a structural error, but DTSC's goals have not been sufficiently well defined to determine whether the structural error is potential or actual.

The definition of "Pathway Exposure Factor for Inhalation" (DTSC 1998a, p. 790) differs (in definition and dimensionality) from the definition in the spreadsheet. The spreadsheet definition is preferable because it maintains equal dimensionality for this quantity for all exposure routes. Based on the values used in the spreadsheet, the receptor is a child for the hazard calculation, although the precise receptor does not appear to be documented. These variations appear to be documentation errors. Only 16.3 hours of the day is accounted for in "moderate activity" and "resting" (documentation) or "resting (sleep)'' and ''heavy activity" (spreadsheet); it is unclear what happens to the rest of the day for this receptor. There is no truncation on the distributions used, so that the length of a day could theoretically exceed 24 hours during the Monte Carlo procedure (allowing an impossible parameter set).

Equation 8 for the pathway exposure factor for inhalation in the residents in the land conversion scenario (DTSC 1998a, p. 793) omits the effect of leaching and vapor emissions, and so is incorrect for soluble and

volatile materials. In view of the length of time involved (20 years), the error introduced is substantial. This has to be considered a structural error in the model.

As mentioned in Chapter 3, DTSC's approach to dust is inadequate for the adjacent residents scenario (DTSC 1998a, p. 792). That discussion is expanded here by looking in more detail at some specific examples that illustrate this inadequacy. There are suitable models in AP-42 for dust-emission rates and for dust-deposition rates (California Air Resources Board model subroutine, as used in industrial source complex (ISC2) model (EPA 1992), or the latest models used in ISC3 (EPA 1995b). Both account for dust-particle size distributions. The fugitive dust model also might be appropriate (EPA 1990b). Using such models would readily allow evaluation of the variation of dust deposition with distance and meteorological conditions, something that is lacking from the DTSC approach. (These models are easy to construct if they are not available prepackaged.) Failure to use such modeling approaches is a structural error for this, the major pathway.

DTSC stated in its discussion of adjacent residents that "the residence is assumed to be very close to the landfill, so no dispersion of dust is included in the model" (DTSC 1998a, p. 785, Section 1.2.1). The committee does not understand this assumption because DTSC does not provide any evidence that even the short distances involved do not result in significant dispersion. After all, the residence is not in the middle of the landfill—it must be to one side of it. Thus, it can be expected that, on average, the wind is blowing in the wrong direction perhaps half the time. That amounts to a neglected factor of 2 (averaging over landfills and residences statewide), even neglecting dispersion. Furthermore, one property of dust is that it settles as a function of distance from the landfill, resulting in a steady change in the importance of the ingestion pathway as distance from the landfill increases.

At the second public meeting, DTSC indicated that a concentration of 50 µg/m³ of dust in air (based on the National Ambient Air Quality Standards) was used as a default value and that dust emission and dispersion had not been modeled because there were no models for dust emission (DTSC, personal commun., November 20, 1998). This state-

ment was very disturbing to the committee, given that dust-emission models are easily accessible in AP-42 (EPA 1997b) and that any consideration of this type of scenario (adjacent resident) clearly calls for adequate modeling of the dust-dispersion pathway (the primary off-site exposure pathway considered). The selection of the National Ambient Air Quality Standard value was apparently also based (according to the information provided in the second public meeting, but not elsewhere documented) on measurements of ambient atmospheric particle loading. However, except in the vicinity of dominating local sources, the great majority of the ambient particle loading in the atmosphere is due to long-range transport. The DTSC treatment of dust can be considered an example of incorrect extrapolation—the measurements of ambient dust loads have been incorrectly extrapolated to the DTSC scenario.

Compounding the incorrect extrapolation are some transcription errors. In the DTSC report (DTSC 1998a, pp. 785, 791, 815), it is stated that the deposited dust mixes to a depth of 0.15 m over a period of 20 years. However, in the spreadsheets for the adjacent resident scenario, 0.015 m and 0.1 m are used for hazard (Off_haz.xls) and risk (Off_risk.xls), respectively. Thus, risk and hazard calculations correspond to different scenarios, and both are different from the scenario that is presented in the DTSC documentation. One of these transcription errors is propagated to the lead spreadsheet, which uses as one of its inputs a dilution factor (0.00098) evaluated in Off_risk.xls.

Dust_conc is defined as the concentration of dust in air over the landfill (DTSC 1998a, p. 792). However, what is required here is the concentration of dust over the resident, not over the landfill. The model used by DTSC should, but does not, differentiate between the landfill and off-site, an extrapolation error. The dilution due to dust dispersion and deposition should be taken into account. In addition, a more subtle point should be taken into account: only one average dust concentration is required for an on-site worker present during the day; for the off-site exposures, two average dust concentrations are required—one for inhalation (probably residents are present more during the night than during the day, on average) and one for deposition. Any reasonable model of dust emission, dispersion, and deposition should take account of such factors, because they will include the strong correlation between dust concentration and meteorological conditions. This could be a mistaken identity error, or an extrapolation error.

V_dep is defined as the deposition velocity for dust to soil (DTSC

1998a, p. 792). Its value is given as 500 m/day (CV 0.30) (DTSC 1998a, p. 815). However, the closest parameter in the DTSC documentation (1998a) is "deposition velocity of air particles," which is given as 690 m/day (CV 1.45), suggesting a transcription error. In addition, the measurements used to derive a value in DTSC (1998a, p. 555ff) are not relevant to deposition of wind-blown dust from a landfill. They are (presumably) measurements of ambient particle-deposition rates (DTSC 1998a, p. 560 and 578), and thus correspond to a very different size distribution from wind-blown dust, so that using this parameter is an error of mistaken identity (or possibly an extrapolation error).

It is confusing to introduce a set of parameters and then redefine them using the same names, as occurs for two pathway exposure factors (PEFs) with the introduction of dilution_dust (DTSC 1998a, pp. 791, 792). Dilution_dust should be introduced immediately in the original definitions of the PEFs. As another example, PEF_inh in equation 2 (DTSC 1998a, p. 792) does not correspond to the spreadsheet definition of the same term (a documentation error).

Section 4.5 of the DTSC report (DTSC 1998a, p. 793) contains many documentation errors. In equation 11, the variables CR_dust and CR_vapor in the numerator on the right have different dimensions from the other terms, and so equation 11 cannot represent any physical process (i.e., the described model is physically impossible). The body weight (BW) is introduced in the denominator, despite the fact that this is already present in the variables CR_ing and CR_der. The same sort of errors occur in equation 12, and in addition the variable Cr_ing used in equation 12 is not defined. It is also incorrect to equate the vapor and dust concentrations (below equation 12). Furthermore, the need for Section 4.5 is unclear, because these intakes are never calculated in the spreadsheets, and correctly so (what is required are route-specific intakes for comparison with route-specific toxicity values).

ABS is defined as chemical-specific absorption through the skin (unitless) (DTSC 1998a, p. 794). (It was incorrectly referenced to Table 4; the correct reference is to Table 3 (DTSC 1998a, p. 812)). This factor should not be the absolute absorption through the skin, but an absorption relative to that occurring in the experiment from which the refer-

ence dose or cancer potency is defined (these might be different). Because the reference dose and cancer potency are generally expressed in terms of external doses (i.e., the dose crossing the envelope of the human body), they already incorporate the absorption factor applicable during the experiments used to define them. Thus using absolute absorption factors corresponds to an extrapolation error.

DTSC's rationale for changing the algorithms in the PEA model for estimating emission rates for volatile and semivolatile organic compounds is incorrect. Selection of the EPA algorithm (EPA 1995a, cited in DTSC 1998a, p. 795) would require a correlation between the scenario evaluated by EPA (an infinite depth of contaminated soil instantaneously emplaced at time zero, with no infiltration of rainwater) and that to be evaluated by DTSC for the workers on a landfill (small quantities of waste added at intervals over a long time interval, with daily cover applied and possible infiltration of rainwater). The two scenarios are so different as to completely invalidate all the calculations for the DTSC scenario based on the EPA algorithm. DTSC has thus incorporated an inappropriate model. In addition, the waste worker spreadsheet (work_org.xls) implements equation 15 (DTSC 1998a, p. 796) incorrectly. It incorporates an extra factor of 0.1 (in the column headed F/CO) that is undocumented and has no physical basis—an implementation error (although of an inappropriate model).

Furthermore, equating the upper limit (concentration) of the chemical in free moisture in soil to the product of solubility and the moisture volume fraction in soil is incorrect (DTSC 1998a, pp. 795 to 796). What is required is just the solubility (S), as correctly implemented in the spreadsheet (a documentation error). The definitions of bulk_density and part_density (DTSC 1998a, p. 796) are logically inconsistent. One equation is for wet bulk density, the other is for dry bulk density, but that distinction is not made. It must be specified whether bulk_density is wet or dry bulk density (this appears to be a documentation error; the propagation of this error in the implementation has not been checked).

The quantities referred to as concentrations in equation 15 (DTSC 1998a, p. 796) are in fact mass fractions. (Concentrations have dimensions of mass/unit volume; mass fraction is dimensionless.) This common error (some would say convention) appears trivial, but is in fact a documentation error—it leads to confusion when attempting dimensional analyses, and so prevents a useful quality-control check.

The calculation for the apparent diffusion coefficient in soil in

equation 16 (DTSC 1998a, p. 797) contains a documentation error. In the numerator of the expression on the right of the equation, the operator connecting the two inner bracketed expressions should be addition, not multiplication. In addition, to correspond with the previous definitions, and with the conventions used throughout the document, there should be a factor of 0.001 liter per cubic meter (L/m³) multiplying the variable Kd in equation 16. However, this convention should not be followed—there should be no conversion factors present in any of the equations describing the models. Symbols in such equations should represent the quantities involved, including their dimensions, not just their numerical values (which depend on the units in which they are measured).

No adequate reference is provided for the mixing height (MH) above the landfill. In Table 5 (DTSC 1998a, p. 817), mixing height is given as 2 m for a 4.54 × 10⁵ m² landfill, with a reference to an earlier 1994 DTSC report. Examination of that reference indicates a default mixing height of 2 m for a residential scenario with lot size of 22 m², where it is more plausible, although again there is no documentation for selecting this value. For the landfill scenario described here, the value is extremely low. A reasonable estimate for mixing height in this situation takes account of typical plume opening angles. The landfill has length of approximately 670 m, so a 2-m mixing height corresponds to a plume opening angle of about 1/335 radian. If that were correct, practically all the dust raised from the landfill would go underneath the noses of the workers. The committee recommends the use of a model that leads to a reasonable estimate of dispersion over an area source. CalTOX explicitly has such a model and ISC2 and ISC3 both handle area sources; it is inexplicable why DTSC did not use such a model. This appears to be a situation where a parameter value has been taken from one scenario and applied to another, without consideration of the difference between scenarios or the appropriateness of the models used, a type of extrapolation error.

For workers on a landfill (DTSC 1998a, p. 798), the method for calculating a hazard index might be inadequate, because the estimated emission

rate and concentration have been averaged over the total exposure time (5.5 years for workers). However, the chosen model for emissions, which is incorrect, corresponds to emission rates that decrease inversely with the square root of time, and so are much higher initially. In the context of its chosen model, therefore, DTSC's averaging method is a structural error. However, as pointed out in Chapter 3 and above, the emission model is incorrect for the scenario evaluated. Emission rates from landfills are likely to vary rapidly and substantially on short (hour to day) timescales as wastes are dumped or surfaces are worked or covered. On longer timescales (days to years), the average emission rates are likely to be much more constant while the landfill is active, but finally decreasing after landfill closure. Such features should be reproduced by appropriate emission models. DTSC should determine whether shorter-term averaging of concentrations for comparison with short-term reference concentrations is required, in addition to the longer-term averages.

The first three sentences of DTSC's discussion of the Monte Carlo analysis (Section 6.2, DTSC 1998a, p. 799) demonstrate a fundamental misunderstanding about probabilistic methods. This misunderstanding has been implemented in the PEA spreadsheets, so that those spreadsheets do not necessarily provide the results that DTSC expects. The problem (Burmaster and Thompson 1995; Burmaster et al. 1995) can be expressed as follows. Suppose a risk estimate R is obtained as a function of various parameters a, b, c, . . . and a concentration C, so that R = f (a, b, c, . . ., C) for some function f. In a probabilistic calculation, the parameters a, b, c, . . . have probability distributions that induce a probability distribution on R for fixed values of C. This corresponds to the physical situation, and the problem is to find the value of C that ensures that some upper confidence limit on R is equal to some target value.

In a deterministic setting, the equation R = f (a, b, c, . . ., C) is, in principle, soluble to obtain C = g (a, b, c . . . . R), allowing a direct calculation of the value of C that corresponds to a given value of R. Unfortunately, while this equation provides an inverse calculation for

fixed values, it does not provide the probabilistic inverse.¹ One cannot perform a Monte Carlo procedure with the distributions for a, b, c . . . . and obtain the "distribution" for C given a fixed R—the whole concept of a distribution for C is indeed meaningless. This becomes more obvious when it is realized that, in a probabilistic calculation, the first equation is simply shorthand to indicate that the distribution function for R is a complex convolution integral over all the distribution functions for a, b, c . . .; and the Monte Carlo procedure is just a way to compute that convolution integral.

The PEA spreadsheets have been constructed to estimate a concentration C given a fixed target value for R and fixed values of the parameters a, b, c . . . . When the Monte Carlo procedure is run in one of these spreadsheets, what is obtained is a set of values for C at the fixed target value for R. In general, the lower percentile of this "distribution" for C does not correspond to the concentration that would produce a distribution for R with upper percentile equal to the target value. All of the PEA spreadsheets, in their current form, satisfy the exception to this general rule (see footnote 1). However, Work_org.xls satisfies that exception through a programming error—by design it should not. In Work_org.xls, DTSC (1998a, pp. 795–796) indicates that there is supposed to be a selection of the smaller of two concentrations C _sat and C_a² to ensure correct calculations when the groundwater and the air within the soil pores are saturated with the chemical. With that test present, the excep-

tion noted in footnote 1 does not apply. While the test is necessary to ensure a physically reasonable model, it is actually absent from the spreadsheet. Furthermore, contrary to the DTSC's statement (1998a, p. 795), there are cases where the test is relevant (the groundwater and soil air are saturated at the calculated soil concentration), even with mean values for parameter estimates for methoxychlor and toxaphene, and surely for other chemicals for some parameter values selected during the Monte Carlo procedure. Applying the same procedures to further chemicals (with possibly very different physical characteristics) may also be expected to result in cases where the groundwater and soil air are saturated at the proposed TTLCs.

In contrast, the CalTOX uncertainty analyses appear to be constructed in a way that allows correct calculations (although the committee has not verified that such correct calculations were performed). Provided all the compartment concentrations remain within certain bounds (imposed, for example, by solubility constraints), all the models within CalTOX are linear in the compartment concentrations. Thus exposure point concentrations, and hence risk estimates, are linear functions of the initial compartment concentrations. Given fixed initial compartment concentrations (or similar inputs proportional to such concentrations, like rates of application), a distribution of risk estimates is calculated, and the required percentile of the risk distribution obtained. The initial concentrations (or similar inputs) may then all be scaled by the ratio (target risk value)/(risk at the required percentile of the distribution). The Monte Carlo procedure should then be repeated with the resultant scaled compartment concentrations, to ensure that none of the concentration bounds are exceeded.

The parameter ET_heavy (time spent in heavy activity) is documented as having a triangular distribution (DTSC 1998a, p. 800, section 6.2.1), and this is repeated in Table 7 (DTSC 1998a, p. 819). In fact, the work_met.xls and the work_org.xls spreadsheets have a constant value of 3.75 hours with no distribution incorporated (an implementation error).

DTSC developed a mathematical model called LeadSpread for estimating human blood-lead (Pb) concentrations resulting from contact with lead-

contaminated environmental media. A blood-lead concentration of 10 micrograms per deciliter (µg/dL) of whole blood was established as a level that will adequately protect adults and children. The model allows estimation of various percentiles of blood lead concentrations associated with a given set of exposures. The model can be used to establish a concentration in soil or waste that will result in an estimated 95% (or other) upper percentile blood-lead concentration of 10µg/dL. The model can provide an estimate of blood-lead concentrations resulting from five exposure pathways: dietary intake, drinking water intake, soil and dust ingestion, dust inhalation, and dermal contact.

The DTSC model falls within acceptable standards for models of this type, although the documentation for the model presented in the DTSC report is incomplete. Although a complete list of model parameters is provided, the structure of the model might not be clear to the uninitiated reader.

The general population is exposed to lead in ambient air, foods, drinking water, soil, dust, and fume. The subsets of the general population at highest risk of health effects from lead exposure are preschool-age children, the fetuses of pregnant women, and occupationally-exposed white males between 44 and 59 years of age. Within these groups, relationships have been established between lead exposure and adverse health effects. For the population including children, exposure to lead occurs primarily via the oral route, with some contribution from the inhalation route, whereas occupational exposure is primarily by the inhalation route with some from the oral route.

Blood lead concentration is an integrated measure of internal dose, reflecting total exposure from all sources and over time. Most data relating human health effects to lead exposure are based on blood-lead concentrations, so regulatory standards are typically based on blood-lead concentrations rather than external dose. EPA, the Food and Drug Administration, and the Centers for Disease Control and Prevention have determined that childhood blood-lead concentrations at or above 10 µg/dL may present risks to children's health. As a result, DTSC is following current practice in proposing TTLCs based on a 10th percentile estimate of the concentrations in environmental media corresponding to a blood-lead concentration of 10 µg/dL for children.

Each of the exposure pathways is represented by an equation relating incremental blood-lead increase to a concentration in waste using contact rates and empirically determined ratios. The contribution from each pathway to blood-lead levels is added to arrive at an estimate

of median blood-lead concentrations resulting from total exposure. Parameter values used in defining exposure equations fall within accepted ranges; however, one parameter value taken from PEA is known to be incorrect—specifically, the dilution factor in surface soil from the original waste (this error results from a transcription error for the soil-mixing depth in the Off_risk.xls spreadsheet).

The committee has reviewed CalTOX and DTSC's modifications very superficially, because of the short time available. However, as with the PEA model, many of the component models within CalTOX appear to be oversimplifications for the scenarios that DTSC is considering. In addition, the committee found numerous errors, in the context of the DTSC scenarios, in the parameter values used in CalTOX. Some of the specific problems with CalTOX and its associated parameters are described below (these are just examples, not a complete set of all the problems that were found).

For calculations of concentrations of materials in soil, CalTOX uses a three-compartment model, and assumes that the time-variation of all concentrations can be adequately modeled by linear first-order differential equations with constant coefficients. Such a model does not correspond to the physical processes, however, at least in some situations in DTSC's scenarios. This mismatch was explicitly realized by the designers of CalTOX, and the model parameters are specifically chosen to match a more physically realistic mathematical model, the Jury model, in order to take into account some of the mismatch (DTSC 1998a, pp. 263–268). The committee is concerned, however, about the adequacy of such matching for DTSC's purposes. It questions whether such approximations are either necessary or desirable and whether DTSC took into account the specific limitations on the model adjustments noted in the CalTOX manuals, and apparent in the simulations described there (DTSC 1998a, pp. 265–267).

As specific examples, the CalTOX parameter value estimates were matched to minimize differences in the surface fluxes and the soil-mass inventories in two of the CalTOX compartments; but the potential resulting three-fold uncertainty in even these quantities is not incorporated in DTSC's calculations. Moreover, although such matching might ensure that some results are reasonably accurate for some pathways (e.g., long-

term estimates for inhalation) in some conditions, the same does not necessarily hold for those pathways under other conditions, or for other pathways of importance. For example,

• Short-term (up to approximately a year) emission rates of vapors might be substantially higher than predicted by CalTOX, resulting in substantial underestimates of short-term exposures that might be toxicologically significant (DTSC 1998a, p. 268).
• Soil concentrations in the top few millimeters of soil, most relevant for the dermal contact pathway, are substantially overestimated by the CalTOX approach (DTSC 1998a, pp. 266–267).

The landfill gas model introduced by DTSC during the modification of CalTOX (DTSC 1998a, pp. 78–104) is also open to question. This model assumes a physically unrealistic continuous flow of landfill gas at a constant rate that is time-independent; it is also physically unrealistic because the variation of flow rate with the mass of material in the landfill has a discontinuity at 1.1 million tons. The committee wonders why the time-dependent EPA landfill gas generation model, for example, was not used. The manual describing this model is available at http: ://www.epa.gov/ordntrnt/ORD/WebPubs/landfill/index.html.

The values and uncertainties of the various parameters used in CalTOX and the other models come from ''The Distribution of California Landscape Variables for CalTOX, February 1996'' (DTSC 1998a, pp. 555 ff.), similar reports for individual chemicals (e.g., Hsieh et al. 1994), and other sources such as those listed as references in DATCAL.XLS and the accompanying DATREF.XLS. However, the values and uncertainties or variabilities reported in these sources occasionally do not correspond to those required; and there appear to be multiple transcription errors. A few examples follow. (Chapter 3 also has a short discussion of some pervasive extrapolation errors in the estimation techniques used by DTSC to estimate some of the parameter values used in CalTOX, for example, partition coefficients for plant tissue and biotransfer factors to meat, milk, and eggs.)

The reported dust-deposition velocity (DTSC 1998a, p. 555) corresponds

to the mean and standard deviation of many individual measurements of deposition velocity, not to the required long-term average velocity at any particular location and the variability between locations. The deposition velocity at any one location will vary substantially from time to time, but that is irrelevant for the modeling required by the DTSC scenario, an error of mistaken derivation. Moreover, there is also a transcription error: 500 m/day (CV 0.30) (DTSC 1998a, p. 816, table 5) versus 690 m/day (CV 1.45) (DTSC 1998a, pp. 560, 578).

The reported mean value for the carbon fraction of residential soil appears to be extremely low for fertile soil, but may be biased by many low measurements in infertile, but residential, soils, as the range is up to a reasonable value. However, for the CalTOX modeling in which this parameter is used, the range of values for fertile soil (backyard gardening pathway) is required, not an average over all residences. This might be an error of mistaken identity or derivation. The reported mean of the carbon content of residential soil is 0.003 (CV 0.367) (DTSC 1998a, pp 560, 572); but in Table 5, it is given as mean 0.003 kg/kg (CV 3.67) (DTSC 1998a, p. 815) (this error occurs twice, in Table 5, Section 2 and Section 5) and thus has been incorrectly transcribed.

DTSC gives the molecular weights of all individual chemicals as a mean and CV, based on several values found in the literature. For example, the molecular weight of trichloroethylene is given as 131.4, with a CV of 0.00039 based on five values (DTSC 1998a, p. 736). Inspection of the original data sources indicates that one value used was simply in error, and the other four values represent the same value written with differing numbers of significant digits. The rote approach taken by DTSC to evaluate a "distribution" for molecular weight undermines the committee's confidence in DTSC's selection of any parameter and of the modeling effort in general. The uncertainty in molecular weight for these chemicals of know structure is negligible, and the DTSC approach does not address the remaining uncertainty.

Many chemical properties are given in Table 2 (DTSC 1998a, p. 808) with a citation to the DATCAL.XLS spreadsheet. The committee has a copy of this file and the associated DATREF.XLS file. Although both files contain some references, in many cases there is no indication as to the original data source from which these values were taken. For example, there is no explanation or reference as to why the CVs for the vapor pressures for chlordane and TCDD are equal to 1.58 and 1.57, respectively, whereas many CVs for other substances are much smaller.

California is proposing to no longer require the use of the waste extraction test (WET) for determining the extractable constituents of hazardous wastes not classified under the Resource Conservation and Recovery Act (RCRA), relying rather on the use of EPA's toxic characteristic leaching procedure (TCLP). The TCLP has long been required by EPA to define the toxic constituents of RCRA hazardous wastes.

In 1972, California's then-new Hazardous Waste Control Act defined "hazardous waste" and "extremely hazardous waste" and, in 1977, California added the requirement that the state develop and adopt criteria and guidelines for the identification of these two waste categories. The California Assessment Manual (CAM) (California Department of Health Services, 1981) prescribed the use of WET as the state's test procedure. The CAM-WET test extracted solid wastes with pH 5 citrate buffer for 48 hr (Table 4-2).

At the federal level, the 1984 amendments to RCRA led to adoption by EPA of a batch extraction test, called the extraction procedure, which was designed to simulate processes occurring in landfills and that might

contribute to the leaching of toxic constituents. EPA subsequently replaced the extraction procedure with the TCLP (55 Fed. Regist. 11798, March 29, 1990). In the standard version of the TCLP, a pH 5 acetate buffer is used in a 18-hr extraction test (Table 4-2).

A comparison of WET and TCLP, using California wastes and waste composites, was provided by DTSC in the Regulatory Structure Update Extraction Test Project Summary Report contained in the DTSC report (DTSC 1998a, p 1078). WET consistently extracted more of 10 elements than TCLP (Table 4-3), with the exception of one mercury result. For several waste-element combinations, WET extract concentrations exceeded TCLP extract concentrations by 1 to 2 orders of magnitude. The major difference in the two extraction procedures is that the citrate buffer in WET leads to chelation of some elements (e.g., lead), and the direct release of elements bound in the solid-phase by dissolving high content metals (e.g., iron) by chelation (DTSC, 1998a). Although WET extracted more of the test elements than TCLP, a comparison of results with municipal solid waste leachate (MSWL) indicated that WET is generally more exhaustive than TCLP, leading to significant overprediction of what is actually present in the leachate for many elements (Table 4-3). On balance, TCLP gave a better representation of what actually leaches from these landfills for most, if not all, elements. Thus WET generally overestimates what leaches out of landfill waste over the lifetime and post-closure period of a landfill, whereas TCLP's results in leaching simulation are more in line with observed leaching behavior. In subsequent tests, citrate, which is used as a buffer in WET but not in TCLP, has not been found to be a constituent of leachate from California landfills. For these reasons, and for the sake of harmonizing with EPA by requiring that only one test in California, DTSC has proposed replacing WET in favor of TCLP for its non-RCRA solid hazardous waste classification testing program.

For an exact simulation of landfill leachates, neither WET nor TCLP provides satisfactory performance for oily wastes, for volatiles that might reach groundwater by diffusion, or for some elements occurring as oxyanions, such as arsenic, chromium, molybdenum, and selenium. Also, neither test adequately addresses questions of speciation for chemicals that can exist in more than one form, such as element, salt, and anion. WET overestimates the leaching potential for many elements in representative California landfill wastes, but there are several exceptions to this, such as cadmium, nickel, and thallium.

Based on the shortcomings of both the WET and the TCLP, and the

fact that both test procedures have beneficial features (exhaustive extractability of WET, simulation more reflective of actual leachate content and acceptability of TCLP), the committee supports the development of a single test protocol to classify California's hazardous waste, and to do so in harmony with the classification test of EPA. Such a test should provide results that can be related to field-realistic exposures, including the uncertainties associated with leaching pathways in the field. Understandably, DTSC may choose not to pursue this effort alone

based upon limitations in needed resources. Such an effort would also be quite time-consuming because of its nationwide implications and the need for extensive testing and validation under a variety of Waste, climatic, and soil conditions. The committee recognizes that the TCLP has nationwide status, use, and acceptability. Harmonizing California's extraction test with that required by EPA would minimize the testing burden on waste disposers in California, who would need to conduct only the TCLP. However, DTSC has not yet provided convincing arguments, either to the committee or, based on written comments, to stakeholders, for the sole adoption of TCLP and elimination of WET. The committee recommends that DTSC conduct an open evaluation of the experimental evidence, including the results of side-by-side testing and the opinions of its own staff, federal EPA counterparts, and stakeholders, before reaching a conclusion on the three possibilities before it: (1) adopt the TCLP as the sole test; (2) continue requiring both the TCLP and the WET; or (3) develop a new test that overcomes the deficiencies of both the TCLP and the WET.

There is one very important aspect of the use of either WET or TCLP assays that DTSC has overlooked in its modeling effort and that it should bear in mind in further evaluations. DTSC is currently using these assays as though they exactly match field conditions; indeed, much of the argument and experimental program has gone into the evaluation of how well each of them matches field conditions. However, a probabilistic approach needs to explicitly introduce the uncertainties in extrapolations such as those from a laboratory assay to the field, and this DTSC has failed to do in considering either the WET or the TCLP methods. Either assay could be used in a probabilistic procedure, although each would have different uncertainties associated with its use. DTSC has expended much effort in a commendable experimental evaluation of WET and TCLP, and the experimental results appear to provide a suitable basis for evaluating the uncertainties associated with the results of those assays with leaching in field conditions. The different biases and/or larger uncertainties associated with certain types of chemicals, or certain types of waste stream, can be built into the probabilistic modeling.

For the (DTSC-designated) category 2 elements of arsenic, antimony, molybdenum, selenium, and vanadium, DTSC proposes to use unadjusted TFLCs for arsenic, molybdenum, and antimony, and to use either the detection limit or develop a new test for selenium and vanadium. Use of the detection limit has the disadvantage of being driven by the state of analytical methodology rather than risk, contrary to the aim

of the DTSC program. Also, it is somewhat arbitrary, for example, in its use of the analytical limit of detection (LOD) rather than analytical limit of quantitation (LOQ) or twice the LOQ. Similarly, DTSC proposes to use twice the estimated quantitation level (2X EQL) in lieu of a SERT when the calculated concentration of the SERT is less than the EQL. In both cases, the committee emphasizes that there is no connection between the sensitivity of chemical analytical methods and the sensitivity of biological receptors, thus, the use of 2X EQL to establish a SERT is also not risk-based.

Analytical methods are continually being improved as new instrumental and other techniques are introduced, and detection limits vary from laboratory to laboratory and sample to sample. Detection limits, and limits of quantification, may be influenced by background. This needs to be taken into account when analyzing for naturally-occurring substances (mercury, selenium, cadmium, etc.), which may vary in background concentration from location to location. It also needs to be taken into account for organic contaminants for which the matrix may contain substances that mimic or interfere with the analyte of interest. This matrix effect may also vary from sample to sample and location to location. Biological sensitivity is fixed by the inherent toxicity of the analytes and response of the organism being exposed to the analyte under specified conditions.

Comparative testing to determine if the use of detection limits as proxy values is protective under reasonable exposure scenarios is lacking. It might turn out that such proxy values are protective, but this can not be determined from the information provided to the committee. DTSC should undertake, for example, a comparison of the SERT values with the EQLs. This should be done by evaluating a range of compounds with different toxic potencies and EQL values to determine the degree of protectiveness.

The proper evaluation of the potential adverse health effects of a substance requires knowledge of the chemical and physical properties of the test material; anticipated human exposure conditions, including

environmental levels, duration, pathway(s) and populations; the nature of the anticipated acute and immediate effects or delayed or chronic effects; and (usually) at least one appropriate nonhuman (e.g., animal) model. Only acute toxicity tests will be addressed in this section.

The interaction between the assessment of risk from acute toxicity and the assessment of risk from chronic toxicity is not entirely clear from the flow chart in the DTSC documentation (DTSC 1998a, p. 36). From the figure, one would assume that the first screen is for chronic toxicity as assessed by the development of TTLCs and SERTs, followed by assessment of risk from the acute toxicity of the chemicals based on acute toxicity assays. However, only 38 chemicals have passed through the first (TTLC) screen and, as noted in previous chapters, there does not appear to be a clearly defined method to either add or delete chemicals from either the TTLC or SERT lists. The chronic toxicity risk assessments are based on reference doses or concentrations, or cancer potency factors that were designed to protect the general population, including sensitive subpopulations. Thus, the thresholds developed based on chronic low-dose exposures of the general population would be expected to be much lower than the thresholds that might be developed for acute toxicity based on almost any acute exposure scenario.

The acute oral toxicity thresholds are based on doses or concentrations calculated to be lethal for half of the test animals (LD₅₀ or LC₅₀ values), divided by a safety factor of 100 and multiplied by an estimated ingestion "rate". The rate given is 5 mg/kg of body weight for children. This is not a rate but rather a dose, although the value is said to be derived from a percentile of the CalTOX parameter corresponding to a rate (the soil ingestion rate). A rate would be a dose per day or some other unit of time. Because only a dose is given, it appears that the threshold is designed to protect someone who, in a one-time, or at least infrequent, situation, actually eats the waste directly, but does not eat it on a daily or regular basis. The acute toxicity threshold so derived could be considered to be protective against lethality for such a one-time ingestion event, but would not necessarily be protective against more subtle toxicity, particularly if the ingestion occurred on a repeated basis. DTSC needs to clarify the purpose of the acute toxicity thresholds, who is to be protected by these thresholds, and whether it expects the exposures to occur one-time or be repeated.

For the acute dermal toxicity thresholds, DTSC provides a better description of the parameter values used in the derivation of the thresh-

olds and correctly uses a dermal contact rate (in millgrams per kilogram per day). A minor point is that the DTSC text refers to oral LD₅₀ values rather than dermal values (DTSC 1998a, p. 72).

DTSC presents the use of the acute oral and dermal toxicity thresholds as though they are based on various acute exposure scenarios (DTSC 1998a, pp. 72–74). This is a reasonable approach, but DTSC has not presented clearly defined goals and appropriate scenarios to meet those goals. For example, is it DTSC's intent to protect the most sensitive subpopulation from death if, in a one-time situation, a member of that population wanders on site and eats the waste? Or is it to protect a person who occasionally wanders on the site and eats the waste once a week? Or is it to protect those who live near the waste site and might inhale vapors and particles emitted from the site on a daily basis? The use of oral and dermal LD₅₀ values is apparently for scenarios in which there is a high-end ingestion or dermal contact with raw waste streams by a child. The parameters selected for oral ingestion and dermal contact rates fail in this purpose, however, through an error of mistaken identity. What are used are upper percentiles of ingestion and dermal contact rate parameters derived for use in CalTOX; but the distributions of those rates for CalTOX should correspond to the variabilities between individuals in long-term average rates. What are required for the acute scenarios are distributions that also include day-to-day variability for individuals. However, because the scenarios are not adequately described, it is not clear with what frequency the estimated dose will be consumed.

For the acute inhalation thresholds, no exposure scenario is presented, merely a rationale that corresponds to a highly unlikely, and maybe impossible, situation. For vapors, the assumption appears to be that persons could be exposed to vapors in equilibrium with fresh waste undepleted by off-gassing (the committee assumes that the temperature of 250°C specified on page 73 is a misprint for 25°C). Although a scenario for a waste worker might be constructed in which such a situation is possible, it is doubtful that there are any such situations involving the general public; moreover, workers should be protected at lower levels by Occupational Safety and Health Administration (OSHA) standards. The basis for the rationale for the particulate inhalation thresholds is even less secure. What scenarios can DTSC suggest that would result in acute exposures that are limited to the OSHA time-weighted average standards, or the long-term National Ambient Air Quality Standard for particulate matter?

The committee found incorporation of different safety or uncertainty factors for the different acute thresholds also to be questionable. For the oral and dermal exposures, a safety or uncertainty factor of 100 is incorporated; for particles, a safety factor of 10 is included, and for vapor exposures, no safety factor is proposed at all. As for the TTLC and SERT derivations, a consistent approach requires DTSC to make explicit its protection goals, and then to evaluate scenarios with parameter values that correspond to those goals.

It is not appropriate to use only acute toxicity tests and short-term exposure scenarios, rather than chronic toxicity or long-term exposures, as the basis of risk assessments for waste classification and disposal. Both types of information have an important place. For chemicals for which there are no TTLC or SERT values, the risk assessments should not be based solely on acute toxicity, that is, by using bioassays with the crudest of endpoints, lethality if chronic toxicity data are available. Such an approach would not give any consideration to reproductive and developmental toxicity, any chronic toxicity (including cancer) or genetic toxicity. At the very least, DTSC should review readily available chronic or other effects data (genetic toxicity, reproductive and/or developmental toxicity) for each of the waste components and compare the concentrations of the components in the waste with the concentrations found to cause no, low, or infrequent effects. Possible sources of such chronic effects information include EPA's maximum contaminant levels for drinking water, oral reference doses or inhalation reference concentrations, or cancer potency factors, and the Agency for Toxic Substances and Disease Registry's minimum risk levels. These values are easily accessed for numerous chemicals and most have been subject to scientific peer review.

If only acute toxicity data are available for the risk assessment, DTSC should follow standard practice and use an additional uncertainty factor to account for the lack of data regarding potential chronic toxicity at concentrations that are lower than those causing acute toxicity. DTSC should take into account the slope of the dose-response curve for acute toxicity data when choosing the uncertainty factor, if such data are available. Failure to use an appropriate uncertainty factor my seriously underestimate the risks associated with chemicals for which only acute toxicity data are available and may result in unprotective thresholds. If the waste contains several chemicals for which chronic toxicity data are available (e.g., several polycyclic aromatic hydrocarbons, which act at common sites to exert their toxicity) then the additive, synergistic, or

antagonistic effects of these chemicals, if known, should also be considered when assessing the risk posed by the waste.

It appears that if a waste does not contain any of the TTLC or SERT chemicals, and is classified as nonhazardous on the basis of its acute toxicity, it is not subject to further scrutiny but may be disposed in nonhazardous waste landfills or by other methods such as recycling or land application. As a result, a waste that may pose serious chronic or mutagenic risks at concentrations far below those that cause acute effects and where long-term exposure may be expected as a result of its disposal, may be inappropriately classified as nonhazardous using DTSC's current or proposed classification system. In essence, the use of acute effects data permits higher (less conservative) risk thresholds for wastes than would be possible if chronic effects data for chemicals without TTLCs or SERTs were required. It appears to the committee that the current and proposed DTSC methods provide distinct disincentives for the identification of chronic effects data for particular wastes or waste constituents, since any such identification is likely to result more stringent regulation.

DTSC should also consider the inclusion of respiratory, ocular, and dermal irritation testing as well as allergic sensitization testing, in its battery of acute toxicity tests. The nuisance factor of odors may also have to be taken into account to meet some goals. Members of a community living close to a waste site are more likely to be aware of and concerned about acute effects related to the irritant and odor properties of the waste than any other type of toxicity. Respiratory irritation might exacerbate existing health conditions such as asthma. If more than a single short-term exposure is anticipated (e.g., in waste workers or those living near a waste-disposal site), the potential for sensitization (allergenicity) may be relevant.

A further problem related to the acute and chronic effects of specific chemicals is that the DTSC approach does not take into account the speciation or chemical form of metal contaminants. This is an arbitrary simplification that is not based on true risks. For example, chromium (III) at low doses is an essential nutrient for humans, whereas chronic exposure to chromium (VI) has been associated with lung cancer in humans; the toxic effects of elemental chromium are relatively unknown. Some consideration of the species and chemical form of the metal contaminants present should be attempted for both acute and chronic risk assessments.

DTSC proposes to protect aquatic organisms by classifying wastes using acute lethality to fish. Two thresholds based on acute lethality (96-hour LC₅₀) to fish of extracts are used to establish the category to which a waste will be assigned. The first threshold, at an LC₅₀ of 30 mg/L, is used to classify a waste as hazardous, and the second threshold, at an LC₅₀ of 500 mg/L, is used to distinguish between nonhazardous and special wastes. The 30-mg/L value is derived from 500 mg/L divided by 18, the 10th percentile estimate for the liner protection factor. The current threshold for classifying waste as hazardous is based on a 96-hour LC₅₀. value of 500 mg/L (22 California Code of Regulations § 66261.24 (a)(6)). DTSC proposes to retain this regulation but to use only a fish acute lethality bioassay to bring wastes into the lower tier of hazardous waste. It appears that even if a waste was not classified as hazardous based on comparing its concentration with a TTLC, it could still be classified as hazardous based on the results of the fish acute lethality test. It is unclear from the DTSC document if or how SERTs will be applied in the classification of wastes in the ecological scenario. It would appear that for wildlife, total concentrations of chemicals in wastes will be compared with TTLCs and a fish acute lethality test will be performed. The fish acute lethality test does not include the potential for bioaccumulation or biomagnification and would not be useful for compounds that are chronically toxic and have great acute to chronic ratios.

The proposed methodology assumes that fish are the most sensitive aquatic organisms. This is certainly not always the case, for some aquatic organisms are more sensitive than fish to a number of compounds. Thus, the proposed screening methodology might not sufficiently protect aquatic life or wildlife that eat aquatic organisms. Also, DTSC does not specify how a waste would fail the bioassay test. Presumably, if the TCLP leachate causes greater than 50% lethality of the fish, the waste will be classified as hazardous. The committee concludes that the use of an acute bioassay using fish would not be sufficient to protect aquatic organisms or animals that might eat aquatic organisms.

Retaining consideration of aquatic toxicity in the screening system is appropriate and is supported by the committee; however, the selection of a threshold value of an LC₅₀ for TTLC for listing a material as hazardous is considered to be somewhat arbitrary and has no scientific justifica-

tion. This is not a risk-based approach. For the approach to be risk-based, DTSC must consider exposure and dose or concentration simultaneously when establishing a risk threshold. The risk presented by a waste is a function of exposure concentration and a threshold for acute effects; thus, setting a single value for a threshold is inappropriate. Although a single value might be predictive, it has not been demonstrated that it will be protective relative to possible aquatic concentrations.