Risk Analysis, Vol. 21, No.

1, 2001

Practical Methods for Meeting Remediation Goals at Hazardous Waste Sites

Terry W. Schulz1* and Susan Grifn2

Risk-based cleanup goals or preliminary remediation goals (PRGs) are established at hazardous waste sites when contaminant concentrations in air, soil, surface water, or groundwater exceed specied acceptable risk levels. When derived in accordance with the Environmental Protection Agencys risk assessment guidance, the PRG is intended to represent the average contaminant concentration within an exposure unit area that is left on the site following remediation. The PRG, however, frequently has been used inconsistently at Superfund sites with a number of remediation decisions using the PRG as a not-to-exceed concentration (NTEC). Such misapplications could result in overly conservative and unnecessarily costly remedial actions. The PRG should be applied in remedial actions in the same manner in which it was generated. Statistical methods, such as Bowers Condence Response Goal, and mathematical methods such as iterative removal of hot spots, are available to assist in the development of NTECs that ensure the average postremediation contaminant concentration is at or below the PRG. These NTECs can provide the risk manager with a more practical cleanup goal. In addition, an acute PRG can be developed to ensure that contaminant concentrations left on-site following remediation are not so high as to pose an acute or shortterm health risk if excessive exposure to small areas of the site should occur. A case study demonstrates cost savings of ve to ten times associated with the more scientically sound use of the PRG as a postremediation site average, and development of a separate NTEC and acute PRG based on the methods referenced in this article.
KEY WORDS: Remediation; preliminary remediation goal; cleanup goals; not-to-exceed concentration; condence response goal; iterative removal

1. INTRODUCTION Preliminary remediation goals (PRGs) are established as risk-based targets to use during analysis and selection of remedial alternatives when assessing risks from hazardous wastes at U.S. Environmental Protection Agency (USEPA) Superfund sites.(1) Chemical-specic PRGs are concentration goals for individual chemicals that exist in contaminated media and are associated with levels of risk that will
1 2

Golden, CO. U.S. Environmental Protection Agency Region 8, Denver, CO. *Address correspondence to Terry W. Schulz, 1218 Pomegranate Lane, Golden, CO 80401-9222; Tschulz-golden@worldnet.att.net.

be adequately protective of human health, ecological health, or both. PRGs are calculated by simply inverting the mathematical equations used in the baseline risk assessment process and determining the contaminant concentration that corresponds to an acceptable level of risk.(1) One key concept for developing PRGs is that the cleanup level should be interpreted in the same manner as the concentration term used in the risk assessment. In the risk assessment, media-specic data are averaged within a spatial area, known as the exposure unit. The exposure unit bounds an area within which a human or ecological receptor moves and contacts contaminated media at random during the exposure 43
0272-4332/01/0200-0043$16.00/1 2001 Society for Risk Analysis

44 duration period. Because of the uncertainty associated with estimating the true average concentration of a chemical at a site, the USEPA recommends that the 95% one-sided, upper condence limit of the arithmetic mean (95 UCLM) of the sample data be used to represent the exposure unit concentration term, rather than the arithmetic mean.(2) Therefore, the PRG should be interpreted as the 95 UCLM of the contaminant concentration within the appropriate exposure unit following remediation. After remediation, some of the contaminant concentration left onsite may exceed its PRG, but the average concentration associated with acceptable long-term exposures should be at or below its PRG. Unfortunately, the PRG has been inconsistently used at Superfund sites. A number of remediation decisions have not utilized the PRG as an average concentration over an exposure unit area, but as a notto-exceed concentration (NTEC). This necessitates the remediation of every sample location where the contaminant concentration exceeds the PRG. This approach to remediation may result in overly conservative and unnecessarily costly clean-up efforts depending on the size and complexity of the site. The use of the PRG in the manner in which it was derivedas an average site concentration for chronic exposureis the most scientically sound application since the health concerns are for long-term exposures. A risk manager must deal with several issues regarding this approach, however. First, over what size area are the original data averaged to compare to the PRG? For example, is it appropriate to average data over 1 acre, 100 acres, or 1,000 acres? Second, which sample locations (i.e., contaminant concentrations at sample points) should be remediated for the site average to be at or below the PRG? Finally, is there a possibility of adverse acute hazards remaining from short-term exposures to those contaminant levels left in an exposure unit on-site that exceed the PRG? It is conceivable that the average contaminant concentration within an exposure unit on a site can be at or below the PRG, yet some smaller areas within the exposure unit may contain concentrations that could pose a shortterm risk to receptors. A practical approach that addresses these concerns is the development of two secondary remediation goals: (1) an NTEC, and (2) an acute-risk PRG applied to any discrete sample point occurring within that exposure unit. Two methods for calculating NTECs and a method for estimating an acute PRG are investigated in this article using actual arsenic and lead data from two EPA Superfund sites. To illustrate the economic con-

Schulz and Griffin sequences of misusing the PRG as an NTEC, a case study is conducted that estimates and compares removal costs and waste volumes for a Superfund hazardous waste site from applying the PRG as the NTEC versus methods described in this article.

2. DATA DESCRIPTION The data source for Pallas Yard data is the health risk assessment for Pallas Yard in Murray, Utah.(3) The Pallas Yard site in Utah is an active railroad siding originally developed around the turn of the century to support the smelter operations in the area. The Pallas Yard site is an area approximately 170 feet wide 4,400 feet long. Fill material from unknown sources was imported as rail bed material. Elevated concentrations of arsenic and lead have been detected in the surface and subsurface soil at the site. Railyard activities are still ongoing, and the risks to current railyard workers are evaluated in the risk assessment. Surface soil (06 inches) lead concentrations from 54 samples collected in 1993, 1996, and 1997 constitute one of the data sets for the present analysis. All of the sample concentrations were detected values; no datum was censored. The data source for three residential neighborhoods in Butte, Montana, is the arsenic human health risk assessment for the Butte Priority Soils Operable Unit.(4) Butte, historically, has been an important mining, milling, and smelting district. The risk assessment uses surface soil data (06 inches) from the risk assessment, and residential yard surface soil data (0 2 cm) from the ButteSilver Bow environmental health lead study(5) that are within the study area. These data are used in the risk assessment to assess risks to residents from elevated levels of lead and arsenic in the soil. Data from the risk assessment used in this article consist of 36, 84, and 34 arsenic sample concentrations from the Lower West Side, 2nd Street, and Timber Butte neighborhoods, respectively. The number of censored samples are 0, 10, and 1, respectively. Censored samples are replaced with half their reported method detection limit for all calculations. 3. CONCEPT OF THE EXPOSURE UNIT Before averaging site data for the purpose of evaluating exposure and risk (and even prior to sample plan design and sample collection), it is important to understand who is being exposed and how they are

Meeting Remediation Goals exposed. Typically, the boundaries of a hazardous waste site are operationally dened by the extent of contamination or by the property boundary lines. Sites may range from less than 1 acre in size to hundreds of square miles. The area within these boundary lines may or may not adequately represent the area where humans or populations of ecological receptors are being exposed and thus over which the environmental data should be averaged to assess chronic exposure and risk over area and time. For this reason, it is important to understand the risk assessment concept of the exposure unit. The exposure unit is dened as the geographic area where exposures occur to the receptor of concern during the time of interest. If the receptor of concern is a resident, she or he will most likely be exposed to contaminated media in a smaller geographic area than a recreational or a migratory ecological receptor. For example, it is not unusual to average spatial data within a 1- to 5-acre exposure unit for residential receptors and a 20-acre exposure unit for recreational and ecological receptors. A smaller exposure unit may be warranted when concerns exist for a contaminant that may elicit adverse effects in a shorter period of time, such as might occur with acute or subchronic exposure events for more potent toxicants. At the Pallas Yard site, the receptors of concern are the railyard workers. The area within the site boundaries is considered to be a plausible exposure unit for long-term exposures to these workers. Therefore, exposure is averaged over the entire site. At the Butte, Montana, site the health concerns are for long-term exposures of children and adult residents to arsenic in soil; therefore the large site is subdivided into smaller neighborhoods and the soil data are averaged separately within each of these smaller exposure units. A key issue in the concept of the exposure unit and the application of PRGs and NTECs to an exposure unit is the quality of the data collected. From a risk assessment perspective, this focuses on how well the data collected represent what the targeted receptor contacts over the specied time period. The USEPA advocates a data-quality-objectives process to ensure adequate site characterization.(6) This process involves the development of site conceptual models to identify contamination sources, contaminated media, and all exposure pathways; the establishment of sampling and remedial objectives; and the methods to achieve those objectives. This data-quality-objectives process can provide measures of condence in (1) the usability of the data within each exposure unit and (2) the representativeness and appropriateness of the

45 PRGs and NTECs calculated from those data for the specic site conditions. 4. METHODS Prior to performing methods to estimate NTECs and acute PRGs, simple descriptive statistics are calculated and probability distribution tting is performed for each of four data setsthe Pallas Yards railyard site and the three Butte residential neighborhoods. The remainder of this section describes the two methods that have been used to calculate NTECs and the method for calculating an acute PRG. The authors acknowledge that additional methods for calculation of NTECs may be possible. 4.1. Bowers Method Bowers, Shifrin, and Murphy(7) presented a method for calculating the condence response goal (same as NTEC) for a lognormally distributed data set. The derivation is involved and the interested reader should see the original paper for further details. Equation (3) of Bowers et al. is solved for the NTEC (c*), found by trial and error or numerically, as the value that results in the minimum absolute difference between the estimated reduction in exposure ( ) and the desired reduction in exposure (PRG/95 UCLM). Equation (3) from Bowers et al. is repeated here: ( 1 2F [ ( 1/ln )ln ( c*/ ) ln ] ( ln ) 2 ( c o )e { 1 2F [ ( 1/ln )ln ( c*/ ) ] } )
2

where F(z) the area under the standard normal curve, the one-sided upper 95% condence limit on the geometric mean (95 UCLGM) [Equation (5) in Bowers et al.(7)], the geometric standard deviation (GSD) of the lognormal distribution dened by the 95 UCLM [Equation (4) in Bowers et al.] and 95 UCLGM (i.e., exp{[2 ln(95 UCLM) ln(95 UCLGM)]1/2}, Equation (7) in Bowers et al., and the contaminant concentration in clean ll.

co

Equation (3) can be solved for the NTEC as a function of , , and . Implementation of the method in a spreadsheet calculation or compiler language is straightforward. An example from Bowers et al.(7) provides a check on proper NTEC calculation. If the PRG 4.5 mg/kg, 95 UCLM 9.657 mg/kg,

46

Schulz and Griffin compared with the PRG. The process is iterated until the recalculated exposure unit average is less than or equal to the PRG. The largest sample concentration remaining in the data set is the NTEC. A spreadsheet is convenient for performing the IR method. Note that if the site average is less than PRG, no remediation for long-term, nonacute exposure is required.

4.3. Calculating an Acute PRG Calculations of NTECs should also consider the probability of adverse effects from short-term or acute exposures. If a receptor preferentially spends their time only in a small area of the exposure unit (such as a play area) where concentrations were much greater than the chronic PRG, an acute PRG should be calculated such that no adverse effects would occur from that acute exposure. At the Butte, Montana, site a chronic PRG of 260 ppm for arsenic in soil was developed to be protective of long-term exposure. This PRG is documented in the Butte priority soils human health risk assessment,(4) and was calculated using the methodologies recommended in the USEPAs Risk Assessment Guidance for Superfund(1) and the site-specic exposure assumptions described in the Butte human health risk assessment previously mentioned. This chronic PRG represents the average arsenic soil concentration residents could be exposed to over their lifetime and not exceed a 1in-10,000 chance of coming down with cancer, which is considered to be protective under USEPA guidelines. At the Butte site, however, several of the residential homes are located next to railroad tracks where high concentrations of arsenic have been detected in the soil, and a concern exists for short-term exposures to children playing in those small areas of high contamination. An acute PRG of 2,564 ppm for arsenic in soil, which would result in an acceptable short-term exposure, was calculated using the sitespecic exposure assumptions described in the Butte human health risk assessment for children 1 to 6 years of age(4) using a shorter exposure duration (e.g., 6 years) and the noncarcinogenic subchronic reference dose for arsenic (Robert Benson, EPA Region 8, memorandum to Chris Weis, Subchronic Reference Dose for Arsenic, September 12, 1995). The acute PRG equation with exposure assumptions is shown in Table I. The other major difference is the exposure unit area over which this value could be applied. Since the concern is for short-term exposures to smaller preferential locations, an exposure

Fig. 1. Chemical of concern (COC) lognormal probability density function, example illustrating Bowers et al.(7) method. Co COC conc. in clean ll; PRG preliminary remediation goal; GM geometric mean; UCLGM upper 95% condence limit on GM; M arithmetic mean; UCLM upper 95% condence limit on M; NTEC not-to-exceed conc., all sample locations greater than NTEC are remediated.

6.062 mg/kg, and co 0.1 mg/kg, one should obtain 2.625 mg/kg, 0.466, and NTEC 14.1 mg/kg. Figure 1 graphically depicts the lognormal probability density function and where various metrics including the NTEC occur for the preceding example. Note that this method of NTEC calculation is not appropriate for data that are not lognormally distributed. The Shapiro-Wilk test for normality of logtransformed data is a stringent test for lognormality. The Shapiro-Wilk test for normality of the natural logs of the data is performed for sample sizes of 50 or smaller. The Shapiro-Francia test is performed for sample sizes larger than 50. In either case a p 0.05 indicates the data are not lognormally distributed. 4.2. Iterative Removal of Hot Spots Method The NTEC can be calculated by iterative removal (IR) of hot spots, assuming the exposure unit average concentration is greater than the PRG. The data are rst ordered from smallest to largest concentrations. Then the highest concentration datum is removed and replaced with the concentration in clean ll. A new exposure unit average concentration is calculated and

Meeting Remediation Goals

Table I. Calculation of Acute Preliminary Remediation Goals
Arsenic at the Butte, Montana site PRGA PRGA TR AT RfDA CF EF IRc EDc Bwc BAF TR AT RfD A ( CF EF ) ( IR c Ed c /BW c ) ( BAF ) Acute preliminary remediation goal (2,564 ppm) Target risk (hazard index 1) Averaging time (365 days 6 years) Acute reference dose (0.006 mg/kg-day) Conversion factor for soil (1E-06 kg/mg) Exposure frequency (350 days/year) Soil ingestion rate for child (200 mg/day) Exposure duration during ages 1 to 6 years (6 years) Body weight for ages 1 to 6 years (15 kg) Site-specific bioavailability of arsenic in soil (0.183)

47 blood lead level used in the Pallas Yard risk assessment for long-term exposureas a level that should not be exceeded for a short-term exposure to a railyard worker. This value and the site-specic exposure assumptions of the Pallas Yard risk assessment(3) were used to calculate an acute PRG of 37,000 ppm lead. The equation and exposure assumptions are shown in Table I. As mentioned above, since the concern is for short-term exposures to smaller preferential locations, a single sample point was considered to be a plausible spatial area over which to apply this PRG. One shortcoming of this acute PRG approach is that the USEPA has not developed very many acute or subchronic toxicity values, thereby limiting its application. It is anticipated that for actual cleanups, consideration of acute toxicity will be a factor at only a small number of sites. 5. RESULTS Exploratory data analysis results are presented in Table II. The large differences between means and medians for all four data sets and the high skewness for three of four data sets indicate nonnormality. The Shapiro-Wilk or Shapiro-Francia test probabilities (S-W-P or S-F-P) indicate only the 2nd Street and Timber Butte data sets should be considered lognormally distributed. Results of application of the two NTEC methods and the acute PRG method to actual site data are presented in Table III. The Pallas Yard chronic PRG of 8,400 ppm was based on acceptable long-term exposures to lead in soil for the railyard workers using the USEPAs adult lead model.(3) If that PRG is misused as an NTEC, then 25 out of 54 sampled locations are indicated for remediation, assuming a clean ll concentration of 100 ppm lead. The Bowers et al.(7) method should not be applied to these data, because they are not lognormal (S-F-P 0.00). The IR NTEC indicates three locations for remediation. If the three sample locations with the greatest lead concentrations are remediated, the largest unremediated sample concentration is 37,000 mg/kg lead. Assuming the data set is representative of lead concentrations for the entire exposure unit, the IR NTEC value is considered to be 37,000 mg/kg lead. An acute PRG of 37,000 mg/kg lead, calculated as described in Section 4.3 and as shown in Table III, indicates four locations for remediation. Since the possibility of acute hazards to smaller preferential locations is an important concern for the risk manager, and the use of the IR method may not be protective in this case, the risk manager

Lead at the Pallas Yard, Utah site PRGA PRGA PbB95 fetal R GSDi PbBo BKSF IR AF EF [ PbB 95 fetal / ( R GSD i ) ] PbBo BKSF ( IR AF EF /365 )
1.645

Acute preliminary remediation goal (37,000 mg/kg) 95th percentile blood lead in fetus (30 g/dl) Mean ratio of fetal to maternal blood lead (0.9) Individual geometric standard deviation (1.54) Baseline blood lead value (2.28 g/dl) Biokinetic slope factor (0.4 g/dl per g/day) Soil ingestion rate (0.05 g/day) Absolute absorption fraction of lead in soil (0.12) Exposure frequency (57.5 days)

unit even as small as a single sample point may be plausible. At the Pallas Yard site, railroad workers were the exposed population of concern. A chronic PRG of 8,400 ppm for lead in soil was calculated using the methodology recommended in USEPAs Recommendations for an Interim Approach to Assessing Risks Associated with Adult Exposures to Lead in Soil(8) and the site-specic exposure assumptions described in the Pallas Yard health risk assessment.(3) The chronic PRG represents the average lead soil concentration at which there is no more than a 5% probability that a pregnant workers blood lead level will exceed 10 g/dl in the fetus, which is considered to be protective under USEPA guidelines. Preferential exposures to small areas of high contamination are possible given the nature of the railroad workers job; therefore, an acute PRG for lead in soil was developed. This acute PRG of 37,000 mg/kg lead was developed using the Occupational Safety and Health Administration (OSHA) occupational blood lead standard of 30 g/dlas opposed to the 10 g/dl

48
Table II. Surface Soil Statistics and Lognormal Distribution Fitting Results
Data/statistics Chemical-risk driver Units of concentration Sample size (n) (Number of nondetects) Minimum concentration Maximum concentration Median concentration Mean arithmetic concentration Standard deviation of mean concentration Coefficient of variation Skewness Shapiro-Wilk probability (n 50) Shapiro-Francia probability (n 50) Lower West Side Arsenic mg/kg 36 (0) 11.8 1,167 232.2 316.4 282.6 0.9 1.1 0.03 2nd Street Arsenic mg/kg 84 (10) 34.8 2,070 195.8 245.9 260.1 1.2 4.8 0.36 Timber Butte Arsenic mg/kg 34 (1) 69.4 1,186 175.6 265.8 228.9 0.9 2.2 0.20

Schulz and Griffin

Pallas Yard Lead mg/kg 54 (0) 34.0 66,000 6,800 10,536 12,739 1.2 2.0 0.00

should select four locations for remediation based on the acute PRG. The chronic PRG of 260 ppm at the Butte, Montana, site is based on acceptable long-term exposures (e.g., 1E-04 cancer risk level) to arsenic in soil for residents.(4) If that PRG is misused as an NTEC for the Lower West Side neighborhood, then 16 out of 36 sampled locations are indicated for remediation, assuming a clean ll concentration of 16.3 ppm arsenic (average arsenic concentration from actual ll analyses). The data are not lognormally distributed (S-W-P 0.03), so the Bowers et al.(7) NTEC is inappropriate.

The IR NTEC indicates remediation for three sample locations, which results in the greatest unremediated sample concentration and IR NTEC value of 741 mg/ kg arsenic. Application of the acute PRG of 2,564 ppm would indicate that no adverse effects would be expected to occur and no remediation of the exposure unit is required. This acute PRG is greater than the calculated IR NTEC, indicating that the IR NTEC goal is protective for adverse, short-term health effects as well as for chronic, long-term health effects. Misapplication of cleanup to the chronic PRG for the Butte 2nd Street neighborhood indicates re-

Table III. Not-to-Exceed Concentrations (NTECs) and Remediation Implications

PRG (mg/kg) 8,400 8,400 37,000 260 260 2,564 260 260 260 2,564 260 260 260 2,564 NTEC (mg/kg) 37,000 Locations requiring remediation (%) 25 (46) 3 (5.6) 4 (7.4) 16 (44) 3 (8.3) 0 (0) 28 (33) 2 (2.4) 0 (0) 0 (0) 13 (38.2) 1 (2.9) 1 (2.9) 0 (0)

Location, chemical Pallas Yard, Lead (n 54)

Method Clean all sample locales PRG Iterative removal Health-based acute PRG Clean all sample locales PRG Iterative removal Health-based acute PRG Clean all sample locales PRG Bowers et al.(7) Iterative removal Health-based acute PRG Clean all sample locales PRG Bowers et al.(7) Iterative removal Health-based acute PRG

Lower West Side, Arsenic (n 36)

741

2nd Street, Arsenic (n 84)

1,099 2,070

Timber Butte, Arsenic (n 34)

788 671

Note: PRG

preliminary remediation goal.

Meeting Remediation Goals

49

Fig. 2. Arsenic lognormal probability density function Butte 2nd Street neighborhood and Bowers et al.(7) method. Co Arsenic in clean ll; GM geometric mean; UCLGM upper 95% condence limit on GM; M arithmetic mean; PRG preliminary remediation goal; UCLM upper 95% condence limit on M; NTEC not-to-exceed concentration 1,099 mg/kg arsenic (not shown); Acute PRG 2,564 mg/kg arsenic (not shown).

Fig. 3. Arsenic lognormal probability density function Timber Butte neighborhood and Bowers et al.(7) method. Co chemical of concern concentrated in clean ll; GM geometric mean; UCLGM upper 95% condence limit on GM; M arithmetic mean; PRG preliminary remediation goal; UCLM upper 95% condence limit; NTEC not-to-exceed concentration 788 mg/ kg arsenic; Acute PRG 2,564 mg/kg arsenic (not shown).

mediation at 28 locations. The data may be considered to be lognormally distributed (S-F-P 0.36). The Bowers et al.(7) method indicates two locations for remediation, while the IR method indicates no locations for remediation. The data set mean is less than the PRG implying the value for the IR NTEC 2,070 mg/kg, the greatest arsenic concentration in the data set. Both NTECs are less than the acute PRG. In this case the risk manager could select two locations for remediation. Figure 2 graphically shows the lognormal probability density function and the Bowers et al.(7) method metrics. Misapplication of cleanup to the chronic PRG for the Timber Butte neighborhood indicates remediaton of 13 locations. The Timber Butte neighborhood data set exhibits lognormality (S-W-P 0.20). The Bowers et al.(7) method and the IR method agree that one location should be remediated. Since the acute PRG for exposure is greater than that calculated for the NTECs, the NTECs will be protective of acute or short-term risks. Figure 3 graphically illustrates the lognormal probability density function and the corresponding Bowers et al.(7) method metrics.

6. DISCUSSION Bowers et al.(7) state that their NTEC method is only appropriate if the data are lognormally distributed. They further state that results of their method can be checked by the IR method when the true mean and standard deviation of a lognormal distribution are known as a result of large ( 200) sample size. The claim is made that their mathematical approach has advantages over the IR method when the sample size is small. Results, not reported here, obtained from simulated synthetic lognormal data sets suggest both methods perform similarly for large and small lognormal data sets. The two methods yield varying results when the lognormal goodness of t of the data set is altered by changing a few large data values. Sensitivity analysis performed on the Timber Butte and 2nd Street data indicated the Bowers et al.(9) method is less sensitive to increases and decreases in the largest sample concentrations than the IR method. Both methods are less sensitive to increases than to decreases in the largest datum. In-

50 creasing the largest datum in the Timber Butte data set one order of magnitude had no remediation consequence. Decreasing the largest datum by 20% and 40% resulted in no required remediation by the IR method and Bowers et al.(7) method, respectively. More sensitivity was observed in the 2nd Street data. Increasing the largest and second largest values by 50% and 500% results in the IR method requiring remediation at one and two locations, respectively. Increasing the largest and second largest values one order of magnitude has no further effect on remediation. The Bowers et al. method was insensitive to the increases in 2nd Street data values. The Bowers et al. method was more sensitive to decreases in concentration of the two largest 2nd Street values, indicating one and no locations for remediation with arsenic concentration decreases of 10% and 40%, respectively. Cohen, Bowers, Lampson, Lampson, and Merrill(9) attempted to improve upon the Bowers et al.(7) NTEC approach by characterizing a condence region for the contaminant-level lognormal distribution parameters. The pair of parameters within the 95% joint condence region resulting in the lowest cleanup goal is claimed to ensure adequate protection with a specied level of certainty. The Cohen et al. method resulted in reported reductions in the Bowers et al. NTEC from 4 to 36.5% for 20 data sets. The Cohen et al. method probably addresses some of the uncertainty associated with small data sets and could be substituted for the Bowers et al. method. An underlying assumption in the development and application of NTECs is that the site data being used adequately represent random, long-term exposures of the targeted receptors to the exposure unit. As discussed in Section 3.0, condence in the usability of the data and the appropriateness of the PRGs and NTECs increase when EPAs data-quality-objectives process is followed during sample design and collection. This is not always the case, however. Often, sample data are biased with more extensive data collection in areas of known contamination (spills, storage tanks, pipelines, fueling depots, etc.) and sparser data collection over the remainder of the exposure unit. Ignoring this bias in the data will result in greater average calculated chemical concentrations within an exposure unit, which, in turn, lead to greater calculated risks and consequently to lower PRGs and NTECs than are actually required to meet a given risk-level goal (e.g., 1E-06). When applying acute PRGs and chronic NTECs as remediation goals, it is important to remember that completion of the remedial action according to

Schulz and Griffin the strategy does not necessarily mean that risks within the exposure unit have been reduced to the desired levels. Hence, postremedial conrmation sampling is recommended to ensure that the remedial action has been successful. When the portion of the contaminated media that exceeds the NTECs has been removed or remediated, both the postremedial conrmation samples and the previous samples from the unremediated locations should be used in a postremedial risk calculation to ensure that the remedial action objectives have been met. Similiar to the initial sampling process, a data-quality-objectives process should be followed for any postremedial sampling. The practice of conducting conrmatory sampling provides an added measure of condence that hot spots were not overlooked and that PRGs and NTECs are adequately protective of health. In addition, for contaminants that have degraded with time, fewer conrmatory samples will exceed NTECs resulting possibly in less remediation being necessary. It should be noted that no consideration is given to spatial correlation of contaminant data in this article and that the whole aspect of NTEC determination presented amounts to hot-spot remediation. If global spatial correlation is present in a data set, such as occurs for many smelter sites, then geostatistics should certainly be employed. Note that even if geostatistics are utilized, the risk manager still must decide what concentration contour to select as an NTEC, hence the subject of this article. The use of geostatistics may not be appropriate on all hazardous waste sites. One of the assumptions of geostatistics is spatial continuity: Sample pairs at close distances have more similar contaminant concentrations than sample pairs at greater distances. The Pallas Yard data exhibit high concentrations of lead in the imported rail bed soil and low concentrations adjacent to the rail bed, indicating that the assumption of spatial continuity has been violated (i.e., a mixture of at least two populations is present, a multivariate situation). Kriging is an averaging process. Kriging these data would incorrectly result in contours of signicant lead concentration far from the rail bed. On most hazardous waste sites sampling has been more extensive in areas of known contamination, usually in an attempt to determine the extent of contamination in these areas. Application of kriging to an exposure unit that has been preferentially sampled in hot areas will indicate more cleanup is required than is necessary to meet a given risk reduction objective, even when declustering techniques are used in an attempt to remove the sampling bias.

Meeting Remediation Goals 7. CASE STUDY A case study of the Butte, Montana, site has been undertaken to examine estimated removal costs and waste volumes associated with the misuse of the PRG as an NTEC versus the use of the PRG as an average and a separate NTEC based on the methods illustrated in this article. Note that the previously developed acute PRG for Butte is greater than all Butte NTECs (Table III) and therefore is not considered further. A detailed explanation of how costs and soil volumes have been obtained is given in CDM Federal Programs Corporation.(10) The location corresponding to each data point greater than or equal to the Bowers et al.(7) or IR NTEC is identied and the extent of the associated waste materials are estimated from historical sampling reports, eld notes, maps, and personal knowledge of the specic source areas. An average removal thickness of 18 inches is assumed. These estimated costs include the cost to excavate, load, and haul the waste/soil to a repository located 5 miles from the removal area. Waste volumes and costs associated with data points greater than or equal to the PRG ( 260 ppm) are estimated from hard data at known locations within the particular Butte neighborhood. Costs are rounded to the nearest $50.00. The results of those estimated removal costs and waste volumes are shown in Table IV. The misuse of the PRG as an NTEC results in estimated removal costs ve to ten times greater than costs associated with the Bowers et al. or IR

51 NTEC. Note also the large cost difference ($63,650 versus $0) attributed to only two sites in the 2nd Street neighborhood, emphasizing the importance of NTEC selection. 8. RECOMMENDATIONS The following steps are recommended to establish the numerical level of remediation for an exposure unit of a hazardous waste site. 1. Using the site-specic data, calculate a chronic PRG for the appropriate exposure unit according to the EPAs Risk Assessment Guidance for Superfund.(1) 2. Perform exploratory data analysis to include descriptive statistics and lognormal distribution goodness of t. 3. If the data are lognormal, determine the num-

Table IV. Not-to-Exceed Concentration Methods, Cost and Removal Implications

Estimated soil volume removed (cu. yds.) 13,300 4,450

Location, no. locations remediated

Method

Estimated removal cost ($) 107,500 31,100

Lower West Side 16 Clean all PRGa 3 Iterative removal 2nd Street 28 2 0 Timber Butte 13 1 1
a

Clean all PRGa Bowers et al.(7) Iterative removal

268,150 63,650 0

38,300 7,650 0 Fig. 4. Flowchart for organizing and documenting multiple decisions for selection of the not-to-exceed concentration (NTEC) of a containment on a hazardous waste site. PRG preliminary remediation goal; S-W-P test Shapiro-Wilk normality test for sample size 50; S-F-P test Shapiro-Francia normality test for sample size 50.

Clean all PRGa Bowers et al.(7) Iterative removal

79,400 5,450 5,450

11,350 800 800

PRG

preliminary remediation goal, rough estimates.

52 ber of locations requiring remediation by the Bowers et al.(7) NTEC. Determine the number of locations requiring remediation by the IR NTEC. Develop a PRG for acute or subchronic exposures if an acute or subchronic toxicity value is available. If the acute PRG is smaller than the Bowers et al. NTEC and IR NTEC, select the acute PRG as the NTEC. If the acute PRG is not smaller than the Bowers et al. NTEC and IR NTEC, and the data are not lognormal (S-W-P or S-F-P 0.05), select the IR NTEC. If the data are lognormal, select the Bowers et al. NTEC. REFERENCES

Schulz and Griffin

4. 5.

6.

7.

The basic ow of the process is shown in Fig. 4. This recommendation results in the development and consideration of three remediation goals: the chronic PRG, the acute PRG, and the NTEC. The average contaminant concentration of the exposure unit should be equal to or less than the chronic PRG following remediation. By remediating sample locations that exceed the selected NTEC, the risk-based objectives of the site cleanup will be achieved at minimal cost. ACKNOWLEDGMENT The views or opinions expressed in this article are those of the authors and do not necessarily represent USEPA policy or guidance. No ofcial endorsement is implied or inferred.

1. U.S. Environmental Protection Agency (USEPA). (1991, December 1). Risk assessment guidance for Superfund: Volume I. Human health evaluation manual Part B, Development of riskbased preliminary remediation goals. Washington, DC: USEPA, Ofce of Emergency and Remedial Response. 2. U.S. Environmental Protection Agency (USEPA). (1992, June 22). Supplemental guidance to RAGS: Calculating the concentration term. Washington, DC: USEPA, Ofce of Solid Waste and Emergency Response. 3. Kleinfelder, Inc. (1997, December 22). Health risk assessment for Pallas Yard Murray, Utah Final Draft. Washington, DC: U.S. Environmental Protection Agency. 4. CDM Federal Programs Corporation. (1997, April 29). Butte priority soils operable unit baseline human health risk assessment for arsenic. Butte Silver Bow Creek /Butte area NPL site, Butte, Montana. Washington, DC: U.S. Environmental Protection Agency. 5. Bornschein, R. (1992, February). The ButteSilver Bow environmental health lead study nal report. Cincinnati, OH: ButteSilver Bow Department of Health and Department of Environmental Health, University of Cincinnati. 6. U.S. Environmental Protection Agency (USEPA). (1992, April). Guidance for data usability in risk assessment (Publication 9285.7-09A). Washington, DC: USEPA, Ofce of Emergency and Remedial Response. 7. Bowers, T. S., Shifrin, N. S., & Murphy, B. L. (1996). Statistical approach to meeting soil cleanup goals. Environmental Science & Technology, 30, 14371444. 8. U.S. Environmental Protection Agency (USEPA). (1996, December). Recommendations for an interim approach to assessing risks associated with adult exposure to lead in soil. Washington, DC: USEPA, Technical Review Workgroup for Lead. 9. Cohen, J. T., Bowers, T. S., Lampson, D. W., Lampson, M. A., & Merrill, D. E. (1997). Quantication of exposure area cleanup thresholds when contaminant levels are uncertain. In American Statistical Association 1997 proceedings of the section on statistics and the environment (pp. 134139). Alexandria, VA: American Statistical Association. 10. CDM Federal Programs Corporation. (1997, April 18). Revised arsenic-contaminated waste volume and removal cost estimate for the Butte priority soils operable unit RI/FS. Washington, DC: U.S. Environmental Protection Agency.