Grantee Research Project Results
Final Report: Experimental Validation and Reliability Evaluation of Multimedia Risk Assessment Models
EPA Grant Number: R825411Title: Experimental Validation and Reliability Evaluation of Multimedia Risk Assessment Models
Investigators: Siegrist, Robert , Dawson, Helen , Sheldon, Andrew
Institution: Colorado School of Mines
EPA Project Officer: Chung, Serena
Project Period: December 15, 1996 through December 14, 1998
Project Amount: $256,174
RFA: Environmental Fate and Treatment of Toxics and Hazardous Wastes (1996) RFA Text | Recipients Lists
Research Category: Hazardous Waste/Remediation , Land and Waste Management , Safer Chemicals
Objective:
Subsurface contamination by toxic organic compounds is a widespread problem in soil and groundwater at industrial and military sites in the United States and abroad. At many sites, baseline risk often is governed by inhalation exposures to volatile organic compounds (VOCs) emanating from contaminated soils, and by ingestion exposures to VOCs that leach into groundwater used for drinking water. The incremental risk due to such a current or future exposure often is estimated using site characterization data and transport models. As shown in the following equations for carcinogenic risk, the estimated mass flux of contaminants to a receptor leads to exposure concentrations (CA or CW) that are directly related to risk:
Risk = CDI*CPF [1]
CDIlh = CA*(IHR*ET*EF*ED) / (BW*AT) [2]
CDIig = CW * (IGR*ET*EF*ED) / (BW*AT) [3]
where CDI = chronic daily intake (mg/kg/d), CPF = cancer potency factor (mg/kg/d)-1 (U.S. EPA, 1989). Chronic daily intakes from inhalation (CDIih) or drinking water ingestion (CDIig) are based on the average chemical concentration in air (CA, mg/m3) and water (CW, mg/L) over the duration of exposure, where IHR = inhalation rate (m3/d), IGR = ingestion rate (L/d),
ET = exposure time (hr/d), EF = exposure frequency (d/yr), ED = exposure duration (yr),
BW = body weight (kg), and AT = averaging time (AT). With an estimated baseline risk, risk reduction often is attempted by treatment. Predicting the degree of treatment needed to achieve a desired reduction in risk is based on back calculation (or iterative calculation) to set an acceptable residual contaminant level that will yield mass fluxes that, upon mixing into receiving media, do not exceed health-based exposure concentrations (e.g., CA or CW).
There are several intermediate and multimedia transport models being used for risk assessment and in some cases, the prediction of the risk reduction benefits of in situ treatment. However, experimental validation of the models with respect to the mass fluxes that occur during simultaneous leaching and volatilization of VOCs from contaminated soils has been lacking. A research effort was undertaken at the Colorado School of Mines to quantitatively evaluate the performance and reliability of multimedia transport-fate models used in assessing human health risk from VOCs leaching and volatilizing from contaminated soil. The performance and reliability of three relatively simple models were explored through experimental and modeling studies. Model performance that can be anticipated during risk assessments as typically completed for sites with VOC contaminated soils was investigated and model reliability was evaluated within the context of probabilistic health risk assessment and the effects of modeling error on risk-based decision making. The incremental increase in overall health risk assessment uncertainty produced by contaminant transport-fate modeling error was evaluated using a probabilistic reliability evaluation framework (see Figure 1). Comparisons of upperbound cancer risk estimates developed using modeled (uncertain) concentration information versus risk estimates developed using perfect (i.e., exact and correct) concentration information were used to identify "acceptable" modeling error and uncertainty tolerances via a formal decision performance analysis. The information developed during this project was designed to assist risk assessors and decision makers who must identify, and comply with, "acceptable" modeling error and uncertainty tolerances in accordance with the U.S. EPA data quality objectives process.
Figure 1. Probabilistic model reliability evaluation framework.
Summary/Accomplishments (Outputs/Outcomes):
Research methods included controlled laboratory leaching and volatilization experiments, computer modeling, deterministic and probabilistic human health risk calculations, and statistical data analysis. The methods and findings of the research are highlighted in this summary report; the details may be found in the dissertation of Sheldon (1999) and in forthcoming publications.
Assembly and Testing of a Multimedia Lysimeter. A multimedia lysimeter was developed to simulate volatilization and leaching processes in VOC-contaminated soil (Figure 2). The apparatus is comprised of an unsaturated soil block underlain by a microporous plate and hanging water column to control soil water content. A dynamic vapor flux chamber is located above the soil block along with an artificial rainmaker to disperse precipitation at a prescribed interval. The airflow rate, humidity, temperature, rainfall, and soil water content are controlled to match desired field environmental conditions. After carefully packing the lysimeter with soil, a known amount of VOCs was added in an aqueous solution during upflow saturation. Follow- ing an equilibration period, drainage of the lysimeter to a baseline unsaturated soil water content was accomplished, and the mass of contaminants recovered in the leachate was determined. The initial VOC mass and concentration in the lysimeter was then determined by difference. A continuous airflow was then initiated along with intermittent rainfall events. During the following days, the VOC concentrations in the air flow moving above the soil surface and in the leachate from the soil were measured by sampling and gas chromatography. To support multimedia modeling efforts, lysimeter environmental parameters were monitored continuously and recorded by a data logging system. Two multimedia lysimeters were fabricated and simultaneous leaching and volatilization results for replicate experiments with 1,1,1-trichloroethane in a sand matrix confirmed the reproducibility of triplicate runs.
Figure 2. Schematic of the CSM multimedia lysimeter apparatus.
Controlled Volatilization and Leaching Experiments. Using the multimedia apparatus, 12 volatilization and leaching experiments were performed involving four VOCs (1,1,1-trichloroethane, tetrachloroethene, trans-1,2-dichloroethene, and toluene), and three soil types (a medium sand, a silty sand, and a heterogeneous soil matrix) yielding results such as those shown in Figure 3. The experimental results included the time-dependent loss of initial VOC mass due to both volatilization into the overlying air flow and into the rainfall-induced leachate. Differences in the rate and extent of intermedia mass transfer are evident and are related to differences in the transport properties of the target compound and the soil matrix (e.g., Koc and KH, and porosity and foc, respectively). The volatilization mass transport was shown to be directly related to the VOC KH and the media air-filled porosity, and inversely related to the VOC Koc and media foc. The leaching mass transport is directly related to the VOCs Cw and the water content of the soil media and inversely related to the VOCs Koc and media's foc.
Multimedia Transport and Risk Modeling. The experiment results were compared with time-average predictions made independently by three commonly used models: EMSOFT, VLEACH, and SOILMOD. (see Figure 3 and Table 1). For this comparison, the experimentally measured volatilization and leaching rates represent "truth," and the discrepancy between the measured quantities and the corresponding model predictions can be considered "errors." For volatilization experiments, average modeling error discrepancies represented underpredictions of actual air concentrations by -30 percent for EMSOFT and -6 percent for SOILMOD. For leaching experiments, average model underpredictions were -16 percent for VLEACH and -44 percent for SOILMOD. The most underprediction of an observed time-average concentration occurred with EMSOFT, for simulation of tetrachloroethene vapor flux from a 0.225 percent organic carbon content soil matrix (-96.8% relative error). The leaching model that displayed the most underprediction was SOILMOD; this discrepancy occurred for the simulation of tetrachloroethene leaching from the heterogeneous soil (-85.9% relative error). Frequency histograms of normalized modeling error were prepared from the suite of experimental model validation results and used to define concentration term uncertainty for the contaminant transport-fate model applications to human health risk assessments (Figure 4).
Figure 3. Experiment results and model predictions for simultaneous volatilization and leaching of 1,1,1-trichloroethane from sand.
Soil | VOC | Lower integration limits (hours) | Upper integration limit (hours) | Experiment time-average concentration (ppb-v) | Modified time-average concentation (ppb-v) | Relative percent error | Normalized relative error1 |
---|---|---|---|---|---|---|---|
Sand | 1,1,1-TCA | 6.00 | 36.00 | 53.48 | 65.38 | 22.26 | 1.22 |
PCE | 6.00 | 21.00 | 29.36 | 15.73 | -46.43 | 0.54 | |
t-1,2-DCE | 6.00 | 14.08 | 330.1 | 570.3 | 72.76 | 1.73 | |
toluene | 6.00 | 24.24 | 135.7 | 90.80 | -33.10 | 0.67 | |
Silty sand | 1,1,1-TCA | 6.00 | 39.12 | 44.26 | 145.5 | 228.6 | 3.29 |
PCE | 6.00 | 84.96 | 214.7 | 6.84 | -96.81 | 0.03 | |
t-1,2-DCE | 6.00 | 24.00 | 300.7 | 231.5 | -23.03 | 0.77 | |
toluene | 6.00 | 63.36 | 21.73 | 12.44 | -42.78 | 0.57 | |
Heterogeneous packing | 1,1,1-TCA | 6.00 | 24.00 | 42.56 | 3.52 | -91.73 | 0.08 |
PCE | 6.00 | 24.00 | 0.95 | 2.68 | 183.1 | 2.83 | |
t-1,2-DCE | 6.00 | 13.15 | 413.1 | 430.7 | 4.27 | 1.04 | |
toluene | 6.00 | 20.57 | 27.07 | 48.80 | 30.28 | 1.80 | |
1 Modeled time-average concentration divided by experiment time-average concentration |
Source Term Error and Uncertainty. The error and uncertainty in quantifying the average concentration or total mass of VOCs in a soil region of interest can propagate into risk estimates through intermedia transfer modeling. An experimental study of VOC source term quantification was completed using trichloroethene (TCE) in the sand soil matrix. For this study, the lysimeter was contaminated in the same fashion as during the other experiments to yield a known mass of TCE in the lysimeter. After equilibration, but before any volatilization or leaching, nine thin-tube sampling probes were driven into the soil simultaneously. Each tube was retrieved and using a microcoring method, duplicate subsamples were taken at each of three depths and extruded directly into hexane. Five subsamples of each hexane extractant were taken and for each of these, five separate VOC analyses were made. The results of this effort revealed that the mass estimated by the discrete samples was biased low accounting for only 5 to 15 percent of that present in the lysimeter. A nested ANOVA revealed that the variances associated with sample location-depth and subsamples per location were dominant (> 95 percent) with limited variance contributed by the hexane extraction or the GC analysis components (< 5 percent). The high bias and spatial variability are believed to be due to nonuniform rapid vapor and aqueous phase losses that occurred during sample handling of the unsaturated sandy soil (with 0.017 wt. percent TOC) despite the careful practices used.
Figure 4. Frequency histograms of normalized relative error observed for two soil VOC volatilization models. Note: the normalized relative error is the unitless ratio of the model-predicted to experimentally observed concentration.
The results of this study support the use of a mass-balance approach versus traditional discrete soil sampling and analysis to quantify the source term for the purposes of the lysimeter volatilization and leaching experiments as described herein. The results also demonstrate the potential bias and variability during characterization at VOC sites, even when extremely careful practices are used.
Multimedia Reliability Evaluation for Human Health Risk Assessment. Multimedia model reliability for human health risk assessment applications is a function of both the form of the modeling error distribution and the tolerances for modeling error established by decision performance goals. Using decision performance goal diagrams, we evaluated the effect on probabilistic risk predictions, of exposure concentration term uncertainty due to modeling error. By comparing statistics of a cancer risk probability distribution calculated using uncertain concentration information to the risk distribution calculated for the same scenario using perfect concentration information, we identified threshold tolerances for concentration term uncertainty in human health risk assessments. Using a new adaptation of the U.S. EPA data quality objectives process, model reliability was evaluated by applying risk-based decision performance goal diagrams (Figure 5). For four typical human exposure scenarios that were evaluated based on model-predicted (i.e., uncertain) vapor and leachate concentrations, upperbound probabilistic cancer risk estimates were within a factor of three to seven times the upperbound risk estimates calculated based on hypothetical perfect (i.e., exact and correct) concentration information. For commonly observed risk management conventions, the performance of all three evaluated models was well within the risk uncertainty tolerance limits that were selected to define acceptable model reliability. It is noted that the application of the methodology reported in this project to reach conclusions for a particular site mitigation scenario must consider the representativeness of the existing model validation database and the applicability of the assumptions used in the decision analysis.
Figure 5. Decision performance goal diagram for evaluating leaching model reliability for use in probabilistic cancer risk assessment for adult benzene ingestion. Note: if the predicted risk cumulative probability distribution passes through the acceptance region, it indicates acceptable model reliability.
Journal Articles:
No journal articles submitted with this report: View all 8 publications for this projectSupplemental Keywords:
volatile organic compounds, VOCs, soil, groundwater, exposure, risk, carcinogen, modeling, risk assessment, leachate, human health., Health, Scientific Discipline, Waste, Water, Ecosystem Protection/Environmental Exposure & Risk, Contaminated Sediments, Geochemistry, Environmental Chemistry, Health Risk Assessment, Chemistry, Fate & Transport, Risk Assessments, fate and transport, ecological risk assessment, fate, sediment treatment, contaminant transport, risk characterization, multimedia risk assessment models, VOCs, contaminated sediment, sediment transport, transport contaminants, reliability evaluation, adverse human health affects, hazardous waste, soils, experimental validation, ecological impacts, toxic environmental contaminants, assessment methods, terrestrial and aquatic fate, assessment technology, ecology assessment models, hazardous waste sites, leaching, human health risk, ecological transferability, exposure assessment, leachateRelevant Websites:
Welcome to Center for Environmental Risk Assessment (CERA) Exit
Progress and Final Reports:
Original AbstractThe perspectives, information and conclusions conveyed in research project abstracts, progress reports, final reports, journal abstracts and journal publications convey the viewpoints of the principal investigator and may not represent the views and policies of ORD and EPA. Conclusions drawn by the principal investigators have not been reviewed by the Agency.