Environmental Risk Assessment Reports - Chapter 31 (end)

CHAPTER 31 Special Topics in Risk Assessment: Models and Uncertainties Stephen G. Zemba and Laura C. Green CONTENTS I. Introduction.................................................................................................551 II. Use of Models in Risk Assessment............................................................552 A. Consider the Relevance of the Model..........................................553 B. Review Input and Output Parameters ..........................................554 C. Check Equations and Calculations...............................................555 D. Perform Reality Checks................................................................556 III. Uncertainty in Risk Assessment.................................................................557 IV. Conclusion..................................................................................................560 References...................................................................................................561 I. INTRODUCTION Risk assessment* is an analytic tool intended to quantify possible threats to the environment and/or the public health. Once an academic instrument played by relatively few analysts, risk assessments are now routinely performed by hundreds of professionals and for many regulatory purposes, at least within the U.S.** Risk-based decisions and regulations abound, and given the current political climate, the use of risk assessment is likely to continue to expand. * As used here, risk assessment means the quantitative assessment of risks to both human health and the environment due to exposure to chemical contaminants present in air, soil, water, and/or food. ** Formal, quantitative risk assessment is somewhat unique to the U.S. In Europe and elsewhere, analyses typically rely less on detailed modeling and extrapolations and more on the semiquantitative judgements of toxicologists and other scientists and engineers. 551 © 2001 by CRC Press LLC 552 A PRACTICAL GUIDE TO ENVIRONMENTAL RISK ASSESSMENT REPORTS Risk assessment is distinguished from other environmental disciplines by its integration of the physical and biological sciences. A thorough understanding of a risk assessment typically requires detailed expertise in a variety of fields. In addition, risk assessment procedures have burgeoned in complexity. Ten or 15 years ago, a typical assessment included at most a few basic pathways — routes from a source to a person — and risk estimates were typically constructed as simple, order-of-magnitude estimates. Nowadays, risk assessments endeavor to account for all rele-vant avenues of exposure, to model in detail the environmental transport and fate of contaminants, and to describe and quantify the inherent variabilities and uncer-tainties in crucial variables. In this chapter, we focus on two current issues in risk assessment. The first topic — the use of models in risk assessment — is motivated by the recent regulatory emphasis on multipathway risk assessment. The desire to quantify the movement of contaminants within the environment has been accompanied by proliferation of fate-and-transport models. For example, by utilizing only a few chemical-specific parti-tioning coefficients, a chain of models can be constructed to trace pollutants from air into water, soil, vegetation, and foodstuffs. Unfortunately, this propagation of models has not typically been accompanied by a commensurate level of testing and validation of their predictions. In addition, a wider audience of (and for) model users has, in some cases, led to improper or at least questionable applications of models. We thus discuss here some critical characteristics of models and suggest procedures that can be used in their selection and review. The second topic we address is uncertainty. Probabilistic methods are an impor-tant advancement in risk assessment methodology; an illustration of a Monte Carlo assessment is included to demonstrate advantages and caveats of the method. Proper characterization of uncertainty is a focus of recent risk characterization policies released by the U.S. EPA (1995). II. USE OF MODELS IN RISK ASSESSMENT Broadly defined, a model is an abstraction used to mimic, describe, and/or predict some aspect of reality. Models may be used to extrapolate from data sets, to inter-polate between data points, or to provide estimates where few or no measurements exist. Models permeate all facets of risk assessment. For example, the characteriza-tion of a series of measurements of contaminant concentrations in groundwater may assume an underlying statistical model. The definition of an exposure pathway requires the conceptualization of the process whereby a contaminant reaches a human or environmental receptor. The linearized multistage method, as another example, is the extrapolation model used by EPA to estimate the carcinogenic potency of a chemical in humans, given, typically, dose-response data from labora-tory animal bioassays at doses vastly greater than those of interest for the risk assessment. Most risk assessors think of models in terms of contaminant fate-and-transport models. Within this category, models range in complexity from simple analytical expressions that consider a few parameters to sophisticated “black box” algorithms © 2001 by CRC Press LLC SPECIAL TOPICS IN RISK ASSESSMENT: MODELS AND UNCERTAINTIES 553 Table 1 Factors to Consider When Renewing Model Usage and Results Is the model appropriate for the physical situation? RELEVANCE INPUT/OUTPUT CALCULATIONS REALITY CHECKS Does the model meet regulatory requirements? Do parameter values seem unusually large or small? Are parameters easily checked against standard, common values? Are units specified, and are they consistent among parameters? Are site-specific values used where possible? Can the results be reproduced from given equations and parameters? Do model results exceed real-world constraints? Are there idiosyncrasies between model predictions and physical expectations? Do model predictions violate conservation of mass? that simulate complex mathematical relationships. Models may be theoretically derived from underlying physical principles, empirically based on statistical infer-ences, or both. The vast number, variety, and complexity of models available can make it difficult to: · Select appropriate models for use in risk assessments · Choose appropriate input values · Review modeling results There are many modeling pitfalls, and even the most experienced users and reviewers must exercise considerable caution. Analysts must be ever mindful both that no model is a perfect representation of the real world, and that considerable expertise (sometimes different from one’s own) may be needed to differentiate between meaningful and meaningless results. Having reviewed many erroneous applications of models in risk assessments, we have developed or used a number of techniques to identify errors and/or inappropriate applications. We offer the follow-ing advice to modelers and reviewers (Table 1). A. Consider the Relevance of the Model Models should be applied only for situations for which they have been designed. Though it is obvious advice, we find it often ignored; perhaps because model users do not always review the derivation or limitations of the model before employing it. For example, a model designed to simulate groundwater flow should not be applied to the unsaturated zone. More often, however, poor judgement in model application involves more subtle errors. For example, Gaussian plume (GP) models are com-monly applied to estimate the dispersion of contaminants in air. Most GP models simulate a plume that proceeds in a straight line from the point of pollutant release. © 2001 by CRC Press LLC 554 A PRACTICAL GUIDE TO ENVIRONMENTAL RISK ASSESSMENT REPORTS In many cases, such an assumption is appropriate. Application of a GP model in settings where winds change direction (such as a valley), however, can lead to the prediction of impacts at erroneous locations. Models should include the essential physics needed to simulate a particular environmental situation. We prefer to use the simplest model possible that contains the basic factors that influence contaminant transport. Care must be taken, however, to insure that all relevant mechanisms have been included (of course, there are many settings in which all relevant mechanisms are unknown or unquantified; only addi-tional research can help remedy such defects). For example, a soil model that neglects water-phase transport may grossly overpredict vapor diffusion rates. Model selection also requires consideration of regulatory requirements. In many situations, agencies recommend the use of specific models. It can be easier (from a political perspective) to apply a less-than-ideal model to avoid costly regulatory review. For example, in permitting of air pollution sources, EPA provides a list of “approved” (though not necessarily fully validated) models that can be used for specific purposes. Use of alternate models instead may involve extensive justifica-tion; depending upon the discretion and/or tastes of the regulators, such alternates may or may not win approval. Models vary greatly in complexity. As a general rule, simple screening models produce less accurate results than more elaborate (refined) models. This does not mean, however, that the most refined model should always be selected. Instead, model sophistication should be matched to the level (and certainty) of available information. Use of overly sophisticated models can produce misleading or inaccu-rate results if they are based on generic default parameters that may bear little or no similarity to site-specific conditions. In selecting and evaluating a model for appropriateness, the best advice we can offer is to gather, read, and assimilate documentation regarding the model’s basis and development. Such documentation may be found in users’ manuals, technical reports, and the scientific literature. From these, one may glean a sense of whether the model’s purposes and strengths are matched by the application at hand. One may also find in the literature additional or alternate models; and it is sometimes instructive to run these and compare results with those of the proffered model. B. Review Input and Output Parameters Models produce one or more outputs given one or more inputs. The GIGO (garbage in/garbage out) principle* is one of the cardinal rules of modeling, and has never been more relevant given the proliferation of user-friendly, menu-driven models that provide default parameters and run with little or no user interaction. While we know of no systematic ways to avoid input/output errors, there are several measures that can be taken to reduce the possibility of errors. First and foremost, check the units. Since models represent mathematical equations, they require consistency among parameters, and generally demand precise specification of parameters. Factors of 1000 errors in using metric units are remarkably common. * Also known as, “you can’t make good applesauce with bad apples;” and note, that while some may find the applesauce made with bad apples to be tasty enough, gourmets won’t be fooled. © 2001 by CRC Press LLC SPECIAL TOPICS IN RISK ASSESSMENT: MODELS AND UNCERTAINTIES 555 Also, be careful in applying conversion factors, since there are unusual (but con-ventional within specified settings) parameters in use in various disciplines. For example, pollutant concentrations in stack gases are often expressed as mass per dry standard volume, which permits comparisons among facilities operating under wide ranges of conditions, but can lead to erroneous calculations of emission rates if used without the necessary conversions. Methods to check units vary according to the type of model. In applying simple algebraic equations in spreadsheets, explicitly write out units by hand using the factor-label method to verify consistency. For computer algorithms, carefully review the users’ guide to make sure that all inputs are specified in the units demanded by the program; be aware of the units specified for output parameters and use them accordingly. In reviewing reports, look for values expressed in suspicious units. For example, rates (flow, emission, etc.) should always be expressed per unit time (although they are frequently not).* As a second step, review input parameters for consistency with your intuition. In some cases, this may require a greater (but useful) assimilation of the metric system. As examples, the sizes of physical objects such as farm fields and surface water bodies should be reasonable. Groundwater should move more slowly than surface water. Densities of liquids and gases should be of the same orders of magnitude as those of water (1000 kg/m3) and air (1.2 kg/ m3), respectively. Values should not be outside allowable limits (e.g., 10,000 g/kg signals an error). Although these compar-isons cannot be done for all parameters, one can learn with practice to recognize a wider range of outliers. For example, by reviewing only a few studies of subsurface transport, it becomes readily apparent that molecular diffusivities of contaminants in air and water are always of the order of 10-5 and 10-9 m2/s, respectively. Third, take steps to ensure that the most appropriate parameters have been selected. Where possible, choose site-specific values that reflect on-site measure-ments or regional characteristics. Be wary of default values and parameters appar-ently chosen arbitrarily from the literature. When assigning parameters, have another person peer review your choices — a second perspective is always useful. As an example, one of our colleagues was charged with selecting half-lives for various organic pollutants in a groundwater modeling study. For one of the pollutants, a half-life of 5 days, as reported in a handbook, was selected. It was readily apparent to another of us, however, that this value was unrealistic in our application, since the pollutant had persisted at the site for many years. Consequently, a longer half-life was chosen and justified. C. Check Equations and Calculations Peer-review is the best method of checking model calculations. Given the volume of calculations they encompass, numerical mistakes in risk assessments are common. We often identify mistakes through reviewing our own and others’ work, and cannot overemphasize the need for checking. We find that electronic spreadsheets, which * Atypical units sometimes reflect conventions and not errors. For example, hydrogeologists express pressures in terms of the equivalent feet of water, while meteorologists express pressures as inches of mercury. © 2001 by CRC Press LLC ... - tailieumienphi.vn