Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
7 Research Needs INTRODUCTION The committee has examined ground water modeling and the use of these models in regulation and litigation. Specifically, the committee was asked to answer two difficult questions: Alto what extent can the current generation of ground water models accurately predict complex hydrologic and chemical phenomena?" and "Given the accuracy of these models, is it reasonable to assign liability for specific ground water contamination incidents to individual parties or make regulatory decisions based on long-term predictions?" This chapter summarizes the committee's recommendations for the di- rection and content of research programs necessary to improve the current state of affairs. Two comments are in order before the recommendations are pre- sented. First, the focus of this study has been the status of ground water models; and therefore associated areas of expertise (e.g., cli- matic scenarios and exposure assessment models), while mentioned, are not given the same consideration as ground water models. Hence, the recommended research, while acknowledging related fields of study, is biased toward ground water models and may not reflect a complete and balanced research program. Second, the questions presented above emphasize mode] accuracy; however, the committee 249
250 GROUND WATER MODELS notes that the accuracy of models should not be equated with the art of accurately applying models. Indeed, simulating the subsurface environment is a mixture of art and science, and an assessment of mode} accuracy is only one element in evaluating the confidence one should have in simulation results. Identifying key or cornerstone issues relevant to a host of policy goals is essential so that limited resources can be devoted to the development of technology necessary to achieve national goals on the environment and economy. Certainly, as a nation we should maintain a leadership role in hydrogeologic studies for a variety of reasons; the application of ground water models in regulation and litigation is only one. Other reasons for maintaining leadership involve the estimation of natural resources and their availability, the evaluation of the safety of disposal of high-level and transuranic wastes in deep geologic deposits, and the understanding of potentially significant changes to our ecosystems (e.g., acid rain and CO2 increases). In general, it is difficult to prioritize specific research requirements for each particular application, and this report does not attempt to do so. If research is needed to improve an aspect of hydrogeologic modeling for application to regulation or litigation, the committee makes no attempt to place that need in the context of other areas of study that will benefit from the research. Certainly, there are whole areas of ground water research that will be omitted, e.g., regional modeling of watersheds and river basins affected by global climate changes. Another consideration that influences the committee's recom- mendations for future research is the present state of the science in subsurface hydrology. It is evolving; indeed it is on the threshold of a significant change in how the subsurface environment is interpreted. Current transport theory developments based on statistical inter- pretations of subsurface deposits may, in time, replace much of the deterministic theory. At issue are the characterization and simula- tion of dispersive phenomena. Central to this issue is the relationship between measurable quantities and parameters for flow and trans- port models. While these fundamental underpinnings to the models of conservative contaminant transport are being revisited, research continues to extend standard deterministic theory to better simu- late a great variety of complex situations. Examples of extensions to deterministically based theory are multiphase flow phenomena, microbiological processes influencing water quality, and coupled geo- chemistry and transport models. Thus extensions of current models
RESEARCH NEEDS 251 to more complex processes and greater spatial dimensionality are being made at the same time that foundational aspects of basic transport theory are being revisited. The state of the practice does not reflect the state of the art, because of the scope of ongoing research and because of the strength with which opposing views are held and debated. The science has not come to grips with the gap between practice and art. Concern exists that until one can predict with confidence the migration of a conservative solute within a heterogeneous medium, one will not be able to convince a great many people of the veracity of reactive solute migration predictions. However, scientists must come to realize that modeling is used to avoid bad decisions as well as to make the best decision. Indeed, the evaluation of good alternatives may be uncertain to the degree that no clear best alternative exists. To the extent that existing field-scale models provide qualitative assessments of good versus bad, they are useful and appropriate. Such a rationale justifies the use of screening models to prioritize sites for further study and possible remediation. Research must be conducted to encourage greater acceptance of screening models and to ensure the proper expenditure of resources they influence. Resources also need to be devoted both to continue fundamental research and to decrease the gap between the state of the art and the state of the practice. There is a recognized need to revise our current concept of mod- eling and modelers. Modeling needs to be redefined as a cost-effective way of interpreting all available data, to the extent that the interpre- tation provided by that modeling effort enables one to be comfort- able in making a decision. Viewed in this way, modeling involves a spectrum of allied technologies that combine to provide the needed interpretation of subsurface events. In such a setting the modeling process would be viewed as a whole, and all subjective decisions af- fecting the modeling process are seen to contribute to an assessment of accuracy. Individuals responsible for mode} applications would be more appropriately described as analysts, rather than modelers, because of the spectrum of technologies to be applied and because of the subjective interpretations required. The preceding remarks guide the scope of the committee's recom- mended research. The committee members, primarily ground water modelers, recognize that evaluation of modeling accuracy is a broad topic influenced heavily by subjective decisions made when climate scenarios are developed, site characterization plans are made and data are analyzed, and subsurface conceptual models are formalized.
252 GROUND WATER MODELS The scope of research in the future must be broadened to formal- ize methods of recording subjective inputs and quantifying accuracy within the modeling process. The objective of mode} validation must be to quantify the accuracy of a mode} prediction for a particular ap- plication. In addition to a core effort to develop accuracy assessment methods, research must improve the methods available to gather and evaluate field data for site characterization, contaminant detection, and contaminant plume monitoring. The focus of a coordinated research program must be on the mode} process and its ability to predict, over the time frame of interest, the behavior of field-scale events. USE OF MODELS There is no doubt that increasingly greater scientific emphasis is being placed on the use of predictive computer models in ground wa- ter hydrology and geochemistry. Early applications of ground water models emphasized qualitative or relative evaluation of several alter- natives. Models were used to better understand the potential impacts of alternative water use or disposal strategies. Water quantity rather than quality was the focus of this modeling, and relative comparisons appear to have been adequate to resolve litigation and regulation questions. With the full allocation or overallocation of ground water resources and the advent of ground water quality regulation, the at- tention of hydrologists has turned to quantitative analysis of water quantity and quality with emphasis placed on contaminant migra- tion. The trend is toward analysis of the interrelationship between quality and quantity of the subsurface water resource and optimiza- tion of various pumping, storage, and remediation designs. The emphasis of most modeling efforts today is on providing an absolute rather than relative performance estimate. Perhaps the most obvious example of this is in the area of storage and disposal of high-level nuclear wastes in geologic repositories (see, for example, Erdah} et al., 1985; Jacobs and Whatley, 1985~. The Nuclear Waste Policy Act of 1982 (Public Law 97-425, 96 Stat. 2201, 42 USC 10101) specifies that the Department of Energy (DOE), the U.S. Nuclear Regulatory Comrn~ssion (USNRC), and the U.S. Environmental Protection Agency (EPA) are responsible for doing the necessary preliminary work to permit the siting and construction of a geologic repository for high-level nuclear wastes in the United States. The only obvious method for predicting the rate of release,
RESEARCH NEEDS 253 geochemical behavior, and rate of transport over a period of 100,000 yr is through computer modeling. Other approaches are possible, but they are at least as uncertain as computer modeling. For example, experiments can be conducted at elevated temperatures to accelerate reactions and thus to simulate longer periods of tune, but there is no guarantee that acceleration resulting from higher temperatures will really simulate long periods of time at low temperature. Another approach is to examine geological sites and ancient archaeological relics for clues as to the behavior of certain chemical elements, but suitable situations are rare for implementing this strategy. All in all, computer modeling probably has at least as good a chance of yielding meaningful predictions as any of the other approaches. A second example is the multitude of governmental agencies and private firms that increasingly rely on computer modeling techniques to investigate, predict, and guide the cleanup of natural waters con- taminated by impurities that have escaped from landfi~Is or from subsurface storage facilities. It appears that the two main objectives in the use of predictive modeling in this area are (1) to optimize the placement of test weld and monitoring wells and (2) to allow inves- tigators to predict the future behavior of a plume of contarn~nation. An obvious application would be to follow a plume of contamination in an aquifer backward in time and space in an effort to determine its original source. The general subject of contamination of ground water is discussed in some detail in a report by the National Research Council (1984) entitled Groundwater Contamination. A third potential use of predictive modeling, which has not yet been widely recognized, ~ to determine what the natural background concentrations might have been in a region prior to any impact by man. This latter application may be particularly useful in establish- ing natural background concentrations of toxic metals in mineralized regions prior to the initiation of mining and milling. There ~ little doubt that the current use of predictive computer models in interpreting and predicting the behavior of contaminants in ground water will continue and, in all probability, will increase. At the same time, as discumed in other chapters of this report, enough has been learned about the weaknesses of such models to justify the significant amount of skepticism that has also developed, both in the scientific community and in the regulatory arena. It is hoped that the proper mix of science and skepticism will be found and that the combination will allow the identification and use of a variety of predictive models that have been adequately tested and found to be
254 GROUND WATER MODELS appropriate, within acceptable limits of error, for a variety of field situations. This is truly a necessity for some situations, such as the disposal of nuclear wastes, that cannot be addressed in any other manner. Emphasis on predictive rather than relative results has created an interest in the uncertainty of predictions. Unfortunately, un- certainty in estimates of ground water system behavior arises from several sources, some of which cannot be quantified. Indeed, there is no known truth to compare against when assessing uncertainty. This is the state of affairs despite the fact that a single conceptual picture of the subsurface environment does exist. Acknowledged sources of uncertainty are (1) ignorance of the true operative and dominant processes or reactions, (2) ignorance of true site characteristics lead- ing to inaccurate boundary and initial conditions, (3) the inability to sample and quantify natural spatial and temporal variability, and (4) the extrapolative rather than interpolative character of predic- tions. The ability to quantify sample variability is complicated by the existence of measurement error, dissimilar data (e.g., sampling method, instrument, and volume), and quasi-periodic or random events. Clearly, sensitivity and uncertainty methods are unable to represent several of the known sources of uncertainty. Recent work has heightened the awareness of the potential un- certainty in ground water mode! results and has led to some caution, or at least warnings, regarding the use of modeling results in the dec~ionmaking process. With regard to the use of deep geologic deposits for the disposal of nuclear wastes, Niederer (1988) believes that certainty is as important as safety. He suggests that the wise de- cision is to place waste where one has confidence in the performance of the geologic setting and not to place it where one merely hopes the performance will be safer. Niederer (1988) also believes that un- certainty in conceptual modem is more disquieting than uncertainty in parameters, especially for flow models. Hm underlying concern is the potential dominance of uncertainty components that are not quantifiable. Confidence and credibility of ground water mode} ap- plications depend on demonstrated applicability in every instance. Research must be undertaken to establish the framework necessary to demonstrate the applicability of models used in formulating or responding to regulation. The objective of such a demonstration is to ascertain the applicability of a given mode} through an assess- ment of accuracy and uncertainty for each situation or problem set of interest.
RESEARCH NEEDS 255 SCIENTIFIC TRENDS AND RESEARCH Three basic objectives inform the recommendations for scien- tific research presented here: (1) to better understand and mode! individual processes and reactions, (2) to translate process-level un- derstanding to sitewide simulation capability, and (3) to integrate the interdisciplinary technology needed to solve ground water contami- nation problems. While our understanding of subsurface processes and reactions has grown significantly in recent decades, something less than a predictive capability exists at this time. Indeed, where process and reaction models exist, field-scale observations of flow and transport have led to the realization that models based largely on laboratory- or caisson-scale studies do not provide a predictive capa- bility at the field scale. It is also apparent that the understanding of models for some processes and reactions is not sufficient for predic- tive purposes in the face of complex, heterogeneous, and anisotropic environments. When process models become accepted, significant efforts are needed to translate the research results into an accepted field-scale technology. Assessments of mode} accuracy and validity at the field scale are an important aspect of this translation from science to application. Finally, interdisciplinary efforts that bring together site geologists, hydrologists, geochemists, geostatisticians, and health physicists are essential if ground water models and allied technolo- gies are to be routinely applied to study and solve contamination problems with confidence. Basic Understanding and Process Models Two paths have been taken toward improving our basic under- standing and developing more predictive ground water models: (1) the further development of mechanistic and deterministic models for individual processes and (2) the development of probabilistic models that recognize the inherent uncertainty in nature and in our ability to characterize and mode} the subsurface environment. Ultimately, both paths have a single objective: to understand basic processes and reactions and their interrelationships. Such an understanding will lead to predictive models of events at the field scale. Physical processes that control or strongly influence contami- nant migration in the subsurface remain an area of intense research. While relatively better understood than geochem~cal and m~crobi- ologica] processes, present conceptual and mathematical models of convection and dispersion do not provide accurate results or inspire
256 GROUND WATER MODELS confidence when applied to highly heterogeneous or otherwise com- plex environments. The probabilistic approach is seen as a way to account for the inherent uncertainty in both the subsurface structure and the knowledge of flow and transport processes. Process Models While considerable progress has been evident in developing mass transport as a practical tool, the hope of routinely using these models in practice lies somewhere in the future. One reason for this state of affairs is the limited ability of most models to account for the important transport processes in a realistic and convincing way. Nowhere is this problem as obvious as with the physical processes accounting for organic compound migration and the chemical and biological processes occurring for a variety of contarn~nants, where considerable effort will be expended to solve a few key problems. The following sections outline the trends of future research designed to improve our understanding of the processes and demonstrate the validity of coupled models. Multiphase Fluid Flow and Transport Models An obvious trend in research is to extend modeling capabilities to new classes of problems. A case in point is the commonly encoun- tered problem of multiphase fluid flow and transport accompanied by dissolved component transport in water. Many of the most common organic contaminants are moderately to strongly hydrophobic. Ex- amples are the chlorinated solvents, various petroleum constituents, pesticides, and PCBs. Modeling of the fate of hydrophobic com- pounds can be complicated because they can form a continuous nonaqueous phase, sorb to aquifer solids, and volatilize to a gas phase. Modeling the transport of hydrophobic materials will require that these complications be incorporated into a solute transport model. When the organic compound forms a nonaqueous-phase liquid (NAPL), it creates three modeling difficulties. First, a significant accumulation of NAPL gives rise to multiphase or immiscible flow, a situation that is poorly understood mechanistically and difficult to describe mathematically. Thus modeling the movement of the NAPL, which is at least partly independent of the movement of the water, creates an added computational burden, if it can be
RESEARCH NEEDS 257 described at all. A general lack of fluid retention characteristics and relative permeabilities for organic compounds or mixtures of organic compounds in the presence of water and air will greatly limit our ability to simulate multiphase fluid migration. Because of interest in the drainage and removal of hydrophobic contaminants, models of hysteresis in soil-fluid properties are essential in correctly simulating the wetting and drainage phenomena of both the organic compound and the water. Second, the presence of an NAPL provides a long-term source for dissolution of contaminants to the aqueous phase. Description of the rate of dissolution requires knowledge of the presence of the NAPL and of the factors controlling its dissolution. Although it is probable that the solubilization is driven by the difference be- tween the aqueous-phase concentration and the maximum solubility, the rate of dissolution is probably controlled by hydrodynamic as- pects of mass transport and the presence of other contaminants. Even when the controlling factors are known, their inclusion into the mode! could increase the computational needs. Finally, model- ing of NAP Es ultimately requires some field verification of NAPEs in subsurface systems. This presents numerous difficulties with re- gard to sampling and interpreting the field-scale environment. Bulk spills or disposals of NAPEs dominated by a single fluid (e.g., fuel of! or trichIoroethene), do exist; however, many cases exist in which the NAPL is a mixture whose behavior in the environment can be quite complicated. Methods of sampling the subsurface and of preserving samples to determine the extent of contamina- tion must acknowledge the variety of contaminants potentially present in soil and fluid samples. Due to the natural heterogene- ity of subsurface environments, NAP Es often are not homogeneously present but are difficult to locate, especially because they can spread out into thin layers. Ultimately, the relationship between flow physics and natural spatial variability will have an impact on the interpreta- tion of field-scale observations through an understanding of viscous fingering, i.e., the balance struck between continuum and channel flow phenomena. Hydrophobic organic compounds also sorb onto or into aquifer and soil solids, especially soil organic matter and clays. Like NAPEs, sorbed materials can be a source of long-term, chronic water con- tamination as they are slowly Resorbed. Solute transport modeling requires that the accumulation of sorbed material be accounted for and that the rate of desorption be described. In addition, realistic
258 GROUND WATER MODELS sorption relations are not necessarily linear (e.g., like partition coef- ficients), which gives rise to much more difficult mathematical and numerical solution requirements for nonlinear terms. For NAPEs and sorbed contaminants, the coupling of their addi- tion to the water with water-phase reactions, such as biodegradation, can create significant complications. For example, microorganisms degrading a dissolving solvent might be located a short distance away from the interface of the water and the NAPL; thus the dissolving compound is exposed to a biological reaction that consumes the con- taminant, allows less contaminant to pass to the rest of the water, and creates an increased driving force for more dissolution. Reac- tions that can occur on a scale (e.g., micrometers to centimeters) much smaller than the mode! grid are among the most significant complications. The effect of including this microscale for a reaction is to introduce another spatial scale to transport models, which in- creases the computational intensity. Additionally, the phenomena controlling reactions (especially biological) for dissolving or Resorb ing contaminants are not easily described. Third, some of the hydrophobic compounds (e.g., the chlorinated solvents) also are volatile and will partition to a gas phase. Thus if there are unconfined conditions and especially if there is gas produc- tion (e.g., with in situ biorecIamation or in situ aeration), some of the volatile contaminants can leave the aqueous and solid phases and go into the gas phase. Modeling of solute transport in such a situation must involve mass balances in the gas phase and description of the transfer rates between the gas phase and other phases. Not only do these requirements add to the computational demands, but they are not easily described with our current knowledge. In summary, modeling that realistically includes hydrophobic components may become significantly more computationally inten- sive because of the need to keep track of nondissolved species, to describe transfer rates between phases, and to mode} on a small scale. Computationally efficient solution techniques, such as quasi- linearization, and the use of local analytical or pseudoanalytical solutions may become a key aspect of successful modeling. Linking Geochemical and Physical Transport Models Considerable success has been achieved in modeling the geo- chemistry of natural waters and in modeling the movement of ground
RESEARaH NEEDS 259 waters. It is logical to take the next step and link an equilibrium geo- chemical mocle} with a ground water transport model. An optimist would say that the product of the linkage should be a mode} that has the capability of predicting chemical changes in the ground water and reactions between the water and the aquifer at each point in space along the flow path. A pessimist would probably visualize such a linkage as being nothing more than the compounding of errors and uncertainties inherent in each of the two separate and still immature models. The truth, at this point in time, lies somewhere between the extremes, but perhaps closer to the pessimist's point of view. The ba- sis for this somewhat negative evaluation is the fact that researchers in geochemistry have yet to demonstrate that any of the popular geochemical models can be fully validated against field or laboratory data. This is not the fault of the models, but instead points to a surprising lack of field and laboratory studies that are designed or are suitable for purposes of validating the theoretical models. Mod- elers tend to go their own way, building impressive computer codes to s~rnulate nature, while field and laboratory workers tend to gather data that are highly relevant for many purposes, but perhaps not for validating models. The lack of validation is far less severe and pervasive in hydrology than it is in geochemistry, but it does exist. The main obstacle in hydrology may be the disparity between the simplifications that are required to write a usable computer code and the great complexities that can exist in real field situations. The most obvious example is the stratigraphic heterogeneity of many real aquifers, in contrast to the perfect homogeneity or the vastly simpli- fied heterogeneity required for modeling. A similar obstacle will face geochemists when field-scale validation is undertaken. Just as hydrologists use simplifying assumptions essential to the creation of a viable conceptual model, geochemists also employ sim- plifying assumptions. Foremost is the assumption of equilibrium thermodynamics determining the aqueous-phase composition. This single assumption influences the form of governing equations and thermochemical databases. Time dependency through dynamic or kinetic reactions is omitted, as are rate constants in the database. When time dependency is observed to be significant in field settings, both the reactions and the associated data will need to be incorpo- rated into either established equilibrium-based codes or entirely new codes. It is apparent that kinetic reactions are important to some contamination events of interest, e.g., the leaching of fly ash and flue gas desulfurization sludge (Warren and Dudas, 1986~.
260 GROUND WATER MODELS Another aspect of equilibrium models of the aqueous phase is the natural assumption that the aqueous phase is in equilibrium with the solid phase (i.e., the porous medium). However, most unconsolidated media do not represent a solid phase in equilibrium with itself. The unconsolidated solid phase may be the result of physical (e.g., floods ant} glaciation) or geochemical processes. Thus the mineral composition of solids that make up porous media is often quite complex. With respect to equilibrium models, one observes that if the mineralogy of an unconsolidated porous medium were dissolved and solids were precipitated according to equilibrium reactions and data, one would not obtain the original mineralogy. This implies that a great deal of care must be taken to correctly conceptualize the geochemistry of ground waters. One must identify that portion of the solid phase that most strongly influences or defines the aqueous- phase speciation and concentration. The solution to the dilemma of constructing fully linked geo- chem~cal and transport codes, and being able to trust the result, lies in part in a close interaction between the modelers and the field and laboratory personnel. We have probably reached the stage of development in modeling at which it ~ Operative to gather more and better empirical data to demonstrate the validity of geochemical and transport models. The problem of how to deal with biological reactions is also of particular concern. As shown earlier, chemical modeling for solute transport often will involve a coupling of equilibrium and kinetic concepts: equilibrium concepts are used to determine which reactions are possible, while kinetic concepts are used to estimate the rate of possible reactions. Biological modeling usually is based solely on kinetics, although at least two types of material must be modeled: the substrate (degradable compound) and the active microorganisms. Several geochemical models that are useful for the equilibrium part of the computation are currently available (see Chapter 4), al- though they do not contain any type of kinetics. However, and very important, these geochemical modem are quite complex and com- putationally demanding to solve for only equilibrium calculations. Incorporation in their present form into solute transport modeling is impractical because of the computing demand. Therefore the avail- able geochemical models with comprehensive reactions and databases do not seem to be appropriate for solute transport modeling. Instead, simpler versions of these comprehensive codes have been developed
RESEARCH NEEDS 261 and incorporated into transport theory. One example is the MI- CROQL:1 code (Westall, 1979) derived from the MINEQL code (Westall et al., 1976) and applied in the TRANQL code (Cederberg et al., 1985~. Ideally, these more streamlined and efficient daughter codes could be tailored to include only the species and components of known importance to the site being studied. This also implies a tailored database specific to the species, components, and reactions of interest. Such an approach is the basis of the FASTCHEM_ package, a coupled transport and geochemistry code (Hostetler et al., 1988~. Yeh and ~ipathi (1989) have concluded that such a sequential iteration is the only practical approach for conducting realistic applications. Kinetic expressions have been incorporated into such a mode! (Kirkner et al., 1985~; however, it remains to be seen if completely general kinetic expressions (e.g., mass transport to and from surfaces) can provide flexible source-sink algorithms for chemical reactions. Of course, great flexibility is needed in such a computational shell because there are so many types of chemical reactions exhibiting kinetic behavior. Biological reactions present a major modeling challenge. One facet of the challenge is the complexity caused by the need to de- scribe the accumulation of attached microorganisms, mass transport from the liquid to the microorganisms, and highly nonlinear reaction rates. In addition, multiple substrates (and sometimes products) and bacteria] transport normally must be modeled. Finally, the model- ing challenge is greatly increased when bacterial growth is sufficient to change the permeability and transmissivity of the aquifer. Then the biological reaction affects the pressure distribution and water flow paths, which in turn can affect the biological reaction. Thus biological reaction and flow characteristics may need to be modeled interactively. Biological reactions also are challenging to model, because the scale of the reactions and the changes in reaction can be very small. For example, biodegradation of low concentrations of biodegradable organic compounds usually goes to completion in travel distances of only a few centimeters. Higher concentrations, lower biodegrad- ability, and limitations from other materials can extend the distance over which biodegradation occurs; nonetheless, biological reactions often occur at a scale much smaller than a normal grid spacing. A significant challenge for the future is efficient incorporation of these small-scale phenomena into models that require larger grid spacings.
262 GROUND WATER MODELS Nonuniform grid spacing and local analytical or pseudoanalytical solutions appear to be good approaches. Before biological and chemical models can be user] routinely in solute transport, the fundamental mechanisms must be studied and better understood, and the models must be tested in controlled field studies. Both steps are difficult and expensive. Probabilistic Methods Methods available in ground water hydrology for obtaining esti- mates of uncertainty involve two general approaches: deterministic porous-media models, where the probability component enters pri- marily through parameter variations, and stochastic porous-media models, where the probability component enters through the treat- ment of the medium itself as well as through parameter variations (Gutjahr, 1988~. These two approaches differ in how probabilities are assigned and incorporated and how the process is modeled, i.e., as deterministic or stochastic. In the latter case, randomness is viewed as an inherent feature of flow and dispersion. That is, the properties of the media are viewed as random processes in space, and stochastic models yield results having stochastic properties. There is a single characterization of the subsurface, and acceptance and application of random field interpretations of the subsurface are a departure from this truth (Gutjahr, 1988~. Data requirements for probabilistic analyses include generic and site-specific values. For example, it is generally accepted that trans- missivity has a generic log normal distribution for any particular type of material (Freeze, 1975; Freeze and Cherry, 1979; Hoeksema and Kitanidis, 1985~' however, this refers to the actual values and not to the mean transm~ssivity. Such generic models lend a structure to the inherent randomne-. Nevertheless, although the form of the distribution is assumed, sampling a highly variable system may still require that a significant quantity of site-specific data be available before the generic distribution can be applied to a site. Methods such as kriging, either standard or nonparametric, are commonly proposed and employed to analyze the spatially distributed yet rel- atively sparse ciata. Measurement error can also be accommodated by these kriging methods. In general, joint probability distribution information (e.g., interrelating transmissivities, porosities, and re- tardation factors) does not exist and must be created by subjectively
RESEARCH NEEDS 263 estimating the character of correlation between variables. In ad- dition, while stochastic models incorporate uncertainty much more directly, they also require specification of the covariance function. This involves both the covariance type and its scale parameters. An important aspect of probabilistic methods is the performance measure used to assess the accuracy of a mode} and hence build con- fidence in the modeling process. The same performance measure should be used for calibrating, validating, and applying the model. Performance and accuracy measures for flow and transport codes should be based on their eventual applications. For example, to predict or extrapolate transport, one needs confidence in the flow model's predicted hydraulic gradient, which directly influences ve- locity; one does not require great confidence in the hydraulic head even though it is more readily measured. Velocity and contaminant flux should be considered performance measures for transport codes. Continued use of measures of mode! performance that do not re- flect the modeling objective will result in an "achievable validation," which may equate to "plausible deniability" in a legal or regulatory setting. Research is ongoing and should continue to develop both prob- abilistic approaches to modeling ground water systems. Both ap- proaches require significantly larger computational resources and substantially more field data than purely deterministic models were thought to require a decade ago. Despite recent advancements, nei- ther the stochastic nor the deterministic method has satisfactorily resolved the role of scale-dependent dispersion. With one or both of these probabilistic approaches, a broadly based technology is re- quired to quantify uncertainty throughout the modeling process for subsurface systems, and a significant effort is needed to demonstrate the relevance of conceptual models, mathematical approaches, and characterization techniques. A holistic methodology is needed to quantify the uncertainty and relate it to measurable and meaningful field parameters. Translation of State of the Art to State of the Practice Research to better understand individual processes or reactions is often conducted in idealized settings that allow one to isolate cause-effect relationships; however, these settings are far from repre- sentative of field settings. Substantial efforts are required to create modeling capabilities applicable at the field scale and to demonstrate
264 GROUND WATER MODELS their relevance and accuracy through calibration and validation ex- ercises. Models used in formulating or responding to regulation must be shown to be applicable and therefore valid. Essentially, accuracy should be established before application of the mode} to a particular site. These efforts are referred to as efforts required to translate the process level understanding to field-scale simulation capability. Ar- eas of research that further this translation process include field-scale code developments, validation or accuracy estimation methods, ad- vances in computer hardware and numeric e] methods, and artificial intelligence or expert systems. Field-Scale Code Developments Several questions are asked about field-scale codes in the legal and regulatory settings. Do they embody specific state-of-the-art process models? Do they include alternative models enabling the study of opposing views? Are they valid for the proposed application? The formalism of answering and documenting the answers to these and many more questions is an aspect of quality assurance (QA) for codes. This topic is treated in Chapter 6. Whether considered research or not, a substantial resource commitment is required to create and maintain quality software. Apart from C2A issues, the development of codes applicable at the field scale is not a trivial undertaking. Flow and transport theory, which is based wholly on laboratory experience, will not necessarily apply at the field scale. Perfectly packed soil columns and highly con- trolled laboratory experiments seek to eliminate confounding effects of competitive processes and enable scientists to better understand individual processes or reactions or specific combinations of processes and reactions. Such controlled and contrived situations often fad! to represent field-scale events. Current research on probabilistic inter- pretations of the subsurface environment is an attempt to quantify the uncertainty in system response arising from quantifiable and unquantifiable spatial variability in the environment. Because of the level of current activity in field-scale modeling and the strength with which opposing views are held5 it is not clear that a single state of the art can be agreed upon for conservative solute transport. Codes that embody state-of-th~art process models are relatively few in number. Because alternative models are so varied in their mathematical and computational structure, virtually no single code embodies alternative models enabling the study of opposing
RESEARCH NEEDS 265 views. Major field-scale experiments conducted in the recent pant (Betson et al., 1985; Mackay et al., 1986; Thurman et al., 1986) have not provided data sufficient to discriminate between alternative the- ories. Thus, the question of validity, especially for extrapolations to predict future events, remains unanswered. Some scientists who have played a dominant role in developing probabilistic theory now advo- cate the use of such modem only by qualified professionals (Freeze et al., 1989~. Probabilistic methods are sufficiently more complex than accepted single-valued deterministic models that caution should be used by the uninitiated, because mode} results may be instrumen- tal in decisions influencing large populations and significant resource commitments. Validation or Accuracy Estimation Methods The validation of a site model, equated here to an assessment of accuracy, requires an acknowledgment of the origins of uncertainty in models of the subsurface. Characterization of actual sites is al- ways uncertain, i.e., incomplete, and consequently, the next data point sampled may reveal a new feature of the site and dramatically alter the established conceptual model. In a very real sense, the truth about a site is never known, and hence absolute comparisons and statements can never be made. What can be addressed are the uncertainty in measured parameters and the influence of that uncer- tainty on the simulation of system behavior. Those aspects of a site that cannot be quantified and simulated with a single mode! must be addressed by ad hoc simulations of equally probable interpretations of the site, i.e., alternative conceptual models. Calibration and validation are areas of research in ground wa- ter modeling. Site-specific data on initial and boundary conditions, along with material and fluid properties, when combined with com- puter software or codes, form a mode} of the site. A calibrated mode} must incorporate or explain all observations of a site. Through calibration and validation, we strive to completely understand the geology, hydrology, and geochemistry of a site. Throughout this pro- cess, it is unportant to acknowledge the relationships between scales of observation and modeling. Clearly, when the spatial and temporal scales of observation and the mode! do not match, one must inter- pret observations to match mode! scales before making comparisons. Calibration is undertaken in two ways: automated as an inverse
266 GROUND WATER MODELS or parameter identification method or ad hoc as a trial-and-error method. Both approaches benefit from advances made in the last decade in the interpretation of field data. Kriging methods enable us to obtain best estimates of interpolated variables (e.g., hydraulic con- ductivity and hydraulic head), and at the same time they enable us to quantify estimation error. While couched in assumptions that char- acterize the connectivity of the physical environment, these methods have significantly broadened the view of what is possible in error and accuracy assessment. Recently, research has been directed toward nonparametric methods that will enable the blending of soft (quali- tative) and hard (quantitative) data. Such methods will also make it possible to blend dissimilar data derived from different measurement methods or sampling techniques. Advances in Computer Hardware and Numerical Methods Current supercomputers and array processors embody an archi- tecture that facilitates vector-based processing, and future architec- tures embody multiple central processing unit (CPU) designs. The availability of hardware is an important influence on future mode! developments. Supercomputers are becoming more widely available throughout government laboratories and universities. Of perhaps greater importance are the powerful desktop systems that are now becoming available. The computational power of these very afford- able systems rivals that of many moderately sizer] mainframes. The speed and memory capabilities of new systems, when combined with multigrid, conjugate gradient, and moving front techniques, have enabled researchers to solve very high resolution problems. Cases requiring a million or more nodes are within reach for both two- and three-dimensional problems; however, only physical processes such as convection and dispersion have been simulated. Hydrogeologists have always been quick to take new ideas and put them to work in models, and this trend is expected to continue as new theoretical ideas develop and new computer hardware becomes available. There are two potential areas that could yield significant scientific returns in modeling. First, the effort to construct more meaningful models, especially those that involve a linkage between geochemistry and hydrologic flow, seems to result in relentlessly in- creasing needs for greater computing resources. In other words, if the mode! is large and difficult to handle, the response of many workers
RESEARCH NEEDS 267 in the field seems to be to "get a bigger computer." It may now be appropriate to reevaluate the approach to the need for more realis- tic models. Instead of simply buying bigger and better computers to handle the new codes, perhaps new mathematical methods for improving the computational efficiency of existing codes should be considered. One possible breakthrough in this area has been pub- lished by Meintjes and Morgan (1985) and Morgan (1987), in which they develop a methodology for optimizing the numerical solution of sets of simultaneous equations developed to describe chemical reac- tions in moving fluids; at present, the method is limited to small sets of equations, but additional mathematical research might make it useful for larger systems. As an aside, mathematical research might be more profitably conducted by individuals and by organizations that are not as richly endowed in computing resources as some of the national centers. Second, another possible way to back away from the need for increased computer power might be to limit the magnitude of the problems to be studied and, with such a limitation, to reduce the overhead of computing resources that must be carried. For example, the databases that are part of very large geochemical models might be selectively reduced in size to be more appropriate for specific problems. Examples could include the removal of thermodynamic data for nonessential radionuclides when modeling the chemistry and movement of plumes of dissolved materials from landfi~Is or the cooling ponds of conventional power plants. Artificial Intelligence and Expert Systems Another important theoretical push in modeling could come from the field of artificial intelligence, more specifically, expert systems. Expert systems are an emerging technology that could have a signif- icant impact on ground water modeling methodologies as they exist at the present time. In essence, an expert system is a computer pro- gram that attempts to capture expert knowledge in a consistent and organized way in order to solve real-worId problems. Commercial and prototype systems have been developed for many different kinds of applications. Earth science applications so far have included a variety of different aspects of geographic information systems (e.g., map design, terrain feature extraction management of geographic databases, and geographic decision support (Robinson and Frank, 1987~), interpretation of data from geophysical logging (Bonnet and
268 GROUND WATER MODELS Dahan, 1983; Smith and Baker, 1983), correlation of lithologic data between boreholes (Rehak et al., 1985), advice in screening areas for potential ore deposits (Dude et al., 1979) or organic chemicals as potential ground water contaminants (Lu~vigsen et al., 1986), and advice in estimating parameters for contaminant transport models (McClymont and Schwartz, 1987~. Different approaches can be used to solve problems. For exam- ple, the earlier chapters stress an approach to predicting contaminant distributions based on formal reasoning methods in a general math- ematical framework. Expert system approaches look at problems using knowledge-based techniques of reasoning (Hayes-Roth et al., 1983~. Thus the key ingredient of an expert system is a knowledge base, which contains judgments, rules of thumb, intuition, and ad- vice about the problem at hand. A further requirement for an expert system (Fenves, 1986) is facilities for manipulating the knowledge base (e.g., displaying, searching, and modifying) and controlling the knowledge base (e.g., extracting information for use). This latter element of an expert system is sometimes known as the inference engine. Expert systems are also distinguished by features such as the following: a significant capability for interacting with the user, ~ user-friendly characteristics that make the operation of the expert system and the computer transparent to the user, and facilities to provide advice, answer questions, and justify con- clusions (Fenves, 1986~. It is not the purpose here to describe the details of expert sys- tems. Information on the different ways of structuring knowledge within a knowledge base, building a system from scratch or with var- ious software tools, and testing a system is provided in introductory textbooks such as those by Goodall (1985), Waterman (1986), and Weiss and Kulikowski (1984~. What follows is a discussion of appli- cations relevant to the problem of contaminant transport modeling and a view of how expert systems could be used in the future. Some Existing Applications A few systems are available that illustrate how expert systems could be used in contamination-related problems. However, all of these systems are essentially prototypes that have just begun to exploit the potential of this important tool. The discussion here will focus on the application of an expert system as an "intelligent
RESEARCH NEEDS 269 assistant" in hazard evaluation and in selecting parameters for use in contaminant transport models. An example of how an expert system can be used in hazard eval- uation is provided by the prototype system DEMOTOX (Lu~vigeen et al., 1986~. This system is designed to help evaluate the potential for ground water contamination caused by a variety of organic con- taminants added to a soil. It works by ranking contaminants using a mobility and degradation index. The index is the ratio between the time required for a contaminant front to travel through a soil zone and the half-life for biodegradation small values indicating a greater contamination potential (Lu~vigeen et al., 1986~. Ranking is a function of not only the mobility ratio but also con- fidence factors that reflect the quality of the available data and expert system estimations, as well as additional confidence constraints sup- plied by the user. This confidence factor, ranging between 0 and 1, is multiplied by the mobility and degradation index to find the final value for classification. Thus in the absence of hard informa- tion about a particular contaminant, it is conservatively ranked as a bigger threat to contaminate ground water. Although currently a prototype, this system contains 200 rules, more than 250 facts, and numerous explanations. It is constructed using expert system too! M.1. A second example in the application of expert systems for hazard evaluation is provided by the work of Law et al. (1986~. The specific application is the determination of ground water flow directions and the permeability of units at a site. The knowledge base for this pro- totype system is a series of if/then production rules and an external function that establishes ground water direction as the solution to a three-point problem. One of the most comprehensive computer systems related to the problem of ground water contamination is Expert ROKEY (Mc- Clymont and Schwartz, 1987~. The main components of this system are a contaminant transport model, two expert systems (EXPAR and EXINS), and a plotting package. The transport model, which is an analytical solution from Domenico and Robbins (1985), describes the transient, thre+dimensional spread of a dissolved contaminant in a unidirectional flow system. The mode} accounts for advection, dispersion, sorption, first-order decay, and time-varying loading of the source. The EXPAR system helps users prepare a set of input data for the model. This operation is coordinated by a set of com- puter forms that serves as the main user/system interface. Each
270 GROUND WATER MODELS form can accept parameters supplied by the user or developed with one of the family of expert systems. These expert systems access production rules, assistance programs (conventional programs that calculate parameters, e.g., hydraulic conductivity from grain size), and appropriate case studies. Like many expert systems, they ex- plain the meaning of questions and why the questions were asked, provide general tutorial information about a process (e.g., disper- sion), and check the overall validity of derived parameters within the context of a problem. The EXINS system is a demonstration prototype to help plan a monitoring strategy for a first-stage field investigation. The strategy is based on (1) data or information used previously in EXPAR, (2) the results from the transport simulation, and (3) specific responses to questions posed to the user. The entire package was designed for users with minimal expertise in the use of a computer and modeling. As such, it represents one of the important potential uses of expert systems in relation to contaminant transport modeling. McClymont and Schwartz (1987) present a detailed discussion of the method and its application to a practical problem. Expert Systems in the Future The systems applications considered so far are conventional in the way they use expert systems. In the future, work with expert systems can be expected to go beyond these applications to look at more fundamental scientific problems. One possibility ~ the use of expert systems as a too} for modeling in what Beck (1987) refers to as "macroscopic logical reasoning." His idea is that for very complex and poorly characterized systems, formal mathematical descriptions using differential equations may not be the most logical way of repre- senting the system. According to Beck (1987), expert systems might be more useful in cases where the system's dynamics are highly non-linear . . ., a theory is in its initial phase of development (e.g., as a verbal conceptual model; crude order must be imposed on a confused and convicting welter of experimental observations; and decision making must be conducted in a setting where a pragmatic, universal shortcut to interdisciplinary communication is a priority. According to Hardt (1986), there is a great deal of similarity between building a simulation mode} and building a knowledge base. Looking at what constitutes the description of a natural problem,
RESEARCH NEEDS 271 Hardt (1986) finds first that the classical approach to mode} formu- lation is "rich with mathematical intuition," and that in formulating and solving the mathematical problem and in interpreting the re- sults, "reality" is being expressed in one, not particularly unique, way. Thus a mathematical mode} evolves as a scientist combines "native common sense" with the thinking tools provided by math- ematics. Hardt (1986) maintains that problems can also be formu- lated and solved very efficiently using cognitive approaches. This nonmathematical procedure involves replacing mathematical equa- tions with qualitative equations. Typically, this approach loses some of the detail of the problem and is less quantitative (Hardt, 1986~. However, if we return to Beck's (1987) message, it may be foolish to think that for poorly described systems the mathematical approach is qualitative. The last area where expert systems may have a role to play is the area of automatic programming. Barstow (1983) examines the potential for constructing software for solving nontrivial problems in the quantitative interpretation of geophysical logs. His prototype mode! was designed to assist users who were not familiar with "tra- ditional computer interfaces in preparing mathematical models for log interpretation, in a matter of minutes. It is beyond the scope of this examination to present an overview of this complex topic, but indications from Barstow's (1983) work are that the impact of an automatic programming system can be quite dramatic. The procedure effectively removes the "programming bottleneck between conceptualization and feedback." The feasibility of using these automatic techniques in areas of contaminant transport modeling is an application that is open for investigation. It is a tantalizing idea to be able to sit down at a keyboard and in a matter of minutes generate a prototype model. This last potential application illustrates that the potential of expert systems in hydrogeology is limited only by our imagination. Interdisciplinary Efforts There is a recognized need to revise our concept of modeling and modelers. A tendency exists to describe ground water models, the application of those models, and the necessary research as a logical progression and, thereby, leave the impression that modelers believe that another generation of models is the answer. In reality, this would perpetuate a myth. Although a logical step in mode! research
272 GROUND WATER MODELS is the development of standard characterizations of accuracy and methods that provide an assessment of accuracy, it is recognized that evaluation and improvement of mode} accuracy are not sufficient. The broader issue to be dealt with is needed research into the art of applying models to accurately simulate the subsurface; the accuracy of a ground water mode! is only one component of this broader topic. There is a need to provide through integrated efforts the interdis- ciplinary technologies necessary to discover, characterize, and solve ground water contamination problems. These allied technologies in- clude measurement technologies needed to characterize sites, initially detect releases, and continuously monitor plume advance. They also include optimization techniques applied to guide sample network de- sign (i.e., location and frequency) and to guide water management practices (e.g., safe withdrawal, minimized cost of supply, or mini- mized cost of containing and treating). Remediation methods and models of their potential effectiveness require an interdisciplinary effort (e.g., in situ treatment using bioremediation methods will be successful only if predicated on a knowledge of contaminant location, i.e., physical behavior) and chemical and biological process models known to accurately describe subsurface response. Risk assessment methodologies, which integrate the probability of occurrence with the consequence of occurrence, are yet another area of interdisciplinary research. Currently, risk assessment methods applied to ground wa- ter systems are simply based and are used for screening or scoping newly discovered problems; however, more sophisticated modeling capabilities are needed for remediation studies. E - entially, the deci- sionmakers are more interested in a quantification of risk than in a quantification of contamination level. The measurement and interpretation of pollution events offer significant challenges to measurement technology. Instrumentation has evolved to measure more accurately the behavior and effects of contaminants. This ~ a result of increasingly available chemical information and recent advances in material science. The use of fiber optics to transmit signals from sensors is one area of improved instrumentation. Standard and nonparametric kriging methods are being used to develop designs for sampling the environment and interpreting field data. Optimization techniques are being used to determine the location and frequency of sampling; however, it must be appreciated that the design is optimum only with respect to sampling objectives, i.e., detection, monitoring, or characterization.
RESEARCH NEEDS 273 One can observe the environment to detect a contaminant, mon- itor a known pollutant, or characterize a site. The design of sam- pling systems depends significantly on the objective. Thus sampling strategies must be based on interdisciplinary knowledge. Detection systems should be clesigned to surround or possibly underlie disposal systems. They might be tailored to only pick up a fingerprint tracer signaling first release. Therefore substantive sampling to determine the character of the contarnunant plume may not be associated with detection sampling. Monitoring of a known plume will have other objectives such as observing the peak concentrations, locating the center of mass of the migrating plume, and determining the mass flux of the contaminant across significant planes in the environment (e.g., zones of capture or remediation and property boundaries). Certainly, monitoring can imply sampling for a broad number of constituents and sampling over an ever-expanding region of contami- nation. Finally, characterization of a site has the objective of defining media~fluid properties, boundary conditions, and initial conditions that govern the flow of water and contaminant migration at the site. One aspect of characterization is sampling to independently define process mode} parameters. The identification of necessary data, suit- able instrumentation, appropriate sample network design, and data interpretation methods is an interdisciplinary effort. Research should proceed toward highly integrated interdisciplinary methods in order to sample the subsurface environment. In situ remediation of contaminated soils and aquifers is a major interdisciplinary activity. Its attractiveness stems from the expense and/or practical problems of excavation followed by proper disposal or incineration. Bioremediation is the most widely applicable strat- egy, because most of the common organic pollutants and some of the inorganic pollutants (e.g., ammonium, nitrate, and sulfate) are amenable to biodegradation, as long as the proper environmental and microbial conditions are present. In situ remediation is particularly advantageous when the contaminants are poorly mobile, because the removal reaction can be close to the source of the contarn~nation. Without in situ reactions, dissolution and flushing of the contarni- nants can require years to decades. The modeling issues and problems discussed in Chapter 4 de- scribe the hydrophobic contaminants and incorporation of chemical and biological reactions that are relevant to any modeling of in situ remediation. In addition, four issues are especially acute during in situ remediation:
274 GROUND WATER MODELS 1. The acIdition and extraction of water through wells or trenches create local nonhomogeneities of head, flow, and solute concentrations. Chemical and biological reactions are likely to be most intense near the nonhomogeneities. Modeling around nonho- mogeneities requires, at a minimum, a tight grid spacing. 2. Flow velocities are often significantly increased in a remedi- ation site in order to flush water and reactants through the ground. The high velocity can alter flow paths and may accentuate the effects on heterogeneities (natural or induced). Therefore modeling that includes heterogeneities is emphasized. 3. The biological and chemical reactions often will alter the permeability of the soils or aquifer, especially near the introduced nonhomogeneities. Thus models must include the interactions of flow and reaction. 4. The mode] must keep track of at least two reacting species: the contaminant and the added material that reacts with the contam- inant. Their removals usually are linked stoichiometrically, but one or both can control the overall reaction rate. Often, many species must be followed, including products, and these species may be affected in very different manners by other mechanisms, such as sorption or volatilization. Another area of interdisciplinary research involves the disposal of liquid hazardous wastes by subsurface injection through wells into deep aquifers. This technique began in the United States in the 1950s and 1960s and was seen as a relatively inexpensive way to prevent pollution of rivers and lakes. Depths of injection typically range from 0.25 to 1 mi below the surface (Gordon and Bloom, 1986~. The liquid wastes most frequently injected into the subsurface are corrosive and reactive liquids, organics, and dissolved metals. In 1983, EPA identified 90 facilities in the United States where 195 wells were being used for disposal of hazardous wastes (Brasier, 1986~. Subsurface injection is the predominant form of hazardous waste disposal in the United States, accounting for 60 percent, or ap- proximately 10 billion gal. In contrast, only 35 percent of hazardous wastes was disposed of in surface impoundments and 5 percent in landfi~Is in 1981 (Gordon and Bloom, 1986~. The predominance of subsurface injection as a method of disposal is largely due to the low cost in relation to other technologies. Until recently, little, if any, treatment of the wastes was required before injection. As with other
RESEARCH NEEDS 275 other methods of waste disposal, usable ground water has been con- taminated by escaping toxic wastes from injection facilities (Gordon and Bloom, 1986~. A majority of the subsurface injection facilities are used by the chemical and petrochemical industries located in Texas, Louisiana, Ohio, Michigan, Indiana, and Illinois. All wells used for injection of hazardous materials are subject to control by the Safe Drinking Water Act (see discussion in Chapter 5) and the Resource Conservation and Recovery Act (see Chapter 5~. Prior to the initiation of injection, a vast array of chemical, phys- ical, geological, and hydrological parameters should be considered. Chemical and physical factors include density, reactivity, viscosity, temperature, content of suspended solids, content of gases, pH, Eh, stability, and volatility. Geological and hydrological factors that should be considered include the permeability and effective porosity of the injection horizon, thickness and integrity of the aquicludes that separate the injection zone from adjacent usable aquifers, possible zones of recharge and discharge, effective porosity, content of clay and other reactivity minerals in the host formation, magnitude and direction of pressure heads, preferred paths of flow, and salinity and reactivity of indigenous water in the formation. The prospect of hav- ing to properly consider such a list of parameters prior to injection would probably cause any potential disposer to hesitate to initiate such a program. The extreme difficulty and cost involved in obtaining adequate field and laboratory data prior to construction of deep-well injection facilities contribute to the increasing use of predictive computer mod- eling. Predictive modeling potentially offers a means to minimize, or at least to optimize, the drilling of numerous test and monitoring wells and possibly to fill existing gaps in knowledge. Prickett et al. (1986) discuss the application of flow, mass transport, and chemical reaction modeling to subsurface liquid injection. They point out that modeling is necessary for estimation of pressure buildup rates at the injection well and of distribution of pressure buildup in the reservoir. With regard to transport of contaminants, it would be desirable to include advection, dispersion, sorption, decay, and biochemical re- action, but at present no mode] can deal with the full complexity of the transport and chemical reactivity of a waste in a deep, high- pressure, high-temperature, high-salinity, subsurface environment. Prickett et al. (1986) suggest that, while it is not possible to truly simulate the transport and reactivity of injected wastes, it should be
276 GROUND WATER MODELS possible to mode} the worst-case scenario of conservative transport of all dissolver! chemicals. Strycker and Collins (1987) state that ad- ditional research is needed in virtually all areas of abiotic and biotic waste interactions before definitive explanations can be given of their long-term fate. Clearly, the deep-well injection of hazardous wastes is an area that could potentially benefit from improvements in our capabilities for modeling transport in ground water. To reach this goal, much research is needed in the coupling of transport and chemical models, so that more realistic predictions of the movement and fate of injected chemicals can be made. POLICY TRENDS AND SUPPORT FOR RESEARCH An EPA study found that existing ground water models do not account for all processes affecting the fate and impact of contami- nants. For example, the flow and transport of organic solvents are influenced by the hysteresis in multiphase soil-fluid characteristics and by biotic and abiotic fate processes; neither is accounted for in existing and available codes. It is thought that existing models lack accuracy when confronted with a high degree of heterogeneity, and, in general, it is believed that data requirements to ensure high levels of confidence in the accuracy of predicted results are prohibitively expensive. It is disturbing to know that models lack accuracy; it is worse not to know the accuracy of the model. Models in support of policy and in response to regulation range from generic to fully mechanistic. Generic models often require no site-specific data, embody no attenuation mechanisms, and charac- terize transport as a one-dimensional flow path. The need to prior- itize or rank disposal sites for cleanup actions in the face of limited resources has led to the application of models requiring little or no site-specific data (Whelan et al., 1987; see vertical-horizontal spread mode! case study in Chapter 5~. While applications of generic mod- els will continue, it would be informative to better understand the relationship between the results of such modeling and actual site per- formance. For example, when generic models are used, are the worst sites always identified as being worst, and are all sites ranked in a hierarchy associated with a real risk ranking? At the other extreme, the need to assess environmental impacts from wastes previously disposed of in complex hydrogeologic systems makes it necessary to improve our understanding of complex systems. Thus complexities of
RESEARCH NEEDS 277 process (e.g., organic compounds, dimensionality, and mathematical formulation heterogeneity, anisotropy, spatial variability, fractured media, and karst systems) must be addressed through continued re- search if we are to be able to realistically portray the risk of future events. The siting regulation for new low-level waste (ELM) disposal sites (10 CFR Part 61) states that "the disposal site shall be ca- pable of being characterized, modeled, analyzed, and monitored" (U.S. Nuclear Regulatory Commission, 1987~. Thus the responsi- bility for being able to simulate site performance is a responsibility of the licensee. Furthermore, it is implied that hydrogeologic sys- tems that cannot be characterized, modeled, analyzed, or monitored with confidence are to be eliminated from consideration. Thus the need to regulate LLW sites does not directly justify research on com- plex hydrogeologic systems. This regulation provides no guidance on measures of confidence; however, all subsurface environments are un- certain or unknown to some degree. A logical question is: What level of confidence is necessary before one can claim an ability to mode} or analyze a site? Methods that quantify confidence in ground water modeling results must be developed for application to any disposal site. As models have begun to influence the assignment of liability and the assessment of long-term hazard, modeling results have begun to be viewed as quantitative rather than qualitative. Modeling results are now frequently compared to regulatory limits, and the methods used to make these comparisons are important to the proper por- trayal of modeling results. Reasonable assurance is a concept that has arisen from the study of the potential for deep geologic systems to provide isolation of high-level radioactive wastes. This term refers to the interval between a realistic assessment of poor performance and a regulatory limit. It represents the interval of safety. If reason- able assurance exists that an event is safe, then it is implied that a comprehensive and defensible analysis supports the finding. If a "bounding performance" estimate indicates good perfor- mance (i.e., does not exceed the regulatory limit), then a realistic analysis providing an estimate of mean and uncertainty ranges is unnecessary. Only in the instance depicted in Figure 7.1, when bounding performance exceeds regulatory limit, does one need to perform a realistic analysis. A realistic analysis is essentially an ef- fort to demonstrate regulatory compliance when realistic rather than bounding models and mode} parameters are employed. Of course,
278 Good Performance Extreme Realistic Performance Estimate ~\\ ~I Uncertainty in Estimate | GROUND WATER MODELS Poor Performance Extreme Regulatory Limit Bounding Performance Estimate 1. ~ Reasonable Assurance FIGURE 7.1 The relationship of reasonable assurance to bounding analysis, regulatory limit, and realistic estimates. when realism ~ introduced, so is uncertainty, and it must be quan- tified to the extent practical. This same logic suggests that after compliance is triggered by conservative modem (used in the prioriti- zation of sites), more realism and certainty should be required if the output of a mode! is used directly to trigger additional regulatory action than if the mode} is used as an interpretative too! to better understand how contaminants migrate. Currently, EPA is adopting an approach for pesticide regulation requiring differential management of pesticide use based on differ- ences in the use, value, and vulnerability of ground water. This implies a recognition of the value to society of using chemicals. It may also signal movement toward acceptance of "de m~nimus"-based regulations, in other words, regulations based on the detection of chemicals at lower levels. Thus the ability to mode} complex envi- ronments and complex contaminants may become more crucial in the future. Because mode} results are being viewed predorn~nantly as quan- titative in regulatory and litigious settings, accuracy and uncertainty are of interest. However, accuracy per se is difficult if not impossible to assess because the subsurface is always to some degree unknown and uncertain. Indeed, the dominant use of the term uncertainty instead of certainty implies the degree to which the environment is unknown and uncharacterizable. Current research seeks, in part, methods to quantify certainty by relating uncertainty in knowledge of the subsurface to uncertainty in predictions of future events. The "truth" of the subsurface environment is not known; therefore re- search toward methods of quantifying uncertainty must treat the
RESEARaH NEEDS 279 influence of both subjective and objective judgments on mode] pre- dictions. One should be aware that in the application of an overly so- phisticated model, or any model, to a situation that does not merit sophisticated modeling, the level of knowledge implied by such model results can be misleading. When mean values and/or distributions of parameters are purely assumed, assumptions may outweigh knowI- edge, and mode} results may imply a level of knowledge or certainty that does not exist. Methods of uncertainty analysis that include the influence of subjective decisions on mode} results will help to ensure the proper use of models by revealing cases where ignorance outweighs knowledge. A number of governmental agencies are active in subsurface en- vironmental studies; however, it is not clear if this contributes to the problem or to the solution of developing theoretically sound and computationally correct ground water models. For example, hydro- geologic studies are among the least funded research topics by the National Science Foundation. This is the case despite the fact that several federal agencies- including the Departments of Defense, the Interior, and Energy, as well as EPA- support a variety of research and application activities that depend on knowledge of the subsurface environment. Issue resolution, legal or regulatory, will not wait until the perfect solution is found. The field of hydrogeology needs to have established and accepted technology, even if flawed, for application to a host of current problems while science advances. However, simply having an accepted technology does not obviate the need for continued ad- vancement. Within the federal bureaucracy, some division exists for those who fund applications and those who perform research-oriented studies. For example, within the USNRC, the bulk of funding to support research is controlled by those responsible for licensing nu- clear facilities. The foundational belief of any group having licensing responsibility must logically be that sufficiently applicable and defen- sible technology exists today to license needed facilities. Support for research issues requiring {ong-term funding and high-risk approaches may not be within their purview. Management by crisis and/or strongly justified large initiatives appears to be the current mode of operation within government. Ini- tiatives such as acid rain, global climate change, the supercollider, RCRA, and Superfund are examples. EPA is one of the few govern- ment agencies that have as a minion the protection and especially
280 GROUND WATER MODELS the improved understanding of our subsurface environment. Most have the responsibility to quantify the impact of their mission on the subsurface. Often they are charged with simply using existing tech- nology to estimate the impacts of waste disposal, remediation, and so on, on ground water aquifers. Frequently, new initiatives encompass a spectrum of technologies, ground water environs being only one component. Acid rain and global climate change are examples of research investments that embrace ground water issues but may not significantly improve our understanding of ground water flow and contaminant transport. Rather, they will improve our understand- ing of linked processes that, when integrated over significant spatial and temporal scales, serve to estimate the overall response of the environment. Such diversified studies do not significantly advance our understanding of basic physical processes such as dispersion or of ways to directly relate mode! parameters to measurable quantities. It is true that ground water models that consider spatial and tem- poral changes appear to be advanced technology when compared to our understanding of geochemical and microbiological phenomena. However, more advanced methods of ground water characterization and modeling are needed in order to understand with confidence where a contaminant ~ in the subsurface so that the effectiveness of bioremediation methods for in situ treatment of contaminants can be estimated. Government research programs studying interdisciplinary problems need to appreciate the complexities of flow and transport phenomena that are not well understood and, as a consequence, are poorly simulated. An interesting evolution seems to have taken place with regard to predictive modeling from the point of view of regulatory agencies. With the development of comprehensive hydrologic models in the 1960s and 1970s, regulatory agencies seemed to accept the predicted results with a certain amount of awe. The potential power of the approach was obvious to even the most nontechnical member of a regulatory board or agency. The same is true of the introduction of comprehensive geochemical models in the 1970s and 1980s. Again, the sheer power of the methodology was obvious and a bit over- whelming. Although regulatory bodies might not fully understand either the input or the output from such models, they seemed to be willing to accept the word of the experts regarding the usefulness of the predictions. However, in the last five years or so, quite the opposite attitude seems to be developing on the part of the regula- tory agencies. An enormous amount of skepticism appears to have
RESEARCH NEEDS 281 developed, with a resulting attitude of "Prove it!" having replaced the more passive and accepting faith of earlier years. At this time, modelers are in the spotlight, and on the spot, to demonstrate that their long-term predictions are worthwhile and meaningful. This new attitude can only be healthy for the science and art of predictive modeling; it will force the scientists to come to grips with the gaps and unknowns that exist, both in the modem themselves and in the field and laboratory data that are required to validate the models. REFERENCES Barstow, D. 1983. A perspective on automatic programming. Pp. 1170-1179 in Proceedings of the Eighth International Joint Conference on Artificial Intelligence, Karlsruhe, West Germany. Beck, M. B. 1987. Water quality modeling: A review of the analysis of uncertainty. Water Resources Research 23~8), 1393-1442. Betson, R. P., L. W. Gelhar, J. M. Boggs, and S. C. Young. 1985. Macrodis- persion Experiment (MADE): Design of a Field Experiment to Investigate Transport Processes in a Saturated Groundwater Zone. EPRI-EA-4082, Electric Power Research Institute, Palo Alto, Calif. Bonnet, A., and C. Dahan. 1983. Oil-well data interpretation using expert system and pattern recognition technique. Pp. 185-189 in Proceedings of the Eighth International Joint Conference on Artificial Intelligence, Karlsrnhe, West Germany. Cederburg, G. A., R. L. Street, and J. O. Leckie. 1985. A groundwater mass transport and equilibrium chemistry model for multicomponent systems. Water Resources Research 21~8), 1095-1104. Domenico, P. A., and G. A. Robbins. 1985. A new method of contaminant plume analysis. Ground Water 23~4), 476-485. Duda, R. O., P. E. Hart, K. Konolige, and R. Reboh. 1979. A Computer- Based Consultant for Mineral Exploration. Final Report, SRI Project 6415, Artificial Intelligence Center, SRI International, Menlo Park, Calif. Erdahl, B. R., J. H. Heiken, and J. Howard. 1985. Workshop on Fundamental Geochemistry Needs for Nuclear Waste Isolation, Los Alamos National Laboratory, N. Mex. June 20-22, 1984. Department of Energy Report CONF8406134, 208 pp. Fenves, S. J. 1986. What is an expert system? Pp. 1-17 in Expert Systems in Civil Engineering, C. N. Kostem and M. L. Maher, eds. American Society of Civil Engineers, Seattle, Wash. Freeze, R. A. 1975. A stochastic conceptual analysis of one-dimensional ground- water flow in non-uniform homogeneous media. Water Resources Research 11~5), 725-741. Freeze, R. A., and J. A. Cherry. 1979. Groundwater. Prentice-Hall, Englewood Cliffs, N.J. Freeze, R. A., G. De Marsily, L. Smith, and J. Massmann. 1989. Some Uncer- tainties About Uncertainty. Pp. 231-260 in Proceedings of the Conference on Geostatistical, Sensitivity, and Uncertainty Methods for Ground-Water Flow and Radionuclide Transport Modeling Held in San Francisco, Cali- fornia, September 15-17, 1987. Battelle Press, Columbus, Ohio.
282 GROUND WATER MODELS Goodall, A. 1985. The Guide to Expert Systems. Learned Information (Europe) Ltd., Abington, England, 220 pp. Gordon, W., and J. Bloom. 1986. Deeper problems, limits to underground injection as a hazardous waste disposal method. Pp. 3-50 in Proceedings of the International Symposium on Subsurface Injection of Liquid Wastes, March 3-5, New Orleans, La. Underground Injection Practices Council, As- sociation of Ground Water Scientists and Engineers, Water Well Publishing Company, Dublin, Ohio. Gutjahr, A. L. 1988. Hydrology. In Techniques for Determining Probabilities of Events and Processes Affecting the Performance of Geologic Repositories, Chapter 5. SAND86-0196, Sandia National Laboratories, Albuquerque, N. Mex. Hardt, S. L. 1986. On the power of qualitative simulation for estimating diffusion transit times. Pp. 46~463 in Proceedings of the 1986 Winter Simulation Conference (held in Washington, D.C.), J. Wilson, J. Henriksen, and S. Roberts, eds. Association for Computing Machinery, New York. Hayes-Roth, F., D. A. Waterman, and D. B. Len at. 1983. An overview of expert systems. Pp. 3-29 in Building Expert Systems, F. Hayes-Roth, D. A. Waterman, and D. B. Lenat, eds. Addison-Wesley, London. Hoekeema, R. J., and P. K. Kitanidis. 1985. Analysis of the spatial structure of properties of selected aquifers. Water Resources Research 21~4), 563-572. Hostetler, C. J., R. L. Erikson, J. S. Fruchter, and C. T. Kincaid. 1988. Overview of the FASTCHEMTM Package: Application to Chemical Transport Prob- lems. EPRI EA-5870-CCM, Vol. 1, Electric Power Research Institute, Palo Alto, Calif. Jacobs, G. K., and S. K. Whatley. 1985. Conference on the Application of Geochemical Models to High-Level Nuclear Waste Repository Assess- ment: Proceedings, Oak Ridge, Tenn., Oct. 2-5, 1984. NUREG/CP-0062, ORNL/TM-9585, U.S. Nuclear Regulatory Commission, Washington, D.C. 126 pp. Kirkner, D. J., A. A. Jennings, and T. L. Theis. 1985. Multisolute mass transport with chemical interaction kinetics. Journal of Hydrology 76, 107-117. Law, K. H., T. F. Zimmie, and D. R. Chapman. 1986. An expert system for inactive hazardous waste site characterization. Pp. 159-168 in Expert Systems in Civil Engineering, C. N. Kostem and M. L. Maher, eds. American Society of Civil Engineers, Seattle, Wash. Ludvigsen, P. J., R. C. Sim, and W. J. Grenneg. 1986. A demonstration expert system to aid in assessing ground water contamination potential by organic chemicals. Pp. 687-698 in Computers in Civil Engineering, Proceedings of the Fourth Conference, W. T. Lenocker, ed. American Society of Civil Engineers, Boston, Mass. Mackay, D. M., D. L. Freyberg, P. V. Roberts, and J. A. Cherry. 1986. A natural gradient experiment on solute transport in a sand aquifer, 1. Approach and overview of plume movement. Water Resources Research 22~13), 2017-2029. McClymont, G. L., and F. W. Schwartz. 1987. Development and application of an expert system in contaminant hydrogeology. Unpublished report for National Hydrology Research Institute, Environment Canada, 206 pp.
RESEARCH NEEDS 283 Meintjes, K., and A. P. Morgan. 1985. A Methodology for Solving Chemical Equilibrium Systems. General Motors Research Laboratory Report GMR- 4971, Warren, Mich., 28 pp. Morgan, A. P. 1987. Solving Polynomial Systems Using Continuation for Engineering and Scientific Problems. Prentice-Hall, Englewood Cliffs, N.J., 546 pp. National Research Council. 1984. Groundwater Contamination. Studies in Geophysics. National Academy Press, Washington, D.C., 179 pp. Niederer, U. 1988. Perception of safety in waste disposal: The review of the Swiss project GEWAHR 1985. Pp. 11-26 in Proceedings of the GEOVAL 1987 Symposium in Stockholm, April 7-9, 1987. The Swedish Nuclear Power Inspectorate, Stockholm. Prickett, T. A., D. L. Warner, and D. D. Runnells. 1986. Application of flow, mass transport, and chemical reaction modeling to subsurface liquid injection. Pp. 447-463 in Proceedings of the International Symposium on Subsurface Injection of Liquid Wastes, March 3-5, New Orleans, La. Underground Injection Practices Council, Association of Ground Water Scientists and Engineers, Water Well Publishing Company, Dublin, Ohio. Rehak, D. R., R. R. Christiano, and D. D. Norkin. 1985. SITECHAR: An expert system component of a geotechnical site characterization work bench. Pp. 117-133 in Applications of Knowledge-Based Systems to Engineering Analysis and Design, C. L. Dym, ed. American Society of Mechanical Engineers, Miami Beach, Fla. Robinson, V. B., and A. U. Frank. 1987. Expert systems for geographic information systems. Photogrammetric Engineering and Remote Sensing 53~10), 1435-1441. Smith, R. G., and J. D. Baker. 1983. The dipmeter advisor system: A case study in commercial expert system development. Pp. 122-129 in Proceedings of the Eighth International Joint Conference on Artificial Intelligence, Karlsrnhe, West Germany. Strycker, A., and A. G. Collins. 1987. State-of-the-Art Report: Injection of Hazardous Wastes into Deep Wells. Report NIPER-230, National Institute of Petroleum and Energy Resources, Bartlesville, Okla., 55 pp. Thurman, E. M., L. B. Barber, Jr., and D. LeBlanc. 1986. Movement and fate of detergents in groundwater: A field study. Journal of Contaminant Hydrology 1~1/2), 143-161. U.S. Nuclear Regulatory Commission. 1987. Low-Level Waste Disposal Licens- ing Program Standard Review Plans. NUREG-1200, Washington, D.C. Warren, C. J., and M. J. Dudas. 1986. Mobilization and Attenuation of Trace Elements in an Artificially Weathered Fly Ash. EPRI-EA-4747, Electric Power Research Institute, Palo Alto, Calif. Waterman, D. A. 1986. A Guide to Expert Systems. Addison-Wesley, Reading, Mass., 419 pp. Weis~, S. M., and C. A. Kulikow,~ci. 1984. A Practical Guide to Designing Expert Systems. Rowman and Allanheld Publishers, Totowa, N.J., 174 pp. Westall, J. C. 1979. MICROQL:1: A Chemical Equilibrium Program in BASIC, EAWAG. Swis~ Federal Institute of Technology, Duebendorf, Switzerland. Westall, J. C., J. T. Zachary, and F. M. M. Morel. 1976. MINEQI~A Com- puter Program for the Calculations of Chemical Equilibrium Composition of Aqueous Systems. Tech Note 18, R. M. Parsons Lab., Massachusetts Institute of Technology, Cambridge, 91 pp.
284 ~ ~ INS Whelan' O., D. L. Strange, J. G. Droppo, Jr" B. L. Steeling, and J. W. Buck. 1987. Ibe Remedl~1 Action Prlorhy System TIPSY: ~tbem~t- ~1 ~rmul~tlons. DOE/RL/87-Og, PAL 620O1 Department of Inert, shlugton' D.C. b, O. T~ and V. S. ~lp~thl. 1989. ^ crklcs1 ev~lustlon of recent deveL opponents in ~drogeoche=~1 transport models of re~ctlve multlcbem~1 components. water Resources ~se~rcb 25~1~, g3-108.
