Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 53
3 Formulating and Applying Models in Ecosystem Management “All models are wrong, but some are useful.” G.E.P. Box. INTRODuCTION The documents the committee was charged to review are largely based on models. Models come in many different shapes and sizes, and the ways they are and can be used to inform management decisions vary enormously as well. Therefore, this chapter provides an overview of formulating and ap- plying models in ecosystem management. It begins with a general overview and then progressively focuses on models used in aquatic, and especially riverine, ecosystems. Because there often is controversy over the appropri- ate role and use of models in decision making, the chapter concludes with discussions of the essential role of model testing and evaluation and use of institutional models for integrating knowledge and management. More de- tailed discussion of the models that underlie the National Flow Study (NFS) are in Chapter 4 and of the ones that underlie the Instream Flow Study (IFS) are in Chapter 5; in addition, a detailed discussion of models for use in regulatory decision making is in a recent National Research Council (NRC) report (NRC 2007a), much of which is relevant to the present case. Modeling is the fallible art of trying to represent enough of the com- plexity and processes of real systems to solve a particular problem. Scien- tifically, such representations provide an ability to assemble more complex understandings of complex real systems than would be possible without such aids. They can be used to develop hypotheses that integrate many aspects of complex phenomena. Moreover, application of models can allow better predictions of the outcomes of proposed actions. This use of models sometimes allows more rapid, less costly, and less risky solutions to practi- 3

OCR for page 53
4 HYDROLOGY, ECOLOGY, AND FISHES OF THE KLAMATH RIVER BASIN cal problems to be developed virtually than direct experimentation allows with the real system, especially for the systems discussed below. Models have become indispensable for managing complex systems ranging from transportation systems (including most airline scheduling) to large building structures, as well as routine wholesaling, retailing, and com- mercial systems by engineers, business managers, and economists. In the physical and environmental sciences, conceptual and quantitative models have been central to the development of new theories and practices, espe- cially in attempts to understand cause-and-effect relations in managed river systems, as well as in predictions of how natural systems will behave. Historically, the scientific use of quantitative models began as early as the 1600s in Galileo’s time, and engineering applications became estab- lished in France before the the beginning of the French Revolution in 1789. Modeling now is the accepted approach for improving the efficiency and effectiveness of efforts to understand and manage complex problems. To improve the likelihood that modeling will deliver on such promises, model development and use commonly follows a fairly standardized process, described in this chapter. Scientific progress results when the hypothetical understanding of the system represented by the model diverges from field observations, leading to improvements in the model, field data, understand- ing of the modeled system, and the model’s predictive powers. Conceptual Versus Simulation Models The science of river restoration is still in its infancy. In most river or wetland systems, there is only a partial understanding of the relation be- tween flows, people, and ecosystems (Castleberry et al. 1996), and therefore science cannot yet provide certain predictions about the consequences of policy and management decisions. For this reason, the concept of “learning by doing” has become an accepted part of management activities in many river basins. A key part of the learning-by-doing process is the development of models that can be tested and refined through monitoring and research programs. Examples where modeling plays a prominent role in ecosystem restoration include the CALFED Bay-Delta Ecological Restoration Program (Healey et al. 2007), the Glen Canyon Adaptive Management Program (Walters et al. 2000), the Comprehensive Everglades Restoration Plan (Og- den et al. 2005), and the Trinity River Restoration Program (USFWS/HVT 1999, Schleusner 2006). For the purposes of this discussion, the committee distinguishes be- tween conceptual models and simulation models. Conceptual models serve to organize knowledge and information in the most general way, whereas simulation models attempt to describe system behavior quantitatively, using a series of deterministic or stochastic relations that link processes together

OCR for page 53
 FORMULATING AND APPLYING MODELS IN ECOSYSTEM MANAGEMENT to explore outcomes of different scenarios. The two types of models are often developed in tandem, conceptual models being used to lay the ground- work for restoration and for developing simulation models and simulation models being used subsequently to examine potential responses of system components. An example of this approach is given in the strategic plan for the CALFED Bay-Delta Ecological Restoration Program (CALFED 2000): Conceptual models are simple depictions of how different parts of the ecosystem are believed to work and how they might respond to restoration actions. These models are explicit representations of scientists’ or resource managers’ tacit understandings and beliefs. Conceptual models are then used to develop restoration actions that have a high likelihood of achiev- ing an objective while providing information to increase understanding of ecosystem function and, in some instances, to resolve conflicts among alternative hypotheses about the ecosystem. The process of adaptive man- agement can be enhanced when conceptual models are developed into simple computer simulations that can be used to explore the consequences of alternative options for restoration. The description implies that conceptual models need not be particularly elaborate or precise; their primary purpose is to provide a framework for testing hypotheses and/or to coordinate research or restoration activities within complex systems. Figure 3-1 shows an example of a conceptual model illustrating the landscape of the Central Valley of California. The components of the landscape are represented by a series of boxes, with links between the boxes indicated by arrows. The arrows imply directional pathways, suggesting that processes or actions in one component of the model have the potential to generate a response in another component of the model. Scientists, resource managers, and landowners can (and often do) argue about the importance of the links, but recognizing their existence arguably is the most important step in developing ecosystem restoration strategies. Simulation models go a step further in representing landscape processes and interactions through computer algorithms and subroutines that quantitatively describe how the physical, biological and engineered components of the system interact in response to changes in state variables, such as water flow, sediment transport, and nutrient loading. Simulation models often fail to replicate landscape, riparian, or aquatic processes completely, but they are nonetheless useful because they permit exploration of general trends or serve to demonstrate the connections among a variety of measurable variables describing the physical and biological systems. Ecological modeling often is difficult to operationalize, but if substantial supporting data are available, such models can successfully replicate ba- sic characteristics as water temperature, cross-sectional profiles, and flow velocity. Often, the most difficult task is to establish direct quantitative

OCR for page 53
56 HYDROLOGY, ECOLOGY, AND FISHES OF THE KLAMATH RIVER BASIN FIGURE 3-1 Conceptual model of the Central Valley, California. Diversions include diversions for agriculture. SOURCE: Kimmerer et al. 2005. Reprinted with permission from the authors; copyright 2005, San Francisco Estuary and Watershed Science. connections between the model that describes the hydrologic and hydraulic properties of the river and the ecological requirements of fishes or other aquatic organisms. Examples of connections between flow and ecological models include applications of model strategies to the Colorado River downstream from Glen Canyon Dam, Arizona. Figure 3-2 shows a flow chart of the Grand Canyon Ecosystem Model, which was developed as part of the Glen Can- yon Adaptive Management Program to examine how changes in the op- eration of Glen Canyon Dam will affect physical, biological, and cultural resources of the Colorado River (Walters et al. 2000). This model is an

OCR for page 53
 FORMULATING AND APPLYING MODELS IN ECOSYSTEM MANAGEMENT FIGURE 3-2 The Grand Canyon Ecosystem Model. 3-2.eps SOURCE: Walters et al. 2000. Reprinted with permission; copyright 2000, Ecology and Society. image—bitmapped, could use improvement fixed executable computer program (in Visual Basic) consisting of separate sub- models that simulate the response of system components (boxes) to changes in reservoir operations, recreation activities and power demand (ovals). The model was developed through a year-long process that involved repeated

OCR for page 53
8 HYDROLOGY, ECOLOGY, AND FISHES OF THE KLAMATH RIVER BASIN meetings with scientists, managers, agency officials, tribal representatives, and advocacy groups, who collectively defined the scope of the problem and key modeling issues. The meetings served to not only parameterize the model but also to provide a mechanism for the various interest groups to express their opinions and reach consensus on the model framework and application. Subsequently, the Glen Canyon Monitoring and Research Center was established to collect and maintain the data and information necessary to test the model and further refine its application to managing the Grand Canyon ecosystem. In this report, the committee is concerned with four specific kinds of models. The first three provide important driving variables for a model of the freshwater dynamics of salmonid fish populations: • A hydrologic model that attempts to reconstruct pre-diversion natural flows of the Klamath River, drawing on historical hydrological data, measured physical relationships, and water balance calculations. • A water temperature model used to simulate average water tem- peratures in a linear fashion down the main-stem Klamath River. • A habitat-suitability model that predicts physical aspects of habitat for aquatic species as a function of stream flow. The fourth model is a fish-population model that simulates salmon spawning, egg incubation, fry and juvenile growth, movement, survival and emigration to the ocean. The first three models were formulated somewhat independently and they address very different questions and concerns. The fish-population model attempts to integrate these models by providing model linkages. This integration is limited by the different time steps among the physical models. TYpES OF MODELS AND MODELING AREAS Hydrologic Modeling Often, hydrologic data needed for planning and design of water re- sources systems are either inadequate or unavailable at locations where the projects are built and operated. In such situations, engineers and sci- entists must rely on models to provide information for decision making. Hydrologic simulation models entail the mathematical descriptions of the components and the response of the hydrologic system (watershed or ba- sin) to a series of events during the desired time. The resulting simulation models describe the various phases of the hydrologic cycle by using the laws of conservation of mass, energy, and momentum. The development

OCR for page 53
 FORMULATING AND APPLYING MODELS IN ECOSYSTEM MANAGEMENT and use of deterministic watershed-simulation models require a thorough understanding of the functions of the various components of the hydrologic cycle, as well as an adequate characterization of the spatial and temporal heterogeneities in the processes and the landscape. Generally, a hydrologic simulation model consists of several sub-models that represent different components of the land phase of the hydrologic cycle. These sub-models usually consist of a set of nested relations that ac- count for inputs, outputs, internal fluxes, and storages of water (Fleming 1974). The relevant hydrologic processes of the land phase of the hydro- logic cycle vary substantially from one region to another. In high-elevation basins in the Pacific Northwest, for example, about half of the annual precipitation falls as snow (Serreze et al. 1999); thus, it is important to monitor seasonal changes in the extent and thickness of snow cover. Simi- larly, in arid and semi-arid regions, the water that potentially goes into the atmosphere via evaporation and transpiration is typically much greater than the water that falls on the surface as precipitation. Seasonal fluxes of water as a result of these processes, as well as groundwater flow and agricultural withdrawals, are particularly important in areas such as the upper Klamath River basin. Since the development of the Stanford Watershed Model during the 1960s by Crawford and Linsley (1962, 1966), many hydrologic simula- tion models have been developed (Singh 1995, Wagener et al. 2004, Singh and Frevert 2006). Hydrologic simulation models for watersheds can be classified in many ways, and Singh (1995) provides a scheme based on process description, scale, and technique of solution. Most classifications use various adjectives to characterize the models according to the model- ing properties. Commonly used adjectives that are relevant for hydrologic modeling in the Klamath basin are given in Table 3-1. The structure of a hydrologic simulation model for a watershed or river basin can be simple or complex, depending on how close the degree of conceptualization of the hydrologic components is to the physical re- ality. Several comparative studies of different hydrologic models can be found in the literature. In 1975, the World Meteorological Organization (WMO) compared several groups of models, including explicit moisture- accounting models, such as the National Weather Service River Forecast System (NOAA 1972); implicit moisture-accounting models (also called tank models); and index models, such as the Antecedent Precipitation Index (API) model (WMO 1975). This study concluded that all models perform equally well on humid basins; that explicit moisture-accounting models are superior in semi-arid and arid areas; and that for poor-quality data, simpler models appear to give “better” results, primarily because the complex mod- els have difficulties in accounting for changes in the soil-moisture balance. The decision regarding the best approach for hydrologic modeling de-

OCR for page 53
60 HYDROLOGY, ECOLOGY, AND FISHES OF THE KLAMATH RIVER BASIN TABLE 3-1 Adjectives Used to Classify Hydrologic Models Adjective Description Black box Process descriptions are based on appropriate mathematical functions fitted to data without any regard to the actual physics of the process Conceptual Process descriptions are based on various conceptualizations of the components of the hydrologic cycle Continuous Process is simulated for a long period, which usually includes many storm events. Moisture accounting is used to simulate the state of the process at the beginning of each event Deterministic Processes can be predicted with certainty without any random component Distributed Process descriptions account for variation of watershed characteristics from point to point Event Given the initial state, the process is simulated only for a single storm event of interest Lumped Process description ignores the spatial variation of watershed characteristics Stochastic Process is governed by random phenomena and the theory of stochastic process is used for its description pends on many factors, including the availability of a modeling code for the problem at hand, data, resources, and time. In the Klamath basin, the con- tribution of groundwater to the total annual runoff may be a critical factor, especially as it influences stream flow recession that carries over from 1 year to the next. In addition, agricultural pumping within the basin might affect the shallow groundwater aquifers, which in turn might affect baseflow. The consideration of the role of groundwater will determine whether a model needs an explicit groundwater component (for example, the MODFLOW model from the U.S. Geological Survey [USGS]). Based on the general requirements as outlined in both the Natural Flow Study (USBR 2005) and the Instream Flow Study (Hardy et al. 2006a), several candidate models could be considered for the hydrologic modeling in the Klamath basin. Table 3-2 presents these models along with some of the key characteristics that might help to choose among them. The selected model code should incorporate the processes needed to model the physical system accurately and to provide the information needed to satisfy model- ing requirements. Typically, the models provide flowcharts for determining whether the features necessary for the particular watershed are included. Figures 3-3 and 3-4 provide examples of flowcharts and conceptual dia- grams for PRMS and MIKE SHE models, respectively.1 1 PRMS is a precipitation-runoff modeling system available from the USGS at http://water. usgs.gov/software/prms.html; MIKE SHE is an integrated hydrologic model developed by the Danish Hydraulic Institute available at http://www.crwr.utexas.edu/gis/gishyd98/dhi/mikeshe/ Mshemain.htm.

OCR for page 53
TABLE 3-2 Models for Coupling with the USGS Three-Dimensional (3-D) Groundwater Model MODFLOW Used in Time Klamath Model Code Type Surface Water Groundwater Scale Spatial Scale Reference Before? GSSHA Distributed, event, 2-D overland flow, 2-D, fully Variable, Gridded Downer et al. No and continuous 1-D stream flow coupled typically 2006 <1 day HEC Lumped, event, Hydrologic Hydrologic Variable, Lumped sub- USACE 2007 No HMS/RAS and continuous methods for methods typically basins, cross overland flow, 1-D sections in <1 day stream flow canals HSPF Lumped, Hydrologic Hydrologic Lumped Donigian et al. No ≤1 day continuous methods methods sub-basins 1995 HYDRO- 2-D overland flow, Up to full 3-D Gridded Therrien et al. Yes? SPHERE 1-D channel flow 2004 MIKE SHE/ Distributed, event, 2-D overland, 1-D Up to full 3-D Gridded Graham and Yes? ≤1 day MIKE 11 and continuous channel Butts 2006 MODHMS Distributed, event, 2-D overland, 1-D Up to full 3-D Gridded HydroGeoLogic No ≤1 day and continuous channel flow MODFLOW Inc. 1997 PRMS Lumped, Hydrologic Hydrologic 1 day Lumped Leavesley et al. Yes continuous methods methods sub-basins 2006 WASH123D Distributed, event, 2-D overland flow, Full 3-D Gridded Yeh et al. 2006 No ≤1 day and continuous 1-D canal flow 61

OCR for page 53
62 HYDROLOGY, ECOLOGY, AND FISHES OF THE KLAMATH RIVER BASIN FIGURE 3-3 Conceptual watershed system represented in PRMS. SOURCE: Leavesley et al. 2006. Reprinted with permission; copyright 2006, Taylor and Francis Group.

OCR for page 53
63 FORMULATING AND APPLYING MODELS IN ECOSYSTEM MANAGEMENT FIGURE 3-4 Schematic representation of a watershed in MIKE-SHE model. SOURCE: DHI 2006. Reprinted with permission; copyright 2006, DHI Group. The calibration of a hydrologic model is an extremely important step. Model results are only as good as the model itself, its input data, and its selected parameters. Models typically have two types of parameters (So- rooshian and Gupta 1995): physical parameters and “process” parameters. Physical parameters represent measurable properties of the watershed, such as area and slope. Process parameters are not directly measurable and depend on the particular scales (temporal and spatial) used in the model. Consequently, such parameters need to be determined through a process of “model calibration.” Two common model calibration criteria may be identified. First, the calibrated model must be able to reproduce the recorded historical data satisfactorily. Second, the parameter values of the calibrated model must be consistent with the watershed characteristics. This consistency can be verified effectively if the model parameters are directly related to measured physical parameters in the watershed. Usually that is the case with highly complex models that attempt to mimic the physical processes. During the

OCR for page 53
80 HYDROLOGY, ECOLOGY, AND FISHES OF THE KLAMATH RIVER BASIN lems to which they are applied. Such problems often can be mitigated by changes in the model (perhaps recalibration) and careful interpretation and communication of model results, which are described later. Model Development Following a statement of the model’s purpose, knowledge of processes and structures thought to be most relevant to the problem is assembled. This knowledge can take the form of empirical relationships, observed locally or in similar circumstances, and relationships derived from fun- damental and well-proven principles. Conservation of mass, energy, and momentum are examples of fundamental principles from which relation- ships can be derived. Empirical relationships are inferred from field data by regression or by other types of fitting to equations. At this point, key variables or parameters that can be measured or evaluated in the modeling and testing process must be identified. Field data can be obtained locally or from locales deemed similar to develop empirical databases. Mathemati- cal forms of these empirical and fundamental relationships are then orga- nized into a coherent representation of the system for the purposes of the problem. Simplifications to some parts of the problem often are required to produce tractable forms. In modeling ecosystem responses to flow, it is essential to recognize spatial and temporal differences in scale because the relationships for individual organisms and hydrodynamic processes change markedly with scale. This simplified representation of the problem sometimes must be sim- plified further to allow solution or approximate solution of the mathemati- cal problem. Numerical methods, such as finite-element or finite-difference techniques, often are used to solve relatively complex mathematical repre- sentations. Frequent checks on the stability and accuracy of the numerical solution often are required. At the end of this step, the model of the system is twice simplified from the original real problem, first to create a mathematical representation of the problem and then to create a solvable mathematical representation. Nevertheless, the result is commonly a far more complex and transparent representation of the problem than would be possible without mathemati- cal aid, and a representation that allows integration of diverse types of scientific knowledge and understandings of the system. Calibration A further empirical phase of model development is model calibration. Calibration consists of adjusting some of the more empirical parameters in the mathematical model to fit data observed from the field. Sometimes

OCR for page 53
81 FORMULATING AND APPLYING MODELS IN ECOSYSTEM MANAGEMENT parameters in component sub-models are adjusted against field data and sometimes parameters in several model components are adjusted together against field data. Sometimes calibration is based on data observed in field conditions elsewhere, if local data are unavailable. Having local field data is greatly preferred under problem-relevant conditions to calibrate empirical parameters. However, field data rarely are available to the extent desired within a time frame relevant for the problem. The adjustment of parameters often is done by experts in modeling the type of system being modeled. Such adjustments sometimes are aided by automated algorithms, particularly when calibration parameters are numer- ous. Because often there are many possible sets of parameter values that “fit” field data, the background and understanding of the modeling experts have an important role in calibration. Usually, parameter calibration is lim- ited within a “reasonable” range based on field and modeling experience for a range of similar conditions. The residual differences between observed field data and the calibrated model represent how well the model fits the field data and provide a form of model test. Calibration residuals are a weak form of model test because the modeler had an opportunity to fit or adjust the parameter values to these data. Thus, when the number of parameters in the model is large or similar to the number of field observations, the utility of calibration residuals for model testing can be small. Model Testing and Evaluation Model testing can consist of a wide variety of techniques intended to evaluate and demonstrate the strengths and limitations of a model for par- ticular purposes (Gass 1983, Kleijnen 1995, Beck 2002, Parker et al. 2002). Ideally, model-testing procedures and protocols are established early in the modeling process (Kauffman et al. 2001). Some common forms of model testing include the following items. Software Tests Software tests can occur at several levels and by several means (Kauff- man et al. 2001). Parts or components of the model can be tested separately, in functional units, and then together as a modeling system. These code tests ideally are done by people other than the authors and can be done by a designated “librarian,” a peer-review process, parallel development teams, or a formal individual or group “walk-through” of the code (Ropella et al. 2002, Grimm and Railsback 2005). When programmers understand that others will inspect and test their code, coding tends to be more reliable.

OCR for page 53
82 HYDROLOGY, ECOLOGY, AND FISHES OF THE KLAMATH RIVER BASIN Numerical Tests Numerical tests are used to ensure that the model’s calculations are stable and correct for some well-known cases and solutions. Complex models can be numerically unstable for some cases, and numerical tests can help establish the limits (Sobey 2001). Routine model applications of common software often rely on software and numerical tests done by the model developer and prior applications of the model. Empirical Tests Comparisons with field data at the component or system scales are useful tests of a model. Such tests are stronger if they are done with data sets different from those used for model calibration and over a wide range of field conditions (wet and dry years, for example). Unfortunately, field data often are sparse and unavailable for complete empirical testing over a wide range of conditions. Such empirical tests against independent field data often are called “model validation” studies, but the sparseness of field data usually means that such tests do not fully demonstrate the “validity” of the model for all relevant field conditions. Empirical model testing never is directly available for model applications for nonexisting conditions, such as conditions in the future with alternative solutions (Gass 1983). An ad- ditional problem is the quality of field data; difficulties and errors in field observations make empirical tests of a model less accurate. Model Comparison Tests A large system model often must simplify components or the overall representation of a system relative to detailed models that might exist of the system or system components. Where the detailed model or model compo- nents provide greater confidence in the representation (sometimes they do not), then comparison between the complex and simplified models can pro- vide some insights and understanding of the relative limitations of the two models. Model comparisons often can be made over a wide range of virtual field conditions, thus avoiding the limitations and expense of comparisons of model and field results. However, model comparisons are weaker tests than good empirical tests. Model-comparison results also often are used to assess the numerical errors in the model solution method. Sensitivity Analysis Sensitivity studies quantify the effects of small changes in model as- sumptions on model results. Such sensitivity results provide insights into

OCR for page 53
83 FORMULATING AND APPLYING MODELS IN ECOSYSTEM MANAGEMENT the probable range of error in model results from such causes. Sensitivity results can be useful for interpreting model results and assessing the data quality needed or desirable from field investigations (Rose 1989, Drechsler 1998, Saltelli et al. 2000, Frey and Patil 2002). Expert Evaluation Almost all model results are evaluated by experts in the problem being modeled. Such expert evaluation occurs in model development, calibration, and application. Errors are frequent in modeling complex systems, and expert inspections are often the most readily available and capable means to identify potential errors. Expert review commonly is done internally by the modeling team through both informal and structured processes. Ad- ditional review by local or external experts on the general type of problem of modeling also can be used. Overall, as noted by Quade (1980), “a particularly dangerous myth is the belief that a policy model can be fully validated—that is, proved correct. Such models can, at best, be invalidated . . . . Thus the aim of the validation [testing] (or rather invalidation) attempts is to increase the degree of confi- dence that the events inferred from the model will, in fact, occur under the conditions assumed. When you have tried all the reasonable invalidation procedures you can think of, you will not, of course, have a valid model (and you may not have a model at all). You will, however, have a good understanding of the strengths and weaknesses of the model, and you are able to meet criticisms of omissions by being able to say why something was left out and what difference including it would have made. Knowing the limits of the model’s predictive capabilities will enable you to express proper confidence in the results obtained from it.” Every decision maker has a mental model or understanding of the problem (Gass 1983). However, these mental models are tested only in- directly by political election, appointment, or promotion processes that place an individual in a decision-making capacity. It should be possible for quantitative models based on scientific and technical information to demonstrate greater levels of credibility to supplement, aid, or improve on decision makers’ mental models and ultimately improve the consideration and selection of decisions. Interpretation and Communication of Results Even a perfect model will be useless if its results are not trusted and used for understanding or solving a problem. Model results and their im- plications must be interpreted and communicated for nonspecialists in the context of the problem. The communication of results must often address

OCR for page 53
84 HYDROLOGY, ECOLOGY, AND FISHES OF THE KLAMATH RIVER BASIN two issues: communication and support of insights and results and dem- onstration of the credibility and limitations of the model and its results. Communication of insights from the results, along with the general degree of confidence in them, often is all that can be provided to busy decision makers. However, the model and its results must also be presented and documented in a form that allows technical workers to understand them more deeply. The formal write-up of the model and its results should aid the clarity and depth of the presentation. Documentation and External Review Documentation facilitates training of model users, supports the cred- ibility and transparency of a model (allowing the work to be externally reviewed), and furthers the education of water managers and modelers re- garding the problem being modeled. Documentation also has an important internal quality-control function. Documenting a model and the thought that goes into documentation helps to ensure that a model works, so that its limitations are understood and can be communicated, and future improve- ments can be identified. Peer or external review can be useful for communicating and establish- ing model credibility. Such reviews always provide some technical value for an ongoing modeling effort. The mere expectation of external review can lead to improvement in the technical discipline and presentation of model- ing. However, a credible model review will almost always find some real or potential flaws, so in an adversarial environment, external reviews can be risky. External review can be conducted in stages throughout the modeling process, at the end of model development, or for specific model applica- tions. External reviews usually are more useful if they are integrated into modeling and application of the model. Although the review process takes time and resources that might have been devoted to additional modeling, at least some level of external review is important for quality and credibility. Establishing Model Credibility A primary aspect of model development, testing, and application is establishing the credibility of the modeling effort (Gass 1983). Credibility can be based on • A model’s agreement with specialist or popular notions regarding the system (face validity) • Credentials of the modeler or modeling organization • Technical procedures and protocols followed in model develop- ment

OCR for page 53
8 FORMULATING AND APPLYING MODELS IN ECOSYSTEM MANAGEMENT • Model documentation produced • Tests conducted on the model and its results • Qualifications of advisers or reviewers of the effort • Outcomes of formal external (peer) review • A (long) period over which the model has been used • Current model use • Diversity of situations for which the model has been used • Authoritative (agency) sponsorship of the model or modeling effort Some of these factors that bolster the perceived credibility of a model may have little to do with its actual technical reliability, but the wide per- ception that a model is credible is required for its results to be trusted. Models developed for applications in an adversarial environment must be pursued with particular care. When a model or its results are expected to enter into legal or political proceedings, an especially systematic, tested, transparent, and articulate modeling effort is required, or an especially as- tute follow-up and clarification is required after the results are released. No amount of effort can ensure that a model is perfect. However, following the systematic model-development and application processes de- scribed above can greatly increase the likelihood that a model will be useful for understanding or developing solutions for problems. INSTITuTIONAL MODELS FOR INTEGRATING KNOWLEDGE AND MANAGEMENT The purpose of applied quantitative modeling for ecosystem manage- ment is to provide information and insights to individuals and groups with decision-making and management responsibilities (Geoffrion 1976). These decision makers are in (sometimes competing) institutions that make and support decisions and operations. The purpose of these results and insights is to improve decisions and provide decision makers with greater confidence in the likely effectiveness of their decisions. Modeling and model results can enter decision making in several ways. • Directly determine a decision. Direct adoption of solutions sug- gested by a model is rare. In a few narrow cases, such as selecting the operation of particular hydropower turbines over short periods, model results directly determine decisions. A few water-distribution systems also are operated largely by model results over short periods, mostly as an aid to system operators. • Provide technical support, along with monitoring data and experi-

OCR for page 53
86 HYDROLOGY, ECOLOGY, AND FISHES OF THE KLAMATH RIVER BASIN ence, for operating decisions. Such support is common for the operation of most large water systems. One or more computer models will be tailored to provide specific information to system operators and managers for hourly, daily, monthly, and longer-term operational decisions. Models provide an ability to estimate field outcomes for locations and times when data are unavailable (such as the future) and provide a timely and less-expensive way to explore operational scenarios under a variety of conditions. • Provide a major direct part of the negotiating and decision-making environment. Especially for routine technical decision-making, model use is common. Models can be tailored for such situations and provide re- sults, which, although imperfect, provide consistent and insightful results for decision makers experienced with a routine problem. For nonroutine decision making, where conflicts are more common and models are less well-tailored to the problem, models have less of a role. The use of models in negotiations is discussed in more detail below and has sometimes been successful. • Model results can inform the background for decision making and the decision issue. More commonly, model results provide background information for decision makers, much as any background technical study provides useful information. • Decision makers, their staffs, and ultimately the public are edu- cated in general through use of models and model results over long peri- ods. For most major water systems, agency staff become educated through development and use of models as well as through direct experience with the system. In the course of such exercises, staff develops an understanding of how the system would perform under a wider variety of circumstances than have been directly experienced. Staff also becomes familiar with the models and their strengths and weaknesses. Modeling staff members often are promoted to middle or senior management, where their reliance on models is less direct but was foundational for their understanding of the system. The design and execution of modeling efforts should consider the decision-making environment that they are intended to inform. Several kinds of decision-making environments and their implications for modeling are discussed below. Technical Decision Making A basic difference exists in the use of models in technical and adver- sarial discourse and decision making. Technical and scientific decision making ideally examines a wide range of solutions, eliminating those whose performance is unpromising until a final small set of promising alternatives

OCR for page 53
8 FORMULATING AND APPLYING MODELS IN ECOSYSTEM MANAGEMENT remains, from which one is chosen. There always is uncertainty in all but the most fundamental knowledge (such as conservation of mass, energy, and momentum). As most applied models are based on assumptions be- yond fundamental knowledge (empirical knowledge and often professional judgment), almost all models are imperfect and will err in some manner. There is no such thing as a “scientifically valid” model unless it is based on fundamental principles (Konikow and Bredehoeft 1992). Like other scientific hypotheses, a model can only be invalidated. A model can never be completely validated. In an applied context, all model results must be interpreted and judged. Uncertainty always exists (Oreskes 2003). When there is consensus on objectives, technical and scientific decision making is quite successful. Quantitative models are routinely used and trusted for major water and environmental decisions every day. National Weather Service models of storms and floods have been tremendously effec- tive for reducing loss of lives and property from storms, even though they are imperfect and their results have significant uncertainties. Outside of environmental applications, quantitative models are relied on for increasing the reliability of buildings and bridges, increasing the efficiency of airline schedules, and countless other practical applications. All of these models re- tain important uncertainties, but they do provide insights and a logical basis for conclusions without which decision making would be more difficult. Adversarial Decision Making Where conflict on objectives exists, typical decision-making processes ask more of quantitative models. Adversarial decision making, which dominates legal and political dis- course, is a contest among alternatives or for and against a particular pro- posed alternative. In such a contest, models and model results supporting an alternative are presented by proponents. Proponents attack or discredit models and model results that do not support the proposition. Adversaries to a particular proposal take the opposite view. In such contests, an uncer- tain model or imperfect model results are often easily discredited. Adversar- ial decision making has difficulty in using models and model results without a preponderance of scientific support (Jackson 2006). Communication of model results is especially important in an adversarial process. In adversarial environments, proponents of the status quo will often call for additional study, detailed modeling, or long periods of data collec- tion, particularly if the models are financed by their opponents. For pro- ponents of the status quo, more studies and modeling are always needed. One of the more productive uses of modeling in adversarial situations is to help reshape understanding of a problem and solutions over a long period. This use poses little urgent threat to the status quo position and allows im-

OCR for page 53
88 HYDROLOGY, ECOLOGY, AND FISHES OF THE KLAMATH RIVER BASIN proved understanding and solutions to be crafted for a future time when the political environment is more fluid, as in the aftermath of a major drought, flood, or lawsuit. Negotiations Use of models and model results often is proposed as part of negoti- ated solutions. The original adaptive management (Holling 1978), “shared vision modeling” (Lund and Palmer 1997, Palmer et al. 1999), and gaming approaches all have in common the use of computer models to represent tradeoffs, certainties, and uncertainties in negotiations among conflicting parties. This decision-making environment lies between pure technical- scientific and adversarial decision making. Where there is broad motivation to come to a consensus agreement and realization that technical support is needed for such an agreement, then models can have a useful, even central, role in negotiated decision making. Quantitative models can have several roles in policy negotiations: • A decision-support system for negotiations. Here, computer mod- els form the central venue and technical arbiter for negotiations, constitut- ing a substantially agreed-on technical basis for discussions and comparison of performance for proposed or crafted alternatives. Typically a “neutral” technical and scientific party creates the model support for negotiations or a process in which technical representatives from major stakeholders come to an agreement on a model representation of the problem. • Model results used directly in negotiations. Here, model results are used in negotiations, as any technical study or document would be used. This approach does not require as much consensus on the technical merits of the work, and allows the modeling to have a more peripheral role in the negotiation deliberations. • Preparation for negotiations. Models and modeling results often are used in preparation for negotiations. Each party can perform internal modeling studies to investigate options from their perspective and those of other parties to the negotiations. These investigations can help to form the basis of proposals and critiques offered during the negotiation process. Sometimes such internal modeling studies are performed during the course of negotiations. • Models used to train technical advisors in negotiations. Actual negotiations often are on time frames too short for new modeling studies to be done. In such cases, past model studies, often accumulated over decades, provide negotiators or technical advisors to negotiators with considerable knowledge of promising and unpromising alternatives, as well as insights and concerns worthwhile during negotiations.

OCR for page 53
8 FORMULATING AND APPLYING MODELS IN ECOSYSTEM MANAGEMENT An adversarial process often follows such a period of negotiation. Even if a negotiation leads to a formal agreement, there are opportunities for further negotiation and adversarial decision making in the implementation of any agreement. Regulatory Environments Agencies are tasked with promulgating and enforcing environmental and water regulations, and enforcing laws and property rights. In an ideal world, field-monitoring data would be abundant, precise, and accurate. However, field data are imperfect, typically sparse, and unavailable for hypothetical future conditions. Thus, for routine regulatory proceedings, field data often are unavailable or insufficient alone to make permitting or enforcement assessments. In such cases, quantitative models can have two roles. First, models can interpolate or extrapolate from existing field data (which often are used to calibrate the model and establish boundary condi- tions) and save the agency and the permittee considerable expense and delay for data collection. Models are used to assess the probable environmental or resource effects and the effectiveness of any proposed mitigation actions. These applications all use model results for a regulatory decision. Another role of modeling is for more formal accounting of environ- mental effects. Here, the model is effectively designated as an accounting standard, eliminating human assessment. For water-rights allocations, mod- els—however imperfect they might be—are almost the only practical means to assess water availability in a complex system. The use of quantitative models as a basis for TMDLs and TMDL allocations is a more modest example of the model developing into a standardized understanding of a system. To some degree, the automation of model-based accounting can provide greater transparency and predictability of regulatory decisions, as presumably any party can run the model. The particular type of resource or environmental regulation also can affect the use of quantitative modeling. Where environmental regulation is based on traditional command and control, including specification of required technology, such as specifying particular wastewater treatment processes or so-called best-management practices, routine model use is less important, although models might be useful for determining which tech- nologies should be required. Where regulations specify only a performance standard, regulated parties can use potentially more economical means to achieve the standard, but monitoring or modeling requirements are increased to make the regulations enforceable. For investment in modeling to be worthwhile compared with investment in monitoring, monitoring must be relatively expensive, and models of performance must be relatively good. For market-based regulations (such as water markets or markets for

OCR for page 53
0 HYDROLOGY, ECOLOGY, AND FISHES OF THE KLAMATH RIVER BASIN TMDL), the use of models as an accounting mechanism becomes attractive because it often is an onerous task to have enough density or accuracy of field monitoring to enforce property rights. CONCLuSIONS Despite their scientific imperfectability, models have a variety of uses for ecosystem management, including hydrologic, hydraulic, water-quality, habitat, biological, and management models. The development of these models and suites of models should address many technical concerns, including issues of scale, and should follow a systematic process of devel- opment and application, including testing. Model development and ap- plication also should be tailored to specific management purposes and decision-making contexts. Despite their potential—and often-realized—usefulness in decision making, not all models or modeling efforts help to solve the problems they are applied to. The systematic process of development and application re- ferred to above needs to take serious account of the appropriate potential applications, utility, and limitations of the models being considered. As a result, the modeling process itself may or may not help with achievement of stated purposes. This point leads to the committee’s discussion in Chap- ter 6 of the need for integrated management systems and efforts, because even the best models and the best data will not help informed decisions to be made unless the right questions are asked about the performance of the entire system and how the separate components influence that system performance.