Chapter 1
General Principles for Developing Landscape Models for Wildlife Conservation
Joshua J. Millspaugh, Robert A. Gitzen, David R. Larsen, Michael A. Larson and Frank R. Thompson III.
Models are abstract descriptions of systems or processes (Starfield and Bleloch 1991, Haefner 1996). In other words, a model is a formal framework for organizing and synthesizing existing knowledge about an ecological system. Models have become pervasive tools in natural resources management, large-scale planning, and landscape ecology (Shenk and Franklin 2001, Scott et al. 2002). Models help address fundamental questions about wildlife habitat relationships and habitat management. For example, models are useful for evaluating the potential impacts of management alternatives (Morrison et al. 1998, Larson et al. 2004, Shifley et al. 2006), predicting species occurrence (Scott et al. 2002), and assessing economic implications of management decisions (Haight and Gobster, this volume).
Landscape models take many forms, including statistical models that quantify relationships and patterns among variables (e.g., Niemuth et al., this volume; Hepinstall et al., this volume), conceptual models that offer a qualitative construct of a system, and simulation models that project landscape features into the future (e.g., He, this volume; Oliver et al., this volume). Landscape models can produce output that is as difficult to analyze and understand as data from the original system. For examining and presenting the results from landscape simulation models, ecologists need tools that facilitate interpretation of complex multivariate patterns (Shifley et al., this volume). For this reason, visualization tools are often used with landscape models because they make complex data easier to understand (McGaughey 1997, 1999).
Because of the usefulness and widespread application of models, researchers and decision makers should be well informed about potential strengths and limitations of these models. Here, we review principles underlying the construction and use of models, with an emphasis on their application to large-scale wildlife conservation planning. In addition to outlining general principles of modeling, we offer advice about using models in an adaptive management framework, addressing uncertainty, and making models useful and transparent. We also encourage a focus on viability and population objectives (Johnson et al., this volume) in modeling and we present a broadened concept of viability for species of conservation concern and game species as an important measure in understanding wildlife response in large landscapes. To communicate results from landscape models, we need tools for visualizing these results. Therefore, we end the chapter by briefly discussing some basic theory, dangers, and utility of visualization software. We refer readers to other relevant papers and books, such as Box (1979), Starfield and Bleloch (1991), Hilborn and Mangel (1997), Starfield (1997), Williams et al. (2002), Shenk and Franklin (2001), and Scott et al. (2002), that further discuss philosophical considerations of modeling in natural resources.
Uses of Models
Modeling has become widespread in natural resources management because models can be incredibly useful and practical tools. Johnson (2001) defined three categories of purposes for models: explanation, prediction, and decision making.
1 Explanatory models are used to describe or decipher the workings of systems. Such models attempt to identify the mechanisms involved in the system.
2 Predictive models are used to forecast future states of systems or results of management actions. Prediction is a common use of landscape models and allows the user to determine the potential impacts of various proposed management actions (e.g., Shifley et al. 2006). The opportunity to ask āwhat if?ā questions is especially attractive to natural resource managers.
3 Decision-support models are used to identify management strategies that will produce desired results. Optimization techniques are one useful example of decision-support models used in planning resource management (Moore et al. 2000).
A given model may be used for more than one purpose. For example, habitat suitability models may be used to investigate the relative importance of key habitat characteristics and simultaneously predict future habitat suitability. Many of the habitat suitability and population models discussed in this book and elsewhere are decision-support models that allow managers to assess the relative trade-offs of management actions.
Philosophy of Modeling
In this section, we summarize general principles that modelers and end users should consider when working with models, regardless of the model purpose. We re-emphasize points frequently made in introductions to modeling, especially Starfield (1997).
Every Biologist Constructs Models
Some biologists view modeling as a mathematical art of little relevance to real-world management problems. However, every biologist constructs models. Every scientist and manager has an intellectual framework of hypotheses about how his or her focal system is organized, what factors drive changes in key resources, how the system will respond to management actions, and what the major uncertainties and holes are in this framework. Whether these scientists and managers admit it, this framework is the basis for a conceptual model that can be translated easily into narratives, diagrams, pictures, equations, and even computer programs (i.e., into quantitative models).
There are multiple potential purposes for formalizing oneās intellectual framework into a model, whether conceptual or quantitative. Regardless of whether one constructs a landscape simulation model or draws a diagram on the back of a napkin, constructing a model forces biologists to confront their assumptions about the system and the support for these assumptions. It prompts them to consider the most critical uncertainties inhibiting scientists and managers from better understanding the system. It can act as a framework for integrating new information and is a tool for more rigorous thought about the system (White 2001). Finally, it forces the biologists to expose hypotheses and assumptions to critiques from others. In the case of complex, high-profile management decisions, a manager may be unable to recommend and defend (perhaps in court) a course of action without well-developed quantitative models (Swartzman 1996, Starfield 1997, Walters and Martell 2004:3ā4).
Models Are Useful Despite a Lack of Data or Understanding
As frameworks for the organization and synthesis of existing information, āall models are wrong, but some are usefulā (Box 1979). Ultimately, we seek a sophisticated, accurate understanding of natural systems, precise estimates of important parameters and their dynamics, and good knowledge about the specific effects of various management alternatives. In such an optimal situation, we might have at least moderate confidence in model predictions, even though there is still significant uncertainty. For example, even biologists who are skeptical about most models are comfortable using predictive results in this situation (e.g., daily weather forecasts produced from atmospheric models) despite knowing that such forecasts are often inaccurate.
However, in wildlife habitat modeling, we usually possess limited data and an incomplete understanding of the system (Holling 1978; Fig. 1-1). Models can be especially useful tools for decision making and for prioritizing efforts to address these gaps in our understanding. The argument that modeling should not be used unless data are adequate is just as misguided as arguing that no new management actions should be tried unless we completely understand the system and can predict the specific effects with high certainty. Managers have to act in the face of uncertainty; models help them make as defensible a choice as is currently feasible. Similarly, researchers have to justify why they are proposing studies of a particular aspect of the resource. Model building helps us evaluate the relative importance of various influences on a system and identify data that should be collected (Starfield 1997; Shifley et al., this volume).
Fig. 1-1 A classification of modeling from Holling (1978). The x-axis represents understanding of a system (from limited to complete), and the y-axis represents the quality and quantity of data (from incomplete to adequate) that are available for use in model-building. Ecological models typically are based on limited data and incomplete understanding of systems, and thus fall in region 3 (Starfield and Bleloch 1991).
Models Should Be Constructed for Specific Purposes
A model can be seen as a structural framework for our current knowledge and as a tool for exploring uncertainties in our knowledge. To create a useful framework or tool, we need clear, specific objectives for the modeling effort. The purpose of the model should determine its structure; scope, resolution, and complexity; its user interface and output; and how it is evaluated (Starfield 1997, Nichols 2001, Kettenring et al. 2006).
In defining the purpose for the model, we should address multiple issues:
1 Who are the intended end users of the model? What are the technical skill levels of these end users?
2 How will the model be used: for evaluating management alternatives, determining high priorities for future research, communicating what we know to other stakeholders, or simply clarifying for our own benefit what we know and need to learn about the system?
3 What spatial and temporal context do we want to explore? For example, do we care about breeding season patterns only, modeling short-term forecasts or long-term dynamics, a specific management area or an ecological province?
4 How will the model be evaluated?
5 Are we building the model for long-term use? How will it be updated as our understanding of the system improves?
Predicting the Future Is a Lofty Goal
Ecological systems are driven by factors with high variability and unpredictability, and observed ecological patterns are shaped partially by random processes (e.g., Hubbell 2001, Fuentes et al. 2006). Modeling experts understand that even the best model rarely can accurately forecast the future condition of natural systems (e.g., Boyce 2001, White 2001, Walters and Martell 2004:10ā11), except sometimes over short time spans. In the face of this variability, the predictive value of models usually comes not in forecasting the expected future condition of a resource, but in projecting a range of potential conditions given the likelihood of different stochastic events (Clark and Schmitz 2001). However, with increasing time, previously undocumented events or misunderstood processes are likely to move the system beyond a range of variability predictable from our current knowledge (but see Brook et al. 2000).
Ther...