Chapter 3 Uncertainty Analysis in Ecological Studies :

Large-scale simulation models are essential tools for scientific research and environmental decision-making because they can be used to synthesize knowledge, predict consequences of potential scenarios, and develop optimal solutions (Clark et al. 2001, Berk et al. 2002, Katz 2002). Modeling is often the only means of addressing complex environmental problems that occur at large scales (Klepper 1997, Petersen 2000). For example, investigations of global climate change (Wobbles et al. 1999), regional assessments of net primary productivity and carbon dynamics (Jenkins 1999, Peters et al., Chapter 7, Law et al., Chapter 9), and landscape analysis of fire spread (Hargrove et al. 2000) rely heavily on simulation modeling at various scales. However, uncertainty in simulation modeling is often overlooked even though it is a fundamental characteristic of modeling that can be caused by incomplete data, limitations of models, and lack of understanding of underlying processes (Beck 1987, Reckhow 1994, Clark et al. 2001, Berk et al. 2002, Katz 2002, Stott and Kettleborough 2002, Urban et al., Chapter 13). If simulation results are to be useful, researchers must show the reliability of the model output by providing information about model adequacy and limitations, prediction accuracy, and the likelihood of scenarios (Clark et al. 2001, Katz 2002). Uncertainty affects every aspect of modeling (Reckhow 1994, Klepper 1997, Jansen 1998, Katz 2002, Stott and Kettleborough 2002, Urban et al., Chapter 13). Data may contain errors that result from problems with sampling, measurement, or estimation procedures (O'Neill and Gardner 1979, Regan et al. 2002). Incomplete data are a common problem, especially in spatial modeling at broad scales. Models are imperfect because they are simplifications of real systems and always have errors in their assumptions, formulation, and parameterization. Moreover, effects of these errors on model adequacy are often insufficiently evaluated (Beck 1987, Reckhow 1994). In fact, most large-scale models are not fully validated, partly