We propose to address a fundamental scaling dilemma that plagues ecologists and resource managers. Many current issues are concerned with processes operating on scales of landscapes or regions. But our conventional knowledge base is comparatively fine-scale. For example, anthropogenic climatic change broaches issues of regional (biome) or even larger-scale importance, but our best empirical understanding of the mechanisms of ecological response is at the level of the individual plant. Resource managers face a similar dilemma: multiple-use or ecosystem management implies scales of watersheds or landscapes, yet our knowledge base for forest management is founded at the level of individual trees or homogeneous stands. In general, scaling involves a trade-off between resolution (grain, or level of detail) and extent (the area or scope of the study). Large extent comes at the expense of fine-grained resolution of detail, and so, studies that focus on details do so over a small spatial domain, while large-scale studies typically sacrifice detail to embrace coarser-resolution patterns. This scaling trade-off can enforce an incompatibility across scales. Large-scale studies typically are based on different conceptual models and different data, as compared to fine-scale studies. While it might be assumed optimistically that models derived from different empirical bases at disparate scales might nonetheless converge at a common scale, in practice such models cannot be compared rigorously because they have too little in common. We propose to develop a suite of simulation models which address questions at different scales while still maintaining a consistent conceptual and empirical basis. By preserving this commonality among models, we can change scale rigorously as needed, and when using alternative models we can be confident that discrepancies in the predictions of these various models are due to explicit assumptions or formulations rather than to unknowable incompati bilities in the underlying data. We will begin with forest gap model, which simulates the establishment, growth, and mortality of individual trees on a small (<0.1 ha) model plot, at an annual timestep. Gap models typically are used to extrapolate tree-level demographics to stand-level (~10 ha) dynamics over successional time periods. We will then use this model to generate and parameterize three derived models (metamodels) that reproduce selected coarse-resolution aspects of the gap model's behavior, but do so with much greater computational efficiency. We propose to demonstrate this approach with a nonlinear stage-structured model, a semi-markovian patch transition model, and a cellular automaton. Each metamodel emphasizes a different aspect of forest dynamics, and so each is amenable to particular kinds of applications. We have selected the forests of the Pacific Northwest (PNW) as a testbed for the development of this methodology, but the resulting methods will be applicable to a diverse range of forest ecosystems. In particular, our gap model is already in use in the northwest, southwest, northeast, and the southeastern United States. The approach should also be applicable to other types of models. We will make our models and documentation available to other users.
