Methods

We developed 337 combinations of vegetation structure classes (Table: Forest structure class definitions) and cover type (Table: Forest cover type classes) to represent existing and potential future vegetation conditions. Cover types were based on the dominant species in the upper-most canopy layer and included several categories of developed land, (e.g., urban, agriculture, etc.). Structure class depended on the size, tree density per unit area, and canopy layering for forests or on the dominant life form and canopy cover for shrublands and grasslands. Our structure classification was carefully designed to address important issues regarding wildlife habitats, fire and fuels, and various commercial forest products. Combinations of cover type and structure class within potential vegetation types formed the basic vegetation state-classes in our models.

Current vegetation data was developed using Gradient Nearest Neighbor (GNN) methods as described by Ohmann and Gregory (2002). This process imputed approximately 1600 inventory plots to 30-m pixels using a statistical relationship between LANDSAT-TM imagery and other geographic data and inventory plots. In general, GNN methods are best at predicting forest structure, (e.g., diameter of dominant and codominant trees) but less accurate for canopy tree species (Ohmann and Gregory 2002). Correlation between predicted and observed quadratic mean diameter of dominant and codominant trees in the Oregon Coast Range was about 0.8, whereas that for tree species richness was about 0.53 (Ohmann and Gregory 2002). Where GNN data were unavailable, we used vegetation composition and structure attributes from Oregon GAP (2006). Cover and structure data were summarized to state-classes within strata of watershed, ownership/land allocation, and potential vegetation type. These estimates of area by state-class by stratum were the initial conditions for our modeling process.

We used state and transition models to project the integrated effects of natural disturbances and management treatments on vegetation. Vegetation composition and structure within plant association strata defined each state. States were connected by transitions that indicated either the effect of successional vegetation development over time, or the effect of disturbance (Hemstrom and others 2004). This approach expanded transition matrix methods and represented vegetation development as a set of transition probabilities among various vegetative states (Cattelino and others 1979, Hann and others 1997, Horn 1975, Keane and others 1996, Laycock 1991, Noble and Slatyer 1980, Westoby and others 1989 ). For example, grass/forb communities might be dominated by closed forest following tree establishment over a period of time or might remain as grass/forb communities following wildfire. Alternatively, management activities or low-severity wildfire may generate more open forest conditions. State changes along the successional, time-dependent paths were usually deterministic, and, without disturbance or management, all the vegetation could ultimately accumulate in one state. Different management scenarios were developed to represent alternative landscape objectives and, hence, management treatments.

We developed and ran our models with the Vegetation Dynamics Development Tool (VDDT) (Beukema and others 2003). VDDT has been used in several landscape assessments and land management planning efforts in the Interior Northwestern United States, (e.g., Hann and others 1997, Keane and others 1996, Merzenich and others 2003) and elsewhere (Hann and Bunnell 2001, Merzenich and Frid 2005). Although VDDT is a non-spatial model, managers and others often need to understand the spatial distribution of vegetation conditions and disturbances. Consequently, we ran models using strata of land ownership and allocation and potential vegetation types within watersheds so that we could display results about the spatial distribution of landscape characteristics without implying pixel or stand-level accuracy. All scenarios were run for 200 years with 30 Monte Carlo simulations to allow the occurrence of rare events and generate estimates of long-term disturbance variability and forest development trends. We compared decadal average area treated with different treatments and disturbances across our three scenarios to examine trends that would have been more difficult to visualize in highly variable annual outputs. Average annual area in various forest types, however, was not as variable and was displayed on a yearly basis.