This study was conducted in a 20.25-km2 region on Sonoma Mountain, with elevation ranging from 170 to 748 m (Sonoma County, California, USA; Fig. 1). The size and position of the study area was chosen to minimize the effect of potentially confounding variables such as solar aspect, climate, infection history and land-use history. This predominantly west-facing section of Sonoma Mountain has a Mediterranean-type climate characterized by cool, rainy winters and warm, dry and often foggy summers. Vegetation is primarily composed of distinct patches of mixed evergreen forest (dominated by bay laurel and oak species) in a matrix of exotic annual grassland (Fig. 1). This simple binary study system of host habitat (bay laurel/oak forest) and non-host habitat (annual grassland) was chosen to minimize the effect of multiple dispersal barriers (non-host land cover types). The impact of Phytophthora ramorum on the vegetation of Sonoma Mountain was first observed in 2000, and ongoing studies show that the pathogen is widely distributed across the region (Cushman & Meentemeyer 2005; Davidson et al. 2005). Although detailed land-use history is not known, the study area has experienced a common history of grazing and some logging over the past century.
To examine the effects of landscape heterogeneity, a map of woodland habitat was derived from Airborne Data Acquisition and Registration (ADAR) multispectral aircraft imagery (August 2001, Positive Systems, Inc., Whitefish, MT, USA). ADAR has a spatial resolution of 1 × 1 m, and collects data from four spectral bands (red, green, blue and near-infrared). Pixels were classified into land cover categories using a supervised approach, where field data were used to develop spectral signatures for pixels of known vegetation composition. These signatures were then used to place the remaining pixels in the image into the appropriate land cover classes using a maximum-likelihood classifier (Erdas Imagine 8.7, Leica Geosystems, Norcross, GA, USA). The map was re-sampled using a nearest-neighbour algorithm to 5-m resolution to increase the computational efficiency of landscape pattern metric calculation (ArcInfo 8, ESRI, Redlands, CA, USA). The final study area map contained approximately 3.6 million cells and was composed of two land cover classes: a host-woodland class and a non-host class (Fig. 1). The woodland class was dominated by bay laurel and coast-live oak, but also contained California black oak, Oregon white oak (Quercus garryana), big leaf maple (Acer macrophyllum), California buckeye (Aesculus californica), toyon (Heteromeles arbutafolia) and a small amount of Douglas fir (Pseudotsuga menziesii). The non-host class was primarily composed of exotic annual grassland (75%), but also included a small percentage of other non-host classes such as farm ponds and seeps, agricultural land (vineyards), exposed land (bare ground) and residential developments. The host-woodland and non-host classes comprised 43% and 57% of the total study area, respectively. Overall pixel classification accuracy was 98% based on field plot data described below.