Appendix A. Procedure used to assess how representative BBIRD sites were of overall conditions of forest fragmentation across the United States.
To evaluate how well the BBIRD study sites represent overall forest conditions across the United States, we compared forest metrics within increasing radial distances around plot and site centers between the BBIRD sites and 200 sets of random landscapes. The number and distribution of sites and associated plots within each randomized set was designed to mimic the total sample of forested sites in the BBIRD database (before we coalesced some sites within a 50 km radius of one another). These included 267 plots distributed among 27 unique forested sites east of the 100th meridian (range = 131 plots/site and 064360 m between site and plot centers), and 119 plots distributed among 13 sites west of the 100th meridian (range = 133 plots/site and 024675 m between site and plot centers). Thus, the number of random sites in each set was limited to 27 in the east and 13 in the west; the number and distribution of plots within each site varied from 131 in the east and 133 in the west; and the distance between these plots could not exceed 64.36 km in the east and 24.68 km in the west. The sequential steps involved in the randomization process are further detailed below. First, the perimeter of the 48 contiguous United States was buffered inward by 30 km. This was done to avoid or reduce the number of water or no data cells that might skew landscape statistics calculated within larger radii (50, 100, 150 km) around site centers that fell close to Canada, Mexico, or a major coastline. Second, analyses were limited to the area within the buffer east or west of the 100th Meridian, depending on region. Furthermore, sampling parameters were defined such that the number and dispersion of plots at each site conformed to the pattern observed among the BBIRD sites in each region (see above). Third, for each run, 27 sites were selected randomly in the east and 13 in the west. For a point to be selected as a random site center, two criteria had to be met: a) it had to fall on a 30 m grid cell representing a forested cover type in the NLCD, and b) at least 10% of the cells within a 1 km radius had to represent forested cover types as well. Once a site center was suitably located, random points representing individual plot centers were selected within the appropriate distance of the site, one at a time, and each evaluated according to the same two criteria as the site centers. In this manner, 27 random sites and their surrounding 260 plots were located in the east, and 13 sites with 165 plots in the west. Afterward, new site centers were assigned to each based on minimum convex polygons enclosing the plots at each site. This process was repeated 200 times, resulting in 5400 random sites and 52000 plots in the east, and 2600 sites and 33000 plots in the west. The data were stored as GIS point coverages, one per region for random plot centers (east: 200*260; west: 200*165) and one per region for random site centers (east: 200*27; west: 200*13). Percent area covered by forest was calculated within 1, 5, and 10 km of all random plot centers and within 50, 100, and 150 km of all site centers. As with landscape analyses for the actual BBIRD sites, water bodies and areas outside the extent of NLCD coverage were not included in calculations of land cover proportions. Additionally, we calculated selected landscape metrics based on forest/non-forest land cover for random and BBIRD landscapes using the Apack (Mladenoff and DeZonia 2002) and ArcInfo (ESRI 2002) programs. Finally, using SAS we developed sets of cumulative distribution functions (CDFs) with ninety percent confidence bands for random and BBIRD landscape metrics overall and by region (see figures below). The CDFs illustrate two landscape characteristics within a spectrum of radii (1, 5, 10, 50, 100, 150 km): proportion of forest area and proportion of forest core area.
The CDFs for the western BBIRD and random sites follow similar trends within the 90% confidence bands across all radii. This suggests that the proportion of forest and core area around BBIRD sites is not significantly different from that around randomly sampled sites across all scales (radii). A similar pattern was observed in the east, with the exception of the 1-km radius. The higher forest and core forest cover within a 1 km radius among random sites versus BBIRD suggests that BBIRD investigators may tend to avoid sites in large contiguous forest that are harder to access regularly for nest counts. For all other CDFs, the curves for random sites tracked the 90% confidence limits for the BBIRD site curves. We therefore conclude that, overall, the proportion of forest and core area around BBIRD sites does not differ from that around randomly sampled sites across the range of radii, meaning that the sites provide an unbiased representation of the extent of fragmentation within a region.