The reported comparisons among forest types were obtained by post-stratifying the continental fragmentation maps. The approach characterized the pixels of Olsen forest types in terms of fragmentation of a coarser scale (forest, nonforest) definition of forest. Results might differ if fragmentation were calculated separately for each Olsen forest type. In this appendix, we compare pre- and post-stratifying and conclude that the differences were not important for 81-km ² windows.

A pre-stratified analysis was accomplished by using the procedures described in the main paper with the following differences. Five Olsen forest types were selected to represent a range of forest amount and configuration from the South America land-cover map. The forest types were montane tropical, seasonal tropical, dry tropical, tropical rain forest, and southern hemisphere mixed. Five new maps of “forest” were then prepared by including only the pixels of a given forest type in each one. The fragmentation model was then applied to the map of each forest type.

Table 3 lists the percentages of forest type area in different fragmentation categories for each of four window sizes, for both the pre- and post-stratified cases. Note that the case of post-stratification and 81-km ² windows corresponds to the main body of results (Table 2). The statistics in Table 3 are displayed in map format in Figs. 67 through 71.

With pre-stratification, the effect of changing scale is generally the same as before; larger windows contain less interior forest and more edge, transitional, and patch forest. However, the differences between pre- and post-stratifying depend on the specific forest type and window size. For the largest window size, for example, all of the montane tropical forest is classified as patch when pre-stratifying, but only 38% of its area is so classified when post-stratifying. The difference is due to the coincidence of this forest type in relation to other forest types in the vicinity. In contrast, the area of patch in tropical rain forest only increases by about 2% when pre-stratifying at the largest window size. Unlike montane tropical forest, the tropical rain forest is less coincident with other forest types, and thus is less affected by choice of stratification rule.

Pre-stratifying reduces the total forest extent to only the forest type in question. If a forest type is coincident with other forest types, then pre-stratifying will increase apparent fragmentation because the “holes” occupied by other forest types are now considered to be fragmenting agents. The seasonal tropical forest and dry tropical woods types illustrate the effect. There will be little change for forest types such as tropical rain forest that tend to occupy a geographic region alone.

Chi-square tests of association between stratification rule and fragmentation category were performed by forest type at each scale. Significance in these tests indicates that the percentage of forest area in a given fragmentation category depends on the stratification rule. None of the five forest types had significant differences at the smallest window size, and the smallest P values were for montane tropical (P = 0.09) and seasonal tropical (P = 0.02) forest. This indicates that all of the forest types tested are locally dominant within 81-km² windows. For tropical rain forest, the stratification rule did not matter at any window size. The stratification rule mattered only for the largest window size in the dry tropical forest type, for the two largest window sizes in the southern hemisphere mixed-forest type, and for the three largest window sizes for the other two forest types.