Raphael and others 



Chapter 18 



Landscape-level Analysis of Habitat in Washington 



Table 11 Attributes afforest cover within 203-ha circles centered on Marbled Murrelet survey sites, western 

 Washington 



'Significance of univanate analysis of variance, based on transformed variables where appropriate. 

 2 Variable was not included in the discriminant analysis. 



type; as the amount increases, the values of the pattern 

 indices increase. Therefore, using planned contrasts we 

 found that mean patch size of old-growth (t = 4.67, P = 

 0.000) and large sawtimber (t = 3.03, P = 0.003) was 

 greater among occupied sites than among detected and 

 undetected sites and that mean shape index was greater as 

 well (t = 3.64, P = 0.000 for old growth; t = 4.24, P = 0.000 

 for large sawtimber). 



To evaluate the relative contributions of the amounts of 

 various forest cover classes and the pattern of those classes 

 over the 203-ha landscapes, we used discriminant analysis to 

 compare attributes among the three detection classes. For 

 this analysis, we used all of the attributes listed in table 11 

 with the exception of proportion other forest (because all of 

 the proportions sum to 1.00 within any 203-ha area, the 

 proportion of other forest is directly implied by the sum of 

 the remaining proportions). This analysis resulted in a single 

 significant discriminant function (chi-square = 48.8, df= 16, 

 P = 0.000); each detection class differed significantly from 

 each of the other classes. The variables that best discriminated 

 among the classes were old-growth proportion, landscape 

 pattern index, old-growth patch size, large sawtimber 

 proportion, and large sawtimber shape index (table 11). 

 Although the average differences among the detection classes 

 were significant, there was considerable overlap among the 

 sites; R 2 was only 17.5 percent and only about 44 percent of 

 the sites could be correctly classified based on the discriminant 

 function (table 12). 



Discussion 



Landscape-level analysis of amount and configuration 

 of forest vegetation can be a valuable tool for assessing the 

 nesting habitat requirements of murrelets. However, the scale 

 of analysis influenced our ability to predict occupancy in a 

 given landscape. We found the forest-cover attributes within 

 a 203-ha circular area surrounding each survey location 

 were useful predictors of occupancy by the Marbled Murrelet. 

 Both the amount and the pattern of various forest-cover 

 classes differ among occupied, detected, and undetected 

 203-ha sites. Given the strong correlations among the forest 

 pattern and amount attributes, the variables describing the 

 amounts of the various cover classes are probably most 

 useful in describing Marbled Murrelet habitat as it occurs in 

 this sample from western Washington. Among the forest- 

 cover classes, old-growth cover, and to a lesser extent, large 

 sawtimber, seem best to predict murrelet occupancy. Sites 

 occupied by murrelets, as evidenced by nests or circling 

 behavior, have a higher proportion of these mature forest 

 classes than do non-occupied sites. 



More definitive analyses must await completion of 

 additional surveys. The present database is not the result of a 

 survey designed to understand the statewide distribution of 

 the species. Instead, it is heavily influenced by one intensive 

 study (Hamer and Cummins 1990, 1991) and by sites selected 

 at the location of proposed timber sales. Therefore, the set of 

 survey sites we analyzed may be biased. Until more systematic 



188 



USDA Farest Service Gen. Tech. Rep. PS W- 152. 1995. 



