Hame: 



Chapter 17 



Inland Habitat Associations in Western Washington 



This was accomplished by estimating the distance between 

 gap edges as if the canopy created vertical shadows on the 

 ground. Trees <9 m tall were not considered a part of the 

 canopy. Mean canopy height was calculated from 10 dominant 

 trees in the plot using a clinometer. 



The percent composition and mean values for mean 

 d.b.h., height, basal area, number of potential nest platforms, 

 moss cover, lichen cover, and mistletoe abundance were 

 calculated for each tree species present on each plot. 



Statistical Model 



Stepwise logistic regression was used to compare the 

 structural characteristics of occupied and unoccupied old- 

 growth stands in Washington. A predictive model for the 

 binary dependent variable, defined as occupied and 

 unoccupied stands, was developed to help define those forest 

 characteristics associated with murrelet nesting habitat. 



Logistic regression methods (SAS Institute, Inc. 1987) 

 were used to develop a model for the binary dependent 

 variable which was defined as occupied and unoccupied 

 stands (Hosmer and Lemeshow 1989). Candidate independent 

 variables were selected for inclusion in the model using the 

 stepwise selection procedure. The P-value chosen for allowing 

 a candidate variable to enter the model was 0.05. This value 

 was also used as the criteria for retaining an independent 

 variable in the model at the conclusion of each step. 



For the statistical analysis, all 38 forest variables were 

 treated as continuous variables except for forest zone. Forest 

 zone was divided into two categories, high-elevation, and 

 low-elevation zones. High-elevation zone included stands 

 located in silver fir (Abies amabilis) and mountain hemlock 

 (Tsuga mertensiana) zones. Low-elevation zone included 

 stands located in the western hemlock (Tsuga heterophylla), 

 western red cedar (Thuja plicaia), Douglas-fir (Pseudotsuga 

 menziesii), and Sitka spruce (Picea sitchensis) zones. In 

 addition, the variable ecozone was analyzed as a separate 

 logistic stepwise model, because at a few sites the ecozone 

 value could not be determined. 



Principal Components Analysis (PCA) (SAS Institute, 

 Inc. 1987) was used to create a correlation matrix of all 

 variables and to consider more complex interdependencies 

 among the independent variables. The correlation matrix 

 was used to gauge the degree of association and 

 interdependence between pairs of variables. This helped 

 determine if one variable could be used in the logistic 

 regression model as a substitute for another highly correlated 

 variable. The PCA was not definitive in identifying higher 

 order dependencies in these data. 



Imponance of Independent Variables 



Four methods were used to subjectively evaluate the 

 relative importance of each variable to the model's ability to 

 predict occupancy and the importance of each variable in 

 describing the differences between occupied and unoccupied 

 sites. The first method was to examine the initial chi-square 

 values of each variable before they entered the model. The 



second technique involved examining the step in which a 

 variable was selected by the model. Variables selected earlier 

 in the stepwise selection procedure had more power in 

 explaining the variation between occupied and unoccupied 

 sites than variables selected later in the procedure or variables 

 not selected at all. The third method involved examining the 

 final chi-square values for each variable used in the model. 

 The last technique examined the stability of a variable as the 

 stepwise selection procedure of the model progressed. 

 Unstable variables experienced large fluctuations in chi- 

 square value as each new variable was selected in the stepwise 

 procedure, because of high colinearity with other variables 

 used in the model. 



Tree Characteristics 



The mean structural characteristics of old-growth trees 

 for the six conifer tree species available for nesting by 

 Marbled Murrelets in Washington were calculated by pooling 

 the values for each variable measured for each tree species 

 across all plots. These variables included mean d.b.h.. mean 

 tree height, basal area, potential nest platforms/tree, percent 

 moss coverage on limbs, percent lichen cover on limbs, and 

 mistletoe abundance. This analysis was used to subjectively 

 compare the structure and suitability of tree species in 

 providing murrelet nesting habitat 



Results 



Landscape Characteristics 



Distance to Salt Water 



Highest detection rates (5.9-9.5 detections/survey 

 morning) in Washington occurred in intervals between 16 

 km and 64 km inland, but declined to 0.85 detections/ 

 morning at distances >63 km from salt water (fig. 1). To 

 date, 98.5 percent of all detections have been recorded <64 

 km inland, but this is partly due to the extensive survey 

 effort that has occurred in this zone. The maximum distance 

 at which birds were detected inland was at an occupied 

 stand 84. 1 km from salt water, located on Irene Creek near 

 the Cascade River Drainage in 1992 and 1993. The next 

 farthest occupied stands were located 72 km and 74 km 

 inland. Of the known occupied stands, 36 percent (n - 3 1 ) 

 were located more than 47 km from the ocean. Nests were 

 located an average of 16 km inland, with a maximum distance 

 of 34 km (n = 6). Of the old-growth stands located between 

 and 63 km inland, 20-54 percent were occupied (fig. 1). 

 The percentage of occupied stands declined sharply after 63 

 km, with only 13 percent of stands occupied >63 km from 

 the ocean. 



Elevation 



In Washington, detection rates declined sharply with an 

 increase in elevation over 1,067 m (fig. 2). The highest 

 detection rates, which ranged from 4.3 to 9.2 detections/ 

 survey morning, were recorded between sea level and 1,067 

 m. Stands located above 1 ,067 m had mean detection rates 



USDA Forest Service Gen. Tech. Rep. PSW-152. 1995. 



167 



