230

DENG ET AL.

VOL. 37, No. 3

trees and roost trees and at random locations in forest
habitat. Detailed vegetations were collected in 12 nest
stands and 12 random plots. Nesting habitat data were
collected at four sample points 25 m from each nest in
each cardinal direction. Also, four sample points were
located in each cardinal direction and at a random dis-
tance between 100 m and 1000 m from the nest tree to
represent available habitat. This sampling was designed
to address nest-site level habitat selection occurring with-
in a hypothetical Grey-faced Buzzard home range. In
each nest stand, four sample points were measured with-
in both nesting and available habitat to increase the num-
ber of points available for use in the assessment of model
stability (Mueller-Dombois and Ellenberg 1974). We cat-
egorized the forest type (conifer, if the proportion of co-
nifers was >70%; mixed conifer/broadleaf, if the pro-
portion of conifer and broadleaf was near equal;
broadleaf forest, if the proportion of broadleaf was
>70%) and slope position (lower, middle, upper third)
at each site, and measured 12 habitat characteristics. At
each plot center, we estimated slope aspect with a com-
pass, slope angle (%) with a clinometer, and relative can-
opy closure (%) with a spherical densiometer. We mea-
sured tree DBH (diameter at breast height in cm) with
calipers, and tree height with a clinometer. We used a
varible radius-plot method (Mueller-Dombois and Ellen-
berg 1974) to estimate basal area (m2/ha) of conifers
and broad-leaf trees. In addition, we recorded nest and
roost tree species, percent height (roost height/tree
height) relative to the height of the tree, and distance
from nest trees to roost trees. We used the variance of
tree height and variance of tree diameters of all trees in
each sample plot as an index of forest structural hetero-
geneity.

Data Analysis. We pooled data among years after find-
ing no difference using a series of Kruskall Wallis tests
(Zar 1984), with sequential Bonferroni adjustments (Rice
1990). We compared forest types and slope position of
eagle sites and random sites using chi-square analysis. We
estimated the mean slope aspect of eagle nests and roosts
using circular statistics (Batschelet 1981), and compared
aspect of eagle nests and roosts with random sites using
a Watson-Williams test (Zar 1984). We assessed univariate
normality of the variables using skewness, kurtosis, and
probability plots. We assessed the equality of variance of
variables between groups using an ,F-max test. We used
logarithmic and square-root transformations to normal-
ize variables and equalize variances for variables that de-
viated from normal distribution. For analyses, we only
used those variables which approximated a normal dis-
tribution and had comparable variances between groups,
either before or after transformation. We tested the null
hypotheses of no difference in variable means between
eagle nest and roost plots and random plots using mul-
tivariate analysis of variance (MANOVA, Stevens 1996).
For the MANOVA, we used Wilk's Lambda to compare
linear combinations of variables between used and ran-
dom sites. If the MANOVA was significant, we tested in-
dividual variables using a series of t tests with sequential
Bonferroni adjustments. We used discriminant analysis
(DA, Stevens 1996) to model data, to estimate which
characteristics contributed the most to differences be-

tween groups. We used cross validation (Capen et al.
1986) to evaluate the stability of the DA model.

We took detailed measurements of all nest trees (age,
height, DBH, canopy and nest height) using the same
techniques used for measuring trees in sample plots. We
estimated nest tree age by extracting a core sample with
an increment borer and counting the rings. We used cir-
cular statistics to estimate mean orientation of the nest
relative to the tree trunk. We used a chi-square analysis
to test for differences in tree species between nest and
random tree distributions. We used paired-sample t tests
(Zar 1984) to compare nest tree height and DBH to ran-
dom trees within nest stands.

RESULTS

Roost-site Characteristics. Of 86 roost sites we
measured, we used 58 (one each for 30 males and
28 females) as independent samples for analysis.
The distribution of forest types at Grey-faced Buz-
zard roosts differed from random sites throughout
the study area (÷2 = 9.06, df = 2, P < 0.05), with
most roosts in the mixed conifer/river birch forest
type. Position of roosts sites on the slope differed
from a random distribution (÷2 = 9.06, df = 2, P
< 0.05). Forty-seven roosts (81%) were located on
the middle third, eight (14%) roosts were on the
upper third, and three roosts were (5%) on the
lower third of the slopes. Mean aspect at roost sites
differed from aspect at random sites (F = 8.94, df
— 1, 112, P < 0.05; mean aspect = 6.8°, mean vec-
tor length = 0.60, angular deviation = 68.5°). Buz-
zards used 11 tree species for roosting, however
over half (58%) of all roosts were located in three
tree species (Korean larch, Chinese pine, and river
birch). Mean roost height was 9.6 ± 0.5 m in the
roost tree. Roost trees averaged 11.2 ± 0.6 m tall
in height with mean DBH of 16.9 ±0.3 cm.

Roost plots differed from random plots (MA-
NOVA; Wilk's Lambda = 0.51, F = 16.8, df = 10,
109, P < 0.05). The Z-tests indicated most variables
differed between roost and random plots (Table
1 ). Higher canopy closure and taller trees best sep-
arated roosts from random plots in the DA (Table
2). The pooled DA correctly classified 78.6% of the
roost and random plots (Table 3).

Nest-site Characteristics. We found 12 Grey-
faced Buzzard nests and used all nest sites as in-
dependent samples for analysis (Fig. 1). The dis-
tribution of forest types at nests differed from
random sites (÷2 = 15.62, df = 2, P < 0.05), with
most nests in the mixed-conifer/river birch forest
type. Position of nest sites on the slope differed
from an independent distribution (÷2 = 11.26, df
= 2, P < 0.05). Eight nests (75%) were located on