McElhone et al. • HABITAT CHANGE AND CERULEAN WARBLERS 
703 
density was the amount of linear edge relative to 
the total land area at each buffered slop 
(McGarigal et al. 2002). We used the four land 
cover classes for the aerial photograph data set 
and weighted a coniferous forest edge as zero and 
a developed or agriculture edge as one. Foresl- 
nonforest edge density for the NLCD data set was 
the amount of linear edge between deciduous/ 
mixed forest and non-forest patches; forest-forest 
edge density was the amount of linear edge 
created by a road splitting deciduous/mixed forest 
patches. We calculated forest-forest edge density 
because Cerulean Warblers may use gaps in forest 
created by roads (Weakland and Wood 2005. 
Perkins 2006). Roads were not included in the 
1992 NLCD classification scheme (Vogclmann et 
al. 2001, Homer et al. 2004) and the 30-m cell size 
of the NLCD may not account for smaller roads in 
the 1992 and 2001 NLCD. Thus, we incorporated 
a general roads layer into both NLCD data sets, 
including roads as small as jeep trails from the 
U.S. Detailed Streets data set (http://www.esri. 
com/meiadata/csriprof80.dtd) following Mc¬ 
Elhone (2009). 
Count data for Cerulean Warblers at stops 
along the selected BBS routes were obtained from 
BBS staff (USGS, Patuxent Wildlife Research 
Center, Laurel, MD, USA). Stop-level count data 
after 1996 were downloaded from www.pwre. 
usgs.gov/bbsapps/index.cfm. We extracted earlier 
data from the BBS field sheets. Cerulean Warbler 
detections were averaged within the 3-5 year time 
bracket surrounding each aerial photograph or 
NLCD year. 
Statistical Analyses .—Statistical analyses were 
conducted using SAS (SAS Institute Inc. 2004) 
with n. - 0.10. Univariate analyses indicated 
variables were not normally distributed, and we 
translonned them using the most appropriate 
method to achieve normality. We used arcsine 
square root transformation on percent of each land 
cover type and percent forest core area, and a 
logarithmic transformation for maximum forest 
patch, edge density, forest-forest edge density, 
and forest-nonforest edge density. A square root 
transformation was applied to average and 
maximum Cerulean Warbler counts because of 
their poisson distribution (Zar 1996). 
We examined long-term changes in habitat over 
the entire BBS survey period by comparing the 
tour land cover classes (deciduous/mixed forest, 
coniferous forest, developed, agriculture) and 
three fragmentation metrics (max size forest 
patch, forest core area, edge density) between 
the early and middle time periods (68 stops on 2 
routes) and between the middle and late time 
periods (240 stops on 6 routes) with ANOVA 
(Ritchie et al. 1998). Each ANOVA model 
included route, period, and stop within route. 
We compared Cerulean Warbler detections be¬ 
tween the early and middle time periods, and 
between the middle and late time periods using 
ANCOVA (Welsh and Ollivier 1998) with 
variables: route, period, and stop within route. 
Covariates were percentage of each land cover in 
each time period. Stop within route was the error 
term in the ANOVA and ANCOVA models for 
testing differences between time periods. We 
repeated comparisons between the middle and 
late time periods using only stops that had a 
Cerulean W’arbler detected during at least one 
period (i.e.. a presence-only analysis). 
Wc examined recent habitat changes by 
comparing three land cover classes (deciduous/ 
mixed forest, coniferous forest, non-forest) and 
four fragmentation metrics (max size forest patch, 
core forest, forest-forest edge density, forest- 
nonforest edge density) between 1992 and 2001, 
and NLCD data for 1,375 stops on 28 BBS routes 
with ANOVA. We also compared land cover and 
fragmentation metrics between the 1992 and 2001 
NLCD data, using only those stops where at least 
one Cerulean Warbler was detected (344 stops 
from 28 routes). 
We compared mean and maximum Cerulean 
Warbler detections between time periods for all 
stops and for those stops at which the species was 
detected using ANCOVA with the variables: 
route, period, and stop within route. Covariates 
were percentage of each land cover in each time 
period. The analysis of all stops along a route 
allowed us to examine habitat changes across 
landscapes where Cerulean Warblers were known 
to occur. Analysis of stops that had at least one 
Cerulean Warbler detected allowed us to examine 
local conditions where Cerulean Warblers actual¬ 
ly were detected. 
RESULTS 
Aerial Photograph Analysis for 1967/197] 
vs. 1982. —Deciduous/mixed forest cover in¬ 
creased from the early to middle time period at 
68 stops on two BBS routes (Table 2). whereas 
agriculture land cover decreased. The amount of 
coniferous forest and developed land cover did 
not change, and none of the fragmentation metrics 
