DECEMBER 1999 NORTHErn SPOTTED OWL DENSITY 277
menziesii), and oak woodlands (Zinke 1988). Species char-
acterizing the oak woodlands included tanoak (Lithocar-
pus densiflorus), California black oak ( Quercus kelloggii)
and Oregon white oak (Q. garryana). Many of the red-
wood and Douglas-fir stands also contained a large com-
ponent of the following hardwoods: tanoak, bigleaf ma-
ple (Acer macrophyllum), madtone (Arbutus menziesii),
California bay ( Umbellular/a californica), and red alder (A1-
nus rubra).
Since the late 1960s, the primary silvicultural practice
has been even-aged management involving relatively
small clearcuts (12-24 ha in size) followed by prompt
replanting. About 97% of the study area consisted of
young forests ranging from 0-80 yr old. Residual trees
(left from past logging operations) were a component of
some forest stands and commonly the largest, oldest trees
present.
METHODS
Within STC lands, Northern Spotted Owl survey
boundaries were established apr/or/based on ownership
patterns, topographic features, vehicular access and oth-
er logistic considerations. The resulting study area was
further subdivided due to geographic and vegetative pat-
terns. In a nearby study area, Franklin et al. (1990) de-
termined that areas exceeding 90-130 km • were suffi-
cient to accurately estimate Northern Spotted Owl
density. Three subregions in our study area met this cri-
terion and hereafter are referred to as Klamath (666
kmg), Korbel (392 km 9) and Mad River (208 kmg; Fig.
1). Other isolated tracts of STC property were too small
to be included as separate subregions. Following Thome
et al. (1999), we created six categories of stand age to
classify habitat: 0-5, 6-20, 21-40, 41-60, 61-80, and •80
yr (Table 1). The 61-80 and •80 yr age classes were com-
bined for this analysis, because there was very little area
of one or both of these age classes in the three subre-
gions.
We surveyed the entire STC study area for Northern
Spotted Owls at least twice each season using a complete
and systematic search protocol from I March-30 August,
1991-97. Prior to initiation of surveys, we inspected the
entire study area using 1:24000 aerial photographs. We
plotted call points at strategic locations that maximized
observer ability to solicit and detect responses from owls.
Call points were usually positioned at relatively high el-
evations with unobstructed forest openings to ensure a
clear and far-ranging broadcast of the call. Solicitations
consisted of playing recorded Northern Spotted Owl calls
or vocalizing imitations of calls for a minimum duration
of 10 min. We used a jet boat to access and survey STC
property bordering the Klamath River. All surveys using
this protocol were conducted nocturnally, beginning no
earlier than dusk. If an owl responded to a nocturnal call,
•ts location was plotted, and a daytime follow up effort
was initiated, where an observer attempted to locate the
roosting owl by pursuing responses made to imitated or
recorded calls (Forsman 1983). We captured owls using
noose or snare poles (Forsman 1983) and banded them
with a USGS band on one leg and a plastic, color-coded
band on the other (serving as a unique identifying mark;
Forsman et al. 1996). Sex and age were determined fol-
lowing Forsman (1981, 1983) and Moen et al. (1991).
We calculated forest stand ages using STC's timber •n-
ventory database in Intergraph's CAD system, integrated
with the Modular Graphics Environment 5.0 (Intergraph
Corporation 1994) geographic information system (GIS).
Forest stands were distinguished based on date of harvest
and polygons were drawn around unique forest stands.
Only GIS data from 1997 were available for analysis
Landscape data from 1997 were considered adequate be-
cause the mean annual percent change in the landscape
(from timber harvest) during this study was 0.7 ñ 0.08
[ñSE], 1.0 _ 0.18 and 0.5 ñ 0.16% for the Klamath,
Korbel and Mad River study areas, respectively.
Not all of the land surveyed was owned by STC, be-
cause other private lands (in-holdings) were common
within our study area, and survey boundaries were set by
topographic features and access points rather than own-
ership boundaries. Since GIS coverage was limited to
STC lands, we were able to assess age-class conditions for
90% (599 km 9) of Klamath, 75% (294 km •) of Korbel
and 70% (145 km 9) of Mad River. Despite this, we believe
the GIS coverage was representative of the entire study
area, since most of the landscape was subjected to the
same historic timber harvesting practices that created en-
tire watersheds with similar aged stands. In addition, the
in-holdings and adjacent lands associated with the Korbel
and Mad River subregions (areas with the least GIS cov-
erage) were virtually all private lands zoned for timber
production. We compared the amount of forest in the
five age classes among the three subregions (Table 2)
using Chi-square analysis (Hintze 1997).
We used the Jolly-Seber (J-S) capture-recapture model
(Jolly 1965, Seber 1965, 1982) that allowed for death and
immigration in open populations. We used program JOL-
LY (Pollock et al. 1990) to calculate J-S estimates of an-
nual abundance (Nt). Because population and density es-
timates on STC lands had never been documented, we
were primarily interested in these parameters from the
modeling. We subjectively chose the reduced parameter
J-S model (model D in program JOLLY) to analyze the
data, because reduced parameter models compute abun-
dance estimates with greater precision than models sat-
urated with parameters (Jolly 1982). Ninety-five percent
confidence intervals were calculated as 1.96 (SE [Nt]).
Goodness-of-fit tests (Pollock et al. 1985) in program
JOLLYwere used to determine if the models fit the data.
When goodness-of-fit tests suggested lack of fit, we used
a variance inflation factor, •, based on quasi-likelihood
theory (Burnham et al. 1987:243-246, McCullagh and
Nelder 1989) to adjust variances in models with overdis-
persed data (Lebreton et al. 1992, Anderson et al. 1994).
The variance inflation factor is calculated as Xg/V where
X 9 was the goodness-of-fit statistic with v degrees of free-
dom. Expected values for • are not, on average, different
from 1.0 with models that fit the data, and do not exceed
•4 in models that attain structural adequacy, but may
need variance inflation measures (values of 6-10 indicate
complete model inadequacy requiring an entirely new
model). If • indicated that variance inflation measures
were necessary, the standard error of each population
parameter was calculated as X/•SE} (Anderson et al.
1994).
Empirical estimates of annual abundance (Nt) fol-
lowed criteria established in Franklin et al. (1990), which