and old-growth forest at nest, roost, and random sites. Habitat variables were ..... 1990). Winters were typically cool and wet, while summers were hot and dry.
HABITAT CONFIGURATION AROUND
SPOTTED OWL NEST AND ROOST SITES
IN NORTHWESTERN CALIFORNIA
by
John E. Hunter
A Thesis
Presented To
The Faculty of Humboldt State University
In Partial Fulfillment
of the Requirements for the Degree
Master of Science
May, 1994
HABITAT CONFIGURATION AROUND
SPOTTED OWL NEST AND ROOST SITES
IN NORTHWESTERN CALIFORNIA
by
John E. Hunter
Approved by the Master's Thesis Committee R.J. Gutierrez, Chairman
Lawrence Fox III
Director, Natural Resources Graduate Program 94/W-289/03/31 Natural Resources Graduate Program Number Approved by the Dean of Graduate Studies Susan H. Bicknell
ABSTRACT
During each breeding season between 1988 and 1992,
nests and daytime roosts were located for all territorial
members of a contiguous population of northern spotted
owls (Strix occidentalis caurina) in northwestern
California. I used guided clustering with Landsat
Thematic Mapper data to map land cover types representing
seral stages of coniferous forest. I produced two maps;
one had six land cover types and the other had only mature
and old-growth coniferous forest. Map accuracy was
estimated to be 76.4% and 83.6%, respectively, and was
determined by comparing land cover map data with randomly
sampled field data. Using these maps and a grid-based
geographic information system, I estimated the amount of
area of each land cover type, habitat heterogeneity, and
the fragmentation, patch number, and patch size of mature
and old-growth forest at nest, roost, and random sites.
Habitat variables were measured within eight concentric
circular plots of 800 - 3600 m radii. I compared the
habitat variables from the 800 m plots among sample
categories because this plot size approximated one-half
distance between the centers of spotted owl territories.
Nest and roost sites were similar, and had less herb and
nonvegetated cover, slightly lower seral stage
iii
heterogeneity, and more mature and old-growth habitat
which was less fragmented and arranged in fewer but larger
patches than random points. Mean amounts of area of
mature and old-growth habitat within 800 m radius plots
were 94.1 ha, 92.0 ha, and 71.8 for nest, roost, and
random sites, respectively. The area of other cover types
were similar between nest, roost, and random sites. Data
from all concentric circular plots were used to estimate
which habitat characteristics changed non-randomly with
increasing distance from owl sites. The area of
herbaceous and nonvegetated, brush, hardwood, and indices
of fragmentation and heterogeneity increased as distance
from owl locations increased. The area of mature and old-
growth forest decreased with increasing distance from owl
sites. My results from the Klamath physiographic provence
of California were similar to results from comparable
studies in Oregon and Washington.
iv
ACKNOWLEDGEMENTS
I am grateful to R. J. Gutierrez for providing me
with the opportunity to participate in this project, and
for his advice, confidence, and support. Many thanks to
Lawrence Fox III and Richard T. Golightly for their
efforts both as committee members and as educators. I
also thank Dave Kirste, John Larson, Mike Martischang,
Kenny Peugh, Greg Schmidt, Kristin Schmidt, and my many
other friends with the Forest Service who helped this
become a reality. Without Antonio Padilla's hard work and
skill in the field I may still be chasing an owl or
recording vegetation data down in some drainage bottom. I
am indebted to Alan B. Franklin for teaching me so much,
and for our many discussions on spotted owl biology. I
thank Kim Adams Hunter for her support and encouragement
which never wavered despite setbacks and frustrations. I
am forever grateful to my parents, James and Cynthia
Hunter, for somehow instilling in me an appreciation for
learning. I also appreciate the help of Ed Biery, William
L. Bigg, David Delaney, Douglas R. Call, David Fix, Colin
Jewett, Richard Smith, Jim Woodford, and the many others
who assisted in various aspects of this project.
v.
Partial funding was provided by the USDA Forest
Service, Pacific Southwest Forest and Range Experiment
Station, Cooperative Agreement No. PSW-90-0013CA.
vi
TABLE OF CONTENTS
Page
ABSTRACT . . . . . . . . . . . . . . . . . iii
ACKNOWLEDGEMENTS. . . . . . . . . . . . v
LIST OF TABLES . . . . . . . . . . . . . . . x
xii
LIST OF FIGURES. . . . . . . . . . . . . .
INTRODUCTION. . . . . . . . . . . . . . . . . 1
STUDY AREA
. . . . . . . . .
. . . . . . . . . 3
MATERIALS AND METHODS. . . . . . . . . . . .
6
Land Cover Mapping . . . . . . . . . . . . . 6
Vegetation Data Collection
. . . . . . . .
8
Landsat Image Processing
• • • • • • •
• 9
• • • • • • . •
• • 10
Accuracy Assessment
. 11
Spotted Owl Data Collection . . . . . Habitat Data Collection and Analysis . RESULTS
. .12
. . . . . . . .
. . . . . . . . . . 17
Land Cover Mapping. . . . . . . . . . . . . 17
Landsat Image Processing. . . . . . . . .
. 17
Accuracy Assessment . . . . . . . Spotted Owl Data •
•
. . . . . . . . . . . . . 22
. . . . . . .
. . . 22
Habitat Analysis DISCUSSION
17
•
•
•
•
•
•
•
•
•
Land Cover Mapping . . . . . . . . vii
•
•
•
• 37
. 37
Page
TABLE OF CONTENTS (CONTINUED) Habitat Configuration . . . . REFERENCES
CITED
37 42
APPENDIXES A.
Variables and methods used for vegetation sampling at random and non-random plots . . 48
B.
Amount (ha) and percentage of area in water around spotted owl nest, roost, and random sites, for eight plot sizes in . . . . 49 northwestern California, 1988-92
C.
Amount (ha) and percentage of area in nonvegetated and herbaceous around spotted owl nest, roost sites, random sites, for eight plot sizes in northwestern . . . . 50 . California, 1988-92. . . .
D.
Amount (ha) and percentage of area in brush around spotted owl nest, roost, and random sites, for eight plot sizes in northwestern California, 1988-92
E.
F.
Amount (ha) and percentage of area in hardwood forest around spotted owl nest, roost, and random sites, for eight plot sizes in northwestern California, 1988-92 . . . . . . . . . . .
. 51
.
52
Amount (ha) and percentage of area in pole and medium conifer around spotted owl nest, roost, and random sites, for eight plot sizes in northwestern California, 1988-92 . . . . . . . . . . . .
53
G.
Amount (ha) and percentage of area in mature and old-growth coniferous forest around spotted owl nest, roost, and random sites, for eight plot sizes in northwestern . . . . . . . 54 California, 1988-92 .
H.
Indices of fragmentation of mature and old-growth coniferous forest around spotted owl nest, roost, and random sites, for eight plot sizes in northwestern California, 1988-92 . . . . . . . . 55
viii
page
TABLE OF CONTENTS (CONTINUED) I.
Indices of heterogeneity around spotted owl nest, roost, and random sites, for eight plot sizes in northwestern California, • • • • 1988-92
ix
56
LIST OF TABLES
Table
Page
1
Definition of land cover classes used
to map the Willow Creek Study Area,
northwestern California, 1990
2
Error matrix for Landsat derived
successional stage map of the Willow
Creek Study Area, northwestern California,
1990. Values are the number of random
points designated as belonging to a
particular cover class relative to the
cover class at corresponding points on
the Landsat derived successional stage
• • map • •
18
3
Error matrix for Landsat derived mature
and old-growth coniferous forest map of
the Willow Creek Study Area,
northwestern California, 1990. Values
are the number of random points
designated as belonging to CF4 or
non-CF4 relative to the cover class
at corresponding points on the Landsat
derived mature and old-growth map . . . . 19
4
Land cover composition (%) of the Willow
Creek Study Area from 1990 Landsat derived
seral stage map and from 1988 USDA Forest
Service timber strata maps . . . . . . 21
5
Landscape characteristics within 800 m radius
plots around spotted owl nest, roost, and
random sites, in northwestern California,
1988-92 . . 24
6
Test statistics and P-values from Conover's
(1971) multiple comparisons of habitat
characteristics from Table 5, from within
800 m radius plots around spotted owl
nest, roost, and random sites in
northwestern California, 1988-92 . . . . 25
LIST OF TABLES (CONTINUED) 7
8
Page
Regression coefficients and slopes for the change in habitat characteristics with increasing concentric circular plot size around spotted owl nest, roost, and random sites in northwestern California, 1988-92
27
Test statistics and P-values for Zar's (1974) comparisons of slopes of the change in habitat characteristics with increasing concentric circular plot size around spotted owl nest, roost, and random sites in northwestern California, 1988-92 .
28
LIST OF FIGURES Figure 1
2
Page Location of the Willow Creek Study Area in relation to California, the Six Rivers National Forest, and the town of Willow Creek . . . . . . . . .
4
Percentage of area in nonvegetated and herbaceous within 8 concentric circular plots around spotted owl nest, roost and random sites, in northwestern California 1988-92 . . . . . . .
29
Percentage of area in brush within 8 concentric circular plots around spotted owl nest, roost, and random sites, for eight plot sizes in northwestern California, 1988-92 . . . . . . . .
30
Percentage of area in hardwood forest within 8 concentric circular plots around spotted owl nest, roost, and random sites, for eight plot sizes in northwestern California, 1988-92 . . . . . . . . . .
31
,
3
4
5
Percentage of area in pole and medium conifer within 8 concentric circular plots around spotted owl nest, roost, and random sites; for eight plot sizes in northwestern California, 1988-92
32
6
Percentage of area in mature and old-growth coniferous forest within 8 concentric plot around spotted owl nest, roost, and random sites, for eight plot sizes in northwestern California, 1988-92 . . . . . . . . 33
7
Mean indices of fragmentation of mature and old-growth coniferous forest within 8 concentric circular plots around spotted owl nest, roost, and random sites, in . . . . 34 northwestern California, 1988-92
xi i
LIST OF FIGURES (CONTINUED) 8
Page
Mean indices of heterogeneity within 8
concentric circular plots around spotted owl
nest, roost, and random sites, in
. . . . 35
northwestern California, 1988-92
INTRODUCTION
Northern spotted owl (Strix occidentalis caurina)
nest and roost sites typically are found at locations with
complex forest structure (Barrows 1981, Forsman et al.
1984, LaHaye 1988, Solis and Gutierrez 1990). Many of
these structural components (e.g., snags, multiple canopy
layers) are common in late seral stage coniferous forests
(i.e., mature and old-growth forests), although these
features can occur in some younger stands (USDI 1992).
Spotted owl nest and roost sites are also found in patches
of late seral stage forest (Blakesley et al. 1992), and
are surrounded by greater amounts and less fragmented
mature and old-growth forest than would be expected by
chance (Ripple et al. 1991b, Meyer et al. 1992, Lehmkuhl
and Raphael 1993). Despite the evidence that owls select
mature and old-growth forest at a variety of spatial
scales, the existence of this relationship in the Klamath
physiographic province has been questioned (California
Forestry Association 1992). Therefore, I evaluated the
influence of habitat configuration on nest and roost site
selection in a contiguous population of northern spotted
owls within the Klamath physiographic province of
northwestern California. I compared the arrangement and
1
2 the amount of area of land cover types around spotted owl
nest and roost sites, and random sites.
STUDY AREA
The 292 km2 Willow Creek Study Area (WCSA) was
located south of Willow Creek, Humboldt County, California
(Figure 1). This area was the site of a long-term
demographic study of spotted owls in which the identity
and locations of all territorial owls was known (Franklin
et al. 1990).
Approximately 90% of the vegetation at the WCSA consisted of Mixed Evergreen Forest (K üchler 1977, Franklin et al. 1990). The overstory consisted of Douglas-fir (Pseudotsuqa menziesii), with a midstory dominated by tanoak (Lithocarpus densiflora), Pacific madrone (Arbutus menziesii), canyon live oak (Quercus chrysolepis), and other hardwood species. Above 1200 m, Klamath Montane Forest ( K ücc hler 1977) dominated, characterized by white fir (Abies concolor), incense cedar (Libocedrus decurrens), and pine (Pinus spp.)
associations. Scattered xeric sites, mostly at lower
ü elevations, consisted of Oregon Oak Forest (
1977),
dominated by Oregon white oak (Ouercus garryana).
Intensive timber harvesting which began in the 1950s,
along with natural environmental conditions, created a
mosaic of seral stages of these vegetation types. During
the study period, approximately 2.1% of the study area was
3
Figure 1. Location of the Willow Creek Study Area in relation to California, the Six
Rivers National Forest, and the town of Willow Creek.
5 logged (unpublished records on file, USDA Forest Service,
Lower Trinity Ranger Station, Willow Creek, CA 95573; USDA
Forest Service, Big Bar Ranger Station, Big Bar, CA 96010;
and Calif. Dept. Forestry, Humboldt Ranger Unit, Fortuna,
CA 95540).
The study area contained rugged, mountainous
terrain and 3 third-order drainages (Franklin et al.
1990). Winters were typically cool and wet, while summers
were hot and dry. Between 1951 and 1980 mean annual
rainfall in Willow Creek was 49 cm, most of which fell
between October and April (USDC 1990). During that same
time period, the temperature in Willow Creek averaged 2.3
degrees C in winter (December to February) and 20.5
degrees C in summer (June to August; USDC 1990). This
area was representative of the Klamath physiographic
province of the northern spotted owl (Franklin and Dyrness
1988, Thomas et al. 1990). Franklin et al. (1990)
provided a more detailed description of the study area.
MATERIALS AND METHODS
Land Cover Mapping
Land cover on the WCSA was mapped using Landsat
Thematic Mapper (TM) digital imagery. I used the
MicroImage (Version 4.0) software package (Terra-Mar
Resource Information Services, Inc., 1937 Landings Drive,
Mountain View, CA 94043) for image classification.
My land cover classification (Table 1) represented
seral stages of coniferous forest (CF). Due to spectral
and structural similarities, nonvegetated areas were
combined with those having vegetation < 2.5 cm diameter at
breast height (dbh; CF1), brush was combined with young
conifer < 12.7 cm dbh (CF2), pole conifer (12.7 cm to 27.8
cm dbh) was combined with medium conifer (27.9 cm to 53.2
cm dbh; CF3), and mature conifer (53.3 cm to 91.4 cm dbh)
was combined with old-growth conifer
91.5 cm dbh; CF4).
Hardwood •(HDW) areas had > 80% of the basal area comprised
of hardwood species > 12.6 cm dbh. Surface water such as
lakes, ponds, rivers, and streams were also mapped. These
broad classes represented cover types that were comparable
to categories derived during previous owl studies at WCSA
(Franklin et al. 1990, Solis and Gutierrez 1990, Blakesley
et al. 1992). Using more narrowly defined habitat classes
6
7 Table 1. Definition of land cover classes used to map the
Willow Creek Study Area, northwestern California,
1990.
Class
Definition.
Water
Water.
CF1
Nonvegetated and herbaceous. Total canopy
closure < 30%. Greater than 50% of ground
cover comprised of forbs, grass, rock, soil,
and woody plants < 2.5 cm dbh.
CF2
Brush. Total canopy closure < 30%. Greater
than 50% of ground cover comprised of brush,
conifer, and hardwood species ranging from
2.5 cm to 12.6 cm dbh.
CF3
Pole and medium conifer. Total canopy
closure k 30%. More than 50% of conifer
basal area comprised of trees ranging from
12.7 cm to 53.2 cm dbh.
CF4
Total canopy
Mature and old-growth conifer. closure k 30%. More than 50% of conifer
basal area comprised of trees k 53.3 cm dbh.
HDW
Hardwood. Total canopy closure k 30%. More
than 80% of basal area comprised of hardwood
trees > 12.6 cm dbh.
8 also would have reduced power and accuracy of comparisons
of use versus availability (White and Garrott 1986).
Vegetation Data Collection
Vegetation data were collected at random and non random locations within the WCSA. Variables measured at
both random and non-random plots were identical (see
Appendix A). Random vegetation plots were located at
random points (n = 57) distributed throughout the study
area. Random points were plotted on 1:24,000 topographic
maps and located in the field using terrain associations
and altimeter readings in conjunction with at least one
compass bearing and distance estimate from a known
location. Four vegetation plots, each 25 m apart and
arrayed in a north-south orientated square pattern, were
arranged around each of the random points. Random data
were used after image classification to assess the
accuracy of the final land cover maps.
Non-random vegetation plots (n = 120) were
purposely located to encompass the full spectrum of
vegetative and physiographic conditions that existed on
the study area. This reference information was used
during image classification to identify clustering areas
and to evaluate spectral classes. Other reference data
included a set of color 1:15,820 aerial photographs
acquired 25 June 1990, and personal knowledge of the study
area.
9 Landsat Image Processing
Scene 5225218174 was acquired by Landsat-5 on 1
May 1990. Before I obtained this imagery it had been
geometrically rectified to fit the Universal Transverse
Mercator Projection (UTM), with each grid-cell resampled
to 25 m (Jensen 1986). This particular Landsat scene was
chosen because it covered the entire study area and was
cloud free. From this scene, I extracted a 785 km 2 area
which contained the WCSA and had UTM coordinates of
4534850 N, 436075 E at the NW corner and 4503975 N, 461450
E at the SE corner.
I used a hybrid approach to image classification,
which combined elements of both supervised and
unsupervised techniques (see Lillesand and Kieffer
1987:687). Guided clustering (Fox and Mayer 1979, Walsh
1980, Fox et al. 1992) with the Euclidean distance
algorithm was used (Richards 1986) to develop spectral
statistics for known areas. These spectral classes with
the maximum likelihood classifier (Jensen 1986, Lillesand
and Kiefer 1987) were used to classify a portion of the
study area, which was then evaluated with reference data.
Those spectral classes which performed well in the
classification were retained. This was an iterative,
trial-and-error process of developing spectral classes and
testing their effectiveness. When spectral classes
adequately defined the target land cover classes while
10 maintaining low spectral variability, I used the maximum
likelihood algorithm for a full classification of the
WCSA. Those scattered grid-cells which remained
unclassified were classified with a supervised Euclidean
distance classifier (Jensen 1986). Following the final
classification, spectral classes were combined into their
respective land cover classes. Two versions of the final
land cover map were produced: a seral stage map contained
all land cover classes, and a mature and old-growth map
contained only areas of CF4.
I only used TM bands 1, 3, 4, and 5 for supervised
classifications. This band combination had been shown to
reduce the amount of redundant spectral information, to
speed processing time, and to provide high separability of
forest cover types (Latty and Hoffer 1980). TM band 6 was
not used because it had unique spatial resolution and
spectral characteristics which made it incompatible with
the remaining TM bands.
Accuracy Assessment
Data for the four plots at each random point were
pooled, and the land cover at each random point was
designated as belonging to a specific class based on the
criteria in Table 1. The land cover class at each point
was compared to the predominate land cover in the nine
grid-cells (75 m by 75 m) around each corresponding point
on the final land cover maps. This three by three grid-
11
cell sampling unit was used to reduce the differences
between vegetation plot data and mapped land cover which
were due to errors in coordinate accuracy of the maps and
navigation errors resulting from locating plots in the
field. Three by three grid-cell sampling units which did
not have a singularly predominate land cover class were
eliminated. I estimated overall map accuracy by the
percent of agreement between the actual land cover at
random points and mapped land cover at corresponding
random points. I constructed error matrices (Story and
Congalton 1986) and estimated overall map accuracies for
both versions of the land cover map. I also calculated
Kappa coefficients, which represented the proportion of
agreement after removing any agreement expected to occur
by chance (Congalton et al. 1983).
Spotted Owl Data Collection
Data on spotted owls were obtained from an ongoing
demography study (Franklin et al. 1990). During each
breeding season (April to August) from 1988 to 1992, the
entire WCSA was surveyed for spotted owls by nighttime
calling.
During daytime searches, roosting owls were
visually located and individually identified. Nests were
located while determining reproductive status. The
methods used to map nest and roost locations on 1:24,000
topographic maps was the same as those used to locate
12 random points. Nest and roost locations for each year
represent all territorial owls within the WCSA.
Individual owls were considered to belong to a specific
territory when they were repeatedly located in a given
area. Survey methods followed Forsman (1983) and Franklin
et al. (1990).
Habitat Data Collection and Analysis
Habitat variables were compared between owl sites
(i.e., used sites) and random sites (i.e., available
sites). Statistically significant differences indicated
disproportional use, and were used to infer selection for
or against those characteristics (Johnson 1980, Peek
1986).
I used Version 4.0 of the IDRISI geographic
information system (GIS; Eastman, J R., Clark Univ.,
Graduate School of Geography, Worcester, MA 01610) to
extract habitat data from land cover maps. Land cover
variables measured at nest, roost, and random sites were
the area (ha) of each land cover type, an index of the
fragmentation of CF4, an index of land cover
heterogeneity, and the number and size of patches of CF4.
I measured the area of land cover types, and the indices
of fragmentation and heterogeneity within 800 m, 1200 m,
1600 m, 2000 m, 2400 m, 2800 m, 3200m, and 3600 m radii
concentric circular plots. With IDRISI the 800 m, 1200 m,
13 1600 m, 2000 m, 2400 m, 2800 m, 3200 m, and 3600 m radii
plots corresponded to 201 ha, 451 ha, 803 ha, 1255 ha,
1807 ha, 2461 ha, 3217 ha, and 4070 ha plot sizes,
respectively. Due to limitations of the GIS, patch
characteristics were measured only within the 800 m radius
circular plot.
I measured the fragmentation of CF4 with a
variation of the fragmentation index introduced by Ripple
et al. (1991a). For each circular plot, I calculated the
mean distance of each non-CF4 grid-cell from a grid-cell
of CF4. Higher mean values represented higher levels of
fragmentation. Lehmkuhl and Raphael (1993) used a similar
version of this index to measure fragmentaion around owl
sites on the Olympic Peninsula, Washington. Using the
proportions of each of the six land cover types within
plots, I calculated habitat heterogeneity using Simpson's
(1949) index of diversity. This heterogeneity index was
more sensitive to the area of each cover type present than
it was to the number of cover types present (Magurran
1988). While the minimum patch size of CF4 that was of
biological importance to spotted owls was unknown, a
priori I excluded patches less than one ha from the
calculation of patch characteristics. This removed the
bias toward very small and more numerous patches which
would have resulted from the presence of isolated single
grid-cells of CF4.
14
Spotted owls may repeatedly nest or roost in the
same general area, and commonly roost near active nests.
This may result in a lack of independence among sites
within territories. If both members of an owl pair were
replaced by another pair in the same year, I considered
subsequent locations in a territory to be independent of
the previous pair. However, turnover of both members of a
pair within a single year was rare during the study period
at WCSA. In addition, because spotted owls are site
tenacious, and often occupy the same home ranges for long
periods of time (Forsman et al. 1984), locations within
any given territory may not be independent between years.
Therefore, for any territory, locations may lack
independence both between years and within a single year.
As a result, for each territory where nesting occurred
during the study period, one nest site was randomly
selected and it alone was subjected to habitat analysis.
I also randomly selected one roost site from each
territory. In order to maintain independence between nest
and roost sites, only roosts from years in which no
nesting occurred were considered.
In use versus availability studies such as this,
the designation of which habitat components were actually
available to the organism could have considerable
influence on the conclusions reached (Johnson 1980).
Because there was markedly different floristic composition
15 at higher elevations within the WCSA, I eliminated from
consideration random points which fell at elevations
greater than the maximum elevation observed at an owl nest
or roost site during the study period. Using this
criteria, 50 random points were selected for habitat
analysis.
Comparisons of land cover variables were first
performed on data from the circular plot which had a
radius which most closely approximated one-half the mean
distance between the centers of spotted owl territories.
This plot size was biologically meaningful because it
represented an estimate of territory size within this
contiguous population of owls. In addition, this plot
size served to reduce overlap between adjacent plots, and
provided data that were independent from the other
concentric circular plots around the same site. Because
some data were not normally distributed, I used
nonparametric Kruskal-Wallis tests (Zar 1974) to compare
nest, roost, and random sites. When Kruskal-Wallis tests
were significant, Conover's (1971) multiple comparisons
were used to determine significance between means. All
tests were considered significant at P 0.05.
The selection of locations used to measure
distances between territory centers was constrained by the
problems of independence discussed above. Therefore, I
only measured distances between 1990 territories; I chose
16 this particular year because it was the midpoint of the
study period. During 1990 each territory had more than
one owl location. Therefore, I chose one location to
represent an estimate of the center of activity for each
territory (Ganey 1991). The order of priority for
selecting this location for each territory was: 1) nest
site; 2) pair roost site; 3) most frequently-used roost
site; 4) female roost site; and 5) male roost site. I
measured the distances between 1990 spotted owl
territories on 1:24,000 scale topographic maps.
If spotted owls use sites with habitat
characteristics different than what is generally available
on the landscape, then these characteristics should change
non-randomly with increasing distance away from owls. I
evaluated this relationship between habitat
characteristics and spatial scale by regressing mean
habitat variables against plot size using data from all
concentric circular plots. The area of each land cover
type within each circular plot was converted to the
percentage of the total area of that plot. To determine
which habitat characteristics were changing non-randomly
around owls, for each variable I compared the slopes of
nest, roost, and random regressions (Zar 1974). Slopes
were considered significantly different at P < 0.05.
RESULTS
Land Cover Mapping
Landsat Image Processing
One-hundred twenty-two unique spectral classes
representing the six land cover classes were identified
during the iterative guided clustering process.
Approximately 4.2% (12.3 km2 ) of the WCSA was not
classified by the maximum likelihood classifier; however,
these areas were successfully classified by the supervised
Euclidean classifier.
Accuracy Assessment
Plots at two random points were not used in the
accuracy assessment because no single land cover type
dominated the three by three grid-cell sampling unit.
Major diagonals in the error matrices (Tables 2 and 3)
show the agreement between accuracy assessment plot data
and land cover map data for both versions of the final
Landsat derived map. Overall map accuracies for the seral
stage map and the mature and old-growth map were 76.4% and
83.6%, respectively. Due to the relatively small sample
size of accuracy assessment data (i.e. only 55 random
points), at the 95% confidence level the lower limits of
overall map accuracy were 64% and 73% (Thomas and Allcock
17
18
Table 2. Error matrix for Landsat derived seral
stage map of the Willow Creek Study Area,
northwestern California, 1990. Values are the
number of random points designated as belonging to
a particular land cover class relative to the cover
class at corresponding points on the Landsat
derived successional stage map.
Random Accuracy Assessment Plot Data
Water Landsat Data
CF1a
CF2
CF3
CF4
HDW
Water
1
0
0
0
0
0
CF1
0
6
0
0
0
0
CF2
0
0
4
0
0
1
CF3
0
0
0
5
2
1
CF4
0
0
0
5
18
2
HDW
0
0
1
0
1
8
CF1--Nonvegetated, herbs, and woody plants < 2.5 cm dbh;
CF2--brush with 2.5-12.6 cm dbh woody plants; CF3--12.7 53.2 cm dbh conifers; CF4--z 53.3 cm dbh conifers; HDW- hardwood trees > 12.6 cm dbh comprising > 80% basal area.
a
19 Table 3. Error matrix for Landsat derived mature and old-
growth coniferous forest map of the Willow Creek
Study Area, northwestern California, 1990. Values
are the number of random points designated as
belonging to CF4 or non-CF4 relative to the cover
class at corresponding points on the Landsat
derived mature and old-growth map.
Random Accuracy Assessment Plot Data
CF4a
Non-CF4
Landsat Data
CF4 Non-CF4 a
18
6
3
28
CF4--z 53.3 cm dbh conifers.
20 1984) for the seral stage map and the mature and old-
growth map, respectively. Kappa coefficients for the
seral stage map and the mature and old-growth map were
0.680 and 0.663, respectively.
The CF3 present at five random points was
misclassified as CF4 on the Landsat derived seral stage
map (Table 2). While I did not quantify this particular
source of classification error, during guided clustering I
noted that spectrally similar areas of CF3 and CF4 were
difficult to separate in shadowed areas. Topographic
shadowing in dissected mountainous terrain had been known
to complicate classification of successional stages of
coniferous forest (Fiorella and Ripple 1993).
The 1990 Landsat derived seral stage map was
different from 1988 1:12,000 USDA Forest Service timber
strata maps of the same area (Table 4; Franklin et al.
1990). The reason there was less area of CF1 on the
Landsat seral stage map may be because the more recent
Landsat map reflected the succession of herbaceous areas
into brush areas. Less area of CF3 on the Landsat seral
stage map may be attributed to topographic shading. More
HDW area on the Landsat seral stage map may be a result of
the spectral similarity between older clearcuts dominated
by well developed tanoak brush and HDW stands dominated by
tanoak trees. Likewise, some areas of very large conifer
reproduction, which is structurally more similar to CF2,
21
Table 4. Land cover composition (%) of the Willow Creek Study
Area from 1990 Landsat derived seral stage map and
from 1988 USDA Forest Service timber strata maps.
Percentage of area in each land cover class
Water
CF1a
CF2
CF3
CF4
HDW
0.3
8.9
14.4
12.8
35.3
28.3
--
20.8
10.0
21.3
34.7
11.1
Source
1990 Landsat
seral stage map 1988 USDA Forest
Service mapsb
CF1--Nonvegetated, herbs, and woody plants < 2.5 cm dbh;
CF2--brush with 2.5-12.6 cm dbh woody plants; CF3--12.7 53.2 cm dbh conifers; CF4--≥ 53.3 cm dbh conifers; HDW- hardwood trees > 12.6 cm dbh comprising > 80% basal area.
a
b
From Franklin et al.
(1990).
22 may have been spectrally confused with the pole timber
component of CF3. An alternative explanation for
differences between the Forest Service timber strata maps
and the Landsat seral stage map is that the timber strata
maps contain inaccuracies.
Spotted Owl Data
Between 1988 and 1992, 50 unique spotted owl
territories were identified within the WCSA. Of the 86
nest sites located, only 33 were selected for habitat
analysis because nesting did not occur in some territories
during the study period. Of the 324 roost sites located,
45 were randomly selected because in some territories
nesting occurred during each year of the study period.
The maximum elevation observed at a spotted owl nest or
roost location during the study period was 1350 m; four
random points above this elevation were not used in
habitat analysis. During 1990, 40 unique territories were
present at WCSA, and were used to measure distances
between territory centers.
Habitat Analysis
Due to the spatial resolution of TM data, it was
not possible to map most of the surface water within the
study area. Therefore, I did not test for differences in
the area of water between owl and random sites. Appendix
23
B shows the amount and percentage of area of water
detected within circular plots.
The mean distance between 1990 spotted owl territory centers was 1579 m (SD = 525, n = 40, range 540 to 3400 m). Because one-half of this distance was 790 m, I used the 800 m radius plots to make initial comparisons in habitat characteristics between nest, roost, and random sites. There was less area of CF1 in 800 m plots around spotted owl nest and roost sites than around random sites, and nest and roost sites had similar areas of CF1 (Tables 5 and 6).
Nest, roost, and random sites did not differ
with respect to area of CF2, CF3, and HDW (Tables 5 and 6).
Spotted owls used nest and roost sites which had more
CF4 than was available throughout the landscape, but nest
and roost sites had similar areas of CF4 (Tables 5 and 6). The area of CF4 within 800 m plots ranged from 42.3 to 162.3 ha,
32.8 to 146.2 ha, and 8.4 to 136.5 ha around
nest, roost, and random sites, respectively.
CF4
fragmentation was also lower around nest and roost sites than around random sites (Tables 5 and 6), with nest and roost sites having similar levels of fragmentation (Table 5).
Spotted owls also used nest and roost sites with
slightly lower habitat heterogeneity, while nest and roost sites were similar (Tables 5 and 6). The observation that spotted owls selected sites
Table 5. Landscape characteristics within 800 m radius plots around spotted owl nest,
roost, and random sites, in northwestern California, 1988-92.
Nest Sites (n = 33) Variablea
Mean
SD
Roost Sites (n = 45) Mean
SD
Random Sites
(n = 50)
Mean
SD
Hb
P-Value
17.0B
15.9
6.28
0.043
Land cover type (ha)
Herb and non-vegetated
9.2A
9.6A
6.8
7.7
Brush
20.8
17.7
21.1
16.9
28.0
24.4
2.81
0.245
Pole and medium
23.3
9.7
26.1
12.2
25.2
13.8
0.79
0.673
Mature and old-growth
94.1A
26.2
92.0A
27.0
71.8B
28.1
15.84
12.6 cm
dbh, measured with a 20 factor basal area prism (Dillworth
1975).
6) Pole and medium conifer basal area - basal area of all
conifers ranging from 12.7 cm to 53.2 cm dbh (see 5).
7) Mature and old growth conifer basal area - basal area
of all conifers z 53.3 cm dbh (see 5).
8) Hardwood basal area - basal area of all hardwoods >
12.6 cm dbh (see 5).
9) Ground cover - percent ground cover of grass and
forbs, woody plants < 2.5 cm dbh, litter, bare soil, and
rock. Ocularly estimated to nearest 10 percent within a
12 m radius circle around plot center.
48
Appendix B. Amount (ha) and percentage of area in water around spotted owl nest, roost,
and random sites, for eight plot sizes in northwestern California, 1988-92.
Plot Radiusa
(m)
Mean
Random Sites
(n = 50)
Roost Sites (n = 45)
Nest Sites (n = 33) SD
%
Mean
SD
%
Mean
SD
%
800
0.2
0.7
0.1
0.1
0.6