A MULTI-SCALE ANALYSIS OF FOREST STRUCTURE
AND VERTEBRATE DIVERSITY
by
Jennifer M. Psyllakis
B.Sc., McGill University, 1997
M.Sc., University of Regina, 2001

DISSERTATION SUBMITTED IN PARTIAL FULFILLMENT OF
THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY
in
NATURAL RESOURCES AND ENVIRONMENTAL STUDIES

THE UNIVERSITY OF NORTHERN BRITISH COLUMBIA
June 2006
© Jennifer M. Psyllakis, 2006

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

Library and
Archives Canada

Bibliotheque et
Archives Canada

Published Heritage
Branch

Direction du
Patrimoine de I'edition

395 W ellington Street
Ottawa ON K1A 0N4
Canada

395, rue W ellington
Ottawa ON K1A 0N4
Canada
Your file Votre reference
ISBN: 978-0-494-28459-9
Our file Notre reference
ISBN: 978-0-494-28459-9

NOTICE:
The author has granted a non­
exclusive license allowing Library
and Archives Canada to reproduce,
publish, archive, preserve, conserve,
communicate to the public by
telecommunication or on the Internet,
loan, distribute and sell theses
worldwide, for commercial or non­
commercial purposes, in microform,
paper, electronic and/or any other
formats.

AVIS:
L'auteur a accorde une licence non exclusive
permettant a la Bibliotheque et Archives
Canada de reproduire, publier, archiver,
sauvegarder, conserver, transmettre au public
par telecommunication ou par I'lnternet, preter,
distribuer et vendre des theses partout dans
le monde, a des fins commerciales ou autres,
sur support microforme, papier, electronique
et/ou autres formats.

The author retains copyright
ownership and moral rights in
this thesis. Neither the thesis
nor substantial extracts from it
may be printed or otherwise
reproduced without the author's
permission.

L'auteur conserve la propriete du droit d'auteur
et des droits moraux qui protege cette these.
Ni la these ni des extraits substantiels de
celle-ci ne doivent etre imprimes ou autrement
reproduits sans son autorisation.

In compliance with the Canadian
Privacy Act some supporting
forms may have been removed
from this thesis.

Conformement a la loi canadienne
sur la protection de la vie privee,
quelques formulaires secondaires
ont ete enleves de cette these.

While these forms may be included
in the document page count,
their removal does not represent
any loss of content from the
thesis.

Bien que ces formulaires
aient inclus dans la pagination,
il n'y aura aucun contenu manquant.

i*i

Canada
R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

APPROVAL
Name:

Jennifer M. Psyllakis

Degree:

Doctor of Philosophy

Dissertation Title:

A MULTI-SCALE ANALYSIS OF FOREST STRUCTURE
AND VERTEBRATE DIVERSITY

Examining Committee:
Chair: Dr. Robert Tait
Dean of Graduate Studies
University of Northern British Columbia
a

. .

ft

Sup^visorr t)r/ Mu^j/ael C0)ingham, Associate Professor
Natural Resources and Environmental Studies Program
University of Northern British Columbia

Committee Member: Dr. Russell Dawson, Associate Professor
Natural Resources and Environmental Studies Program
Canada Research Chair, Avian Ecology
University of Northern British Columbia

Com
>r. Katherine Parker, Associate Professor
Natural resources ana Environmental Studies Program
Ian McTaggart Cowan Muskwa-Kechika Research Professor
University of Northern British Columbia

committee lviemner: Ur. Koger Wheate, Associate Professor
Natural Resources and Environmental Studies Program
University of Northern British Columbia

Committee Member: Dr. Bruiio Zumbo, Professor
Department of Educational aiiii Counselling Psychology and,
Special Education
University of British Columbia

External Examiner: Dr. Alton Harestad, Professor
Department of Biological Sciences
Simon Fraser University

Date Approved:

rTuk\€

(pQOU

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

Abstract
Predicting the occurrence of species is a central problem in ecology and conservation.
The ability to accurately predict species occurrence requires an understanding of the relations
among species and the environment defined at appropriate spatial and temporal scales.
Through my research, I test the hypothesis that forest structure can reliably predict the
occurrence of vertebrate species and vertebrate-species assemblages using logistic regression
models and classification and regression tree (CART) analysis. With the results, I assess the
potential of using forest structure as a surrogate measure for monitoring species diversity. I
also investigate how deriving species presence from different types of detection data (i.e.,
audio and visual detections versus sign) and using different measures of forest structure
affect prediction accuracy and model selection. To assess prediction accuracy, I use the area
under the receiver-operating characteristic curve (ROC) for logistic regression models and a
classification matrix of predicted and observed group membership for CART analysis. In
addition, I use spatially and temporally independent data to validate single-species models.
Models constructed using presence derived from detections of sign resulted in higher
prediction accuracy, probably due to lower spatial uncertainty. Models for single species (n =
101) had good prediction accuracy (ROC > 0.70) only 56.4% of the time and few models
retained good accuracy when validated with spatially and temporally independent data. Only
spatial uncertainty appeared to systematically affect ROC values when sources of uncertainty
(i.e., identification, spatial, or temporal) were examined with ANOVA. CART analysis
successfully predicted 45.8% of group membership of plots. Together these results suggest
that spatial uncertainty and measuring structural characteristics of forests at the appropriate
spatial scale, for the species being modelled, have the largest effect on model outcome.

iii

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

Models developed cannot be assumed to be transferable to different areas or applicable in
different years. Overall, forest structure did not accurately predict species presence or species
groups well and, therefore, is not a suitable surrogate measure for species occurrence or
monitoring diversity of vertebrate species.

iv

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

Table of Contents
Abstract...........................................................................................................................................iii
Table of Contents............................................................................................................................ v
List of Tables................................................................................................................................ vii
List of Figures..................................................................................................................................x
Acknowledgements.......................................................................................................................xii
Chapter 1. Introduction................................................................................................................... 1
Background.................................................................................................................................. 1
Measuring and monitoring biodiversity.....................................................................................1
Surrogate approaches..................................................................................................................2
Modelling the occurrence of species..........................................................................................4
Single species.......................................................................................................................... 4
Multiple species...................................................................................................................... 5
Scale.........................................................................................................................................6
Frequentist versus Bayesian approaches................................................................................... 7
Summary and outline of chapters.............................................................................................. 8
Chapter 2. Considerations for the validation of species-habitat models................................... 11
Abstract...................................................................................................................................... 11
Introduction................................................................................................................................12
Methods...................................................................................................................................... 18
Study area...............................................................................................................................18
Habitat data............................................................................................................................18
Species detections.................................................................................................................21
Model construction and predictive evaluation....................................................................22
Results........................................................................................................................................25
Red squirrel........................................................................................................................... 25
Pileated woodpecker.............................................................................................................28
Discussion..................................................................................................................................28
Chapter 3. Using forest structure to predict the occurrence of vertebrate species................... 34
Abstract......................................................................................................................................34
Introduction................................................................................................................................35
Methods......................................................................................................................................40
Study area.............................................................................................................................. 40
Habitat measurements.......................................................................................................... 42
Vertebrate sampling..............................................................................................................46
Analysis..................................................................................................................................53
Results........................................................................................................................................ 59

v

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

Species detection................................................................................................................... 59
Model validation and accuracy............................................................................................72
Detection uncertainty and species traits.............................................................................. 72
Discussion..................................................................................................................................75
Structure as a surrogate measure..........................................................................................75
Variables, scale and spatial context.................................................................................... 77
Validation.............................................................................................................................. 79
Multi-species inventory....................................................................................................... 83
Conclusion.................................................................................................................................85
Chapter 4. Using forest structure to predict the occurrence of vertebrate assemblages..........88
Abstract......................................................................................................................................88
Introduction................................................................................................................................89
Methods......................................................................................................................................93
Study area.............................................................................................................................. 93
Habitat measurements.......................................................................................................... 95
Vertebrate sampling..............................................................................................................96
Data analysis......................................................................................................................... 98
Results...................................................................................................................................... 104
Plot groupings..................................................................................................................... 104
Predictions using structure................................................................................................. 112
Discussion................................................................................................................................ 123
Chapter 5. General Conclusions................................................................................................ 129
Structure as a surrogate measure............................................................................................ 129
Uncertainty...............................................................................................................................132
Validation.................................................................................................................................133
Summary..................................................................................................................................133
Literature cited.............................................................................................................................136
Appendix I. Explanation for electronic files and data CD........................................................155
Appendix II. List of vertebrate species (200 species), with associated number of detections
.......................................................................................................................................................167
Appendix III. List of vertebrate species used in non-metric multidimensional scaling
ordination analysis...................................................................................................................... 172
Appendix IV. Comparison of cluster analysis results for groupings of plots based on
vertebrate-species co-occurrence (X, lettered groups) and structural characteristics (Y,
numbered groups)........................................................................................................................ 174
Appendix V. Indicator species analysis to identify vertebrate species most closely associated
with the 11 groups identified with cluster analysis of plots based on species co-occurrence in
the Williams Lake Study Area from 2001-2004...................................................................... 179

vi

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

List of Tables
Table 2.1. Matrix of prediction classifications describing the possible outcomes of presence
or absence and the associated indices that can be used to describe predictive performance.. 14
Table 2.2. List of candidate variables used to construct competing models for Red Squirrel
{Tamiasciurus hudonicus', TAHU) and Pileated Woodpecker {Dryocopus pileatus', PIWO). A
structural variable was considered a candidate if it related directly to known associations
between the species and its habitat...............................................................................................23
Table 2.3. Model results predicting the occurrence of Red Squirrel (Tamiasciurus hudonicus)
in the Williams Lake Study Area using audio and visual detections or sign detections. Using
sign as the response variable resulted in the highest pseudo-R2 and highest ROC value
26
Table 2.4. Models predicting occurrence of Red Squirrel (Tamiasciurus hudonicus) in the
Williams Lake Study Area using different sources for independent data variables. The model
using plot measures and Vegetative Resource Inventory (VRI) data resulted in the lowest
AICc and highest ROC value........................................................................................................ 27
Table 2.5. Model results predicting the occurrence of Pileated Woodpecker {Dryocopus
pileatus) in the Williams Lake Study Area using audio and visual detections or sign
detections. Using sign as the response variable resulted in the highest pseudo-R2 and highest
ROC value...................................................................................................................................... 29
Table 2.6. Models predicting occurrence of Pileated Woodpecker {Dryocopus pileatus) in
the Williams Lake Study Area and different sources for independent data variables. The
model using plot measures and Vegetative Resource Inventory (VRI) data resulted in the
lowest AICc and highest ROC value............................................................................................ 30
Table 3.1. Echolocation frequency and species groupings for bats expected to occur in the
Williams Lake Study Area. Identification of the recorded echolocation calls of bats was
based primarily on the minimum frequency and slope characteristics. For some species we
also used maximum frequency. Species were confirmed present in the study area if we
captured them in mist-nets. Because we could not record reference calls for species not
captured, we could not confirm presence and, therefore, did not assign that species to an
echolocation group (NA), although characteristics of calls for some species are distinctive. 52
Table 3.2. Example of the candidate model set to describe the occurrence of Red-naped
Sapsucker {Sphyrapicus nuchali) within the Williams Lake Study Area at the plot level. Full
lists of the candidate variables used for 55 vertebrate species are included in Appendix I
(electronic)..................................................................................................................................... 56
Table 3.3. Number of vertebrate species, summarized by order, detected by different
monitoring methods during model development phase (2001-2004) in the Williams Lake
Study Area. The total number of detections are presented in brackets below the number of
species. Detections by species and by year are included in Appendix I (electronic)...............60

v ii

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

Table 3.4. The 55 species detected in the Williams Lake Study Area that met sample
prevalence criteria (>10% and <90%) from 2001-2004, with the exception of Southern Redbacked Vole (see text for details, page 62). Prevalence is reported as a percentage of all plots
(n = 228) and stands (n = 97), which were not altered by harvesting through the course of
data collection, where presence was detected. Latin names are available in Appendix II
63
Table 3.5. Ten species detected in the Williams Lake Study Area, from 2001-2004, that were
not modelled to predict occurrence because of high prevalence at the plot and stand scale (i.e.,
presence was detected at >90% of plots). Latin names can be found in Appendix II
65
Table 3.6. Example of model results to predict the occurrence of Red-naped Sapsucker
(Sphyrapicus nuchalis) in the Williams Lake Study Area. Variables, coefficients, and
receiver-operating characteristic curve (ROC) values are presented for the best plot and stand
models and their temporal and spatial validation counterparts. All model results for 55
vertebrate species can be found in the electronic Appendix 1....................................................67
Table 3.7. Summary of area under the receiver-operating characteristic curve (ROC) values
for each of the 55 species detected in the Williams Lake Study Area. Each species modelled
met our detection prevalence criteria from 2001-2004 at the plot and / or stand scale. ROCs
are only reported for the top model with the lowest AICc.Not all species could be modelled at
both the plot and stand scale (-)....................................................................................................69
Table 3.8. Most common variables found in the top models used to predict occurrence of
vertebrate species in the Williams Lake Study Area. The direction of the species’ association
with the variable (selection with positive coefficient or avoidance with negative coefficient)
is also reported. Structure data were derived from measures taken at each plot, as determined
from a GIS, or as calculated using a classified landscape image and FRAGSTATS
(McGarigal and Marks, 1995)...................................................................................................... 71
Table 3.9. Summary of prediction accuracy (ROC values) for plot and stand models, and
their spatial and temporal validation results. Models were constructed to predict the
occurrence of vertebrate species in the Williams Lake Study Area. Not all models could be
spatially and / or temporally validated because of insufficient detections in the new plots or
for a particular year. The mean difference in prediction accuracy between 2001-2004 models
and the 2004 spatial and temporal validation model was negative for each model set..........73
Table 3.10. Summary of the calibration plot regression slopes for stand and plotmodels (e.g.,
Fig. 3.4) constructed to predict species occurrences in the Williams Lake Study Area, and
their spatially and temporally validated counterparts. Regression slopes describe the observed
proportion of occupied sites versus predicted probability of occurrence for vertebrate species.
Percentage of slopes that were non-significant, near 1 (significant), and significantly different
from 1 are reported. Those models with slopes near 1 can be deemed well calibrated and,
therefore, reliable...........................................................................................................................74

v iii

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

Table 3.11. Summary of ANOVA results to test if species’ traits affected the accuracy of
predictions (ROC values) for the occurrence of vertebrate species. ROC values were used
from models constructed with data collected in the Williams Lake Study Area from 20012004................................................................................................................................................ 76
Table 4.1. Methods used to detect species presence in the Williams Lake Study Area and the
time of year surveys were conducted from 2001-2004............................................................. 97
Table 4.2. Variables used in the classification and regression tree (CART) analysis to predict
group membership for plots in the Williams Lake Study Area. Plots were grouped based on
the co-occurrence of vertebrate species detected from 2001-2004........................................ 103
Table 4.3. Pearson correlation coefficients for ordination axes resulting from non-metric
multidimensional scaling. Ordination was based on species co-occurrence at plots in the
Williams Lake Study Area from 2001-2004............................................................................ 105
Table 4.4. Pearson correlation coefficients for ordination axes resulting from non-metric
multidimensional scaling. Ordination was based on structural characteristics measured at
plots in the Williams Lake Study Area...................................................................................... 107
Table 4.5. Pearson correlation coefficients for plot coordinates based on species and
structure using non-metric multidimensional scaling ordinations. Species occurrence was
determined from 2001-2004 in the Williams Lake Study Area............................................... 108
Table 4.6. Relationship between vertebrate species presence and absence with ordination
axes as determined by non-metric multidimensional scaling. Axis one is defined by distance
to edge (-) and edge density (+), axis 2 by distance to water (+), and axis 3 by distance to
edge (+) and edge density (-). Variables are described in Table 4.2. Species with no clear
relationship to any of the variables defining ordination space are not listed (n = 37). If species
were not located in a particular quadrant of the 2-dimensional space, then the relationship
was interpreted as avoidance. Latin names for all species are listed in Appendix II
109
Table 4.7. Correct classification matrix of observed class against predicted class for CART
analysis. Structural characteristics of plots were used to predict group membership and
groups were defined based on similarities in species co-occurrence from 2001-2004 at plots
in the Williams Lake Study Area. The 11 classes were defined using cluster analysis. Overall
classification accuracy was 45.8%. Number of correct classifications are in bold for each
class...............................................................................................................................................119
Table 4.8. Ranking of importance for structural variables used in CART analysis to predict
membership of plots grouped based on species co-occurrence from 2001-2004 in the
Williams Lake Study Area; 100 = most important, 0 = least important..................................120

ix

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

List of Figures
Figure 2.1. The location of the Williams Lake Study Area (■) in central British Columbia,
Canada. Williams Lake is mapped for reference (+). We established 243 plots (•) across a
range of variation in structural characteristics and surveyed them for species presence from
May 2001-January 2004.............................................................................................................. 19
Figure 3.1. The distribution of sites within the Williams Lake Study Area in central British
Columbia, Canada (approximately 30km south of Williams Lake). Multiple plots were
located within each site. Development sites (Aand ■) included 243 plots used for 2001-2004
model construction. In the summer of 2004, we established an additional 123 plots at spatial
validation sites (•) and continued to monitor 91 plots at temporal validation sites (*) to assess
the prediction accuracy of these models with independent data. The shaded area of the map
indicates the Sub-Boreal Pine-Spruce (SBPS) biogeoclimatic zone, the unshaded area is
Interior Douglas-fir (IDF).............................................................................................................41
Figure 3.2. Example of plot (A) layout within the Williams Lake Study Area. Most plots
were connected with a 150-m transect, although some were connected with a 300-m transect.
The boundary for Vegetation Resource Inventory (VRI) polygons that were used to define
forest stands are highlighted with white...................................................................................... 43
Figure 3.3. Schematic of the vegetation sampling scheme used to assess the vegetation and
other structures present in each of the sampling plots................................................................44
Figure 3.4. Example of a calibration plot for the Red-naped Sapsucker (Sphyrapicus nuchalis)
plot-level model. The expected distribution of a well-calibrated model has a slope (P) of 1(—)
to describe the relationship between observed proportion of occurrence and predicted
probability of occurrence. The slope of the 2001-04 model was close to 1 (closed circles, — ,
P = 0.915, p = 0.004). The 2004 spatial validation data had a slope of less than 1 (open
circles,
, P = 0.889, p = 0.146) indicating that the model is misspecified for the spatially
independent data............................................................................................................................ 68
Figure 4.1. Plot (o) layout in the Williams Lake Study Area (★; inset) approximately 30 km
south of Williams Lake, British Columbia (+; inset). Each plot was connected by a 150- or
300-m transect............................................................................................................................... 94
Figure 4.2. Dendrogram from hierarchical cluster analysis of plots in the Williams Lake
Study Area by species co-occurrence detections from 2001-2004........................................ 110
Figure 4.3. Dendrogram from hierarchical cluster analysis of plots by characteristics of
forest structure at plots located in the Williams Lake Study Area...........................................I l l
Figure 4.4. Non-metric multidimensional scaling ordination (NMS) of plots based on
presence and absence data for vertebrate species detected at plots in the Williams Lake Study
Area from 2001-2004. Separate figures are presented for each pair of ordination axes (A, B,

x

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

C). Cluster analysis groups are depicted by different symbols. Plots shown closer to one
another are more similar in species composition...................................................................... 113
Figure 4.5. Non-metric multidimensional scaling ordination (NMS) of plots based on
characteristics of forest structure at plots in the Williams Lake Study Area. Separate figures
are presented for each pair of ordination axes (A, B, C). Cluster analysis groups are depicted
by different symbols. Plots shown closer to one another are more similar in forest-structural
characteristics............................................................................................................................... 116
Figure 4.6. Dendrogram for the classification and regression tree analysis. Plots were
grouped based on species co-occurrence from 2001-2004 in the Williams Lake Study Area.
Group membership was predicted using forest structural variables. Nodes to the left of each
decision criteria meet the stated condition, whereas nodes to the right do not meet the stated
criteria. The numbers along each decision branch indicates the number of cases that did or
did not meet the decision criteria. The upper number within terminal nodes (red) is the
number of cases assigned to the predicted class, the lower number within terminal nodes is
the group name. Not all cases are correctly classified. Overall classification success is
presented in Table 4.7..................................................................................................................121

xi

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

Acknowledgements
Over the duration of my studies there have been several people who have contributed
to my success and learning. I thank my supervisor, Dr. Michael Gillingham, for intellectual
support and challenges and the overall opportunity to work on this research. He also
extended great understanding and flexibility for the unexpected “life” experiences I went
through over the past 5 years. To my committee, Drs. Russ Dawson, Kathy Parker, Roger
Wheate, and Bruno Zumbo, I very much appreciated your insights and dedication. I don’t
think one could ask for a better, more conscientious group of people to work with. As well, I
am very grateful to my external examiner Dr. Alton Harestad for his thoughtful questions and
input.
Several individuals assisted with the collection of field data, as well as with
compiling the literature review; without their help I am sure I would still be doing vegetation
plots. Specifically, thanks to Karl Bachmann, Shannon Barker, Dan Baxter, Erin Boyle,
David Gummeson, Marilyn Hagedom, Erin Heibert, Christopher Hotson, Heather Jack,
Elena Jones, Heather Knight, Lana Miller, and Nelson Thomas. Others contributed to the
literature review that I relied on extensively to build candidate models including Mark
Bidwell, Erin Boyle, Jacqueline Dockery, Simon Goring, Heather Peet, and Danielle
Thompson. On campus, Nancy Elliot and Rob Higgins were my pillars of support through
course work and throughout my degree. A special thanks to Jermey Ayotte, Dave Gustine,
Brian Milakovic, and Andrew Walker for many animated discussions about modelling and
word formatting. Most of all, I am grateful to my family, who always supported and believed
in me. Especially my husband, Peter, for his incredible patience and tolerance, much of my
success I owe to him.

xii

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

Chapter 1. Introduction
Background
Activities driven by anthropogenic values have contributed to habitat loss and
fragmentation resulting in a global decline of many species (Hunter, 1990; Harper and
Hawksworth, 1995). Protecting biodiversity in reserves will not fully mitigate the effects of
anthropogenic activities on biodiversity loss (Lindenmayer and Franklin, 2002; Beazley et al.,
2005). In addition, many species are adapted to various levels of disturbance that should
allow human use concurrent with conservation objectives. Therefore, biodiversity assessment
and monitoring have become an important part of management planning, such as for forest
harvesting (Hunter, 1999; Noss, 1999).
Forests provide habitat necessary for approximately 65% of terrestrial taxa (World
Commission of Forests and Sustainable Development, 1999), as well as considerable
economic and social resources to humans. With estimates of less than 12% of the world’s
forests remaining intact outside the boreal forests (Bryant et al., 1997), a better understanding
of the relations among species distribution and forest structures, both in the short and long
term, will provide the knowledge to help mitigate the loss of species from forest
environments. This understanding is a relevant and necessary component of efforts that will
lead to an improved balance between economic, social, and ecological values, which includes
the conservation of biodiversity.

Measuring and monitoring biodiversity
Biodiversity is a relatively new word in the scientific and political arenas and holds
several meanings, often dependent on perspective. In the scientific community it is generally
1

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

considered synonymous with ‘biological diversity’ defined as encompassing all genetic,
species, and ecosystem variability (Norse et al., 1986). Because this definition is so broad,
researchers must state explicitly the component(s) of biological diversity of interest, as the
focus of research is rarely, nor practically, inclusive of all three levels of biodiversity. Even
species diversity is difficult to quantify, as accurately measuring all species in an area is not
easily done. Most studies, therefore, focus on a subset of species for estimates of species
richness (i.e., the number of species within an area) or other surrogate measures of
biodiversity.

Surrogate approaches
A good surrogate measure of biodiversity is sensitive to environmental change, both
physical and biological, and data are reasonably easy to sample and analyse. Surrogates may
also be selected because of economic importance or because of their usefulness as a planning
or management tool (Hannon and McCallum, 2001). Surrogate measures can broadly be
categorised into coarse-, medium-, and fine-filter approaches.
Coarse-filter approaches focus at the level of ecosystems, processes, or habitats with
the goal of maintaining biodiversity within them across broad areas (Hunter, 1990; Noss and
Coopenider, 1994). The assumption of a coarse-filter approach is that complete
representation of environmental variability and the preservation or emulation of processes
that contribute to this variability will maintain species diversity. Less emphasis is placed on
full knowledge of species biology and species richness and, therefore, this approach
recognises that for many species this knowledge is unknown. In forest management,
measurements of structural diversity, including stand complexity, composition, connectivity,
and heterogeneity, are proposed as suitable surrogate measures of species diversity in a

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

coarse-filter approach (Kimmins, 1997; Onal, 1997; Lindenmayer et al., 2000). Structural
diversity increases as variation in tree-species composition and tree size, across both
horizontal and vertical spaces, increases (Zenner and Hibbs, 2000; Staudhammer and LeMay,
2001; McElhinny et al., 2005). If a relationship exists between structural diversity and
vertebrate-species diversity, measures of structural diversity should be correlated with
measures of species richness when the unit of area is the same for both measures (Araujo et
al., 2001).
Medium-filter approaches (meso-filter, sensu Hunter, 2005) focus at a stand scale and
centre on specific elements of habitat (Bunnell et al., 1999). For example, several studies
have identified the importance of dead and dying trees, coarse-woody debris, riparian areas,
and deciduous species to forest-dwelling species in western forests (e.g., Zabel and Anthony,
2003). The retention and enhancement of these structural elements may contribute to the
maintenance of species diversity. Monitoring the distribution and abundance of these habitat
elements may thus be a sufficient means to ensure that sensitive species are maintained in an
industrial landscape (Bunnell et al., 1999; Lindenmayer et al., 2000).
Fine-filter approaches focus on the population dynamics, presence and absence, or the
specific needs of a plant or animal species (Hunter, 1991; Hansen et al., 1999). Changes in
the population or distribution of a fine-filter surrogate should indicate changes to medium- or
coarse-filter objectives (e.g., loss of a certain ecosystem type or habitat elements) and are,
therefore, not exclusive of other approaches. Designation of indicator species, keystone
species, umbrella species, and rare or endangered species can all be categorised as fine-filter
approaches. The selection of the fine-filter surrogate depends upon the context, goals, and
objectives of the study or problem (Caro and O’Doherty, 1999).

3

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

The general disagreement of the success of fine-filter approaches (Simberloff, 1998;
Caro and O’Doherty, 1999) and a concern that coarse-filter approaches will not be sensitive
enough (Reyers et al., 2000) have resulted in the recommendation that a combination of
approaches is necessary (Hansen et al., 1999; Sarakinos et al., 2001). Regardless of the
approach, or complement of approaches, the task remains to test and validate indicators and
surrogates to ensure they are telling us what we think they are (Lindenmayer, 1999; Noss,
1999).

Modelling the occurrence o f species
Single species
Niche, island-biogeography, and metapopulation theories have all played important
roles in predicting species occurrence at various spatial and organisational scales. Niche
theory predicts species occupancy at a site given specific habitat conditions relative to the
species’ physiology, morphology, behaviour, and ecology (Wiens, 1989). In this context,
habitat is defined as the area that provides the resources (e.g., food, water, and cover) and
environmental conditions (e.g., temperature and precipitation) that support an individual or
population of a given species, its survival, and successful reproduction (Morrison et al.,
1998). The set or range of environmental features that allow a species to survive and
reproduce is one way to describe a species’ niche (Grinnell, 1917). Other definitions of a
species’ niche include defining the functional role of a species in a community (Elton, 1927)
or as a multidimensional hypervolume where numerous axes represent individual resources
or other important factors (Hutchinson, 1957). In investigations of the relations among
species and physical aspects of habitats, assumptions about the existence of, or role of, a
species within a community should be avoided (Morrison, 2001). Therefore, delineating the

4

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

physical habitat and biological components of a species’ niche allows independent focus on
the interactions among species and habitat, and species with other species (Leibold, 1995).
Predictions of species occurrence may be based on both the physical and biological
components of the species’ environment. Predictions relative to the physical components of
habitat, described in terms of spatial extent, vegetation structure, and vegetation species
composition (Morrison and Hall, 2002), are traditionally approached through a variety of
single-species models that build on knowledge of life history, habitat selection, and / or
foraging theory (reviewed in Morrison et al., 1998).

Multiple species
Predictions of species occurrence relative to the biological component of a species’
niche require an understanding of the presence or absence of prey species, predators,
facilitators (i.e., a species dependent on another species for the efficient acquisition of
resources), or competitors. To determine the importance of biological contributors to species
distributions, a multi-species approach to predictions and analyses is necessary. Multi-species
approaches include categorising species into guilds, compiling species-habitat matrices, and
community-structure models (reviewed in Morrison et al., 1998, Root et al., 2003).
To avoid assumptions based on species interactions, species groups can also be
defined based on their occurrence together in space and in time. These groupings are simply
defined as assemblages (Fauth et al., 1996). Assembling species into groups that share
similar characteristics is a compromise to deal with the impracticality of considering all
species at the same time or all species individually. The general aim of forming species
groups, regardless of how the group is defined, is to make predictions that are more general
than for individual species, but not so general, or unrealistic, as making predictions for all

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

species combined (Wilson, 1999). Species selected as representatives of the species
categorised within groups may then be monitored with fine-filter approaches.

Scale
Measures of species diversity are inherently scale-dependent (Magurran, 2003). For
example, species-area curves (i.e., species richness plotted against area; Arrhenius, 1921;
Coleman, 1981) repeatedly plotted for data collected at different spatial scales do not produce
constant slopes, but rather show systematic variation dependent on spatial scale of
measurement (Crawley and Harral, 2001). Thus, for any study examining biological diversity,
the specific level of diversity being studied and the scale must be defined. Determining the
appropriate spatial and temporal scale of measurement should coincide with the goals and
objectives relevant to the study, ecological processes, and the physical and behavioural traits
of an organism. How a habitat patch is defined in space (i.e., pattern, area, and isolation)
depends on the ecological scale at which the species uses the habitat patch (Vos et al., 2001).
For example, an ideal habitat patch situated 500 m from a similar habitat patch is very
isolated for a terrestrial mammal with a home range of less than a hectare. For a terrestrial
mammal with a home range of several 100 ha, however, this habitat patch is not isolated. The
assessment of isolation is further confounded by the structure of the vegetation that surrounds
the habitat patch and the flexibility of the organism to tolerate sub-optimal conditions. The
probability of a species occurring in a suitable habitat patch will, therefore, relate to the
effective isolation distance, juxtaposition of habitat types, the combined area of the suitable
habitat in relation to the species’ area needs, and the ability of the species to exploit resources
in the surrounding habitat matrix. Determining the scale that is ecologically relevant for the

6

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

species is an important step in the classification of the landscape and subsequent prediction
of species occurrence; therefore, forest structure should also be described at multiple scales.

Frequentist versus Bayesian approaches
There are several statistical approaches to modelling the prediction of species’
occurrence, many of which have been applied only in recent years. Review articles (e.g.,
Guisan and Zimmerman, 2000; Rushton et al., 2004) summarize approaches and how
advances in new technologies have made ecologists play a “catch-up game” to determine
how characteristics of data may affect accuracy and interpretation of model results. Although
a review of individual techniques is beyond the scope of this introduction, a brief overview of
two paradigms, frequentist and Bayesian, is important.
The debate between frequentist (i.e., classical) and Bayesian statisticians is not new
(Clark, 2005). Frequentist statistics are those typically thought of as probabilistic techniques
that rely on falsification of hypotheses, whereas Bayesian approaches use prior information
to develop the formulation of competing hypotheses. Recently, information-theoretic
approaches, a paradigm rooted in Bayesian statistics, have become more prevalent in
ecological studies (Rushton et al., 2004). Comparison studies have reported that informationtheoretic approaches (Burnham and Anderson, 2002) often had a better ability than a
frequentist approach to fit the data and furthered ecologic understanding of the system
studied (e.g., Greaves et al., 2006). In instances, however, where the system is not well
studied, or the investigator cannot use prior knowledge to formulate competing hypotheses,
information-theoretic approaches are undermined. Therefore, there are situations where
frequentist or a combination of Bayesian and frequentist approaches are appropriate (e.g.,
Boone and Krohn, 1999; Rushton et al., 2004).

7

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

Summary and outline o f chapters
Accurately predicting the occurrence of species is a central problem in ecology. As
well, predicting species occurrences based on surrogate measures for biodiversity is an
important aspect of sustainable forest management. Overall, our ability to predict species
occurrence requires an understanding of the relations among species and the environment
defined at appropriate spatial, temporal, and organisational scales relative to the goals and
objectives of the study. In forest management, approaches such as ecosystem representation
and objectives aimed at mimicking natural-disturbance regimes, contribute to coarse-level
biodiversity objectives. This study focuses on determining the relationship between the
presence of a species or group of species (fine-filter) and structural elements of the stand
(medium-filter). The results of my research are intended to contribute to both ecological and
conservation problems.
My dissertation is divided into 3 main chapters. In Chapter 2 ,1 investigate the effect
of detection method and different sources of structure data on model selection and validation.
Different methodologies can result in different ways of determining species presence. Further,
forest managers maintain databases of forest-inventory measures that are commonly derived
through photo interpretation, whereas species may be responding to structure at local scales.
It is, therefore, important to understand if these differences affect model results and
inferences that are drawn. My specific objectives include determining: 1) the effect that
detection type (i.e., sign, audio, visual) has on selection of model variables and predictive
accuracy; and 2) how data sources (i.e., local or photo interpreted measures) affect prediction
accuracy.

8

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

In Chapter 3 ,1 examine the role that physical aspects (i.e., forest structure) of a
species’ environment play in determining the presence or absence of individual vertebrate
species. If forest structure predicts species occurrence well, then there is the potential for
efficiently incorporating its use as a surrogate measure in species-diversity monitoring
programs. Because species use landscapes at different scales (i.e., small or large territories
and a range of mobility), this approach may be more appropriate for a subset of species with
specific life-history traits. Therefore, my specific objectives include determining: 1) the
viability of using attributes of forest structure to predict the occurrence of vertebrate species;
2) the prediction accuracy of models when validated with spatially and temporally
independent data; and 3) if statistical artefacts (e.g., prevalence or detection uncertainty) or
specific traits of species (e.g., mobility or territory size) systematically affect prediction
accuracy.
In Chapter 4 , 1 examine how the physical aspects of the environment explain presence
or absence of assemblages of vertebrate species. If biological aspects of species occurrence,
which are not accounted for by structural aspects of habitat, can be accounted for by
grouping species, forest structure may still be useful as a surrogate measure for monitoring.
Specifically, in this chapter my objectives include determining: 1) the reliability of plot
groupings based on species co-occurrence and forest structural characteristics of plots; 2) the
correlation among species-based and structure-based plot groupings; and 3) whether forest
structure can be used to predict group membership.
Examining these objectives in a multi-scale framework will contribute to the fields of
community ecology, landscape ecology, and conservation biology as well as the management
goal of assessing a potential surrogate approach to monitoring species. By taking a multi-

9

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

scale approach to predicting the occurrence of species and analysing the relations among
species and structural elements of forests, my research will contribute to an improved
understanding of the complexity and variation of species responses to the physical and
biological environment.
Chapters are presented as individual, stand-alone manuscripts intended for
submission for peer-review publication; therefore, there is a certain amount of overlap among
chapters, particularly in the methods sections. I conclude with a brief summary (Chapter 5) of
the results from the preceding chapters.

10

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

Chapter 2. Considerations for the validation of species-habitat models1
Abstract
The multitude of approaches to wildlife-habitat modeling reflect the broad objectives
and goals of various research, management, and conservation programs. Validating models is
an often overlooked component of using models effectively and confidently to achieve the
desired objectives. Statistical models that attempt to predict the presence or absence of a
species are often developed with logistic regression. In this paper, we review principles of
validating logistic regression models, measures of prediction accuracy assessment, and
potential sources of prediction errors in a forest-management context. Based on our work in
central British Columbia, Canada, we use two species, Red Squirrel (Tamiasciurus
hudsonicus) and Pileated Woodpecker (.Dryocopus pileatus), to examine the prediction
accuracy of logistic regression models. Because different types of detections may increase
spatial or temporal uncertainty in empirical models, we use either audio and visual
observations or sign (e.g., forage, nests) as the response variable to compare model results.
We also compare models using data from Vegetation Resource Inventory (VRI; the regional
inventory used by the Province of British Columbia for timber quality and quantity
developed from photo interpretation and ground measurements), local plot measurements
collected as part of this study, and a combination of the two as explanatory variables in the
statistical models. Using detections of sign as the dependent variable resulted in models with
higher predictive accuracy for both species, but the difference was not as great for Red
Squirrel, with small home-range sizes, as for Pileated Woodpeckers, that use landscapes at

1 This chapter is written in the first person plural to recognize the contribution o f others to the work. It has been
submitted for publication with the authorship Psyllakis, J.M and M.P Gillingham to the Proceedings for
Monitoring the Effectiveness o f B iological Conservation, Richmond, BC.

11

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

much larger scales. The final models selected, based on low Akaike’s Information Criterion
(AIC) and predictive accuracy, included a combination of locally measured independent
variables and VRI data as explanatory variables. Our results suggest that detection type may
affect model outcomes and relatively small investments in data acquisition can improve
predictive accuracy. We discuss considerations for the development and validation of
statistical models intended for use in biodiversity monitoring.

Introduction
The worldwide rate of species decline has resulted in international agreements (e.g.,
Convention on Biological Diversity; United Nations Environment Programme, 1992),
national strategies (e.g., Canadian Biodiversity Strategy; Minister of Supply and Services
Canada, 1995), and provincial guidelines (e.g., Landscape Unit Planning Guide; Province of
British Columbia, 1999a) aimed at preserving species diversity in reserves, as well as
landscapes managed for industrial purposes. Forests provide habitat necessary for thousands
of species, as well as considerable economic and social resources to humans. Activities
driven by anthropogenic values have contributed to habitat loss and fragmentation resulting
in a global decline of many species from forest environments (Hunter, 1990; Harper and
Hawksworth, 1995). A strategy for conserving, or minimizing impacts on, biodiversity is also
required as part of most sustainable forest certifications (e.g., Sustainable Forestry Initiative,
2004; Forest Stewardship Council, 2005).
Models that build on the relationships between species and their environments
provide an important tool for biodiversity monitoring. To be effective, however, specieshabitat models need to be explicitly tested (Guisan and Zimmerman, 2000; Scott et al., 2002)
and model validation is a vital component to confidently implement monitoring objectives

12

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

(Ottaviani et al., 2004). The process of validation can increase the understanding of specieshabitat relationships (Fleishman et al., 2002), the limitations of the statistical model
application, and whether or not the model is appropriate for its intended use (Rykiel, 1996).
Statistical approaches to species-habitat modeling vary with modeling objectives and
available types of data (Guisan and Zimmermann, 2000). Logistic regression is often the
preferred method to model species presence or absence in relation to habitat variables (Manel
et al., 1999; Pearce and Ferrier, 2000). The resulting logistic equation predicts the probability
of species presence given independent variables and parameters (i.e., the intercept and P
coefficients). Validation of logistic regression models usually focuses on the accuracy of
predictions (Fielding and Bell, 1997; Johnson, 2001) and is judged on: 1) reliability - the
accuracy of the predicted likelihood of occurrence; and 2) discrimination - the ability of the
model to accurately distinguish between occupied and unoccupied sites (Pearce and Ferrier,

2000).
Calculating the area under the receiver-operating characteristic curve (ROC) is a
favoured measure used to assess the predictive accuracy of logistic models, when presence
and true absence data are available (Fielding and Bell, 1997; Pearce and Ferrier, 2000). The
ROC value is calculated by plotting the number of sites where presence is correctly predicted
divided by the total number of positive sites (sensitivity), against the fraction of incorrect
cases where presence is predicted (1-specificity; Table 2.1) across available thresholds. The
area under the resulting curve is an estimate of predictive accuracy not biased by threshold
probabilities (i.e., p > 0.5 designated as presence; Fielding and Bell, 1997) or species
prevalence (i.e., one outcome greatly outnumbers the other; Manel et al., 2001). A ROC
value is interpreted as the percentage of time that a random selection from the positive class

13

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

Table 2.1. Matrix of prediction classifications describing the possible outcomes of presence
or absence and the associated indices that can be used to describe predictive performance.

Classification matrix

Recorded
Present

Recorded
Absent

Predicted
Present

A

B

Predicted
Absent

C

D

Definitions of the four indices of performance

Sensitivity

= A / (A + C)

Specificity

= D / (B + D)

False positive fraction

= B / (B + D)

False negative fraction

= C / (A + C)

14

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

will have a higher predictive score than a randomly drawn case from the negative class
(Deleo, 1993). ROC values of 0.5 indicate the explanatory variables do not improve
discrimination beyond random assignment and 1.0 indicates perfect discrimination. A value
below 0.5 indicates the model performs more poorly with the explanatory variables than
without them. Values between 0.5-0.7 are considered to have low discrimination ability, 0.70.9 are good, and >0.9 are considered excellent (Manel et al., 2001).
Statistical models developed to monitor biodiversity are often applied to areas beyond
the location where data were collected (Mac Nally, 2002). To have the highest confidence in
the reliability and discrimination of the model, an external data set (independent from the
data used to build the model) should be used in validation (Guisan and Zimmerman, 2000;
Fleishman et al., 2002). Obtaining an independent data set, however, is often not feasible.
Withholding data to test the model or using a resampling technique are alternatives to using
an independent data set (Fielding and Bell, 1997; Boyce et al., 2002), but will still result in
optimistic prediction accuracy (Verbyla and Litaitis, 1989; Fielding and Bell, 1997; Pearce
and Ferrier, 2000).
Further assessment of the sources of prediction error can lead to improved
understanding of the ecological associations between the species and its habitat as well as the
utility of the model. Prediction errors can occur because of errors in specifying the model,
inappropriate statistical assumptions, measurement errors, and uncertainty related to natural
variation (Elith and Burgman, 2002; Fielding, 2002). Two potential sources of error related
to specification error come from bias in detection type and misappropriate inclusion or
exclusion of explanatory variables.

15

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

Species presence can be established from a variety of detection types including visual
detections, auditory detections, tracks, scat, and forage sign. Visual detections have little or
no spatial or temporal uncertainty associated with them. In contrast, auditory detections are
temporally certain, but potentially can have a high degree of spatial uncertainty given that,
for some species, calls can travel several hundred meters. Mobile species may call while in
flight adding additional uncertainty as they may be in transit between activity areas of their
home ranges. Feces, tracks, dens, and nests are all exact in space, but vary in their temporal
certainty. Some signs (e.g., cavity nests, dens) are very persistent on the landscape and
habitat structure can change significantly around the sign. It may be unclear if the location is
still useful to the species it is associated with, because individuals may not be present at the
time of detection. Uncertainty in the response data potentially leads to unexplainable
variation in the model and reduced reliability and discrimination (Pearce and Ferrier, 2000).
Determining species absence is more ambiguous, and may require that an alternative
modeling and validation approach is adopted for presence-only data (e.g., ecological niche
factor analysis, see Hirzel et al., 2002; MacKenzie et al., 2002; Ottaviani et al., 2004).
Misappropriate inclusion or exclusion of explanatory variables also leads to
prediction error of species presence or absence. Factors that influence species distribution
often include variables that are not typically measured in association with studies of specieshabitat relationships (e.g., intra- and inter-species interactions). In the context of forest
management, habitat data for use as predictor variables may be obtained from data that are
available for forest-harvest inventories. Vegetation Resource Inventory (VRI) is an inventory
methodology adapted by the British Columbia (BC) Provincial government (Province of
British Columbia, 2002); the primary objective of the inventory is to assess the quantity and

16

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

quality of timber. These data are available for much of the forested landbase of BC. The
limits associated with the type of environmental data available or acquisition of appropriate
data, both for statistical model construction and its use in monitoring activities may, however,
influence the effectiveness of achieving the desired model objective(s).
Statistical model assessment also includes an evaluation of the variation explained by
the independent variables. In the logistic model, this cannot be calculated in the same way as
for other regression models and there are several alternative measures. Pseudo-R is the
recommended measure to describe variance explained for logistic regression (Menard, 2002)
and is measure of proportional reduction in the absolute value of the log-likelihood when
2

,

variables are included (Nagelkerke, 1991). What constitutes a “good” pseudo-R value is
unclear, but it can be useful when comparing across models that do not have the same
response data (e.g., comparing between detection types). A model with a low pseudo-R may
still have high predictive accuracy.
In this paper we investigate how two types of potential model error affect the
predictive accuracy and evaluation of species-habitat models: 1) the type of detection used to
determine species presence and, 2) different sources of data for explanatory variables. We
use two species as examples of how differences in the response variable and habitat data
source may affect model selection and discrimination, Red Squirrel (Tamiasciurus hudonicus)
and Pileated Woodpecker (Dryocopus pileatus). We selected these species for examples
because they are relatively common in our study area, their sign is distinctive (i.e., there is
little uncertainty in species identification from sign), and the Red Squirrel uses the landscape
at small-spatial scales and is not highly mobile in short-time periods, whereas the Pileated

17

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

Woodpecker uses the landscape at relatively large scales and is highly mobile in short-time
periods.

Methods
Study area
The data were collected from May 2001 through January 2004 near the northern
extent of the Interior Douglas-Fir (IDF) Biogeoclimatic zone (Meidinger and Pojar, 1991).
The study area was located approximately 30 km south of Williams Lake, British Columbia,
hereafter the Williams Lake Study Area (Figure 2.1). Elevations within the study area ranged
from 800-1200 m above sea level and was characterised by stands of closed- and opencanopy Douglas-fir (Pseudotsuga menziesii). At high elevations, or where crown fires have
occurred in the past, Lodgepole Pine (Pinus contorta) was common and at low elevations
forest stands were intermixed with non-forested grassland and wetland communities. There
were localised stands of hybrid White Spruce (Picea engelmannii x glauca) and Trembling
Aspen (Populus tremuloides) throughout the study area. Fire was an important disturbance
process in our study area historically, but is now actively suppressed. Insect outbreaks
continue to influence stand dynamics. Forest harvesting and cattle grazing are the
predominant anthropogenic disturbances.

Habitat data
We established 243 plots to encompass a range of variation in structural
characteristics over a breadth of spatial scales (Figure 2.1). Plots were connected by 150-m or
300-m transects for a total of nearly 42 km of transects. Each transect was flagged to ensure
that the same route was followed on successive visits. We collected extensive vegetation data

18

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

-iT
HWY 97
Jkm

gravel roads

Figure 2.1. The location of the Williams Lake Study Area (■) in central British Columbia,
Canada. Williams Lake is mapped for reference (+). We established 243 plots (•) across a
range of variation in structural characteristics and surveyed them for species presence from
May 2001-January 2004.

19

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

for all plots using standardised methodologies modified from several sources (summarized in
Gillingham and Parker, 2001). Modifications to the standard inventory were made so that
data were relevant from both the perspectives of wildlife and forest inventory, as well as for
the examination of spatial variation.
We used surveyor tapes, laid out on perpendicular axes through plot centre, to
measure shrub species and cover, canopy gaps, and coarse-woody debris along the intercept.
For coarse-woody debris, we recorded the diameter of the piece perpendicular to where it
crossed the axis, the tree species, decay class (Maser et al., 1979), and any sign of use by
wildlife. At 5, 2-m radius plots located 11.28 m from plot centre on each axis and at plot
centre, we measured coverage for litter, coarse-woody debris, herb species, moss and lichens,
shrub species, sapling species, bare ground, and rock. Within a 5.64-m radius plot (0.01 ha)
around plot centre, we tallied all trees and stumps. Within an 11.28-m radius (0.04 ha) we
tallied all trees >30 cm diameter at breast height (dbh) and snags. For trees, we recorded dbh,
height, species, health, evidence of wildlife use, and whether or not the tree was standing or
not rooted. We recorded general information for each plot including canopy closure around
plot centre (average of 4 measurements taken on each axis; Robert E. Lemman model C
densiometer, Bartlesville, Oklahoma), aspect, slope, canopy stratification and complexity,
disturbance history (evidence of fire, cattle grazing, logging), any wildlife species detected
while taking vegetation measures or their sign, and the elevation above sea level. We were
supplied Vegetation Resource Inventory (VRI) data by the forest company that operates in
our study area (Tolko Ltd., Vernon, British Columbia).

20

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

Species detections
Vocal and visual detections of Red Squirrels were recorded during point-count
surveys and encounter-transects surveys. We also recorded foraging sign, middens, and nests
during encounter transects and intensive-plot searches. Vocal and visual detections were
recorded for Pileated Woodpeckers during playback surveys, point counts, and encounter
transects. Foraging sign and nest cavities were recorded during encounter transects and
intensive-plot searches.
Red Squirrels frequently vocalise and are often recorded during point-count surveys
(Mattson and Reinhart, 1996; Bayne and Hobson; 2000). Therefore, during point-count
surveys, which began no earlier than 30 min before sunrise and were ceased no later than 4 h
after sunrise (Ralph et al., 1993) from late-May until early-July, we recorded visual and vocal
presence of Red Squirrels. Point counts lasted 6 min and began 1 min after arriving at the
centre of a plot. Distance and direction to detection from plot centre were estimated and
recorded as either <50 m, 50-75 m, or >75 m. One visit per plot was made over a 7- to 10day period by a different observer. Direction of travel along the transect was changed on
successive visits to reduce bias associated time of day.
We conducted playback surveys, broadcasting recordings of calls and drumming, for
the 7 species of woodpeckers expected to occur in the study area, including Pileated
Woodpecker. The call playback technique attempts to solicit woodpecker responses to
broadcasted recordings (Johnson et al., 1981). Woodpecker surveys began no earlier than 30
min before sunrise and ended by 1100 h from mid-May until mid-June at every other plot
(minimum distance of 300 m between playback stations). When a woodpecker was detected,
distance and direction to bird from plot centre were estimated and recorded as for point

21

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

counts. When point counts were conducted in conjunction with playbacks, playbacks were
always conducted after the passive listening period of the point count. Poor weather, such as
high winds, rain, and fog can inhibit both bird behaviour and observer ability; therefore,
point-count and playback surveys were only conducted in appropriate weather conditions.
Each plot was surveyed by point count and playback 3 times per year for a total of 9 visits.
Along the transects connecting plots, we conducted encounter-transect surveys of
unlimited width. If a species was detected along the transect, a Global Positioning System
(GPS) waypoint was recorded with species ED, distance and bearing from waypoint, as well
as a sign code (singing, call, visual, den/nest, track, remains, browse, or feces). In addition,
extensive searches were made around all plots for animal sign. Each plot was searched
intensively, to a 50-m radius, for all sign of vertebrates (i.e., visuals, calls/song, nests, dens,
feeding, feces, remains). Detections from all methods were entered into a database and
georeferenced.

Model construction and predictive evaluation
We used logistic regression (Hosmer and Lemshow, 2000) to examine the
relationship of species occurrence to vegetation structure and composition. Candidate
variables were selected from potential variables measured and supplied (i.e., VRI data). We
considered a variable a candidate if it was, or was related to, an aspect of the species’ habitat
requirements (Table 2.2). If variables existed that were measures of the same characteristic,
but from a different data source (e.g., plot percent gap and VRI sum of crown closure), the
variable with the highest correlation coefficient with the species presence was included in a
candidate model. Therefore, the final candidate model set included combinations of variables
expected to influence the occurrence of Red Squirrel and Pileated Woodpecker. We only

22

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

Table 2.2. List of candidate variables used to construct competing models for Red Squirrel
(Tamiasciurus hudonicus', TAHU) and Pileated Woodpecker (Dryocopus pileatus; PIWO). A
structural variable was considered a candidate if it related directly to known associations
between the species and its habitat.

Data Source
Plot Measures

Vegetation Resource Inventory

Variable

Species

Main canopy height
Percent gap
Percent shrub cover
Percent herb cover
Coarse-woody debris volume (m3)
Percent suspended CWD
Live tree basal area
Dead tree basal area
Diameter breast height >30cm
basal area
Deciduous stems per ha
Frequency Douglas-fir
Frequency Spruce
Coniferous stems ha"1

TAHU, PIWO
TAHU, PIWO
TAHU
TAHU
TAHU, PIWO
TAHU
TAHU, PIWO
TAHU, PIWO
TAHU, PIWO

Shrub crown closure
Adjusted live basal area
Herb cover percent
Adjusted canopy closure
Douglas-fir cover
Aspen cover
Spruce cover
Adjusted leading species height
Structure class (categorical)

TAHU
TAHU, PIWO
TAHU
TAHU, PIWO
TAHU, PIWO
TAHU, PIWO
TAHU, PIWO
TAHU, PIWO
TAHU, PIWO

TAHU, PIWO
TAHU, PIWO
TAHU, PIWO
TAHU, PIWO

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

considered plots in which structure was not altered (e.g., harvested) over the duration of the
study (n = 228).
To test for collinearity among independent variables, we calculated variance inflation
factor values for all variables in the model after linear regression (Neter et al., 1985).
Variation inflation increases with increasing collinearity among variables and results in
overestimates of variance explained. Although there is no set rule for a variance inflation
factor indicating a collinearity problem, we adopted a value of 5 or above, which corresponds
with a tolerance score of 0.2, a recommended threshold (Menard, 2002). If collinearity was
indicated, we reran our model using only one of the indicated problem variables and
compared outcomes. We retained the variable that resulted in the highest pseudo-R and
predictive accuracy. We calculated the pseudo-R (Nagelkerke, 1991) as our measure of
variation explained and ROC values to estimate predictive accuracy. We classified ROC
values between 0.5-0.7 as low, 0.7-0.9 as good and >0.9 as high model prediction accuracy
(Manel el al., 2001). For comparing models with different response variables (i.e., audio or
visual detections versus sign detection), we considered the highest pseudo-R2 and highest
ROC value as the best model.
Once we determined which detection type resulted in the best model (i.e., highest
pseudo-R2 and highest ROC value), we then assessed competing models (i.e., different
combinations of independent variables). Competing models were ranked using Akaike’s
Information Criteria (AIC; Akaike, 1973; Burnham and Anderson, 2002). AIC model
selection estimates the information loss between the probability distribution with the true and
the probability distribution associated with the model that is to be evaluated. Choosing the
model with the lowest expected information loss between the true model and the

24

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

approximating model is asymptotically equivalent to choosing a model that has the lowest
AIC value (Burnham and Anderson, 2002). We applied a correction to the AIC value to
account for small sample sizes (AICc) and determined the Akaike weight (w,), the likelihood
of the model given the data (Burnham and Anderson, 2002). For those competing models, the
lowest AICc value and highest Akaike weight defined the best model for each set of predictor
variables. After the final models were selected, we reviewed cases for high leverage and
studentised residual values to determine if any cases were disproportionately driving the
model relationship (Menard, 2002). We used the program Stata (version 8.2; StataCorp, 2003)
for all statistical analyses and employed the DESMAT procedure (Hendrickx, 2001) for
design matrices involving categorical variables.

Results
Red squirrel
We recorded audio and visual detections of Red Squirrel at 178 plots and detected
sign at 205 plots (219 plots in total). Audio and visual detections were primarily made during
point-count surveys. Sign detections included foraging sign, middens, remains, and nests.
The best model was constructed with sign as the response variable (Table 2.3). This model
explained approximately 16% more variation and the prediction accuracy improved by 8%,
but both the sign and audio visual models had good predictive accuracy (sign ROC = 0.88,
audio and visual ROC = 0.80; Table 2.3). Several of the same variables were included in both
models, specifically, structure class, percent gap, and frequency Spruce. Comparing the
results of competing models for only sign as the response variable, the best model was
constructed with a combination of plot-level data and VRI data (AICc = 181.55; Table 2.4).

25

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

Table 2.3. Model results predicting the occurrence of Red Squirrel (Tamiasciurus hudonicus)
in the Williams Lake Study Area using audio and visual detections or sign detections. Using
sign as the response variable resulted in the highest pseudo-R2 and highest ROC value.

Response variable

Independent variables

Pseudo-R2

ROC

Audio Visual

Structure Class
Percent Gap
Frequency Spruce

0.20

0.80

Sign

Structure Class
Percent Gap
Frequency Spruce
Dead trees basal area
Shrub crown closure

0.36

0.88

26

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

Table 2.4. Models predicting occurrence of Red Squirrel (Tamiasciurus hudonicus) in the
Williams Lake Study Area using different sources for independent data variables. The model
using plot measures and Vegetative Resource Inventory (VRI) data resulted in the lowest
AICc and highest ROC value.

Data Source

Independent variables

AICc

AAIC

w,

ROC

VRI and Plot

Structure class
Percent gap
Spruce stems ha'1
Dead tree basal area
Shrub crown closure

181.55

0

0.93

0.88

VRI

Adjusted canopy
closure
Structure class
Spruce cover

187.31

5.28

0.067

0.86

Plot

Percent gap
Spruce stem ha'1

212.36

29.28

<0.01

0.76

27

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

The multi-source model was highly favoured as the best model with an Akaike weight of
0.93, or a 93% likelihood of being the best model; however, prediction accuracy did not
improve dramatically (VRI ROC = 0.86, multi-source ROC = 0.88; Table 2.4).

Pileated woodpecker
We recorded audio and visual detections for the Pileated Woodpecker at 63 plots and
sign at 33 plots (85 plots in total). Sign detections included foraging and nest cavities. The
best model was constructed using sign as the response variable (Table 2.5). There was little
overlap among explanatory variables included in the two models. Sign detection as the
response variable dramatically improved model performance over audio and visual measures.
The variation explained improved by 14% and predictive accuracy improved from poor
(ROC = 0.66) to good (ROC = 0.82; Table 2.5). Comparing the results of competing models
using only sign as the response variable, the best model was constructed with a combination
of plot and VRI data (AICc = 179.55; Table 2.6). The likelihood of the multi-source model
being the best approximating model was 66% and predictive accuracy improved by 10%
(Table 2.6).

Discussion
For both the Red Squirrel and Pileated Woodpecker, logistic models using sign
detections outperformed those with audio and visual detections. Measurement uncertainty
from several sources may explain this result. Firstly, it is sometimes difficult to associate
vocalisations to specific locations because of the distance over which sound travels and
measurement error assigning the detection to a spatial location. Secondly, vocalisations may
be made while the individual is in transit between parts of its home range, as is the case for

28

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

Table 2.5. Model results predicting the occurrence of Pileated Woodpecker (Dryocopus
pileatus) in the Williams Lake Study Area using audio and visual detections or sign
detections. Using sign as the response variable resulted in the highest pseudo-R2 and highest
ROC value.

Response variable

Independent variables

Pseudo-R

ROC

Audio Visual

CWD volume
Percent gap
Douglas-fir stems ha"1
dbh >0cm basal area

0.06

0.66

Sign

Percent Gap
Structure class
Douglas-fir cover
Main canopy height
Adjusted live basal area

0.21

0.82

29

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

Table 2.6. Models predicting occurrence of Pileated Woodpecker (Dryocopus pileatus) in
the Williams Lake Study Area and different sources for independent data variables. The
model using plot measures and Vegetative Resource Inventory (VRI) data resulted in the
lowest AICc and highest ROC value.

Data Source

Independent variables

AICc

AAIC

Wi

ROC

Multi-Source

Percent Gap
Structure class
Douglas-fir cover
Main canopy height
Adjusted live basal area

179.55

0

0.66

0.82

VRI

Adjusted canopy closure
Douglas-fir cover
Adjusted live basal area

181.16

1.611

0.30

0.72

Plot

Main canopy height

185.21

5.66

0.04

0.65

30

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

the Pileated Woodpecker. Finally, the observer must be at the location at a moment in time
when the individual is near the same location. For species with large home ranges, the
probability of an observer being in the same location, in space and time, is lower than that for
a species with a smaller home range. It is possible, therefore, that we may have not detected
presence. Given our survey methods for sign, it was unlikely that we would not detect
presence if it was there.
Sign detections are often associated with a specific component of a species’ habitat
requirements (e.g., foraging or nesting sites). Logistic models with sign data as the response
variable likely had higher predictive efficiency because foraging and nesting substrates are
often selected for at smaller scales, within the context of a home range. For example, Red
Squirrels have home ranges of 1-3 ha (Obbard, 1987), easily within a single stand of similar
habitat characteristics. Audio and visual detections are likely recorded within the same stand
as foraging and other life history activities take place; therefore, it is not surprising that
model results are similar for the different detection types for Red Squirrel. The additional
variables included in the sign model for Red Squirrel, shrub cover and dead wood basal area,
may be more related to the associated effects of middens on ground vegetation and the
availability of cones. Potential temporal uncertainty associated with changes to vegetation
structure and composition from disturbance (e.g., alteration because of harvesting practices)
around sign detections may prove a greater issue in long-term monitoring studies.
The Pileated Woodpecker has a large home range; pairs in the Pacific Northwest use
between 300 and 600 ha, while unpaired birds used up to 1400 ha (Bull and Holthausen,
1993). Pileated Woodpeckers select large snags and logs to forage on and large diameter
trees for nesting (Bull and Holthausen, 1993; Carey et al., 1991). These structural

31

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

components may be localised or clustered within a highly variable home range, resulting in
better predictive accuracy of models for sign. Further, because we employed a survey
methodology that was designed to solicit responses from individuals if they were in the area
(i.e., to minimise false absence detections) we may have inadvertently attracted an individual
from its original location and the associated characteristics of that original location. It is
unclear which is more detrimental to the predictive efficiency of logistic species-habitat
models and their use in a biodiversity monitoring program, false absence or spatial
uncertainty related to an individual’s movement.
Overall, differences in logistic-model results, and the potential difference in a
monitoring program designed around them, emphasize the importance of assessing sources of
potential model error and the predictive efficiency. For a species that uses landscapes at
relatively small scales (e.g., Red Squirrel), sign and visual or audio detections are likely to be
within the same area and not have a large effect on a monitoring program. In contrast, there
was little overlap among variables included in the different statistical models for Pileated
Woodpecker and the predictive efficiency varied greatly. It may be more appropriate to use
complementary models over a general model that has weak predictive accuracy and to give
careful consideration to the biology of the species. Ultimately, the decision of which is the
most appropriate approach will depend on what poses the highest risk to the species (e.g.,
loss of critical habitat features) and the goals and objectives the statistical model is intended
to help achieve.
In this paper we examined the effects of two kinds of potential model error: detection
type and availability of structure data. We showed that: 1) different types of detection can
potentially affect model results and assessment of accuracy; and, 2) that adding a small

32

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

amount of locally measured data can improve predictive efficiency dramatically. Measuring
these data may have relatively low costs. We conducted our analyses using our original data
set; using independent data to validate models is the preferred approach to address all aspects
of validation. It is, therefore, likely that we have reported optimistic predictive accuracy
(Chatfield, 1995). Additionally, although the area under the ROC curves is a robust measure
of prediction accuracy, its use in the biological sciences is relatively recent. Some caution
should be taken when using the method outside of its original development (Kraemer, 1988)
although the approach is highly recommended for presence and absence data (Ottaviani et al.,
2004).
Overall, assessing prediction accuracy is only one component of model validation.
Validation needs to be an iterative process so that confidence is maintained in the model’s
usefulness through continued monitoring (Johnson, 2001). Other factors, outside of those
used in predictive-habitat models, may ultimately have greater impact on the presence of
species (e.g., invasive species, climate change). A better understanding of the relations
among species distribution and forest characteristics, both in the short- and long-term, will
provide knowledge to help mitigate the loss of species from forest environments. Ultimately,
assessing model performance will allow for informed trade-offs and lead to improved
effectiveness of biodiversity monitoring. In conclusion, every monitoring program for
biodiversity that uses species-habitat models should make effort to ensure that models are
valid for their intended use. Clear articulation of model objectives and a thorough
consideration of appropriate types of data collection, a standard of acceptable error, and
independent data for evaluation of error will improve the effectiveness of biodiversity
monitoring programs.

33

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

Chapter 3. Using forest structure to predict the occurrence of vertebrate species1
Abstract
In forested environments, mitigating negative impacts of forest-harvesting activities
and monitoring biodiversity are common requirements of sustainable-use certification and of
various laws and regulations. Managers require an understanding of how species respond and
persist within the dynamics of changing forest environments so that management strategies
can retain and recruit structural aspects necessary for the persistence of populations. We
tested several structural models to predict the presence or absence of a range of vertebrate
species (n = 55) and to determine the potential of using forest structure to monitor species
distributions. We validated models with temporally and spatially independent data. Some of
the models had good predictive accuracy that was retained when validated and thus have
application in terms of implementation as management tools. Modelling success varied,
however, depending on whether plot or stand data were used; many models included
variables related to spatial relationships of structures. Few models were reliable when applied
to the independent data; therefore, our results indicate that models cannot be assumed to be
applicable in different years or applied outside the area where the model was developed, even
with similar spatial and temporal contexts. We did not find robust relationships necessary to
guide management targets for retention and recruitment of specific forest structures.
Therefore, using these structural models as surrogates for monitoring species occurrence is
limited. Monitoring of structure should still be included as part of biodiversity monitoring
programs because preservation of structures known to be negatively affected by harvesting
(e.g., dead wood, large trees, closed canopies, continuous forests) contributes to local and
1 Throughout this chapter the first person plural is used to reflect the contributions o f others to this research. The
manuscript will be submitted with the authorship: Psyllakis, J.M. and M.P. Gillingham.

34

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

landscape heterogeneity and has been shown to affect species presence in this study and
others.

Introduction
Industrial forest-management activities can contribute to the loss and fragmentation
of habitat, change in tree-species composition, and the introduction of non-native species.
These factors are linked to the decline of species from forest environments at local to global
scales (Groombridge, 1992; Berg et al., 1994; Harper and Hawksworth, 1995; Noss, 1999;
Hunter, 2001). Protecting biodiversity in reserves is not sufficient to mitigate the decline
(Hunter, 1990). Therefore, efforts to mitigate loss of biodiversity are also included in plans
for the sustainable use of natural resources (Rosenzweig, 2003). Through sustainable forestmanagement policies and certification, protection of biodiversity and ecological integrity are
mandated (e.g., United Nations Environment Programme, 1992; Province of British
Columbia, 1995; Montreal Process Working Group, 1999; Canadian Council of Forest
Ministers, 2003), thus forest-land managers require tools to assess biodiversity and to ensure
that operational activities have minimal effects on biodiversity.
Biodiversity is a broad concept that spans spatial, temporal, and organizational scales
(see Chapter 1; Bunnell and Huggard, 1999; Purvis and Hector, 2000; Willis and Whittaker,
2002; Magurran, 2003). Species diversity is a commonly measured component of
biodiversity (Purvis and Hector, 2000; Magurran, 2003). Even the narrow focus on the
measurement of species diversity is problematic as it is impossible to measure or monitor all
species. To overcome the difficulty of measuring all species directly, surrogate measures are
used with an assumption that whatever is being measured is representative of a larger aspect
of diversity (Noss, 1990; Caro and O’Doherty, 1999; Margules and Pressey, 2000).

35

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

Approaches to surrogate measures may be habitat-based, species-based, or a
combination of the two, and will vary depending on the spatial scale of the conservation
goals (Poiani et al., 2000; Groves et al., 2002). To reflect the scale of the approach, both
spatially and biologically, surrogate measures are often characterised as fine-, medium-, or
coarse-filter (see Chapter 1; Hunter, 1990; Noss, 1996; Lindenmayer and Franklin, 2002;
Hunter, 2005). In the context of forest management, fine-filter approaches are typically
species-based and applied at local scales; medium- and coarse-filter approaches are typically
habitat-based and applied at stand to landscape scales (Noss, 1996; Hunter, 2001;
Lindenmayer and Franklin, 2002; Hunter, 2005).
Examples of fine-filter approaches include monitoring or managing population trends
or distribution of single species (Hansen et al., 1999; Hunter, 2001). Fine-filter species may
be rare or endangered or be representative of other aspects of the ecosystem or community
(e.g., indicator, flagship, keystone, or umbrella species; Hunter, 2001). Fine-filter approaches
have been criticised because of a general lack of knowledge and understanding of the
relationships among species and species diversity, the expense and time that is required to
monitor or manage single species, and the questionable contribution that protecting the
identified species has on the rest of the ecosystem (Simberloff, 1998; Ricketts et al., 2002;
Caro et al., 2004).
Medium-filter (also mesofilter sensu Hunter, 2005) surrogate approaches focus on
specific elements of forest stands that represent critical habitat for the persistence of some
species (Bunnell et al., 1999; Lindenmayer et al., 2000; Hunter 2005). For example, in
forested environments several studies have identified the importance of dead and dying trees,
coarse-woody debris, riparian areas, and deciduous species to forest-dwelling species (e.g.,

36

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

Bunnell et al., 1999; Martin and Eadie, 1999; Bowman et al., 2000; Zabel and Anthony, 2003;
Mazurek and Zielinski, 2004). The retention and enhancement of these structural elements
may, therefore, contribute to reducing the risk of extirpation of species dependent on these
structures for some aspect of their life history from managed forests (Lindenmayer and
Franklin, 1997; Bunnell et al., 1999; Lindenmayer et al., 2000).
Coarse-filter surrogate approaches centre on the preservation of representative
habitats and ecosystems on the landscape through time. An underlying assumption is that
complete representation of environmental variability, and the preservation or emulation of
processes that contribute to this variability, will maintain species diversity and thus their
processes and functions (Hunter, 1990; Noss and Cooperrider, 1994; Hunter, 2001). Coarsefilter approaches are criticised because of a lack of congruence between environmental and
species diversity (Reyers et al., 2000; Araujo et al., 2001) and a sensitivity to missing rare,
endemic, and sensitive species (Margules and Pressey, 2000; Hunter, 2001; Noon et al.,
2003). The general disagreement of the success of fine- and coarse-filter approaches
(Simberloff, 1998; Caro and O’Doherty, 1999) has resulted in the recommendation that a
combination of approaches is necessary (Noss, 1990; Noss and Cooperrider, 1994; Poiani et
al., 2000; Sarakinos et al., 2001; Lindenmayer and Franklin, 2002).
The use of habitat-based surrogates at the coarse- and medium-filter level has been
encouraged because complete knowledge of species’ biology is not required and it is usually
less costly to assess characteristics of habitats than species and populations (Noss, 1996;
Bunnell et al., 1999; Lindenmayer et al., 2000; Ricketts et al., 2002). Habitat-based
approaches are particularly appealing to forest managers because measures such as forest
cover, vegetation species composition, and stand age are intuitive and directly linked to

37

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

harvest-based forest inventories. Regardless of the approach, the surrogate measure should
have a strong relationship with its intended non-measured counterpart.
Linking species presence to forest structural elements for the purpose of biodiversity
monitoring incorporates aspects of both species-based and habitat-based surrogate
approaches and is a compromise between the expense of monitoring a single species and the
possibility of missing species needs in coarse-filter approaches. Using forest structure as a
predictor of species richness has received considerable recent attention (e.g., Lahde et al.,
1999; Diaz et al., 2005; Loehle et al., 2005; Oxbrough et al., 2005). Generally, however,
forest structure remains relatively untested as a surrogate measure for biodiversity
monitoring. The development of statistical models linking species occurrence to structural
characteristics of forests may also provide clear targets for management that have direct
impacts on structural aspects of forests. Further, predictions on how species will respond to
changing structure need to be made if a better understanding of the causal effects of
management on diversity, an important aspect of adaptive management (Walters, 1986), are
to be incorporated into management. If linkages are strong and predictions of species
presence are accurate, monitoring forest structure could contribute to assessing the effects of
management on species diversity.
Examining the linkages between species presence and forest structure can be
approached with statistical modelling. Before implementing the use of statistical models as
part of management programs, however, the models should be validated with spatially and
temporally independent data (Rykiel, 1996; Guisan and Zimmerman, 2000; Ottaviani et al.,
2004; Guthery et al., 2005). Validation of logistic regression models, a common method of
modeling species presence and absence data, usually involves assessment of two aspects of

38

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

prediction accuracy: 1) reliability - the accuracy of the predicted likelihood of occurrence;
and, 2) discrimination - the ability of the model to accurately distinguish between occupied
and unoccupied sites (Fielding and Bell, 1997; Pearce and Ferrier, 2000; Johnson, 2001).
Statistical artefacts (e.g., sample prevalence; Manel et al., 2001; McPherson et al., 2004), as
well as biological attributes of species (Karl et al., 2000; Scott et al., 2002; McPherson et al.,
2004; Seoane et al., 2005) are linked to prediction accuracy. An assessment of systematic
biases in model predictions is, therefore, required if models are to be compared among
different species.
The overall goal of our study was to examine the possible use of structure-based
measures as explanatory variables to predict the presence and absence of vertebrate species,
as a surrogate approach to monitoring species diversity. Although vertebrates represent a
relatively small portion of species diversity in most ecosystems (Redak, 2000; Spence, 2001),
several standardized protocols exist for their measurement (Province of British Columbia,
1991; Heyer et al. 1994; Wilson et al., 1996) and there is a wide range of studies linking the
presence of vertebrate species to structural elements of habitat (e.g., snags, basal area, large
trees, coarse-woody debris; Keisker, 2000; Bull, 2002). Specifically, our objectives were to
determine: 1) the viability of using attributes of forest structure to predict the presence of
vertebrate species; 2) the validity of model predictions using spatially and temporally
independent data; and, 3) if statistical artefacts (e.g., sample prevalence, detection uncertainty)
or biological traits of vertebrate species (e.g., mobility, territory size) consistently influence
the prediction accuracy of these models.

39

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

Methods

Study area
Our study took place near the northern extent of the dry-warm subzone of the Interior
Douglas-Fir (IDF) Biogeoclimatic zone (Meidinger and Pojar, 1991) approximately 30 km
south of Williams Lake, British Columbia, Canada (Williams Lake Study Area; Figure 3.1).
Data were collected from May 2001 through January 2004 for original model development
(development sites). For spatial validation data, the study area was extended in the summer
of 2004 within the IDF to the south-east extent of the moist-cool subzone of the Sub-Boreal
Pine Spruce (SBPS) Biogeoclimatic zone (Figure 3.1).
The IDF was characterised by stands of closed- and open-canopy Douglas-fir
(.Pseudotsuga menziesii). At higher elevations within the IDF (>1000 m above sea level) or
where crown fires had occurred in the past, Lodgepole Pine (Pinus contorta) was common.
At lower elevations (-850 m above sea level) grassland communities and wetland
communities were dispersed throughout the forested landscape. There were localised stands
of hybrid White Spruce (Picea engelmannii x glauca) and Trembling Aspen (Populus
tremuloides) throughout the IDF. Even-aged lodgepole pine stands dominated the SBPS
landscape, as large-scale fires historically occurred frequently. Moist sites were dominated
by White Spruce (Picea glauca). Wetlands were also common throughout this zone.
Livestock grazing, primarily in the IDF, but also the SBPS, and forest harvesting were
predominant anthropogenic disturbances and insect outbreaks continued to influence stand
dynamics of both biogeoclimatic zones. The mean annual temperature is 4.2°C (range = -1.3
to 9.6 °C; Environment Canada, 2002).

40

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

G ravel R o ad

km

Main R oad

Figure 3.1. The distribution of sites within the Williams Lake Study Area in central British
Columbia, Canada (approximately 30km south of Williams Lake). Multiple plots were
located within each site. Development sites (▲ and ■) included 243 plots used for 2001-2004
model construction. In the summer of 2004, we established an additional 123 plots at spatial
validation sites (•) and continued to monitor 91 plots at temporal validation sites ( A ) to
assess the prediction accuracy of these models with independent data. The shaded area of the
map indicates the Sub-Boreal Pine-Spruce (SBPS) biogeoclimatic zone, the unshaded area is
Interior Douglas-fir (IDF).

41

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

Initial plot layout and stratification were accomplished using a Geographic
Information System (GIS) with forest cover and orthophoto layers. We established 243 plots
within 107 stands across a range of structural variation within the IDF in 2001-2002 (Figure
3.2). We excluded plots from our final analysis if industrial activity resulted in alteration of
forest structure (e.g., complete removal of trees or thinning; n = 15). Therefore, the final
sample included 228 plots within 97 stands. Stands were defined based on forest inventory
polygons of homogeneous vegetation characteristics as delineated on forest inventory maps
(Tolko Ltd., Vernon, BC).
In 2004, we established an additional 123 plots within the IDF (n = 32) and SBPS
{n = 91) for collection of spatial validation data. Twenty-three of the plots in the SBPS were
very close to the IDF boundary and had a large component of Douglas-fir; therefore, the new
plots were distributed along a gradient of Douglas-fir dominance through the transition
between IDF into SBPS zones. In 2004, we also continued visiting a subset of the plots
established in 2001-2002 for temporal validation of models (n = 90). All plots were spaced at
least 150 m apart, but occasionally 300 m separated plots where roads, landings, or water
interrupted the transect.

Habitat measurements
We measured local habitat variables using a combination of methods. Shrub species
and cover, canopy gaps, and coarse-woody debris were measured along the intercept of 2,
48-m transects laid perpendicular through plot centre (Figure 3.3). For coarse-woody debris,
we recorded the diameter of the piece perpendicular to where it crossed the axis, the tree
species (if possible), decay class (Maser et al., 1979), and any signs of use by wildlife. At 5,
2-m radius plots located at 11.28 m away from plot centre on each axis and at plot centre,

42

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

300

0

300m

Figure 3.2. Example of plot (A) layout within the Williams Lake Study Area. Most plots
were connected with a 150-m transect, although some were connected with a 300-m transect.
The boundary for Vegetation Resource Inventory (VRI) polygons that were used to define
forest stands are highlighted with white.

43

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

2-m radius subplots: percent
cover in layers from ground to
canopy.

12-m extensions for
coarse-woody debris
>30 cm diameter

Plot Centre: 4 densiometer
readings; general plot
attributes

24-m transects used for
coarse-woody debris
(>7.5 cm), shrub intercept
and canopy gap
distribution
5.64-m radius circle: trees and
snags >7.5 cm diameter at breast
height (dbh); number o f trees
<7.5 cm dbh but > 1.3 m tall;
estimate o f canopy coverage

11.28-m radius plot: live trees
>30 cm diameter at breast height
(dbh); snags >7.5 cm dbh

Figure 3.3. Schematic of the vegetation sampling scheme used to assess the vegetation and
other structures present in each of the sampling plots.

44

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

we measured the percent coverage for litter, coarse-woody debris, herb species, moss and
lichens, shrub species, sapling species, bare ground, and rock. Within a 5.64-m radius plot
around plot centre, we tallied all trees and stumps >7 cm diameter at breast height (dbh) and
recorded tree species, dbh, and height. We tallied all trees >30-cm dbh and snags within an
11.28-m radius. We recorded general information for each plot including canopy closure
around plot centre, as an average of 4 measurements taken on each axis (Robert E. Lemmon
model C densiometer; Bartlesville, Oklahoma), aspect, slope, canopy stratification and
complexity, disturbance history (evidence of fire, grazing, logging), and the elevation (in m)
above sea level. We averaged vegetation data collected at multiple plots measured within
stands, as defined by forest inventory polygons, to create stand-level variables.
Because the study area is currently undergoing a major mountain pine beetle
(Dendroctous ponderosae) outbreak, we also included beetle presence as a potential predictor
variable for woodpecker species, Black-capped Chickadee (Poecile atricapillus), and
Townsend’s Warbler (Dendroica townsendi). These data were derived from our plot surveys
as well as provincial aerial forest health surveys (Province of British Columbia, 2000). Using
the ArcView GIS (version 3.2a; ESRI, 2000) extension Patch Analyst 3.1 (Rempel and Carr,
2003), we created a 75-m buffer around plot centres and intersected this layer with the
provincial data layer that contained the insect pest distribution data. In cases where provincial
maps and our plot data disagreed, we reviewed transect data and helicopter survey data
provided to us by the forest company tenured in the study area (Tolko, Ltd., Vernon, BC) to
resolve this discrepancy.
To obtain landscape-level attributes, we subsetted a Landsat 7 (30-m resolution)
image of the study area (July 2002) and used PCI Works GIS software (version 7.0; PCI

45

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Geomatics Corp., 2001) to complete a supervised maximum-likelihood classification. Six
habitat classes were identified: water, nonforest, early serai, shrub and Aspen, moderate
retention conifer, and conifer. We used colour airphotos, orthophotos, and the vegetation data
collected at plots to seed areas for training and to assess the accuracy of the classification.
We assessed accuracy of the classification by determining the number of correctly classified
pixels from a randomly selected subset. Water, nonforest, and conifer classes had the highest
accuracy (97%, 94%, and 88%, respectively). Early serai, shrub and Aspen, and moderate
retention conifer were less often classified correctly (67%, 78%, 67%, respectively). Overall
classification accuracy was 81%.
We calculated landscape metrics using the Patch Analyst 3.1 (Grid) extension
(Rempel and Carr, 2003) for ArcView GIS (version 3.2a; ESRI, 2000) to interface to the PC
raster version of FRAGSTATS 2 (McGarigal and Marks, 1995). We intersected the
classification layer with a buffer created around plot centres at 3 scales: 2 ha, 50 ha, and 300
ha. We selected these extents based on available information of species home-range size
(Gillingham, 2003). We determined centroids for each stand polygon and repeated this
process with 50- and 300-ha buffers around each centroid. Distance from plot centres and
polygon centroids to water, high-contrast edge (e.g., meadow - forest), and roads were
estimated using the GIS.

Vertebrate sampling
For each vertebrate observation, we used a handheld Global Positioning System (GPS;
Garmin eTrex GPS, Olathe, Kansas) to obtain coordinates or used known plot coordinates to
import detections into a GIS (ESRI, 2000). If necessary, detections were corrected with an
estimated bearing and distance to the individual. To ensure that observations were associated

46

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited without perm ission.

with the correct plot- and stand-structural characteristics, we calculated distance to nearest
plot and associated each detection to the nearest plot (<75 m) and to the stand polygon. We
used a variety of techniques to determine the presence of vertebrate species.

Point counts
We conducted variable-radius point counts at each of the plots (Reynolds et al., 1980;
Ralph et al., 1995) 3 times throughout the breeding season (mid-May - early July) in each
year. Point counts began no earlier than 30 min before sunrise and were ceased no later than
4 h after sunrise when there was little or no wind or rain and temperature was at least 3°C
(Robbins, 1981; Province of British Columbia, 1999b). A minimum of 10 days passed
between each successive visit. Observers were rotated between plots (1 visit per observer per
plot per year) and direction of travel along the transect was changed between surveys to
reduce bias associated with the observer and time of day. We recorded all birds detected
during a 6-min recording period that began after a 1-min settling period. Distance and
direction to bird from plot centre was estimated and recorded as either within 50 m, 50-75 m,
or outside 75 m. Training was conducted prior to the onset of surveys and opportunistically
throughout the field season to calibrate distance estimates among observers.

Woodpecker surveys
We conducted woodpecker playback surveys from mid-May until mid-June at plots a
minimum of 300 m apart (usually every other plot), 3 times per year. We began surveys no
earlier than 30 min before sunrise and ended no later than 1100 h, in appropriate weather
conditions as for point counts. A pre-recorded cassette tape of the calls and drumming of the
7 species expected to occur in the study area was broadcasted starting from the smallest

47

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited without perm ission.

species and ending with the largest species (Downy Woodpecker, Picoides pubescens; Rednaped Sapsucker, Sphyrapicus nuchal is', Hairy Woodpecker, Picoides villosus; Three-toed
Woodpecker, Picoides tridactylus\ Black-backed Woodpecker, Picoides arcticus\ Northern
Flicker, Colaptes auratus\ Pileated Woodpecker, Dryocopus pileatus) in an attempt to invoke
woodpecker responses (Johnson et al., 1981). Passive sampling was also conducted at each
plot in conjunction with point-count surveys. When point counts were conducted in
conjunction with playbacks, playbacks were always conducted after the passive listening
period of the point count was finished.

Encounter transects and time-constrained searches
We conducted encounter-transect surveys with unlimited width along transects
between plots throughout each field season and in 2 winters (see below). If a species was
detected along the transect, a GPS waypoint was recorded with species ID, distance and
bearing from waypoint, as well as a sign code (singing, call, visual, den/nest, track, remains,
browse, or feces).
In 2001, we conducted time-constrained searches for amphibians and reptiles at both
plots and along transects. These searches involved lifting cover and replacing cover objects
with a restriction on the amount of time spent searching (Scott, 1994). We also searched
ponds and riparian areas for presence of amphibians (all life stages). These searches
produced relatively few detections for the effort spent; therefore, in 2002 - 2004 we
incorporated search methods (e.g., lifting cover objects) into encounter-transect surveys.
Auditory detections for vocal amphibian species were also recorded opportunistically during
owl playback and surveys near marshes at dusk.

48

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout p erm ission .

Owl surveys
We established several roadside-calling stations for owl-playback surveys at locations
that maximised the coverage of stands where we had established our plots. Calls for each owl
species expected to occur in the study area were broadcasted 3 times, progressing from
smallest species to the largest species (Northern Pygmy Owl, Glaucidium gnoma\ Northern
Saw-whet Owl, Aegolius acadicus; Boreal Owl, Aegolius funereus; Short-eared Owl, Asio
flammeus\ Long-eared Owl, Asio otus ; Barred Owl, Strix varia\ Great-homed Owl, Bubo
virginianus\ and Great Grey Owl, Strix nebulosa), using a prerecorded CD and megaphone.
As with woodpecker playbacks, the broadcasted call is intended to invoke a vocal response
or attract an individual (Fuller and Mosher, 1981). Surveys were conducted from late April
until late May, as well as on visits during winter-tracking surveys (see below). Passive owl
detections were recorded throughout the season. These data were entered into the species
database as were coordinates obtained during encounter-transect surveys.

Remote cameras
We used 7 TrailMaster TM1000 (Goodson & Associates, Inc. Lenexa, KS) activeinfrared monitoring systems and 3 TrailMaster TM550 passive-infrared systems to record the
presence of medium and large vertebrates in our plots. The 2-piece active-infrared trail
monitor used an infrared beam across the trail between the transmitter and receiver (30-m
range). When the beam was broken for the specified length of time (0.25 s), a camera was
triggered to photograph the area. All events were logged on the receiver and photographs
were indexed to specific times. The passive-infrared trail monitor was a single unit that
detected the combination of heat-and-motion in the area it was monitoring. The area of
sensitivity formed a wedge radiating outward in front of the monitor. We constrained the

49

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

wedge, which was 20-m deep and spread 150° wide, with electrical tape so that it was
restricted to the field of view of the camera. Endotherms, generally squirrel-sized or larger,
that moved within this wedge registered as an event and a photograph was triggered.
Cameras were placed along game trails and at game-trail intersections, latrine sites,
across logs, and other positions where evidence of species presence existed, but specific
species was not determined (e.g., mustelid or canid scat). At or near previously documented
scat of these types, cameras were placed in locations for a minimum of 12 days, which is the
recommended minimum based on the travel behaviour of weasels through their home ranges
(Zielinski and Kucera, 1995).

Small-mammal trapping
We conducted live trapping for small terrestrial mammals in all years. Collapsiblelive traps (H. B. Sherman Traps, Inc. Tallahassee, FL) were placed at 25-m intervals along
the primary and secondary axes of all plots (4 traps; Figure 3.3) in each year. In 2002, an
additional 4 traps were placed 25 m from the plot centre in 45° increments between axes.
Each trap location was pre-baited for a minimum of 24 h prior to deploying live traps. Traps
contained a small piece of carrot to provide moisture, oats and sunflower seed for energy,
and a small wad of cotton bedding for warmth (Jones et al., 1996; Province of British
Columbia, 1998).
We trapped over 3 nights, opening the traps at dusk and checking them beginning at
dawn. In 2002, we marked all animals captured with ear tags; in other years we clipped hair
on individuals to identify recaptures. In 2002, we also conducted day trapping at
approximately 50% of the plots following the same routine, but traps were checked between
4 and 7.5 h after opening. Traps were not opened in unseasonable cool overnight weather.

50

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Bat detections
In 2002-2004, we surveyed for bats at plot centres using Anabat™ broadband bat
detectors coupled with Zcaim recording units (Titley Electronics, Ballina NSW, Australia),
which record echolocation calls directly to a flash disk. We placed Anabat™ detectors 1.3 m
off the ground, at an angle of 45° and directed north. At sites where transects between plots
could be safely navigated at night, plots were surveyed for bat activity in a transect method.
Bat activity was recorded for a minimum of 30 min at each plot before moving the detector
to a new plot along the transect. Sampling was conducted between 2000 and 0200 h on 3
consecutive nights, unless weather conditions precluded sampling (rain or strong wind). On
each successive night we switched the direction traveled along the transect so that plots were
visited at different times. During the 3-night period, 1 plot was sampled throughout the night
for the entire survey period to obtain a reference of overall nightly bat activity and variation
among nights (Hayes, 1997). Where terrain or vegetation made it difficult and dangerous to
travel at night, detectors were placed at plot centres and programmed to record data from
2000 to 0600 h for 4 consecutive nights.
We mist-netted bats to obtain reference calls for the identification of echolocation
calls recorded using discriminant function analyses (DFA; Statistica 6.0, StatSoft, Inc. 2003;
O’Farrell and Gannon, 1999). Because of similarity in echolocation calls for species
belonging to the genus Myotis (Thomas et al., 1987; Corben and O'Farrell, 1999) and the
variability by calls of big brown (Eptesicus fuscus) and silver-haired bats (Lasionycteris
noctivigans; Fenton and Bell, 1981; Thomas et al., 1987), we grouped detections into 1 of 3
groups (long-eared myotis, little brown or long-legged myotis, or big brown/silver-haired;
Table 3.1).

51

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited without perm ission.

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

Table 3.1. Echolocation frequency and species groupings for bats expected to occur in the Williams Lake Study Area.
Identification of the recorded echolocation calls of bats was based primarily on the minimum frequency and slope
characteristics. For some species we also used maximum frequency. Species were confirmed present in the study area if we
captured them in mist-nets. Because we could not record reference calls for species not captured, we could not confirm
presence and, therefore, did not assign that species to an echolocation group (NA), although characteristics of calls for some
species are distinctive.
Minimum
frequency

Species membership

Slope characteristic

Confirmed presence in
Study area by capture

Group

<30-35 Khz

Western Long-eared Myotis
(Myotis evotis)
Little Brown Myotis
(M. lucifugus)
Long-legged Myotis
(M. volans)
Yuma Myotis
(M. yumansis)
California Myotis
(M. califomicus)
Big Brown Bat
(Eptesicus fuscus)
Silver-haired Bat
(Lasionycteris noctivagans)
Hoary Bat
(Lasiurus cinereus)
Spotted Bat
(Euderma maculatum)

Steep slope with little or no
curve

Yes

1

Steep slope, usually less time
between calls

Yes

2

Yes

2

No

NA

No

NA

Yes

3

Yes

3

Shallow sweep, little change
in frequency

No

NA

Maximum frequency 12 Khz

No

NA

>35-40 Khz

>50 Khz

20-30 Khz

17-20 Khz
5-8 Khz

Steep slope, Maximum
frequency >90 Khz

Varies from sweeping to
steep

Winter tracking
We conducted snow-tracking surveys along all transect routes connecting plot centres
twice in January and March of 2002 and once through January and February 2004 at 4 areas
accessible by vehicle (plot n = 93). We considered conditions suitable for tracking after a
significant snowfall (accumulation of >5 cm), with stable temperatures, and low winds
(Beauvais and Buskirk, 1999). Poor snow conditions in 2002-2003 and 2003-2004 winters
precluded additional surveys. We began our surveys no earlier than 24 h after snowfall and
made our best attempts to cover as much of the transects in as short a time as possible. Each
track observed was identified to species and its location on the transect was recorded and
georeferenced as for detections from encounter-transects surveys. We also recorded all
species detected visually or vocally.

Intensive plot surveys and miscellaneous detections
Each plot was intensively searched to a radius of 50 m for all animal sign (visuals,
calls/songs, nests, denning, feeding, feces, prey remains); cover objects were lifted and
replaced where appropriate (Crump and Scott, 1994). We searched plots once each year
during the collection of vegetation data, which occurred primarily in the month August. As
part of a concurrent study, we conducted encounter-transect surveys and intensive-searches
around marshes and were provided with a number of detections from helicopter surveys
conducted by Ducks Unlimited Ltd.

Analysis
We converted our data detections to presence and absence, given the use of several
different methods to detect species presence (Magurran, 2003). We then used logistic

53

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

regression (Hosmer and Lemshow, 2000) to examine the relationship of species occurrence
to vegetation structure and composition. We only considered species that were detected at a
minimum of 10% of the plots or stands surveyed, but not more than 90% because of the
effect sample prevalence can have on model assessment and prediction accuracy (Manel et
al., 2001; Liu et al., 2005; Seoane et al., 2005). For each species, we selected a subset of
structural variables based on habitat associations as determined by a review of the literature
(Gillingham, 2003) and examined correlations among the variable subset and species
occurrence. If variables existed that were measures of the same characteristic (e.g., percent
Aspen cover and deciduous stems ha'1), the variable with the highest correlation coefficient
with species presence was included in the candidate model. We constructed a set of candidate
models using the final subset of structural variables for both plot presence and stand presence.
We carefully considered variable combinations to minimize data mining (Guthery et al.,
2005); one model included all the uncorrelated candidate variables (i.e., global model).
We tested a priori candidate models for multicollinearity among independent
variables by calculated variance inflation factor scores after linear regression (Neter et al.,
1985; StataCorp, 2003). We adopted a variance inflation factor of 5 or above, which
corresponds with a tolerance score of 0.2 as a cut-off for variable inclusion (Menard, 2002).
If multicollinearity was indicated, we reran our model using only one of the indicated
problem variables and compared outcomes using a log-ratio test (Menard, 2002; StataCorp,
2003). We performed Box-Tidwell transformation of the variables to determine the
relationship between the logit of predictor and response variables (Box and Tidwell, 1962;
Menard, 2002). If a non-linear relationship was identified, we considered transforming the
variable (arcsine or natural log) or removing the variable(s) from the candidate list. Variables

54

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

were removed if there was little contribution to the model or if no interpretable
transformation rectified the problem. Therefore, for each of the 55 species for which we had
sufficient data, we had a set of several candidate models including different combinations of
relevant structural variables that were not correlated at both stand and plot scales (Table 3.2)'.
We ranked competing a priori models using Akaike’s Information Criteria (AIC)
with a correction to account for small sample sizes (AICc; Akaike, 1973; Burnham and
Anderson, 2002). AlC-model selection estimates the information loss when the probability
distribution with the true model is approximated by the probability distribution associated
with the model that is to be evaluated. Choosing the model with the lowest expected
information loss between the true model and the approximating model is asymptotically
equivalent to choosing a model that has the lowest AIC value (Burnham and Anderson,
2002). We determined the Akaike weight (w,), the likelihood of the model given the data
(Burnham and Anderson, 2002), for all models within a competing set. Models were
eliminated if the variables were complete subsets of the highest ranked model and there was
little change in the maximized log-likelihood (Burnham and Anderson, 2002) or the area
under the receiver-operating characteristic curve (ROC) value was <0.70 (i.e., poor; Manel et
al., 2001; see below). We examined the model set with AAIC <2.0, as models within this
range are considered to be equally plausible (Burnham and Anderson, 2002), to determine
the best model for validation. Given our restricted criteria for use of variables in model
development, we used this cut-off as a minimum limit because the AIC method ranks all
models in the candidate list and will identify a “best model” even if all proposed models are

1 The literature review used to derive candidate models, variable lists, and full model results can be found in
Appendix I (electronic).

55

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Table 3.2. Example of the candidate model set to describe the occurrence of Red-naped
Sapsucker (Sphyrapicus nuchali) within the Williams Lake Study Area at the plot level. Full
lists of the candidate variables used for 55 vertebrate species are included in Appendix I
(electronic).
Model ID

Model

1

Percent Gap, Distance to edge, Proportion shrub and Aspen (50 ha),
Stumps ha'1

2

Percent Gap, Distance to edge, Percent Aspen, Stumps ha'1

3

Percent Gap, Distance to edge, Proportion shrub and Aspen (50 ha)

4

Distance to edge, Proportion shrub and Aspen (50 ha)

5

Distance to edge, Proportion shrub and Aspen (50 ha), Stumps ha'1

6

Percent Gap, Percent Aspen, Distance to edge

7

Percent Gap, Percent Aspen, Distance to edge, Stumps ha'1

8

Percent Aspen, Distance to edge, Stumps ha'1

9

Distance to edge, Stumps ha"1

10

Percent Gap, Distance to edge

11

Percent Gap, Distance to edge, Stumps h a 1

12

Percent Aspen, Distance to edge

13

Percent Gap, Proportion shrub and Aspen (50 ha)

14

Percent Gap, Proportion shrub and Aspen (50 ha), Stumps h a 1

15

Proportion shrub and Aspen (50 ha), Stumps ha'1

16

Proportion shrub and Aspen (50 ha)

17

Percent Gap, Percent Aspen, Stumps ha'1

18

Percent Gap, Percent Aspen

19

Percent Aspen, Stumps ha'1

20

Percent Gap, Stumps ha"1

21

Percent Aspen

56

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

“bad” (Guthery et al., 2005). If multiple models remained in the final model set, we
considered them to be competing best models and validated each model. After the final
model(s) were selected, we reviewed cases for high leverage and studentised residual values
to determine if any cases were disproportionately driving the model relationship (Menard,
2002).

Model assessment
We examined two aspects of prediction accuracy, discrimination and reliability, for
spatial and temporal validation. Discrimination refers to a model’s capacity to correctly
classify occupied and unoccupied sites, whereas reliability measures the agreement between
predicted probabilities of occurrence and observed sites occupied (Pearce and Ferrier, 2000).
When presence and absence data are available, calculating the area under the receiveroperating characteristic curve (ROC) is a favoured measure used to assess the discrimination
ability of a logistic model (Fielding and Bell, 1997; Pearce and Ferrier, 2000). The ROC
value is calculated by plotting the number of sites where presence is correctly predicted
divided by the total number of positive sites (sensitivity), against the fraction of incorrect
cases where presence is predicted (1 - specificity) across available thresholds (Deleo, 1993).
The area under the resulting curve is an estimate of predictive accuracy not based on
threshold probabilities (i.e., p > 0.5 designated as presence; Deleo, 1993; Fielding and Bell,
1997) or species prevalence (i.e., one outcome greatly outnumbers the other; Manel et al.,
2001). A ROC value is interpreted as the percentage of time that a random selection from the
positive class will have a higher predictive score than a randomly drawn case from the
negative class (Deleo, 1993). ROC values of 0.5 indicate the explanatory variables do not
improve discrimination beyond random assignment and 1.0 indicates perfect discrimination.

57

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

A value below 0.5 indicates the model performs more poorly with the explanatory variables
than without them. We classified ROC values between 0.5 - 0.7 as low, 0.7 - 0.9 as good and
>0.9 as high model prediction accuracy (Manel et al., 2001).
To determine model reliability (i.e., the agreement between predicted probabilities of
occurrence and observed occurrence), we examined calibration plots (Cox, 1958; Miller et
al., 1991; Pearce and Ferrier, 2000). Calibration plots are calculated by plotting the median
probability for predictions, divided into 10 equal classes (x-axis), against the proportion of
occupied sites within each class (y-axis; Pearce and Ferrier, 2000). The expected distribution
is equivalent to a slope of 1 through the origin, where the proportion of occurrences equals
the median for each class. In a reliable model, observed proportion of occupied sites equals
the median predicted value and thus regression of the points results in a slope of 1 (Pearce
and Ferrier, 2000) in the calibration plot. We considered models reliable, therefore, if
regression slopes were not significantly different from 1.

Prediction accuracy bias
To determine whether sample prevalence, species traits, or detection uncertainty
systematically affected prediction accuracy, we examined correlations between sample
prevalence (%) and ROC values for all models and used ANOVA and f-tests to test for
differences in ROC values among groups. We tested ROC values within groups for
normality, as well as for homogeneity of variance. If a significant difference was indicated by
ANOVA, we conducted post-hoc analysis using Tukey’s HSD test (Zar, 1999).
We classified migration strategies as resident, short-distance migrants, and
neotropical migrants; territory size as small (<10 ha), medium (10 - 50 ha), and large (>50
ha); and mobility as limited (terrestrial, small-bodied), moderate (terrestrial, medium- or

58

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

large-bodied), and high (volant). Spatial uncertainty was classed as either <10 m from plot
centre, between 10 m and 75 m, or >75 m from plot centre. This coincides with error
associated with obtaining a GPS waypoint (approximately 10 m), our general vegetation
measures at the plot level and stand level, as well as the largest distance category for point
counts (>75 m). Species that typically call while in flight or have calls that travel long
distances (e.g., Sandhill Crane, Grus canadensis) were given a rank of 3 (i.e., >75 m) and
species primarily detected by visual detections or sign were assigned 1 (<10 m; e.g., southern
Red-backed Vole, Clethrionomys gapperi). We classified most songbirds as a 2 because
these species are primarily detected by song or calls, but do not have great audio range.
Temporal uncertainty was classified as none (i.e., visual or vocal detections that are in real
time) or possible (e.g., tracks or scat that can persist for extended periods). Likewise,
identification uncertainty was classified as low, for species very unlikely to be misidentified,
or possible, for species with tracks, calls, or appearances close to that of other species. We
did not include detections of species where observers noted unconfirmed identification;
therefore, identification uncertainty is a substitute for potential observer error. We used the
program STATA (version 8.2; StataCorp, 2003) for all statistical analyses.

Results

Species detection
We recorded 38,389 observations for 191 species (Table 3.3) from May 2001 to the
beginning of January 2004. During the summer of 2004, we recorded 9,419 observations for
148 species at temporal and spatial validation plots. Two-hundred species were observed in
all years (Appendix II; annual detections by species are located in the electronic Appendix I).

59

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

Table 3.3. Number of vertebrate species, summarized by order, detected by different monitoring methods during model development
phase (2001-2004) in the Williams Lake Study Area. The total number of detections are presented in brackets below the number of
species. Detections by species and by year are included in Appendix I (electronic).
Reptilia
6

Amphibia
6

Aves
204

Mammalia
64

Total Species
280

Total Species
Detected

2

3

155

39

200

Point Count

0

0

112
(21,424)

7
(492)

119

Remote Camera

0

0

3
(5)

10
(171)

13

Helicopter

0

0

21
(268)

2
(3)

23

Marsh Survey

1
(1)

3
(66)

89
(1,231)

13
(95)

106

Plot Survey

0

2
(27)

87
(1,974)

17
(828)

106

Mist-netting

0

0

0

5
(37)

5

Owl Playback

0

0

4
(29)

0

4

Species Included in
Literature Review

ON

o

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

Table 3.3. Detections by taxa and method (continued).

Reptilia
0

Amphibia
0

Aves
31
(516)

Mammalia
1
(5)

Total Species
32

Anabat

0

0

0

31
(563)

3

Track Plate

0

1
(4)

0

32
(324)

3

Encounter Transect

2
(4)

4
(103)

91
(1,134)

24
(5,824)

121

Small-mammal Trap

0

1
(1)

0

12
(2,668)

13

Woodpecker
Playback

os

1 Confirmed species identification; other detections placed into echolocation frequency groups,
2 Confirmed species identification; others grouped into Sorex spp. or M icrotus spp.

Point counts yielded the highest number of individual detections and encounter transects
resulted in the highest number of species detected from a variety of taxa (Table 3.3).
Intensive-plot surveys and marsh surveys also resulted in detections for a large number of
species from a variety of taxa. In 2002, the capture rate at each trap for Southern Red-backed
Vole was much greater than any other year (2001 = 53.4%, 2002 = 87.2%, 2003 = 44.7%),
possibly indicating a population high (Cheveau et al., 2004). Further, this was the only year
that we trapped in plots using 8 traps per plot and conducted a second round of day-time
trapping. During population peaks, dispersal to lower quality habitats is more likely because
of density-dependent processes such as competition (Holt, 1987). Because our use of
presence and absence data could obscure relationships with preferred structures, we did not
include capture data from 2002 to model the probability of Southern Red-back Vole
occurrence.
We had sufficient data to construct species-habitat models for 55 species at the plot
and/or stand level, 27.5% of the species detected (Table 3.4). Ten species were not modelled
because of high abundance (i.e., detected at greater than 90% of the plots; Table 3.5).
Because some species could not be modelled at both the plot and stand level because of low
(<10%) or high (>90%) prevalence (Liu et al., 2005; Vaughan and Ormerod, 2005), we
examined 101 model sets in total. We detected 7 of 32 (21.9%) species listed as management
concern (e.g., threatened status); only the Sandhill Crane was detected often enough to model.
We present an example of how we interpreted model results and then provide a
summary of all models. The Red-naped Sapsucker (Sphyrapicus nuchalis) is associated with
deciduous forests, frequently adjacent to water and other edges. We were able to validate

62

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Table 3.4. The 55 species detected in the Williams Lake Study Area that met sample
prevalence criteria (>10% and <90%) from 2001-2004, with the exception of Southern Redbacked Vole (see text for details, page 62). Prevalence is reported as a percentage of all plots
(n = 228) and stands (n = 97), which were not altered by harvesting through the course of
data collection, where presence was detected. Latin names are available in Appendix n.

Class

Species
Western Toad

Aves

Alder Flycatcher
American Crow
American Redstart
American Robin
Barred Owl
Black-backed Woodpecker
Black-capped Chickadee
Brown Creeper
Brown-headed Cowbird
Cassin’s Vireo
Clay-colored Sparrow
Cedar Waxwing
Common Raven
Common Yellowthroat
Downy Woodpecker
Dusky Flycatcher
Golden-crowned Kinglet
Gray Jay
Hammond’s Flycatcher
Hairy Woodpecker
Hermit Thrush
Least Flycatcher
Mountain Chickadee
Northern Flicker
Northern Waterthrush
Northern Saw-whet Owl
Olive-sided Flycatcher
Orange-crowned Warbler
Pileated Woodpecker
Red Crossbill
Red-naped Sapsucker
Ruffed Grouse
Sandhill Crane
Song Sparrow
Spruce Grouse
Townsend’s Solitaire
Townsend’s Warbler

Plot
Prevalence
%
13.2

Stand
Prevalence
%
37.1

20.6
9.7
12.7
82.9
3.1
8.8
78.5
49.6
32.5
84.2
7.9
5.3
28.1
15.8
10.1
27.6
77.6
48.3
34.2
34.2
63.2
21.9
88.6
50.9
16.7
0.9
42.5
63.2
43.4
33.8
39.5
32.9
16.2
12.7
7.5
40.8
18.9

35.1
24.7
17.5
95.9
18.6
21.7
89.7
67.0
47.4
85.1
17.5
11.3
75.3
29.9
26.8
51.6
89.7
75.3
53.6
58.8
77.3
36.1
93.4
81.4
24.7
13.4
69.1
79.3
79.4
57.7
62.9
66.0
57.7
25.8
17.5
56.7
29.9

63

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Table 3.4. Species modelled and sample prevalence (continued).

Class

Species
Tree Swallow
Three-toed Woodpecker
Warbling Vireo
Wilson’s Warbler
Winter Wren
Western Wood-Pewee
Yellow Warbler

Mammalia Deer Mouse
Southern Red-backed Vole
Yellow-pine Chipmunk
Snowshoe Hare
Ermine
Long-tailed Weasel
Moose
Black Bear
Coyote
Lynx

Plot
Prevalence
%
14.0
9.7
52.6
31.1
14.5
35.1
26.3

Stand
Prevalence
%
29.9
22.7
70.1
57.7
26.8
53.6
48.5

50.9
70.2
32.0
68.0
9.7
13.6
76.8
55.3
22.4
10.1

70.1
83.3
43.3
77.3
19.6
23.7
85.6
78.3
49.5
20.6

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Table 3.5. Ten species detected in the Williams Lake Study Area, from 2001-2004, that were
not modelled to predict occurrence because of high prevalence at the plot and stand scale (i.e.,
presence was detected at >90% of plots). Latin names can be found in Appendix II.
Taxa
Aves

Mammalia

Species
Chipping Sparrow
Dark-eyed Junco
Evening Grosbeak
Pine Siskin
Red-breasted Nuthatch
Ruby-crowned Kinglet
Swainson's Thrush
Yellow-rumped Warbler
Mule Deer
Red Squirrel

Total
detections
all years
1650
2403
1209
863
1868
2021
1959
1863
2255
2449

65

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

models to predict presence and absence of Red-naped Sapsucker for both plot and stand data.
At the plot level, percent canopy gap (P = 1.251; Table 3.6), distance to forest edge (P =
-0.007; Table 3.6), and proportion shrub and Aspen landscape class in the surrounding 50 ha
(P = 3.413; Table 3.6) were included in the best model. The relationship of all variables was
consistent across years with the exception of percent canopy gap in 2004 (P = -1.236; Table
3.6). Predictive accuracy varied, but was good in 2002 and 2004 (ROC = 0.692 - 0.773;
Table 3.6). Likewise, the predictive accuracy of this model with spatially independent data
was good (ROC = 0.737; Table 3.6). The stand model for Red-naped Sapsuckers included
percent Aspen (P = 0.041; Table 3.6) and edge density in the surrounding 50-ha area (P =
0.012; Table 3.6). The relationship with edge density was consistently positive, but
relationship with percent Aspen varied among years. Predictive accuracy was good for the
spatial validation model (ROC = 0.752; Table 3.6) and in all years except 2001 (2001 ROC =
0.659; Table 3.6). Calibration plots indicated 2001-2004 models were well calibrated, but the
spatial validation models overestimated the predicted probability of occurrence (Figure 3.4).
Fifty-seven of the 101 models (56.4%) had good or excellent predictive accuracy.
Stand models performed best, with 60.4% having good or better predictive accuracy (i.e.,
ROC > 0.70) compared to 52.1% of plot models (Table 3.7). Of the 38 variables used in all
plot models, the most often included were: percent canopy gap (n = 19), distance to edge (n =
19), distance to water (n = 14), edge density in the surrounding 50 ha (n = 10), deciduous
stems ha'1 (n = 7), large tree basal area (n = 6), dead tree basal area (n = 5), and percent
Douglas-fir (n = 5). Common variables within stand models included percent canopy gap (n
= 25), distance to edge (n = 16), edge density in the surrounding 50 ha (n = 12), percent
Aspen (n = 12), percent Douglas-fir (n = 11), and main canopy height (n = 10; Table 3.8).

66

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Table 3.6. Example of model results to predict the occurrence of Red-naped Sapsucker
(Sphyrapicus nuchalis) in the Williams Lake Study Area. Variables, coefficients, and
receiver-operating characteristic curve (ROC) values are presented for the best plot and stand
models and their temporal and spatial validation counterparts. All model results for 55
vertebrate species can be found in the electronic Appendix I.

P-coefficient

Spatial ROC

9
00120012004
1.251

2001

2002

2003

2004

2.096

0.456

0.558

-1.236

-0.007

-0.005

-0.013

-0.007

-0.011

3.413

1.195

2.995

2.058

5.501

-1.398

-3.441

-1.713

-1.325

-1.232

ROC

0.753

0.692

0.748

0.692

0.773

Percent Aspen

0.041

-0.037

0.080

-0.006

-0.072

Edge density
(50 ha)
Constant

0.012

0.007

0.014

0.014

0.006

-2.371

-2.731

-4.421

-3.188

-1.574

ROC

0.716

0.659

0.761

0.726

0.702

Model

Variable

“Plot

Percent canopy
gap
Distance to forest
edge
Proportion Aspen
and shrub (50
ha)
Constant

20012004
0.753

0.737

0.716

0.752

2004

67

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

1. 0 1
0

o
c

0l D
O
o
O
L_

c
o
t
o
a
o

T3
0

fc

0

w

-Q

o

Predicted Probability of Occurrence

Figure 3.4. Example of a calibration plot for the Red-naped Sapsucker (Sphyrapicus nuchalis)
plot-level model. The expected distribution of a well-calibrated model has a slope (P) of 1(—)
to describe the relationship between observed proportion of occurrence and predicted
probability of occurrence. The slope of the 2001-04 model was close to 1 (closed circles, — ,
P = 0.915, p = 0.004). The 2004 spatial validation data had a slope of less than 1 (open
circles,
, P = 0.889, p = 0.146) indicating that the model is misspecified for the spatially
independent data.

68

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Table 3.7. Summary of area under the receiver-operating characteristic curve (ROC) values
for each of the 55 species detected in the Williams Lake Study Area. Each species modelled
met our detection prevalence criteria from 2001-2004 at the plot and / or stand scale. ROCs
are only reported for the top model with the lowest AICc.Not all species could be modelled at
both the plot and stand scale (-).

Class
Amphibia
Aves

Plot
Stand
Species___________________ ROC________ROC
Western Toad
0.749
0.733
Alder Flycatcher
American Crow
American Redstart
American Robin
Barred Owl
Black-backed Woodpecker
Black-capped Chickadee
Brown Creeper
Brown-headed Cowbird
Cassin’s Vireo
Clay-colored Sparrow
Cedar Waxwing
Common Raven
Common Yellowthroat
Downy Woodpecker
Dusky Flycatcher
Golden-crowned Kinglet
Gray Jay
Hammond’s Flycatcher
Hairy Woodpecker
Hermit Thmsh
Least Flycatcher
Mountain Chickadee
Northern Flicker
Northern Waterthrush
Northern Saw-whet Owl
Olive-sided Flycatcher
Orange-crowned Warbler
Pileated Woodpecker
Red Crossbill
Red-naped Sapsucker
Ruffed Grouse
Sandhill Crane
Song Sparrow
Spmce Grouse
Townsend’s Solitaire
Townsend’s Warbler
Tree Swallow
Three-toed Woodpecker

0.629
0.589
0.629
0.710
0.850
0.752
0.749
0.681
0.617
0.769
0.706
0.637
0.776
0.674
0.597
0.671
0.712
0.782
0.712
0.660
0.797
0.734
0.697
0.624
0.573
0.753
0.619
0.673
0.848
0.595
0.786
0.619
-

0.683
0.611
0.680
0.780
0.542
0.681
0.757
0.704
0.867
0.637
0.589
0.550
0.721
0.723
0.574
0.792
0.804
0.671
0.703
0.758
0.719
0.755
0.803
0.629
0.863
0.796
0.707
0.633
0.716
0.627
0.714
0.851
0.810
0.718
0.759
0.682
0.733

69

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Table 3.7. Area under the receiver-operating characteristic curve (ROC) values for species
modelled (continued).

Class

Species
Warbling Vireo
Wilson’s Warbler
Winter Wren
Western Wood-Pewee
Yellow Warbler

Mammalia Deer Mouse
Southern Red-backed Vole
Yellow-pine Chipmunk
Snowshoe Hare
Ermine
Long-tailed Weasel
Moose
Black Bear
Coyote
Lynx

Plot
ROC
0.753
0.648
0.714
0.791
0.587

Stand
ROC
0.698
0.610
0.787
0.683
0.656

0.660
0.635
0.733
0.884

0.686
0.789
0.701
0.839
0.780
0.629
0.738
0.719
0.668
0.743

-

0.689
0.675
0.709
0.769
0.821

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Table 3.8. Most common variables found in the top models used to predict occurrence of
vertebrate species in the Williams Lake Study Area. The direction of the species’ association
with the variable (selection with positive coefficient or avoidance with negative coefficient)
is also reported. Structure data were derived from measures taken at each plot, as determined
from a GIS, or as calculated using a classified landscape image and FRAGSTATS
(McGarigal and Marks, 1995).
Model
level
Plot

Stand

Percent canopy gap
Distance to edge
Distance to water
Edge density 50 ha
Deciduous stems ha'1
Large tree basal area
Dead tree basal area
Percent Douglas-fir

Plot measures
GIS
GIS
Landscape
Plot measures
Plot measures
Plot measures
Plot measures

# Selection
associations
10
4
2
5
6
5
3
5

Percent canopy gap
Distance to edge
Edge density 50 ha
Percent Aspen
Percent Douglas-fir
Main canopy height

Plot measures
GIS
Landscape
Plot measures
Plot measures
Plot measures

14
5
10
10
11
7

Variable

Data Source

# Avoidance
associations
9
15
12
5
1
1
2
0
11
11
2
2
0
3

71

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Model validation and accuracy
We collected sufficient independent-presence data from spatial validation plots to
validate 43 of the 101 models. Of these spatially validated models, predictive accuracy was
good or excellent for 12 plot and 12 stand models (55.8% in total). We temporally validated
44 models with independent data; predictive accuracy was good or excellent for 12 plot and
13 stand models (56.8%). On average, predictive accuracy declined in temporally and
spatially validated models compared to the accuracy of predictions for the years from which
data were collected (Table 3.9). Generally, predictive accuracy was slightly better for
temporal data.
Our assessment of model reliability with data used to construct the model, indicated
that 60% of plot and 84.6% of stand models were well calibrated (i.e., plotting the predicted
probabilities and proportion of occurrence resulted in a slope close to 1). Model reliability
was poor, however, when assessed with spatially and temporally independent data. Only
21.1% of plot level and 3.4% of stand level models had slopes of 1 with spatial
validation data (Table 3.10; e.g., Figure 3.4). Results were slightly better for temporal
validation models where 21.1% of plot and 32.1% of stand models had slopes of 1 (Table
3.10).

Detection uncertainty and species traits
Prevalence in 2001-2004 was not related to ROC values at the plot (n = 48, r - 0.088,
p = 0.553) or stand (n = 53, r = 0.131, p = 0.117) level; however, there was a weak negative
correlation between sample prevalence and spatial (n = 43, r - -0.319p = 0.007) and
temporal (n = 44, r = -0.385, p = 0.018) ROC values in 2004 models.

72

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

Table 3.9. Summary of prediction accuracy (ROC values) for plot and stand models, and their spatial and temporal validation results.
Models were constructed to predict the occurrence of vertebrate species in the Williams Lake Study Area. Not all models could be
spatially and / or temporally validated because of insufficient detections in the new plots or for a particular year. The mean difference
in prediction accuracy between 2001-2004 models and the 2004 spatial and temporal validation model was negative for each model
set.

2001-2004

Spatial Validation

ROC > 0.70
Mean
(SE)

Model level

n

Mean
(SE)

n

Plot

48

0.704
(0.011)

25

0.764
(0.010)

Stand

53

0.712
(0.010)

32

0.758
(0.008)

ROC

Difference

Mean
(SE)

Mean
(SE)

19

0.730
(0.022)

-0.0006
(0.026)

23

0.698
(0.024)

-0.051
(0.025)

n

Temporal Validation
ROC

Difference

Mean
(SE)

Mean
(SE)

19

0.748
(0.028)

-0.006
(0.024)

25

0.726
(0.024)

-0.023
(0.026)

n

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

Table 3.10. Summary of the calibration plot regression slopes for stand and plot models (e.g., Fig. 3.4) constructed to predict species
occurrences in the Williams Lake Study Area, and their spatially and temporally validated counterparts. Regression slopes describe
the observed proportion of occupied sites versus predicted probability of occurrence for vertebrate species. Percentage of slopes that
were non-significant, near 1 (significant), and significantly different from 1 are reported. Those models with slopes near 1 can be
deemed well calibrated and, therefore, reliable.

Plot Models
2001-2004

Spatial
Validation

Temporal
Validation

2001-2004

Spatial
Validation

Temporal
Validation

n = 48

n= 19

n = 19

n = 53

n = 23

n = 25

No Slope

31.6

68.4

68.4

28.1

93.2

64.3

Slope = 1

57.9

21.1

21.1

34.4

3.4

32.1

Slope ^ 1

10.5

10.5

10.5

37.5

3.4

3.6

Slope

--4

4^

Stand Models

Prediction accuracy was not related to identification or temporal uncertainty for any
model level; however, the spatial uncertainty of detections did significantly affect ROC
values of plot models {Fqas) = 3.75, p = 0.031), but not stand models (F(2,5o>= 2.58, p =
0.086). Post-hoc analysis using Tukey’s HSD test of significance indicated that the difference
was between detections >75 m from plot centre ( x = 0.655, SE = 0.017) and those between
10 and <75 m (x = 0.717, SE = 0.014, p = 0.043). There was no relationship between
prediction accuracy of 2001-2004 models and migration strategy, territory size, or mobility
(Table 3.11).

Discussion
Structure as a surrogate measure
Use of habitat measures, such as forest structure, to model species occurrence,
abundance, or population dynamics is a fundamental aspect of both theoretical and applied
ecological science. It is well accepted that habitat heterogeneity (e.g., MacArthur and
MacArthur, 1961) and certain forest elements (e.g., coarse-woody debris, snags, large trees)
are associated with species diversity (Mazurek and Zielinski, 2004; Loehle et al., 2005).
Measuring aspects of environmental diversity, such as structural complexity, has become
preferred as a surrogate measure of diversity over species-based surrogates (Margules et ah,
2002; Faith et al., 2004), and is particularly appealing to forest-land managers because of the
ease of incorporating measures into forest inventory databases (McElhinny et al., 2005)
relative to separate monitoring programs for species. Structural diversity has been shown to
correlate with diversity of several taxa (e.g., mammals, Mac Nally et al., 2001; amphibians
and reptiles, Loehle et al., 2005; spiders, Oxbrough et al., 2005). To be effective as a

75

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Table 3.11. Summary of ANOVA results to test if species’ traits affected the accuracy of
predictions (ROC values) for the occurrence of vertebrate species. ROC values were used
from models constructed with data collected in the Williams Lake Study Area from 20012004.
Plot
Mean
F
(SE)

Grouping

n

Migration Strategy
Resident

23

0.708
(0.016)

Short-distance
migrant

13

Neotropical
migrant

Stand
Mean
F
(SE)

P

n

0.874

21

0.708
(0.015)

0.694
(0.019)

13

0.724
(0.020)

12

0.705
(0.026)

13

0.711
(0.019)

30

0.711
(0.014)

13

0.720
(0.012)

Medium (10 50 ha)

12

0.686
(0.024)

16

0.703
(0.021)

Large (>50 ha)

6

0.703
(0.033)

7

0.701
(0.028)

6

0.698
(0.040)

6

0.707
(0.012)

Moderately
mobile

5

0.733
(0.027)

7

0.727
(0.024)

Highly mobile

37

0.713
(0.012)

40

0.729
(0.031)

Territory size
Small (<10 ha)

Mobility
Limited
mobility

0.14

0.46

0.49

0.633

0.613

P

0.20

0.819

0.39

0.679

0.38

0.685

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

surrogate measure of species occurrence, however, models that use forest structural
characteristics must have a strong relationship with the probability of species occurrence
across the range of the model’s intended use (Rykiel, 1996; Lindenmayer, 1999). The
discrimination accuracy and reliability of most of the models we developed were not
satisfactory and, therefore, we conclude that forest structural characteristics, by themselves,
are not effective as surrogate measures for predicting the occurrence of a broad range of
vertebrate species or to guide management targets for structural retention or recruitment at
the scales we tested.
Of those species modelled (n = 55) only 52% of plot and 60% of stand models had
acceptable prediction accuracy (i.e., ROC >0.70; Table 3.9). Because we were specifically
looking for relationships with structural elements, misspecification of the models (Burnham
and Anderson, 2002) may explain the overall poor performance. Because the goal of our
study was to examine the possible use of structure as a surrogate measure, the selection of
habitat variables was narrow by necessity. Most of the species we modelled, however, are
found across broad ranges (e.g., continental) and thus likely have broad niche breadths
indicating flexibility in habitat selection and less dependence on specific habitat elements.
For several of the species that we modelled, however, discrimination (Table 3.4) and
reliability were relatively high. For those species, structure may act as a useful surrogate, but
as a part of a complementary program to ensure monitoring of a broad range of taxa
(Lindenmayer and Franklin, 2002).

Variables, scale and spatial context
The complexity among species associations with structure, the scale and spatial
relationship with structure, and the landscape context make generalisations of models and

77

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

structural variables difficult. The variables most often included in plot models (i.e., distance
to water, large tree basal area, and dead tree basal area; Table 3.8) were included less often in
stand models suggesting that the influence of these structures on species presence is localized.
Forested areas near riparian zones offer a unique set of characteristics and these habitats are
often diverse (Naiman and Decamps, 1997). Likewise, the influence of single large trees and
dead trees can influence diversity at a relatively small scale (Mazurek and Zielinski, 2004).
For example, nesting habitat for weak cavity excavators that require trees in specific stages
of decay to excavate their nest site (e.g., Red-breasted Nuthatch, Sitta canadensis), or
secondary cavity users that inhabit abandoned cavities may be quite flexible in the selection
of foraging habitats. The availability of a single nest tree represents a requirement that is on a
much smaller spatial scale than foraging; therefore, the availability of a single nest site may
influence species presence at the plot level, but not the larger spatial scale of the stand where
foraging may take place.
As was the case in the example for the Red-naped Sapsucker, many of the plot and
stand models included variables related to the spatial relationship of the plot or stand (e.g.,
distance to edge or water) or aspects of the surrounding landscape (e.g., edge density and
percentage forest composition; Table 3.8) indicating that landscape structure and
composition influenced species presence (e.g., Saab, 1999; Hagen and Meehan, 2002).
Distance to edge and edge density, although related, are not exactly the same variable.
Distance to edge was measured to the closest forest / non-forest boundary. Non-forest may
have been a road, clearcut, meadow, or water body. Edge density was a measure of all
contrasting landscape class boundaries within a defined area (usually 50 ha), and thus
includes less distinct edges such as the contrast between unharvested-conifer forest and

78

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

partially-harvested forest. Because different species likely perceive edges and edge contrasts
differently, it is difficult to make a generalized statement of how to treat these 2 variables;
however, because harvesting practices contribute to fragmentation of habitats, species that
avoid edges are most likely detected in areas with low edge density (i.e., continuous
landscape classes). Therefore, species that are sensitive to (avoid) edge or edge density may
require specific management planning to ensure continuous areas of forest. If structure is to
be used as a surrogate, spatial context and scale should be included in forest-inventory
databases, which is not typically done at present.
We found that canopy closure and some measure of deciduous forest cover were
commonly included in both plot and stand models (Table 3.8). There was a fairly even split
among species associated with open-canopy forests and closed-canopy forests. Canopy
closure is an easily obtained structural variable, from ground surveys and photo interpretation.
Because harvesting activities consistently open canopies, identifying species with negative
association with open canopies, and determining if there are thresholds to occupancy, could
provide forest managers with a useful target for stands across the landscape. In landscapes
dominated by coniferous-forest cover, deciduous stands and mixed-woods can be areas of
high diversity (Stelfox, 1995). Other research has documented the importance of deciduous
species to cavity nesting birds (Martin and Eadie, 1999) in the region and this research
reinforces the importance of deciduous stands, specifically Aspen, to a host of species.

Validation
More problematic than the difficulty in drawing general recommendations linking
specific structural characteristics to a broad array of species was that prediction accuracy
remained good or better for only 58% of the models when temporally and spatially validated

79

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

(Table 3.9). If models that predicted species presence well (i.e., ROC > 0.70) are used to
meet diversity-monitoring objectives, the relationships to structure must be robust with good
or better prediction accuracy across spatial and temporal scales (Rykiel, 1996). Stand-level
models performed slightly better than plot models with the data used to develop models;
however, for models that were externally validated, plot models were more likely to retain
higher predictive accuracy. Overall, stand-level modelling is probably more appropriate for a
broader range of species because of the confounding effects of small-scale variation in
structure and spatial uncertainty for detections at the plot level.
When model reliability was assessed with calibration plots, temporal validation
models performed better than the spatial validation plots, emphasizing that caution is
required if models are to be applied outside the geographic area in which they were
developed. Declines in prediction accuracy for temporal validation may be a result of yearto-year variance in prevalence and a higher probability of false negative detection during one
year of data collection (Manel et al., 2001; Tyre et al., 2003; Gu and Swihart, 2004;
McPherson et al., 2004). Further, a species’ relationship to habitat can fluctuate through time
because of various aspects of population dynamics and density (Wiens, 1989; Haila et al.,
1996; Green and Stamps, 2001).
Prediction errors may also arise from data collection error and other statistical
artefacts, species interactions, regional variability in habitat, and intraspecific variability
(Fielding and Haworth, 1995; Fielding and Bell, 1997; McPherson et al., 2004). Because we
used the same sampling methodologies throughout the course of the study and assessed
accuracy using area under the ROC curve, statistical artefacts were likely minimized
(McPherson et al. 2004). Further, there was no correlation between ROC values and

80

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

prevalence within the set minimum (>10%) and maximum (<90%), with the exception of
2004 ROC values, suggesting that this criteria was sufficient. The negative correlation
between species prevalence and ROC values in 2004 was unexpected, because prevalence
usually has a positive relationship with predictive accuracy (Manel et al., 2001; Seoane et al.,
2005). Species that were less common in the SBPS biogeoclimatic zone, however, may have
been more selective for habitats similar to those sampled in the IDF (i.e., narrower habitat
breadth) resulting in better ROC values. Conversely, species that were common in the SBPS
may have broader niche breadths and thus did not respond to basic differences in structure
between the 2 zones, reducing predictive accuracy. This result was also supported by the
reduced reliability of spatially validated models (i.e., slope <1.0). Slope departed from the
expected 45° pattern for most calibration plots when applied to independent data; therefore, it
is likely that the species we modelled respond to different structural elements in the IDF and
SBPS, or that structural characteristics of forests alone are not the best predictors of species
presence across this range of conditions. Presence for some species may, therefore, be related
more to other factors such as population legacies, stand history, or species interactions and
not characteristics of the habitat structure (Fielding and Haworth, 1995; Tyre et al., 2001).
Species traits, such as environmental specialization, regional distribution,
detectability, and body size have been linked to prediction errors in other studies (Karl et al.,
2000; Scott et al., 2002; McPherson et al., 2004; Seoane et al., 2005); however, we did not
find any differences among migratory strategy, territory size, or mobility and predictive
accuracy. Our results may contradict other studies because of the spatial extent of our study
area and the general characteristics of species modelled (Karl et al., 2000). Although we
established our plots and sampled across a range of variation in structural features, on a

81

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

regional scale we sampled a narrow range of environmental conditions. The species that were
modelled were those that met our prevalence criteria (>10% or <90%) and, therefore,
inherently may not be representative of the full range of biological traits, especially those
associated with rarity (Kunin and Gaston, 1993; Gaston and Blackburn, 1996; Davis et al.,
2000). Only spatial uncertainty at the plot level systematically affected ROC values, which
likely reflects error associated with the measures of structural characteristics and the actual
location of the individual.
Van Home (2002) argued against validation of species-habitat models because
specific models can only make predictions for specific populations at specific times and,
therefore, species presence cannot be tested with new data. Although results support Van
Home’s supposition, as most models had reduced discrimination and reliability when
validated, this premise is also potentially dangerous. Non-validated species-structure models
may be tenuously implemented at best, although structure performed poorly at predicting the
occurrence of species, even with data used to develop models. Likewise, others have
cautioned strongly against the dependence on species-habitat models in general because they
can only be considered correlative and do not indicate causal mechanisms or processes
(Morrison, 2001; Mitchell, 2005). Species-habitat models are implemented effectively as
management tools, however, in a variety of ways including habitat suitability models and
resource selection functions. The weakness of such approaches is that it is simply not
possible to measure the life-history aspects or conduct intensive studies for all species. Forest
structure may be a suitable surrogate if used only to make general guideline
recommendations and not species-specific targets, more in line with a coarse-filter approach

82

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

(Simberloff, 1999; Lindenmayer and Franklin, 2002) and not as a monitoring tool of species
occurrence.

Multi-species inventory
Conducting a multi-taxa inventory contributes to baseline occurrence data for future
monitoring inventories, identifies species for which methodologies are not sufficient to detect
presence, identifies species that may be of management concern and helps to form specific
management goals as they relate to biodiversity objectives (Manley et al., 2004). By
compiling detections for species from a variety of methods, species that may be under­
represented in samples of a single approach because of methodological biases may be
reduced. Point counts, encounter transects, and intensive-plot searches all achieved high
numbers of species detections. As expected, the greatest diversity of vertebrate species in our
study area was represented by birds, and point counts resulted in a high proportion of species
and observations. Encounter-transect surveys were effective in that we detected a breadth of
species from a variety of taxa. Overall, the multi-taxa inventory was successful in that we
detected a high proportion of the species expected to occur (71.4%).
Amphibians and reptiles had very low rates of detection (Table 3.3). We targeted
amphibians and reptiles by lifting potential cover objects during time-constrained searches,
encounter-transects surveys, and intensive-plot searches (Heyer et al., 1994). We opted not to
place artificial cover boards or arrays of drift fences and pitfall traps because we did not want
to alter the interpretation of structural aspects of stands artificially or potentially influence
non-target species (e.g., attraction or avoidance). Given that our encounter rates were so low,
however, we do not recommend our methods for successful monitoring of species from these
taxa. Because most amphibians and reptiles have reasonably limited mobility, however, if the

83

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited without perm ission.

objective was purely for monitoring purposes, using artificial structure would not affect
interpretations regarding species presence of amphibians specifically.
Our method of measuring bat activity was also not satisfactory because calls could
only be categorized into species groups (Table 3.1) and, therefore, we could not construct
single-species models. Bats are difficult species to monitor because they are volant,
nocturnal, and echolocate (Kunz, 1988), but often represent a large portion of mammalianspecies richness (up to 22% in our study area) and thus are an important component of
vertebrate diversity. Activity levels and echolocation groups, however, could be monitored
and modelled and whether or not that was sufficient would depend on the objectives of the
study (e.g., Baxter et al. in press). Further research is needed to identify the habitat
requirements for individual bat species using different methods (e.g., telemetry study) until
methods to identify species by call are refined.
Other than black bear (Ursus americanus), an omnivore, we detected few mammalian
or avian predators. Although predatory species tend to have large home ranges with naturally
lower densities and, therefore, are expected to be detected less frequently (Tyre et al., 2003;
MacKenzie, 2005), there is growing evidence of the role that predators play in increasing
diversity (Soule and Terborgh, 1999; Shurin and Allen, 2001; Hebblewhite et al., 2005;
Sergio et al., 2005). Our study area was a heavily managed, multi-use landscape (i.e.,
industrial forestry, cattle grazing, hunting and recreation uses) and has high road density,
which affects the distribution of several predators that may have been more abundant
historically (e.g., Grizzly Bear, Ursus arctos\ Grey Wolf, Canis lupus). Overall, the species
community present in the Williams Lake Study Area may not be representative of the full
range of goals for a program with the objective of conserving biodiversity, and therefore,

84

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

restoration activities may need to be considered as part of a comprehensive biodiversity
program.
Species of management concern, either because of sensitivity to disturbance, limited
distribution, or rarity, were not well represented in our sample. This result was not
unexpected (Margules and Pressey, 2000; Noon et al., 2003), but it does highlight a weakness
of the general approach of targeting a breadth of species (Manley et al., 2004). More
intensive survey methodologies are required to determine the presence and absence of rare or
cryptic species to assess predictive accuracy of structural models. Many of the rare species
and species of concern should be included in fine-filter management approaches (Hansen et
al., 1999; Hunter, 2001). Therefore, not detecting them in a broad survey should not
necessarily be a criticism, but the approach cannot be assumed to encompass these species.

Conclusion
In an industrial forest landscape, where structural aspects are changing frequently,
land managers require an understanding of how species will respond and persist within the
dynamics of the changing forest environment so that strategies can be implemented to retain
and recruit structural aspects necessary for the persistence of populations. Some of the
models that we tested had good predictive accuracy that was retained when validated and
thus have application in terms of implementation as management tools (e.g., Red-naped
Sapsucker). However, because structure did not predict the presence of the majority of
vertebrate species, there was not a single set of forest structure measures that predicted
several species. Because we did not find robust relationships that are necessary to guide
management targets for retention and recruitment of specific structures, using these models
as surrogates for species occurrence is limited. Models based on stand structural

85

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

requirements meant to meet specific targets will need further examination and testing (e.g.,
volume, coarse-woody debris, snag density, etc.). Given the variability in species’ responses,
it may be most efficient to study the response of species that appear most closely linked to
structure and sensitive to the loss of specific attributes.
In our results, modelling success varied depending on whether plot or stand data were
used and many models included variables related to spatial relationships of structures
suggesting that relationships with structure are complex across species and scales. Other
surrogate approaches have had varying success, but are generally poor predictors of diversity
at stand scales (McElhinny et al., 2005). We did not examine the relationship between
vertebrate richness and structure, but given the complexity of relationships with scale and the
different scales that species use across landscapes, this will be an important research question
to pursue before surrogate measures are adopted with confidence. In the interim, because the
preservation of structures known to be negatively affected by harvesting (e.g., dead wood,
large trees, closed canopies, continuous forests) contributes to local and landscape
heterogeneity and has been shown to affect species presence in this study and others
(Lindenmayer et al., 2000; Mazurek and Zielinski, 2004; Hunter, 2005), insuring that these
structures are present on the landscape should still be a part of programs that have goals of
maintaining biodiversity.
There is an abundance of literature on recommendations for the validation of
ecological models (e.g., Rykiel, 1996; Pearce and Ferrier, 2000; Boyce et al., 2002; Vaughan
and Ormerod, 2005), and a pervasiveness of the use of predictive models to describe the
relationships of species to their environments (see review Guisan and Zimmerman, 2000);
however, models are frequently accepted without full validation of the predictive

86

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

performance (Guthery et al., 2005). Few models in our study were reliable when applied to
the independent data; therefore, our results indicate that models cannot be assumed to be
applicable in different years or applied outside the area where the model was developed, even
when the spatial and temporal context is relatively close. Until models provide confident
predictions of multiple-species occurrence, as well as response to habitat alteration, land
managers may find that monitoring species directly is a more beneficial contribution to
biodiversity monitoring (e.g., Manley et al., 2004).

87

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Chapter 4. Using forest structure to predict the occurrence of vertebrate assemblages1
Abstract
Land-use practices can have significant impacts on biodiversity. In industrial forests,
efficiently measuring, monitoring, and mitigating impacts on biodiversity are major
challenges for forest managers. Several studies have reported on the relationship between
characteristics of forest structure and the occurrence of single species or taxonomic groups
such as birds. Few studies, however, have examined the relationship between forest structure
and species occurrence across taxonomic groups. If linkages between forest structure and
multiple species can be made, forest structure may be used as a surrogate measure to monitor
the potential effects of management on species occurrence. In this paper, we examined
whether or not species, from different taxonomic groups, could reliably be grouped together
based on their co-occurrence in space and time using non-metric multidimensional scaling
and cluster analysis. We determined species presence and absence at 225 plots over a 3-year
period (2001-2004). We used presence and absence records and structural characteristics of
forests to group similar plots. Using groupings based on species co-occurrence, we used
classification and regression tree analysis to determine if structural characteristics of forests
could be used to predict occurrence of group members. Plots could be statistically defined
using both characteristics of forest structure and species co-occurrence. There was high
variation within groups, however, suggesting the ecological significance of groupings was
weak. Although there were correlations among species groups and forest structure, prediction
of group membership using structural characteristics was poor (45.8%). The structure

1 This chapter is written in the first person plural to recognize the contribution o f others to the work. The
manuscript will be submitted with the authorship Psyllakis, J.M. and M.P. Gillingham.

88

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

variables identified with correlation analyses were also included in models used to predict
group memberships. These variables included distance to water and edges and characteristics
of larger scale surroundings (e.g., edge density). Overall, our results suggest that forest
structure by itself is a poor surrogate of species co-occurrence and that the spatial aspects of
structure were important determinants of species occurrence.

Introduction
Scientists find it difficult to quantify and monitor the effects of various activities that
alter habitat on biodiversity largely because of difficulties in measuring biodiversity (Purvis
and Hector, 2000; Sarkar and Margules, 2002). Because it is not feasible to measure all
aspects of biodiversity, or even some aspects (e.g., species richness), surrogate measures are
often used as alternatives to describe and monitor biodiversity (Noss, 1990; Sarkar and
Margules, 2002). The use of surrogate approaches is critical to biodiversity planning and
conservation across spatial and temporal scales, because information for complete species
inventories over all areas is not available (Margules and Pressey, 2000; Faith, 2003). For
example, the occurrence of certain indicator species (e.g., Mac Nally and Fleishman, 2004)
and areas with high environmental heterogeneity (e.g., Faith et al., 2004) are correlated with
increased species richness. Using environmental heterogeneity is often less expensive to
measure, monitor, and forecast change for than measuring species or a subset of species
directly. Further, information regarding habitat types can be obtained over large areas with
the aid of remote-sensing technology. Therefore, as a substitute for measuring species
richness, using indicator species or environmental heterogeneity as a surrogate for species
richness has frequently been applied to various management purposes, such as the
designation of reserves (Myers et al., 2000; Fleishman et ah, 2005).

89

R ep ro d u ced with p erm ission o f the copyright ow ner. Further reproduction prohibited without perm ission.

Species-based surrogates have been criticised because robust relationships have not
been identified consistently (Caro and O’Doherty, 1999; Margules et al., 2002). Surrogates of
environmental heterogeneity have been criticised because of unclear relationships among
species and habitat, as well as a lack of representation for species interactions (e.g.,
competition or prey availability). Testing of both species-based and habitat-based approaches
remains to be done to confirm that relationships are stable through time (Lindenmayer and
Franklin, 2002). The lack of reliable approaches, however, does not absolve land managers
from minimising impacts to diversity and many levels of government have pledged to
implement programs specifically to mitigate the effects of human activities on biodiversity
loss (e.g., United Nations Environment Programme, 1992; Province of British Columbia,
1995; Montreal Process Working Group, 1999; Canadian Council of Forest Ministers, 2003).
Habitat-based approaches are appealing in an industrial forests context, where many
stand-structure characteristics are already measured, monitored, and managed. There is
considerable interest from forest managers to link structural characteristics to species
occurrence and / or richness (Lindenmayer and Franklin, 2002; Kavanagh and Stanton, 2005;
McElhinny et al., 2005). Because forest-management activities directly alter structural
characteristics of forests, there are several advantages to developing surrogate monitoring
programs around forest structure. Structure is an aspect of habitat that managers have some
control over and various forms of vegetation databases are maintained for harvest forecasting,
standardized techniques exist for the measurement of structural characteristics, and
characteristics are described at the same spatial and temporal scales as management.
Therefore, using forest structure as a surrogate for species richness or species occurrence
may be a cost-effective and efficient alternative to species-monitoring programs (Lahde et al.

90

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

1999; Lindenmayer and Franklin, 2002; McElhinny et al., 2005). Overall, a successful
monitoring program could contribute to sustainable management of forest resources by
ensuring the conservation of species diversity through time. To accomplish this goal,
however, requires strong relationships among structure and species occurrence and the
testing of these relationships across spatial and temporal scales (Lindenmayer and Franklin,
2002; Fleishman et al., 2005).
Using forest-structure variables to predict species richness has shown promise in
different forest types (e.g., Eastern US, Loehle et al. 2005; South America, Diaz et al., 2005).
Other studies report that models using a combination of forest management and physical
properties of stands (e.g., elevation, aspect) and scale variables are better predictors of
species richness than structure type measurements alone (e.g., du Bus de Wamaffe and
Dufrene, 2004). Regardless, species richness alone does not provide detailed information for
species representation across taxa or different taxonomic groups (Magurran, 2003; Schulze et
al., 2004), therefore grouping species based on some defined aspects may provide better
inference to the effects of management activities. Species may be grouped based on:
functional groups (e.g., Cummins, 1974); environmental responses or trophic criteria (Gaines
et al., 1989); the shared use of specific resources (e.g., guilds, Root, 1967); or co-occurrence
in space and time (e.g., species assemblages, Kavanagh and Stanton, 2005). These
terminologies are often used interchangeably, although there are distinct differences and
several authors have attempted to clearly define the terms (e.g., Fauth et al., 1996; Blondel,
2003). Groups based on species assemblages may be a preferred approach because no
assumptions are made regarding the role a species plays in the ecosystem or specific types of
interactions with other species, which are often unknown.

91

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Multiple-species approaches require multivariate techniques to first determine the
relationship among species co-occurrence and then predictive models can be used to
determine if there are variables that accurately predict the occurrence of groups or group
members. Applying multivariate statistical techniques to relate species groups based on their
co-occurrence have been widely used in community ecology (Jongman et al., 1995; McCune
and Grace, 2002). There are several challenges to defining meaningful groups, including a
lack of statistically objective ways to define group boundaries. There remains debate whether
or not it is ecologically feasible to distinguish distinct groups versus continuums of
occurrence with overlap (Blondel, 2003). Therefore, it is advisable to use complementary
techniques to define groups and follow up with separate techniques to test the statistical
significance of groupings (McCune and Grace, 2002). This approach still requires that
ecological interpretations are made cautiously as statistical and biological significance are
not always one and the same (McCune and Mefford, 1999).
Linking variables directly affected by forest-management activities, specifically
measures of forest structure, to the occurrence of species assemblages may provide forest
managers with a useful tool to monitor, manage, and mitigate the effects of industrial harvest
on species diversity. Determining relationships among species may account for some
unexplained variation when examining structural variables and, therefore, account for
biological influences on species distributions. In this paper, we used presence and absence
data collected during intensive species inventories (Chapter 3) to determine if: 1) forest plots
could be reliably grouped based on vertebrate-species assemblages (i.e., based on co­
occurrence in space and time) as well as on structural characteristics of the vegetation; 2)
species-based and structure-based groupings were correlated with one another; and 3) forest-

92

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

structure characteristics could be used to predict group membership based on vertebratespecies assemblages. We specifically tested the hypothesis that forest structure can reliably
predict the occurrence of vertebrate-species assemblages and, therefore, has the potential to
be used as a surrogate measure to implement forest management targets and to monitor
species distributions within an operational forest landscape.

Methods

Study area
We collected data from May 2001 through January 2004. Our study area, in central
British Columbia, Canada was located approximately 30 km south of the community of
Williams Lake, hereafter the Williams Lake Study Area (Figure 4.1). Land uses included
industrial forestry, free-range cattle ranching, and a broad diversity of all-season recreation
(e.g., hunting, snowmobile touring). The Williams Lake Study Area, at the northern extent of
the Interior Douglas-Fir (IDF) Biogeoclimatic zone (Meidinger and Pojar, 1991), was
characterised by stands of closed- and open-canopy Douglas-fir (Pseudotsuga menziesii) and
lodgepole pine (Pinus contorta) at higher elevations (e.g., >1000 m above sea level). At
lower elevations (~ 850 m - 1000 m above sea level) forests were intermixed with grassland
and wetland communities. There were localised stands of hybrid White Spruce (Picea
engelmannii x glauca) and Trembling Aspen (Populus tremuloides) throughout the study
area. Harvesting activities in stands dominated by Douglas-fir were characterized by
multiple-pass selective harvests; where other species were dominant, clear-cuts were more
frequent. A major insect outbreak (mountain pine beetle; Dendroctonus ponderosae)
influenced stands across the study area. Mean annual temperature is 4.2°C (range = -1.3 to
9.6 °C; Environment Canada, 2002).
93

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

Jkm

10

Figure 4.1. Plot (o) layout in the Williams Lake Study Area (★ ; inset) approximately 30 km south of Williams Lake, British
Columbia (+; inset). Each plot was connected by a 150- or 300-m transect.
VO

-P.

Habitat measurements
We established 243 plots spaced systematically 150 m or 300 m apart along transects
and measured local habitat variables using a combination of methods. Shrub species and
cover, canopy gaps, and coarse-woody debris were measured along the intercept of 2, 48-m
transects laid perpendicular through plot centre. For coarse-woody debris, we recorded the
diameter of the piece perpendicular to where it crossed the axis, the tree species (if possible),
and decay class (Maser et al., 1979). At 5, 2-m radius plots located at 11.28 m away from
plot centre on each axis and at plot centre, we measured the percent coverage for litter,
coarse-woody debris, herb species, moss and lichens, shrub species, sapling species, bare
ground, and rock. Within a 5.64-m radius around plot centre we tallied all trees and stumps
>7.5 cm diameter at breast height (dbh) and recorded tree species, dbh, and height; we
recorded trees <7.5 cm dbh as live or dead saplings. We extended the radius to 11.28 m and
tallied any additional trees >30 cm dbh and snags. We recorded general information for each
plot including aspect, slope, canopy stratification and complexity, disturbance history
(evidence of fire, grazing, logging), and the elevation (in m) above sea level. We averaged
vegetation data collected at multiple plots measured within stands, as defined by forest
inventory polygons, to create stand level variables.
To obtain spatial measures (e.g., distance to water), proportion of cover, and
landscape metrics, we subsetted a Landsat 7 (30-m resolution) image of the study area (July
2002) and used PCI Works GIS software (version 7.0; PCI Geomatics Corp., 2001) to
complete a supervised maximum-likelihood classification. Six habitat classes were identified:
water, nonforest, early serai, shrub and Aspen, moderate retention conifer, and conifer. We
used colour airphotos, orthophotos, and the vegetation data collected at plots to seed areas for

95

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

training and to assess the accuracy of the classification. We assessed accuracy of the
classification by determining the number of correctly classified pixels from a randomly
selected subset. Water, nonforest, and conifer classes had the highest accuracy (97%, 94%,
and 88%, respectively). Early serai, shrub and Aspen, and moderate retention conifer were
less often classified correctly (67%, 78%, 67%, respectively). Overall classification accuracy
was 81%.
We calculated landscape metrics using the Patch Analyst 3.1 (Grid) extension
(Rempel and Carr, 2003) for ArcView GIS (version 3.2a; ESRI, 2000) to interface to the PC
raster version of FRAGSTATS 2 (McGarigal and Marks, 1995). We intersected the
classification layer with a buffer created around plot and stand centres and measured
characteristics in the surrounding 50 ha and 300 ha. Distance from plot centres to water and
high-contrast edge (e.g., meadow-forest), and roads were estimated using the GIS. For a full
description of methods and plot layout, see Chapter 3.

Vertebrate sampling
We used a variety of techniques to determine the presence of vertebrate species
(Table 4.1; see Chapter 3 for a full description of vertebrate sampling methods). We
georeferenced all vertebrate observations and imported them into a GIS (ESRI, 2000).
Observations >75 m from a plot centre were excluded to reduce possible effects of spatial
uncertainty on structure association (see Chapter 3).
Because of the variety of methods used to collect species-occurrence data, we
converted species detections to presence and absence for each plot (Magurran, 2003). We
deleted species detected at fewer than 5% of the plots, to maximize the detection of patterns
of structure, if present (McCune and Grace, 2002). Species that had high spatial uncertainty

96

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

Table 4.1. Methods used to detect species presence in the Williams Lake Study Area and the time of year surveys were conducted
from 2001-2004.

v©

Target species

Method

Timing

References

Song Birds

Variable-radius point
counts

3 surveys at each plot (midMay - early July)

Reynolds et al., 1980; Ralph et
al., 1995; Robbins, 1981

Woodpeckers

Call playback

3 surveys at each plot (midMay - mid-June)

Johnson et al., 1981

Owls

Call playback

Late April - early May;
Winter opportunistically

Fuller and Mosher, 1981

Small Mammals

Live trapping;
Track plates

July - August

Jones et al., 1996; Province of
British Columbia, 1998

Bats

Bat detectors

July August

Hayes, 1997; O’Farrell and
Gannon, 1999

Birds and mammals
Medium - large
vertebrates

Winter tracking
Remote cameras

November - February
May - August

Beauvais and Buskirk, 1999
Zielinski and Kucera, 1995

All vertebrate taxa

Intensive searches

July - August

Crump and Scott, 1994;
Wilson et al., 1996

All vertebrate taxa

Encounter transects
and Time-constrained
searches

May - August

Scott, 1994; Wilson et al.,
1996

associated with their detection were also deleted (n = 5), either because of their frequent
detection while in flight (e.g., Evening Grosbeak; Coccothraustes vespertinus\ Red Crossbill,
Loxia curvirostra\ Tree Swallow, Tachycineta bicolor) or because their calls may have
travelled long distances (e.g., Barred Owl, Strix varicv, Sandhill Crane, Grus canadensis',
Chapter 3). Therefore, we used 66 species in all subsequent analyses (Appendix III). Because
of the high number of zeros in the resulting matrix of species occurrence by plot, we
transformed the data using Beals smoothing (Beals, 1984). Beals smoothing replaces the
matrix cells with a probability of occurrence for that species at that location based on its joint
occurrence with the other species present at that plot. Likewise, for our structure-data matrix,
we transformed data using general relativization by row and column totals (McCune and
Grace, 2002). Relativization transforms matrix elements by a row or column standard (e.g.,
maximum, sum, mean, etc.) and is useful when different units of measure (e.g., basal area
and percent cover) need to be standardized for comparison (McCune and Grace, 2002). We
removed plots that had substantial change in structure because of industrial harvesting (n =
15) and examined the species composition at other plots for outliers (defined as >3.0 standard
deviations; Tabachnik and Fidell, 2001) and considered them for removal. We concluded that
3 plots were notable anomalies because of increased numbers of non-forest vertebrate species.
Two of these plots occurred in open, marshy meadows, and the third at the edge of a recent
harvest. Thus, our final analysis consisted of 225 plots and 66 vertebrate species.

Data analysis
Community-level data often violate assumptions of parametric approaches, although
some multivariate approaches are robust to violation of some assumptions (Jongman et al.,
1995). Because inference must be drawn from sets of correlative predictors, confidence in the

98

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

interpretation increases when results are consistent using different methods and data are
compared to randomisations of the data (Jongman et al., 1995; McCune and Grace, 2002). To
describe species assemblages and determine their relationship to structural variables, we used
non-metric multidimensional scaling ordination (NMS; Shepard 1962a, b; Kruskal, 1964a, b),
cluster analysis (Lance and Williams, 1967), multiple response permutation procedures
(MRPP; Zimmerman et al., 1985) and classification and regression trees (CART; Breiman et
al., 1984). We used the Sprenson distance measure in all analyses where a distance measure
was necessary as it has repeatedly been shown to be an effective measure regardless of
approach and consistency is recommended when using multiple approaches (Faith et al.,
1987; McCune and Grace, 2002). We completed most analyses with the software PC-Ord
(version 4.37; McCune and Mefford, 1999); we used Statistica (version 6.1; StatSoft, Inc.,
2003) for CART analysis and correlation analyses of environmental and species ordination
coordinates.
Ordination techniques arrange entities along single or multiple axes summarizing the
continuous trend within data; sites with similar species composition or environmental
characteristics are depicted closer together in this ordination space. NMS differs from other
ordination techniques in that it does not carry assumptions of linearity among variables, it
preserves similarity distances in ranked order (i.e., nonparametric) and tends to linearise
distances in species and environmental space, and is not constrained to any specific distance
measure or relativization method (Clarke, 1993; McGune and Grace, 2002). NMS is,
therefore, often a preferred method to use with community data (Clarke, 1993; McCune and
Grace, 2002).

99

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited without perm ission.

The final solution for NMS ordination is accepted after comparing several runs with
real data to randomizations of data through Monte-Carlo simulations (McCune and Grace,
2002). Two measures are used to evaluate the structure of the ordination results: stress and
instability. Stress, analogous to a goodness-of-fit measure, is the deviation from monotonicity
when distance is compared between the original species space and distance in the reduced
ordination space (McCune and Grace, 2002). Stress is typically in the range of 10 - 20 for
community data (McCune and Grace, 2002), but at the upper end of this scale cautious
interpretation should be made as plots are relatively easy to misinterpret (Kruskal, 1964a;
Clarke, 1993). Instability is a measure of change in stress at each iteration. Stable, low stress
solutions indicate strong data structure.
To derive ordination results, we conducted 40 runs with real data and 50 runs with
presence and absence data randomized among plots. We repeated each run with 1 to 6 axes
(i.e., dimensions in ordination space). Stress was compared for each ordination result with
different number of axes. The starting configuration coordinates for the final run, used to
determine stress and instability values for the final solution, were derived from the solution
with the number of axes where stress did not decrease substantially when fewer axes were
used. We repeated this procedure for ordinations of species detected at plots and structure
measures at each plot to determine if plot structure driven by species assemblage was
comparable to pattern derived from structural characteristics. To evaluate the effectiveness
of the ordination, we calculated the coefficient of determination (R ) between the distances of
plots in ordination space relative to the distances in the original, unreduced space (McCune
and Grace, 2002).

100

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

To elucidate discrete groupings of the plot sample units we used cluster analysis.
There are several approaches for clustering and results vary with the approach (e.g.,
agglomerative or divisive; McCune and Grace, 2002), distance measure, and group linkage
method used (McCune and Grace, 2002). We used an agglomerative-hierarchical approach
(Breiman et al., 1984) where all sample units are grouped from the bottom up with Sprenson
distance, as in the ordination, and a flexible beta linkage of -0.25. A flexible beta linkage of
-0.25 has been shown to minimize increases in errors in the sum of squares of distances from
each individual to the centroid of its group (Lance and Williams, 1967) and is recommended
for ordinations that use Sprenson distance (McCune and Grace, 2002). We selected the
number of groups based on a compromise between total information remaining (analogous to
R2 in multiple regression approaches) and ecological interpretation. We tested statistical
difference among groups with MRPP. MRPP tests the hypothesis that there is no difference
among groups and is also a non-parametric approach. It provides a statistic of how different
groups are, analogous to effect size, given as the chance-corrected, within-group agreement
(A) and a p-value. A is equal to 1 when all items are identical within groups and 0 when
heterogeneity within groups equals expectation by chance.
To infer relationships of vertebrate assemblage to structural variables measured at
plots, we first examined the correlation of environmental variables with the plot coordinates
on each ordination axis for species-based and structure-based ordinations (Pearson’s r).
Second, we assessed correlations between the coordinates for the species-based and
structure-based ordinations. To compare the visualization of the continuous plot structure
derived from NMS ordination and the discrete groupings defined with cluster analysis, we
overlaid cluster groupings on the ordination diagram, thus providing a hybrid explanation of

101

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

patterns observed within the data structure. We repeated the group definition for both
species-based and structure-based clusters.
To describe the structural characteristics that distinguished groups, we used
classification and regression trees (CART) analysis (Breiman et al., 1984). CART analysis is
similar to discriminant analysis, but is more flexible partially because it is a nonparametric
approach and thus is not influenced by deviations from normal distributions, data
transformations, or outliers (Breiman et al., 1984, De’ath and Fabricius, 2000). Comparative
studies have concluded that it performs as well or better with predictive classifications as
logistic regression, discriminant analysis, and artificial neural networks (Selker et al., 1995;
Vayssieres et al., 2000; Karels et al., 2004) and given the ease of interpretability of the results,
CART analysis may be preferential in many instances (Olden and Jackson, 2002; Worth and
Cronin, 2003). Other advantages result because CART analysis makes several individual
selections of where to split the data. Therefore, the amount of available information is
maximised because collinear variables can all be included and used as surrogates when data
are missing and variables can be included at multiple locations in the tree elucidating
complex relationships (Breiman et al., 1984, De’ath and Fabricius, 2000; Vayssieres et al.,
2000). CART analysis, however, can overfit data (McCune and Grace, 2002). Therefore, it is
important to carefully consider variables to be included in the analysis, and determine the
method of determining the split decisions of the tree and as well as when to stop splitting
(Breiman et al., 1984). We were specifically interested in the ability of forest-structure
variables to predict species group membership; therefore, our variable list was selective
(Table 4.2). Because we reduced our variable set to structural variables at the outset, we used
an exhaustive search for all possible univariate splits. With this method, all possible splits for

102

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

Table 4.2. Variables used in the classification and regression tree (CART) analysis to predict group membership for plots in the
Williams Lake Study Area. Plots were grouped based on the co-occurrence of vertebrate species detected from 2001-2004.
Variable Name
Total coarse-woody debris volume
Canopy gap
Shrub cover
Shrub height
Forb cover
Saplings ha'1
Deciduous stems
Conifer stems
Total basal area
Dead wood basal area
Large tree basal area
Douglas-fir basal area
Pine basal area
Spruce basal area
Distance to edge
Distance to water
Edge density
Canopy stratification

ou>

Abbreviation Units
nr1
total_cw
%
P_gap
%
P_shrub_
m
P_sh
%
P_2_forb
Count
P_sap_ha
Count
P_dec_st
P_con_st
Count
ha'1
M_ba_ha
ha'1
dead_ba_
ha'1
large_ba
ha'1
Fd
ha'1
Pi
ha'1
sx
m
d_edge
m
d_water
ed_50ha
m ha'1
m
p_diffa

Measurement Technique
Plot transects
Plot transects
Plot transects
Plot transects
Average of 5 * 2 m cover plots
5.64-m radius plot
5.64-m radius plot
5.64-m radius plot
5.64-m radius plot
11.28-m radius plot
11.28-m radius plot
5.64-m radius plot
5.64-m radius plot
5.64-m radius plot
GIS
GIS
GIS
5.64-m radius plot

each predictor variable at each node are examined and the split producing the largest
improvement in goodness-of-fit is used, for which we used the Gini goodness-of-fit measure
(Breiman et al., 1984). To reduce the possibility of overfitting with large classification trees
(i.e., several splits), we used the smallest tree (i.e., fewest splits) with classification errors
nearest to the lowest number of classification errors for the largest tree (Breiman et al., 1984;
De’ath and Fabricius, 2000). We used similar criteria for CART analysis as in our cluster
analysis in that we compromised between variation explained and ecological interpretability.
We set prior probabilities for group membership equally (n = 11; p = 0.91) because we did
not assume higher risks for misclassifying plots belonging to different groups (Breiman et al.,
1984). Therefore, to determine the potential of using forest structure as a surrogate measure
for the occurrence of species groups, our analyses were a complementary approach of several
techniques.

Results

Plot groupings
We accepted a 3-dimentional solution for NMS ordinations based on both species
occurrence and structural characteristics. For the species-based ordination, correlation
analysis with the structural variables and ordination axes indicated the most influential
variables for species occurrence were edge density, distance to edge, and distance to water
(Table 4.3; stress = 11.61; instability = 0.0001, iterations = 116;/? = 0.020). Total variance
explained by all axes was 92.6% indicating that the ordination represented similarity among
plots well. Proportion of variance represented by each axis based on R between distance in
the ordination space and distance in the original space indicated that the first axis captured

104

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

Table 4.3. Pearson correlation coefficients for ordination axes resulting from non-metric
multidimensional scaling. Ordination was based on species co-occurrence at plots in the
Williams Lake Study Area from 2001-2004.

Variable

Axis 1

Axis 2

Axis 3

Distance to edge

-0.501**

0.290**

0.449**

Distance to water

-0.313**

0.587**

0.216*

Edge density

0.484**

-0.215*

-0.555**

* p <0.05
** p <0.001

105

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

most of the variance (49.3%); the second and third dimension contained 16.5% and 26.8%,
respectively.
For the structure-based ordination, tree-species composition dominated correlations
with the 3 axes, either by basal area by species or counts of conifer or deciduous stems
(Table 4.4; stress =13.47; instability = 0.0001; iterations = 96, p = 0.020). Variance captured
by the first, second and third axis was 30.1%, 14.4%, and 42.1%, respectively, for a
cumulative total of 86.6% variance captured. Plot coordinates in ordination space were
highly correlated for all axes between the species-based and structure-based ordinations
(Table 4.5), indicating that both ordinations resulted in plots within similar positions in
ordination space. Based on the presence or absence of species within plots when overlaid in
continuous ordination space, several species showed no relationship to any of the variables
most strongly correlated with the ordination axis (n = 37). Some species, however, showed
avoidance or selection for characteristics on one or more axes (n = 29; Table 4.6). For
example, presence of Brown-headed Cowbird (Molothrus ater) was positively associated
with edge density on both axis 1 and axis 3 and negatively associated with distance to water
on axis 1 (i.e., found closer to water).
Cluster analysis of plots based on species composition and structural characteristics
resulted in the identification of 11 groups. Groups based on species composition had 27%
unexplained variance (Figure 4.2; Appendix IV)1. Plots within the same groups had species
occurrences more similar to one another than to plots in other groups. The cluster analysis
based on structural characteristics had 30% unexplained variance (Figure 4.3; Appendix IV).
MRPP analysis indicated that both groupings, species and structure, were statistically
significant. The species grouping, however, had less within-group variance than structure1 Species most associated with each group are presented in Appendix V.

106

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

Table 4.4. Pearson correlation coefficients for ordination axes resulting from non-metric
multidimensional scaling. Ordination was based on structural characteristics measured at
plots in the Williams Lake Study Area.

Variable

Axis 1

Axis 2

Axis 3

Percent gap

0.494**

-0.500**

-0.259**

Percent shrub cover

0.051

0.108

-0.600**

Saplings ha'1

-0.495**

-0.009

0.297**

Deciduous stems ha'1

-0.247**

0.478**

-0.314**

Coniferous stems ha'1

-0.624**

0.226**

0.397**

Basal area Douglasfir

-0.204*

0.325**

0.668**

Basal area Pine

-0.505**

-0.111

-0.143*

Basal area Spruce

-0.321**

0.342**

-0.463**

Distance to edge

-0.443**

-0.151*

0.479**

Distance to water

-0.027

-0.509**

0.390**

Edge density

0.472**

0.095

-0.369**

* p <0.05
** p <0.001

107

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

Table 4.5. Pearson correlation coefficients for plot coordinates based on species and
structure using non-metric multidimensional scaling ordinations. Species occurrence was
determined from 2001-2004 in the Williams Lake Study Area.

Structure-based

Axis 1

Species-based
Axis 2

Axis 1

-0.335

0.101

-0.331**

Axis 2

-0.299

0.163*

0.300

Axis 3

0.503

0.024

-0.030

**

Axis 3

**

* p <0.05
** p <0.001

108

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

Table 4.6. Relationship between vertebrate species presence and absence with ordination
axes as determined by non-metric multidimensional scaling. Axis one is defined by distance
to edge (-) and edge density (+), axis 2 by distance to water (+), and axis 3 by distance to
edge (+) and edge density (-). Variables are described in Table 4.2. Species with no clear
relationship to any of the variables defining ordination space are not listed (n = 37). If species
were not located in a particular quadrant of the 2-dimensional space, then the relationship
was interpreted as avoidance. Latin names for all species are listed in Appendix II.
Species
Moose
Alder Flycatcher
Black-blacked Woodpecker
Brown-headed Cowbird
Brown Creeper
Coyote
Clay-colored Sparrow
Common Yellowthroat
Downy Woodpecker
Dusky Flycatcher
Gray Jay
Hairy Woodpecker

Axis 1
d_water (+)
ed_50ha (+)
ed_50ha (-)
ed_50ha (+)
ed_50ha (+)
ed_50ha (+)
ed_50ha (+)
ed_50ha (-)
ed_50ha (+)

Snowshoe Hare

ed_50ha (-)

Least Flycatcher

-

Lynx
McGillivary’s Warbler
Meadow Vole
Ermine
Long-tailed Weasel

ed_50ha (-)
ed_50ha (-)
ed_50ha (+)
ed_50ha (-)
ed_50ha (+)

Northern Flicker
Northern Waterthrush
Olive-sided Flycatcher
Song Sparrow
Yellow-pine Chipmunk
Townsend’s Warbler
Three-toed Woodpecker
Red Fox

ed_50ha (+)
ed_50ha (+)
ed_50ha (+)
ed_50ha (+)
ed_50ha (-)
ed_50ha (-)
ed_50ha (-)

Winter Wren
Western Wood-peewee

-

ed_50ha(+)

Axis 2
d_edge (+)
d_water (-)
d_water (-)
d_water (-)
d_water (-)
d_edge (-)
d_water (-)
d_edge (+)
d_water (+)
d_edge (-)
d_water (-)
d_edge (+)
d_edge (-)
d_water (-)
d_water (+)
d_edge (-)
d_water (-)
-

Axis 3
ed_50ha (-)
ed_50ha (+)
d_edge (+)
ed_50ha (+)
ed_50ha (-)
ed_50ha (+)
-

-

ed_50ha (-)

d_edge (+)
d_edge (+)
d_edge (-)
d_water (-)
d_edge (+)
d_water (-)
d_edge (-)
d_water (-)

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

-

-

"

"

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

i

F igure 4.2. Dendrogram from hierarchical cluster analysis o f plots in U . Williams Lake Study Area by
Symbols indicate groups of associated plots. Plots in the same groups have species composition more si:

:ies co-occurrence detections from 2001-2004.
r to one another than to plots in other groups.

110

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

CD

'cc
E
0)SCO

CO

E

oc

Ail M

w
o
CL

Figure 4.3. Dendrogram from hierarchical cluster analysis of plots by characteristics of forest structure at plots located in the Williams Lake Study Area.
Symbols indicate groups of associated plots. Plots in the same groups have structural characteristics more similar to one another than to plots in other groups.

Ill

based groups (species groups: A = 0.292, p <0.01; structure groups: A = 0.090, p <0.01).
Species groups were distinct when overlaid on the ordination plot (Figure 4.4A, B, C);
groups defined by similarity in structural variables were not distinct on any of the ordination
planes (Figure 4.5A, B, C). These results indicate that cluster analysis and ordinations based
on species occurrence support each other, but the same can not be said of the structure cluster
and ordination analyses. Overall, overlap between species-based and structure-based groups
ranged from 16.7-50.0% (Appendix IV).

Predictions using structure
Using the groups defined by species co-occurrence, predicting group membership
with CART analysis resulted in low classification success (45.8%) and ranged widely among
groups (0% - 100%; Table 4.7). Seven structural variables were used to discriminate among
plots; edge density and distance to edge were ranked as most important (i.e., explained more
of the variation, appeared at the top of the tree, and occurred multiple times; Table 4.8). The
CART dendrogram depicts which variable defines the split at each branch, the number of
cases in the group, and classification success (Figure 4.6). Edge density defined the first split
among groups, indicating it was the most important variable discriminating among groupings.
Moving to the left branch of the tree, edge density and saplings ha"1led to 2 terminal nodes
that correctly predicted 8.5% of group 5 members and group 11 perfectly (Table 4.8);
although there were misclassifications included within these splits. The dendrogram thus
illustrates the variable splits that define the groups and how impure the groups are on each
branch.
Distance to water and edge density led to one terminal node and the correct prediction
of 50% of group 8 plots, to the right outer branch of the dendrogram. Adding edge density

112

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

A
Group

•v

o5
♦ 6
o7
• 8
□9

♦•
♦ V

%V
•o
CM

A1
*2
v3
*4

■ 10
x 11

0V<

<n

3

• •

Axis 1

Figure 4.4. Non-metric multidimensional scaling ordination (NMS) of plots based on
presence and absence data for vertebrate species detected at plots in the Williams Lake Study
Area from 2001-2004. Separate figures are presented for each pair of ordination axes (A, B,
C). Cluster analysis groups are depicted by different symbols. Plots shown closer to one
another are more similar in species composition.

113

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

B
Group

A1
±2
v3
▼4
o5
♦6

o7
• 8

□9
■ 10
x 11

oo

(fl

V♦
♦ V

3

* ° o
°

Q j

’ $%b

»

A

AO

OA

♦v

O□
o•

Axis 1

Figure 4.4. NMS ordination plots for plots based on presence and absence data of species
assemblages (continued).

114

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

c
Group

A1
*■

2

v3
▼4

^5
♦ 6

o7

*8
□9
■ 10
X

11

co
(O
3

AO

Oo
□□

Axis 2

Figure 4.4. NMS ordination plots for plots based on presence and absence data of species
assemblages (continued).

115

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

S tru c tu r e G ro u p

a1
*2
v3
▼4
o5
♦ 6

♦ •

o7
• 8

a*

□9

*■

■ 10
x11

□ T

Axis 1

Figure 4.5. Non-metric multidimensional scaling ordination (NMS) of plots based on
characteristics of forest structure at plots in the Williams Lake Study Area. Separate figures
are presented for each pair of ordination axes (A, B, C). Cluster analysis groups are depicted
by different symbols. Plots shown closer to one another are more similar in forest-structural
characteristics.

116

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

B
Structure Group

A1
a2
v3
▼4
o5
♦ 6

°7
*8
□9
■ 10
x 11

Axis 3

<>♦

□□
A V

O♦

□□

OV V
AA

* ♦
*
W

a&a

i

Axis 1
Figure 4.5. NMS ordination for plots based on characteristics of forest structure (continued).

117

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

C
Structure Group

A1
*2
v3
▼4
o5

♦6
o7
•8
□9
■ 10

Axis 3

X11

A

AO***

VO
• V

AA

□ <>

Axis 2

Figure 4.5. NMS ordination for plots based on characteristics of forest structure (continued).

118

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

Table 4.7. Correct classification matrix of observed class against predicted class for CART analysis. Structural characteristics of plots
were used to predict group membership and groups were defined based on similarities in species co-occurrence from 2001-2004 at
plots in the Williams Lake Study Area. The 11 classes were defined using cluster analysis. Overall classification accuracy was 45.8%.
Number of correct classifications are in bold for each class.

Predicted
Class
1
2
3
4
5
6
7
8
9
10
11

Class
1

Class
2

Class
3

Class
4

Class
5

0
4
0
21
0
0
1
1
0
1
1

0
7
0
1
0
0
1
1
1
5
0

0
7
0
16
0
0
0
3
1
3
1

0
2
0
9
0
0
0
0
0
0
0

0
2
0
23
5
4
5
16
0
8
1

Observed Class
Class
Class
Class
7
8
6
0
0
0
1
0
7
1
0
0
2
0

0
0
0
0
0
0
12
0
1
0
0

0
0
0
2
0
1
5
10
0
2
0

Class
9

Class
10

Class
11

0
0
0
1
0
0
5
1
5
2
0

0
2
0
1
0
3
0
0
0
7
0

0
0
0
0
0
0
0
0
0
0
3

Percent
Correct
0.0
43.8
0.0
56.3
8.5
63.6
92.3
50.0
35.7
53.8
100

Table 4.8. Ranking of importance for structural variables used in CART analysis to predict
membership of plots grouped based on species co-occurrence from 2001-2004 in the
Williams Lake Study Area; 100 = most important, 0 = least important.

Variable
Distance to edge
Edge density
Distance to water
Saplings ha"1
Percent gap
Shrub cover Percent
Forb cover percent

Ranking
100
93
76
59
56
45
42

120

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

204
ed_50ha<=145.36

121

p_sap_ha<=700.

p_d_water<=182.58

ed_50ha<=293.48

:=34.073

p_gap<=.55417

p_sh<=.57688

[»] [?J
Figure 4.6. Dendrogram for the classification and regression tree analysis. Plots were grouped based on species co-occurrence from
2001-2004 in the Williams Lake Study Area. Group membership was predicted using forest structural variables. Nodes to the left of
each decision criteria meet the stated condition, whereas nodes to the right do not meet the stated criteria. The numbers along each
decision branch indicates the number of cases that did or did not meet the decision criteria. The upper number within terminal nodes
(red) is the number of cases assigned to the predicted class, the lower number within terminal nodes is the group name. Not all cases
are correctly classified. Overall classification success is presented in Table 4.7.

and percent canopy gap led to the correct prediction of 43.8% of group 2 and 56.3% of group
4. In the centre branch, the addition of distance to edge led to the correct prediction of 92.3%
of group 7; the addition of forb cover predicted 63.6% of group 6; and finally, shrub height
distinguished 53.8% of group 10 and 35.7% of group 9. The presence of only 9 terminal
nodes, instead of the 11 groups identified with cluster analysis, reflected the lack of
successful predictions for any members of groups 1 and 3. The right and centre branch are
distinctive from the left branch in that they each contained a spatial measure near the top of
the branches (i.e., high variable importance). Overall, the results from the CART analysis
indicate that although there was a correlative relationship between groups classified with
species co-occurrence and structural characteristics, the relationships were not strong enough
to accurately predict the groups.
Given the results of the ordination and correlation analyses, plots could reliably be
grouped based on both structure and the occurrence of species. Plot groupings from both
approaches had similarities indicated by significant correlation among coordinates in
ordination space. Three variables (edge density, distance to water, and distance to edge)
consistently appeared in each analysis suggesting that both spatial context and larger scale
aspects of structure were the best correlates of species occurrence. Overall, however, these
relationships were not strong enough to reliably predict the membership of plots to groups
based on species co-occurrence. Therefore, although species groups were distinctive,
indicating that species co-occur with some degree of pattern, the structural characteristics we
measured were not sufficient to describe the relationships.

122

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

Discussion
Plots could be grouped into significant assemblage types as well as structural types. Both
approaches resulted in groups that had considerable variation, more so for the structurally
defined groups. Consequently, the biological significance of the groupings is suspect
(McCune and Mefford, 1999). Further, poor predictive ability of forest structure and the
overall weak relationships between species-based groupings of plots and structure-based
groupings of plots suggest that these patterns are driven by factors other than those we
measured (e.g., biological factors such as competition, prey availability, other environmental
variables). Therefore, reliance on structure as a surrogate for predicting the occurrence of
species cannot be recommended.
To facilitate comparison among groupings, we selected the same number of groups for
each set of groups based on explained variation and simplicity. Improved grouping for
structure groups may have been achieved if fewer classes had been selected. Other
researchers have used clustering techniques to define structure groups in our study area by
cumulative distribution plots of basal area ha'1 and tree stems ha"1 (Moss, 2002). The
clustering techniques were used to distinguish differences between narrow- and wide-ranging
diameter distributions and were implemented for management purposes, although the
classification had considerable variation within and among groupings (Farnden et al., 2003).
The use of such groupings may carry lower risks when used to characterize structure alone,
especially with supporting vegetation inventories, but not if being used as a surrogate
approach to monitoring species presence.
It has been debated whether or not it makes ecological sense to define community units
or species assemblages into discrete entities because the groups may be too simplistic and do

123

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

not take into account ecological factors such as competition and stand history (Allen and
Hoekstra, 1992; McIntosh, 1995; Heino et al., 2003). For vertebrate species, as opposed to
vegetation structure composition differences, alternative techniques using individualist
concepts relating species along continuums may be more ecologically meaningful. Data
needed to develop predictive models for several species are lacking, however, and models are
often limited by low predictive ability (Chapter 3). These data are also difficult and
expensive to obtain. For example, in our study area, where extensive inventories were
conducted for species presence, several vertebrate species were not detected often enough to
model and few species were modelled with good predictive accuracy (Chapter 3). Grouping
species by co-occurrence should account for species interactions that play a significant role
in species presence (Caswell, 1976; Gotelli and Graves, 1996) and if structure predicted
groups well, management and monitoring objectives could be more efficiently implemented
(Angermeirer and Schlosser, 1995). Because the single-species models had few overlapping
variables, generalisations about structural characteristics for several species were not
practical (Chapter 3). Further, taking an assemblage-level approach would be advantageous
only if species were reliably grouped and structure measures predicted groupings well, which
was not the case.
We limited our selection of variables to these that would normally be measured in forest
inventories and are affected by forest-management activities. Other variables may help to
improve the unexplained variation in groups. Other studies have reported that non-structural
variables (e.g., aspect, moisture, elevation; du Bus de Wamaffe and Dufrene, 2004; stand
history, Tyre et al., 2001), ecological interactions, and scale influenced the variation
explained in predictive models (Heino et al., 2003; du Bus de Wamaffe and Dufrene, 2004;

124

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

Oxbrough et al., 2005). Our study area is heavily managed by multiple users (e.g., harvesting,
ranching, recreation). Therefore, disturbance history, types of disturbance, and disturbance
intervals may help predict the occurrence of species. Although non-structural variables and
stand-history variables may be relatively easy to obtain, the lack of overall relationships
between species co-occurrence and structural variables alone suggest that forest structure is
limited in terms of a surrogate approach for monitoring and that the biological influences
override those of structure.
Small-scale variation in plot structure may have further confounded group definition.
Most species that we included in our analysis are highly mobile (i.e., capable of flight) or
have large home ranges. Species groupings are thus less likely to be influenced by smallscale variation. Further, species may not have grouped well because of a wide range of
environmental tolerances and geographic distributions. Many of the species that were
included in the assemblages are found across provincial or continental scales; therefore, we
may not have sampled over a broad enough environmental range. Overall in our analyses, we
found groups were statistically distinct, but ecological significance was difficult to interpret
and without strong relationships to predictive variables, assemblage-level approaches are not
feasible or more efficient than single-species approaches (Chapter 3).
There was significant correlation among plot coordinates in ordination space for
species-based and structure-based groupings. The variables most strongly correlated to each
ordination axis, however, differed (Table 4.3; Table 4.4). The variables most highly
correlated with the species ordination were related to spatial aspects of plot location (i.e.,
distance to water, distance to edge) and aspects of the surrounding habitat (i.e., edge density).
In contrast, the structure-based ordination variables related to tree species composition were

125

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

highly correlated to axes. Further, when we overlaid the cluster-analysis groups onto the
plots in ordination space, the species ordination had much more distinct groupings than
structure ordination (Figures 4.3, 4.4, respectively). This suggests that although the processes
driving the correlations may be related, characteristics of forest structure are not good
surrogates for those processes. Again, scale of measurement and landscape interactions
confound the interpretation of results. It was important, however, to determine the
relationship at the scales we measured because this scale is representative of management
practices. Further, we needed to standardize our measures of species co-occurrence. Overall,
these results suggest that species more likely responded to the co-occurrence of other species
or conspecifics and spatial aspects of plots than to structural characteristics alone. Settling
near conspecifics may provide benefits. For example, it may provide assessment of habitat
suitability (Desrochers and Magrath, 1993; Doligez et al., 2002), reduced risk of predation
(Stephens and Sutherland, 1999), and reduced territorial defence costs (Meadows, 1995).
Increasing the geographic scope and management regimes may improve groupings and
predictive models relative to structural characteristics, but it is not likely to improve aspects
of monitoring diversity at the operational scale of forest harvesting.
Defining species assemblages and predicting their occurrence for a single taxonomic
group, particularly birds (e.g., Diaz et al., 2005) and invertebrates (e.g., Oxbrough et al.,
2005), have proven successful in other studies. Conservation assessment and diversity
monitoring, however, should not be based on a single taxonomic group. In fact, even all
vertebrate groups comprise a relatively small proportion of biodiversity. Spanning taxonomic
groups may have confounded structural relationships because of differences in the scale that
species use to respond to structure. For example, Coyote (Canis latrans) and Meadow Vole

126

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

(Microtus pennsylvanicus) were both positively associated with edge density (Table 4.6;
Appendix 1), but perceive their habitat at very different spatial scales. Although many
researchers advocate multiple taxonomic approaches as desirable for diversity monitoring
programs (Fleishman et al., 2005; Heino et al., 2003), and the approaches have shown
correspondence between indicator species of birds and invertebrate richness (Fleishman et al.,
2005), modelling for individual taxa may be required for the development of effective
programs. Other approaches that use species richness as the response variable in predictive
models (e.g., Loehle et al., 2005) require caution because it is difficult to distinguish loss of
sensitive species and ensure the full representation of regional or local species when richness
is used (Lindenmayer, 1999; Magurran, 2003).
Habitat-based surrogates are appealing because they do not require the direct
measurement of species occurrence and can usually be described cost effectively. Predicting
the occurrence of species rapidly and cost effectively for landscapes that are changing
frequently because of management practices, however, is a challenging issue. Successful
approaches have been proposed and tested for some species, but these studies tend to more
intensely measure habitat use or use abundance as a response variable (e.g., Florida Scrub
Jay, Aphelocoma coerulescens\ Carter et al., 2006). The cost of obtaining these data is often
prohibitive. If similarities in structural association could be identified for groups of species,
predicting occurrence and response to management practices could be made even more
efficient. Although we found statistically distinct groups, we did not find that they were
predicted well by structure.
Other studies have analyzed prediction errors and questioned the underlying habitatrelationship models used to predict occurrence (Boone and Krohn, 1999), or more broadly

127

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission.

the use of environmental diversity as a surrogate for species diversity (Araujo et al., 2001). In
our study there were significant correlations among structure and species groups, but the
relationships were weak. Given the lack of predictive success and variance within groups,
using structure alone to monitor species assemblages is not a feasible management option in
our study area. With the results of our study we agree with Kavanagh and Stanton (2005) and
Manley et al. (2004) that it is better to monitor species directly than to rely on habitat
surrogates that are not proven reliable.
Ensuring that surrogate measures are thoroughly tested and robust through time in the
dynamic environments of forests will also require long-term monitoring data. Continued
species monitoring and research on the development of surrogate approaches in forested
environments would benefit from adaptive management (i.e., experiments with controls and
replicates; sensu Walters, 1986). Although recording presence and absence of different
species does not necessarily provide insight as to how to improve management because
causal mechanisms are not identified (Stone and Porter, 1998), these data collected over the
long term through monitoring could potentially identify the responses of species to
modification of structure and, therefore, go beyond correlative inference.

128

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Chapter 5. General Conclusions
Accurately predicting species occurrence is an important aspect in the studies of
ecology and conservation management. Economic and social values often conflict with
ecologic values on forested landscapes. With efficient biodiversity monitoring programs,
forest managers may be able to better incorporate a balance of competing goals. Linking
variables directly affected by forest management activities, specifically measures of forest
structure, to the occurrence of species and assemblages may provide forest managers with a
useful tool to monitor, manage, and help mitigate the effects of industrial harvest on species
diversity. Through my research, I investigated using forest structure to predict the occurrence
of single species and species groups, as well as how methods of detection may influence
model outcomes. Unique aspects of the research I conducted include simultaneous collection
of distribution data for multiple species, the use of external data to validate single-species
models, and the extremely detailed structure data collected over multiple scales. Overall, the
success of forest structure as a surrogate measure of vertebrate presence was poor. Therefore,
the hypothesis that structure can be used in place of direct species monitoring can not be
accepted.

Structure as a surrogate measure
I was able to model a small portion of the species detected (n = 55). Although this
was a low percentage (<30%) of the overall species detected in the study area, it can be
considered successful in that modelling this number of species by taking a species-by-species
approach would have taken considerably more time, effort, and other resources. By
conducting concurrent inventories, significant amounts of data were collected relatively

129

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

efficiently. Further, the presence and absence data provide baseline distribution and habitat
association information that can be used to improve future modelling efforts. Research
designed for a single species, however, often has much better results in terms of model
outcomes, likely because of more intensive measurement of habitat use, either through direct
observation or radio-telemetry.
In my research, I found that different types of detection (e.g., sign, audio, visual) led
to inclusion of different habitat variables in models and different levels of model
discrimination (i.e., ROC values; Chapter 2). I believe this result is a consequence of species
being more selective of habitat for certain activities that may result in different types of
detection. For example, the Pileated Woodpecker may be highly selective in using a tree for
foraging or nesting, which results in sign detections, but not very selective in choosing
habitat to commute through, which may result in visual or audio detections. Future inventory
efforts may consider modifying methodologies to increase the detection of species use of
specific forest structures deemed critical habitat elements and should always include the
detection type for the potential of separate analysis.
Relating species presence to habitat characteristics is not without its critics because of
the correlative nature of relationships, interactions with others species, and various aspects of
behaviour (see review Mitchell, 2005). Regardless, describing species-habitat relationships
remains an extremely important research and management goal. Other researchers have
found significant relationships between species richness and physical characteristics of the
environments and forest structure in particular (e.g., birds; Diaz et al., 2005; herptofauna;
Loehle et al., 2005; spiders; Oxbrough, et al., 2005). Using species richness as a response
variable, however, may not reveal losses or population declines of specific species. For

130

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

example, areas subjected to intermediate disturbance (Connell, 1978) and ecotones (e.g.,
riparian areas; Naiman and Decamps, 1997) often have high diversity, but the species
represented may not include species that are sensitive to disturbance or require large areas of
contiguous forests. Using species richness, therefore, may miss sensitive species that are
important components of the suite of species that occur in an area (Magurran, 2003).
Ensuring that all species are represented across the landscape and that viable populations are
maintained remains the overall goal of most management programs aimed at conserving
biodiversity. More sensitive species could be included in a fine-filter approach and, therefore,
complementary approaches should continue to be recommended (Lindenmayer and Franklin,
2002).
Neither single-species (Chapter 3) nor multiple-species (Chapter 4) models resulted in
clear relationships with forest structure. There were several variables, however, that were
repeatedly used, independent of the approach or statistical methodology. Specifically,
distance to edge, distance to water, and edge density were common predictors of species or
species-group occurrence. Further, several single-species models combined variables that
described forest structure at multiple scales and the variables most often included in plot
models (i.e., distance to water, large tree basal area, and dead tree basal area) were not as
frequently included in stand models, suggesting that the influence of these structures on
species presence was localized (Chapter 3). Overall, because the scale at which different
species use habitat varies so widely, generalisations among presence and forest structure
across species are difficult to infer (Chapter 3 and Chapter 4). The results of my research, and
that of others, highlight the importance of spatial context and scale of habitat as factors
related to habitat use (e.g., Scott et al., 2002). Spatial context, however, is not a routinely

131

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

described component of structure when used for forest management purposes. Further
research on the influence of spatial configuration of elements of forest structure (e.g., large
trees, dead trees) and the spatial context of habitats (i.e., distance to specific habitat features)
should be a priority to improve the potential of management recommendations regarding
harvest patterns, retention, and recruitment of important elements as the current emphasis on
structure without context appears misguided.

Uncertainty
Examining potential sources of error is frequently overlooked, but is an important
aspect of model validation (Barry and Elith, in press). My examination of potential sources
of error indicated that spatial uncertainty was the most problematic. Models derived from
detections that were spatially certain (e.g., sign) resulted in models with better discrimination
(i.e., higher ROC values; Chapter 2) and models for species that were associated with high
spatial uncertainty at the plot level (e.g., Sandhill Crane) had poorer discrimination (i.e.,
lower ROC values; Chapter 3). Because variation in forest structure was high within plots,
accurately determining the location of a species seems particularly crucial to model
performance. For species that are rarely detected with high levels of spatial certainty, plotlevel monitoring is not likely appropriate without high confidence in occurrence data.
Methods that illicit responses from individuals that are highly mobile (e.g., owls and
woodpeckers) may further confound determination of the relationship with habitat elements
as individuals may move before they are detected (these occurrence points could be modelled
at larger landscape scales). I did not find a relationship between temporal uncertainty and
model discrimination, which is probably because of the short lifespan of most measures of

13 2

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

presence through sign (e.g., tracks and scat) relative to changes in forest characteristics from
disturbance (e.g., harvest, fire, or windthrow).

Validation
With the presence and absence data collected, species-structure models were
developed for 55 species, some at multiple scales. Discrimination and reliability of several of
these models, however, were weak (Chapter 3). Further, when the models were validated
with spatially and temporally independent data, most models failed to meet the criteria of a
“good” model (i.e., had poor discrimination and / or reliability). Discrimination and
reliability tended to be less when spatially validated compared to temporal validation,
suggesting that site history may be an important factor influencing species presence (e.g.,
Tyre et al., 2001) and that models cannot be applied across an environmental gradient in my
study area (Van Home, 2002). This also reinforces the association of spatial uncertainty and
poor model performance. Habitat models have often been used for management purposes
without adequate validation (Rykiel, 1996), and this research casts further doubt on that
practice. Given the dynamic nature of forests and species populations, continued assessment
of models through time should be a required aspect of any program that uses species-habitat
models as a surrogate measure of presence.

Summary
The extensive presence and absence data across taxa, vegetation data collection at
multiple scales, examination of relationships between structure and single species and species
assemblages, and the comparative analysis among techniques allowed for a thorough
examination of the potential of structure as a surrogate measure of species occurrence. One

133

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

outcome of this approach was the demonstration that multiple-species monitoring can be
achieved efficiently and cost-effectively. The external validation of models allowed for
confident conclusions to be drawn regarding the ability of the model to accurately predict the
occurrence of species. Finally, using individual species presence and absence, as opposed to
species richness, explicitly tested the potential of forest structure as a surrogate measure to
monitor occurrence for a broad range of species. Few studies have examined such a breadth
of species concurrently and several of the species modelled are not frequently studied. The
model results highlight the complex relationship among species presence and spatial aspects
of forest structure and overall, they provide a strong starting point for the continued
development of biodiversity monitoring objectives.
Species that show strong avoidance of edge and selection for continuous stands are
likely to be negatively impacted by the current management practices in the study area and
may be species of concern. Species with strong relationships with certain structural elements
could be tested for their potential as indicator species. Within broader biodiversity
management objectives, such as ecosystem management and representation and single­
species management, the models developed in my research can complement a comprehensive
approach to biodiversity monitoring.
As much as the above aspects are strengths of my research, they also identify
important weaknesses of the approach. Presence and absence data are often criticised because
they do not provide information on important demographic parameters such as fecundity and
mortality (Magurran, 2003); however, these data are much more costly and time consuming
to collect. Using several different methodologies to collect data made it difficult to quantify
effort across methods for different taxa, and therefore, the potential of using another measure,

134

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

such as relative density was limited. Variables that described spatial context and models that
included variables measured at multiple scales were frequently identified as best models for
several species and groups of species (Chapter 3 and 4). This was not unexpected, but
because of characteristics of species, such as mobility and territory size that result in the
perception of landscapes at different scales (e.g., Vos et al., 2001), generalisations were
difficult across single-species models (Chapter 3) and limit the utility of the surrogate
approach. Finally, my focus on vertebrate species may be criticized because, overall,
vertebrates represent a small portion of species diversity. Indeed, future studies may benefit
from the examination of taxa that are more diverse, use structure at smaller scales, and are
not as flexible in their habitat selection (e.g., Redak, 2000). However, the availability of
standardized methodology, relatively easy identification, and potential “buy-in” of
stakeholders are benefits of using vertebrates. Further, in other studies, vertebrates have been
shown to be effective surrogates of species richness (e.g., Grelle, 2002).
Future research should examine presence-only modelling methodologies (e.g.,
ecological-niche factor analysis; Hirzel et al., 2002) that may account for differences in
methodologies and uncertainty in detections. An extension of the study area to include areas
that are not impacted by management activities may also help to elucidate important
relationships between species occurrence and structure, although in my study area this option
is limited because of extensive impacts of multiple users. Finally, an analysis of single taxon
or subset of a taxon (e.g., birds or neotropical migrants) may provide further insights into
management recommendations as birds, in particular, are often associated with specific
elements of forest structure (e.g., MacArthur and MacArthur, 1961; Hagen and Meehan,
2002; Diaz et al., 2005).

135

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Literature cited
Akaike, H., 1973. Information theory and an extension of the maximum likelihood principle.
In Petrov, B. N„ Caski, F. (Eds.), Proceedings of the Second International
Symposium on Information Theory. Budapest, Akademiai Kiado., pp. 267-281.
Allen, T.F.H., Hoekstra, T.W., 1992. Toward a unified ecology. University of Chicago Press.
Chicago, IL.
Angermeirer, M.J., Schlosser, I.J., 1995. Conserving aquatic biodiversity: beyond species
and populations. American Fisheries Society Symposium 17, 402-414.
Araujo, M. B., Humphries, C. J., Densham, P. J., Lampinen, R., Hagemeijer, W. J. M.,
Mitchell-Jones, A. J., Gasc, J. P., 2001. Would environmental diversity be a good
surrogate for species diversity? Ecography 24, 103-110.
Arrhenius, O., 1921. Species and area. Journal of Ecology 9, 95-99.
Barry, S., Elith, J., in press. Error and uncertainty in habitat models. Journal of Applied
Ecology.
Baxter, D.J.M., Psyllakis, J.M., Gillingham, M.P., O’Brien, E.L., in press. Behavioural
response of bats to perceived predation risk while foraging. Ethology.
Bayne, E.M., Hobson, K.A., 2000. Relative use of contiguous and fragmented boreal forest
by North American red squirrels (Tamiasciurus hudsonicus). Canadian Journal of
Zoology 78, 359-365.
Beals, E.W., 1984. Bray-Curtis ordination: an effective strategy for analysis of multivariate
ecological data. Advances in Ecological Research 14, 1-55.
Beauvais, G.P., Buskirk, S.W., 1999. An improved estimate of trail detectability for snowtrail surveys. Wildlife Society Bulletin 27, 32-38.
Beazley, K., Smandych, L., Snaith, T., Austin-Smith, P. Jr., Duinker, P., 2005. Biodiversity
considerations in conservation system planning: a map-based approach for Nova
Scotia, Canada. Ecological Applications 15, 2192-2208.
Berg, A., Ehnstrom, B., Gustafsson, L., Hallingback, T., Jonsell, M., Weslien, J., 1994.
Threatened plant, animal, and fungus species in Swedish forests: distribution and
habitat association. Conservation Biology 8, 718-731.
Blondel, J., 2003. Guilds or functional groups: does it matter? Oikos 100, 223-232.

13 6

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Boone, R.B., Krohn, W.B., 1999. Modelling the occurrence of bird species. Are the errors
predictable? Ecological Applications 9, 835-848.
Bowman, J., Sleep, D., Forbes, G. J., Edwards, M., 2000. The association of small mammals
and coarse woody debris at log and stand scales. Forest Ecology and Management
129,119-124.
Box, G.E.P., Tidwell, P.W., 1962. Transformation of the independent variables.
Technometrics 4, 531-550.
Boyce, M.S., Vernier, P.R., Nielsen, S.E., Schmiegelow, F.K.A., 2002. Evaluating resource
selection functions. Ecological Modelling 157, 281-300.
Breiman, L., Friedman, J.H., Olshen, R.A., Stone, C.J., 1984. Classification and Regression
Trees. Chapman and Hall, New York, NY.
Bryant, D., Nielsen, D., Tangley, L., 1997. The Last Frontier Forests: Ecosystems and
Economies on the Edge. World Resources Institute, Washington, DC.
Bull, E.L., 2002. The value of coarse woody debris to vertebrates in the Pacific Northwest.
USDA Forest Service. General Technical Report PSW-GTR-181, Portland, OR.
Bull, E.L., Holthausen, R.S., 1993. Habitat use and management of pileated woodpeckers in
northeastern Oregon. Journal of Wildlife Management 57, 335-345.
Bunnell, F.L., Huggard, D.J., 1999. Biodiversity across spatial and temporal scales: problems
and opportunities. Forest Ecology and Management 115, 113-126.
Bunnell, F.L., Kremsater, L.L., Wind, E., 1999. Managing to sustain vertebrate richness in
forests of the Pacific Northwest: relationships within stands. Environmental Reviews
7, 97-146.
Burnham, K.P., Anderson, D.R., 2002. Model selection and Inference. A Practical
Information-Theoretic Approach, second ed. Springer, New York, NY.
Canadian Council of Forest Ministers, 2003. Defining Sustainable Forest Management in
Canada: Criteria and indicators 2003. Canadian Forest Service, Natural Resources
Canada, Ottawa, ON.
Carey, A.B., Hardt, M.M., Horton, S.P., Biswell, B.L., 1991. Spring Bird Communities in the
Oregon Coast Range. In: Ruggiero, L.F., Aubry, K.B., Carey, A.B., Huff, M.H., (Eds).
Wildlife and vegetation of unmanaged Douglas-fir forests. General Technical Report
PNW-GTR-285, USDA, Portland, OR, pp 123-140.
Caro, T.M., Engilis, A. Jr., Fitzherbert, E., Gardner, T., 2004. Preliminary assessment of the
flagship species concept at a small scale. Animal Conservation 7, 63-70.

137

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Caro, T.M., O’Doherty, G., 1999. On the use of surrogate species in conservation biology.
Conservation Biology 13, 805-814.
Carter, G.M., Stolen, E.D., Breininger, D.R., 2006. A rapid approach to modeling specieshabitat relationships. Biological Conservation 127, 237-244.
Caswell, H., 1976. Community structure: a neutral model analysis. Ecological Monographs
46, 327-354.
Chatfield, C., 1995. Model uncertainty, data mining and statistical inference. Journal of the
Royal Statistical Society, Series A, 158, 419-466.
Cheveau, M., Drapeau, P., Imbeau, L. Bergeron, Y., 2004. Owl eruptions as an indicator of
small mammal population cycles in the boreal forest of eastern North America. Oikos
107, 190-198.
Clark, J.S., 2005. Why environmental scientists are becoming Bayesians. Ecology Letters 8,
2-14.
Clarke, K.R., 1993. Non-parametric multivariate analyses of changes in community structure.
Australian Journal of Ecology 18, 117-143.
Coleman, D.B., 1981. On random placement and species-area relations. Mathematical
Biosciences 54, 191-215.
Connell, J.H., 1978. Diversity in tropical rain forests and coral reefs. Science 199, 13021309.
Corben, C., O'Farrell, M.J., 1999. Techniques for the effective use of Anabat in identifying
free-flying bat species. Anabat System Manual. O’Farrell Biological Consulting, Las
Vegas, NV.
Cox, D.R., 1958. Two further applications of a model for binary regression. Biometrika 45,
562-565.
Crawley, M.J., Harral, J.E., 2001. Scale dependence in plant biodiversity. Science 291, 864868 .

Crump, M.A., Scott, N.J. Jr., 1994. Visual Encounter Surveys. In: Heyer, W.R., Donnelly,
M.A., McDiarmid, R.W., Hayek, L.C., Foster, M.S. (Eds.), Measuring and
monitoring biological diversity: standard methods for amphibians. Smithsonian
Institution Press, Washington, DC, pp. 84-92.
Cummins, K.W., 1974. Structure and function of stream ecosystems. Bioscience 24, 631-641.

138

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Davis, K.F., Margules, C.R., Lawrence, J.F., 2000. Which traits of species predict population
declines in experimental forest fragmentation? Ecology 81, 1450-1461.
De’ath, G., Fabricius, K.E., 2000. Classification and regression trees: a powerful yet simple
technique for ecological data analysis. Ecology 81, 3178-3192.
Deleo, J.M., 1993. Receiver operating characteristic laboratory (ROCLAB): software for
developing decision strategies that account for uncertainty. In: Proceedings of the
Second International Symposium on Uncertainty Modelling and Analysis, IEEE
Computer Society Press, College Park, MD, pp. 318-325.
Desrochers, A., Magrath, R.D., 1993. Environmental predictability and remating in European
Blackbirds. Behavioral Ecology 4, 271-275.
Diaz, I.A., Armesto, J .J ., Reid, S., Sieving, E., Willson, M.F., 2005. Linking forest structure
and composition: avian diversity in successional forests of Chiloe' Island, Chile.
Biological Conservation 123, 91-101.
Doligez, B., Danchin, E., Clobert, J., 2002. Public information and breeding habitat selection
in a wild bird population. Science 297, 1168-1170.
du Bus de Wamaffe, G., Dufrene, M., 2004. To what extent can management variables
explain species assemblages? A study of carabid beetles in forests. Ecography 27,
701-714.
Elith, J., Burgman, M., 2002. Predictions and their validation: rare plants in Central
Highlands, Victoria, Australia. In: Scott, J.M., Heglund, P.J., Morrison, M.L. (Eds).
Predicting Species Occurrences: Issues of Accuracy and Scale. Island Press,
Washington, DC, pp. 303-313.
Elton, C., 1927. Animal Ecology. Sidgwick and Jackson. London, UK.
Environment Canada., 2002. Canadian Climate Normals or Averages 1971-2000: Williams
Lake A Station. Available at
http://climate.weatheroffice.ec.gc.ca/climate_normals/index_e.html accessed
December 2006.
Environmental Systems Research Institute (ESRI). 2000. Arc View GIS Version 3.2a.
Environmental Systems Research Institute, Inc. Redlands, CA.
Faith, D.P., 2003. Environmental diversity (ED) as surrogate information for species-level
diversity. Ecography 26, 374-379.
Faith, D. P., Ferrier, S., Walker, P. A., 2004. The ED strategy: how species-level surrogates
indicate general biodiversity patterns through an 'environmental diversity' perspective.
Journal of Biogeography 31, 1207-1217.

139

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Faith, D.P., Minchin, P.R., Belbin, L., 1987. Compositional dissimilarity as a robust measure
of ecological distance. Vegetatio 69, 57-68.
Famden, C., Moss, I.S., Earle, T., 2003. Prototype field guide for identifying stand structure
classes in the Cariboo Forest Region. Lignum Limited, Williams Lake, BC.
Fauth, J.E., Bernardo, J., Camara, M., Resetarits, Jr., W.J., Van Buskirk, J., McCollum, S.A.,
1996. Simplifying the jargon of community ecology: a conceptual approach. The
American Naturalist 147, 282-286.
Fenton, M.B., Bell, G.P., 1981. Recognition of species of insectivorous bats by their
echolocation calls. Journal of Mammology 62, 233-243.
Fielding, A.H., 2002. What are appropriate characteristics of an accuracy measure? In: Scott,
J.M., Heglund, P.J., Morrison, M.L. (Eds). Predicting Species Occurrences: Issues of
Accuracy and Scale. Island Press, Washington, DC, pp. 271-280.
Fielding, A.H., Bell, J.F., 1997. A review of methods for the assessment of prediction errors
in conservation presence-absence models. Environmental Conservation 24, 38-49.
Fielding, A.H., Haworth, P.F., 1995. Testing the generality of bird-habitat models.
Conservation Biology 9, 1466-1481.
Fleishman, E., Mac Nally, R., Fay, J.P., 2002. Validation tests of predictive models of
butterfly occurrence based on environmental variables. Conservation Biology 17,
806-817.
Fleishman, E., Thomson, J.R., Mac Nally, R., Murphy, D.D., Fay, J.P., 2005. Using indicator
species to predict species richness of multiple taxonomic groups. Conservation
Biology 19, 1125-1137.
Forest Stewardship Council, 2005. Forest Stewardship Council Regional Certification
Standards for British Columbia: Main Standards. Available at:
http://www.fsccanada.Org/SiteCM/U/D/48B4F585905BF469.pdf accessed May 2006.
Fuller, M.R., Mosher, J.A., 1981. Methods of detecting and counting raptors: a review. In:
Ralph, C.J., Scott, J.M. (Eds.), Estimating Numbers of Terrestrial Birds Studies in
Avian Biology 6, pp. 235-246.
Gaines, W.L., Cushing, C.E., Smith, S.D., 1989. Trophic relations and functional group
composition if benthic insects in three cold desert streams. Southwest Naturalist 34,
478-482.
Gaston, K.J., Blackburn, T.M., 1996. Rarity and Body Size: Importance of Generality.
Conservation Biology 10, 1295-1298.

14 0

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Gillingham, M.P., 2003. Lifeforms, structure and biodiversity: species literature review.
March 2003. 651 pp. Available at http://web.unbc.ca/biodiversity/ accessed March
2006.
Gillingham, M.P., Parker, K.L., 2001. Summary report for the first year of the Project on
Lifeform Classification. Contribution agreement report to Lignum, Ltd. 10 Dec. 21 pp.
+ 337 pp. appendices. Available at http://web.unbc.ca/biodiversity/ accessed March
2006.
Gotelli, N.J., Graves, G.R., 1996. Null Models in Ecology. Smithsonian Institution Press.
Washington, DC.
Greaves, R.K., Sanderson, R.A., Rushton, S.P., 2006. Predicting species occurrence using
information-theoretic approaches and significance testing: an example of dormouse
distribution in Cumbria, U.K. Biological Conservation 130, 239-250.
Green, C.M., Stamps, J.A., 2001. Habitat selection at low population densities. Ecology 82,
2091-2100.
Grelle, C.E.V., 2002. Is higher-taxon analysis an useful surrogate of species richness in
studies of Neotropical mammal diversity? Biological Conservation 108, 101-106.
Grinnell, J., 1917. The niche relationship of the California thrasher. Auk 34, 427-433.
Groombridge, B., (Ed.), 1992. Global Diversity - Status of the Earth's Living Resources.
Compiled by the World Conservation Monitoring Centre. Chapman & Hall, London,
UK.
Groves, C.R., Jensen, D.B., Valutis, L.L., Redford, K.H., Shaffer, M.L., Scott, J.M.,
Baumgartner, J.V., Higgins, J.V., Beck, M.W., Anderson, M.G., 2002. Planning for
biodiversity conservation: putting conservation science into practice. BioScience 52,
499-512.
Gu, W., Swihart, R.K., 2004. Absent or undetected? Effects of non-detection of species
occurrence on wildlife-habitat models. Biological Conservation 116, 195-203.
Guisan, A., Zimmerman, N.E., 2000. Predictive habitat distribution models in ecology.
Ecological Modelling 135, 147-186.
Guthery, F.S., Brennan, L.A., Peterson, M.J., Lusk, J.J., 2005. Information theory in wildlife
science: critique and viewpoint. Journal of Wildlife Management 69, 457-465.
Hagen, J.M., Meehan, A.L., 2002. The effectiveness of stand-level and landscape-level
variables for explaining bird occurrence in an industrial forests. Forest Science 48,
230-241.

141

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Haila, Y., Nicholls, A.O., Hanski, I.K., Raivio, S., 1996. Stochasticity in bird habitat
selection: Year-to-year changes in territory in a boreal forest birds assemblage. Oikos
76, 536-552.
Hannon, S.J., McCallum, C., 2001. Using the focal species approach for conserving
biodiversity in landscapes managed for forestry. Sustainable Forest Management
Network Synthesis Paper. Available at
http://www.biology.ualberta.ca/faculty/susan_hannon/uploads/pdfs/white_paper_foca
l_spp.pdf accessed April 2006.
Hansen, A.J., Rotella, J.J., Kraska, M.P.V., Brown, D., 1999. Dynamic habitat and
population analysis: an approach to resolve the biodiversity manager’s dilemma.
Ecological Applications 9, 1459-1476.
Harper, J.L., Hawksworth D.L., 1995. Preface. In: Biodiversity Measurement and Estimation.
D.L. Hawksworth (Ed.), The Royal Society. Chapman and Hall. London, pp. 5-12.
Hayes, J.P., 1997. Temporal variation in activity of bats and the design of echolocationmonitoring studies. Journal of Mammology 78, 514-524.
Hebblewhite, M, White, C.A., Nietvelt, C.G., McKenzie, J.A., Hurd, T.E., Fryxell, J.M.,
Bayley, S.E., Paquet, P.C., 2005. Human activity mediates a trophic cascade caused
by wolves. Ecology 86, 2135-2144.
Heino, J., Muoka, T., Mykra, H., Paavola, R., Hamalainen, H., Koskenniemi, E., 2003.
Defining macroinvertebrate assemblage types of headwater streams: implications for
bioassessment and conservation. Ecological Applications 13, 842-852.
Hendrickx, J., 2001. Contrasts for categorical variables: update. Stata Technical Bulletin 59,
2-5.
Heyer, W.R., Donnelly, M.A., McDiarmid, R.W., Hayek, L.C., Foster, M.S. (Eds.), 1994.
Measuring and monitoring biological diversity: standard methods for amphibians.
Smithsonian Institution Press, Washington, DC.
Hirzel, A.H., Hausser, J., Chessel, D., Perrin, N., 2002. Ecological-niche factor analysis: how
to compute habitat-suitability maps without absence data? Ecology 83, 2027-2036.
Holt, R.D., 1987. Population dynamics and evolutionary processes: the manifold roles of
habitat selection. Evolutionary Ecology 1, 331-347.
Hosmer, D.W., Lemeshow, S., 2000. Applied Logistic Regression. 2nd eddition. J. Wiley
and Sons, Toronto, ON.

142

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Hunter, M.L. Jr., 1990. Wildlife, Forests, and Forestry: Principles of Managing Forests for
Biologic Diversity. Prentice-Hall. Englewood Cliffs, NJ.
Hunter, M.L. Jr., 1991. Coping with ignorance: the coarse-filter strategy for maintain
biodiversity. In: Kohm, K. (Ed.), Balancing the Edge of Extinction. Island Press,
Washington, DC, pp. 266-281.
Hunter, M.J. Jr., 1999. Maintaining biodiversity in forested ecosystems. Cambridge
University Press. Cambridge, UK.
Hunter, M.L. Jr., 2001. Fundamentals of conservation biology. 2nd ed. Blackwell Science,
Malden, MA.
Hunter, M.L. Jr., 2005. A mesofilter conservation strategy to complement fine and coarse
filters. Conservation Biology 19, 1025-1029.
Hutchinson, G.E., 1957. Concluding remarks. Cold Harbor symposium on quantitative
biology 22, 415-427.
Johnson, D.H., 2001. Validating and evaluating models. In: Shenk, T.M., Franklin, T.B.
(Eds.), Modeling in Natural Resource Management: development, interpretation, and
application. Island Press, Washington, DC, pp. 105-119.
Johnson, R.R., Brown, B.T., Haight, L.T., Simpson, J.M., 1981. Playback recording as a
special avian censusing technique. In: Ralph, C.J., Scott, J.M., (Eds.), Estimating
numbers of terrestrial birds. Cooper Ornithological Society Studies in Avian Biology
6, Allen Press, Lawrence, KS, pp. 68-75.
Jones, J., McShea, W. J., Conroy, M. J., Kunz, T. H., 1996. Capturing mammals. In: Wilson,
D.E., Cole, F.R., Nichols, J.D., Rudran, R., Foster, M.S., (Eds.), Measuring and
Monitoring Biological Diversity: Standard Methods for Mammals. Smithsonian
Institution Press, Washington, DC, pp. 115-155.
Jongman, R.H.G., ter Braak, C.J.F., van Tongeren, O.F.R., 1995. Data analysis in community
and landscape ecology. Cambridge University Press, Cambridge, UK.
Karels, T., Bryant, A.A., Hik, D.S., 2004. Comparison of discriminate function and
classification tree analyses for age classification of marmots. Oikos 105, 575-587.
Karl, J.W., Heglund, P.G., Garton, E.O., Scott, J.M., Wright, N.M., Hutto, R.L., 2000.
Sensitivity of species habitat-relationship model performance to factors of scale.
Ecological Applications 10, 1690-1705.
Kavanagh, R.P., Stanton, M.A., 2005. Vertebrate species assemblages and species sensitivity
to logging in the forests of north-eastern New South Wales. Forest Ecology and
Management 209, 309-341.

143

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Keisker, D.G., 2000. Types of wildlife trees and coarse woody debris required by wildlife of
north-central British Columbia. Working Paper. 50/2000. Research Branch, Ministry
of Forests, Victoria, BC.
Kimmins, J.P., 1997. Biodiversity and its relationship to ecosystem health & integrity.
Forestry Chronicle 73, 229-323.
Kraemer, H.C., 1988. Assessment of 2 x 2 associations: Generalisation of signal detection
methodology. The American Statistician 42, 37-49.
Kruskal, J.B., 1964a. Multidimensional scaling by optimizing goodness of fit to a nonmetric
hypothesis. Psychometrika 29, 1-27.
Kruskal, J.B., 1964b. Nonmetric multidimensional scaling: a numerical method.
Psychometrika 29, 115-129.
Kunin W.E., Gaston, K.J., 1993. The biology of rarity: patterns, causes, and consequences.
Trends in Ecology and Evolution 8, 298-301.
Kunz, T.H. (Ed.), 1988. Ecological and Behavioral Methods for the Study of Bats.
Smithsonian Institution Press, Washington, DC.
Lahde, E., Laiho, O., Norokorpi, Y., Saksa, T., 1999. Stand structure as the basis of diversity
index. Forest Ecology and Management 115, 213-220.
Lance, G.N., Williams, W.T., 1967. A general theory of classification sorting strategies. I.
Hierarchical systems. Computer Journal 9, 373-380.
Leibold, M.A., 1995. The niche concept revisited: mechanistic models and community
context. Ecology 76, 1371-1382.
Lindenmayer, D.B., 1999. Future directions for biodiversity conservation in managed forests:
indicator species, impact studies, and monitoring programs. Forest Ecology and
Management 115, 277-287.
Lindenmayer, D.B., Franklin, J.F., 1997. Managing stand structure as part of ecologically
sustainable forest management in Australian Mountain Ash Forests. Conservation
Biology 11, 1053-1068.
Lindenmayer, D.B., Franklin, J.F., 2002. Conserving Forest Biodiversity: a Comprehensive
multiscaled approach. Island Press, Washington, DC.

Lindenmayer, D.B., Margules, C.R., Botkin, D.B., 2000. Indicators of biodiversity for
ecologically sustainable forest management. Conservation Biology 14, 941-950.

144

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Liu, C., Berry, P.M., Dawson, T.P., Pearson, R.G., 2005. Selecting thresholds of occurrence
in the prediction of species distributions. Ecography 28, 385-393.
Loehle, C., Wigley, T.B., Shipman, P.A., Fox, S.F., Rutzmoser, S., Thill, R.E., Melchiors, M.
A., 2005. Herptofaunal species richness responses to forest landscape structure in
Arkansas. Forest Ecology and Management 209, 293-308.
Mac Nally, R. 2002. Improving inference in ecological research: issues of scope, scale, and
model validation. Comments on Theoretical Biology 7, 237-256.
Mac Nally, R., Fleishman, E., 2004. A successful predictive model of species richness based
on indicator species. Conservation Biology 18, 646-654.
Mac Nally, R., Parkinson, A., Horrocks, G., Conole, L., Tzaros, C., 2001. Relationships
between terrestrial vertebrate diversity, abundance and availability of coarse woody
debris on south-eastern Australian floodplains. Biological Conservation 99, 191-205.
MacArthur, R.H., MacArthur, J.W., 1961. On bird diversity. Ecology 42, 594-598.
MacKenzie, D.I., 2005. What are the issues with ‘presence/absence’ data for wildlife
managers? Journal of Wildlife Management 69, 849-860.
MacKenzie, D.I., Nichols, J.D., Lachman, G.B., Droege, S., Royle, J.A., Langtimm, C.A.,
2002. Estimating site occupancy rates when detection probabilities are less than one.
Ecology 83: 2248-2255.
Magurran, A.E., 2003. Measuring biological diversity. Blackwell Press, Oxford, UK.
Manel, S., Dias, J., Ormerod, S.J., 1999. Comparing discriminant analysis, neural networks
and logistic regression for predicting species distributions: a case study with
Himalayan river bird. Ecological Modelling 120, 337-347.
Manel, S., Williams, H.C., Ormerod, S.J., 2001. Evaluating presence-absence models in
ecology: the need to account for prevalence. Journal of Applied Ecology 38, 921-931.
Manley, P.N., Zielinski, W.J., Schlesinger, M.D., Mori, S.R., 2004. Evaluation of a multiplespecies approach to monitoring species at the ecoregional scale. Ecological
Applications 14, 296-310.
Margules, C.R., Pressey, R.L., 2000. Systematic conservation planning. Nature 405, 243-253.
Margules, C.R., Pressey, R.L., Williams, P.H., 2002. Representing biodiversity: data and
procedures for identifying priority areas for conservation. Journal of Biosciences
(suppl. 2) 27, 309-326.

145

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Martin, K., Eadie, J.M., 1999. Nest webs: A community-wide approach to the management
and conservation of cavity nesting forest birds. Forest Ecology and Management 115,
243-257.
Maser, C., Anderson, R.G., Cromack, K., Jr., Williams, J.T., Martin, R.E., 1979. Dead and
down material. In: J.W. Thomas (Ed.). Wildlife habitats in managed forests: the Blue
Mountains of Oregon and Washington. USDA Agricultural Handbook No. 553,
Portland, OR, pp. 78-95.
Mattson, D.J., and Reinhart, D.P., 1996. Indicators of red squirrel (Tamiasciurus hudsonicus)
abundance in the whitebark pine zone. Great Basin Naturalist 56, 272-275.
Mazurek, M.J., Zielinski, W.J., 2004. Individual legacy trees influence vertebrate wildlife
diversity in commercial forests. Forest Ecology and Management 193, 321-334.
McCune, B., Grace, J.B., 2002. Analysis of Ecological Communities. MjM Software
Design, Gleneden Beach, OR.
McCune, B., Mefford, M.J., 1999. PC-ORD. Multivariate analysis of ecological data. MjM
Software Design, Gleneden Beach, OR.
McElhinny, C., Gibbons, P., Brack, C., Bauhus, J., 2005. Forest and woodland stand
structural complexity: Its definition and measurement. Forest Ecology and
Management 218, 1-24.
McGarigal, K., Marks, B.J., 1995. FRAGSTATS: Spatial pattern analysis program for
quantifying landscape structure. General Technical Report PNW GTR-351. USDA,
Portland, OR.
McIntosh, R.P., 1995. H.A. Gleason’s “individualistic community concept” and theory of
animal communities: a continuing controversy. Biological Reviews 70, 317-357.
McPherson, J.M., Jetz, W., Rogers, D.J., 2004. The effects of species’ range sizes on the
accuracy of distribution models: ecological phenomenon or statistical artefact?
Journal of Applied Ecology 41, 811-823.
Meadows, D., 1995. Effects of habitat geometry on territorial defense costs in damselfish.
Animal Behaviour 49, 1406-1408.
Meidinger, D., Pojar, J. (Eds.), 1991. Ecosystems of British Columbia. Special Report Series.
6. BC Ministry of Forests, Victoria, BC.
Menard, S. W., 2002. Applied Logistic Regression Analysis. 2nd Edition. Sage University
Paper Series on Quantitative Applications in the Social Sciences, 07-106. Thousand
Oaks, CA.

146

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Miller, M.E., Hui, S.L., Tierney, W.M., 1991. Validation techniques for logistic regression
models. Statistics in Medicine 10, 1213-1226.
Minister of Supply and Services Canada, 1995. Canadian Biodiversity Strategy: Canada’s
Response to the Convention on Biological Diversity. Catalogue No. En21-134/1995E,
Hull, QC.
Mitchell, S.C., 2005. How useful is the concept of habitat? - a critique. Oikos 110, 634-638.
Montreal Process Working Group, 1999. The Montreal Process: Criteria and indicators for
the conservation and sustainable management of temperate and boreal forests, Second
edition. Montreal Processes Liaison Office, Canadian Forest Service, Natural
Resources Canada, Ottawa, ON.
Morrison, M.L., 2001. A proposed research emphasis to overcome the limits of wildlifehabitat relationship studies. Journal of Wildlife Management 65, 613-623.
Morrison, M.L., Hall, L.S., 2002. Standard terminology: toward a common language to
advance ecological understanding and application. In: Scott, J.M., Heglund, P.J.,
Morrison, M.L. (Eds.), 2002. Predicting Species Occurrences: Issues of Accuracy and
Scale. Island Press, Washington, DC, pp. 43-52.
Morrison, M.L., Marcot, B.G., Mannan, R.W., 1998. Wildlife-habitat relationships: concepts
and applications. 2nd Ed. The University of Wisconsin Press, Madison, WI.
Moss, I.S., 2002. Lignum Limited stand structure key, second approximation, 17 stand
structure classes. Report to Lignum Ltd., Williams Lake, BC.
Myers, N., Mittermeier, R.A., Mittermeier, C.G, da Fonseca, G.A.B., Kent, J., 2000.
Biodiversity hotspots for conservation priorities. Nature 403, 853-858.
Nagelkerke, N.J.D., 1991. A note on general definition of the coefficient of determination.
Biometrika 78, 691-692.
Naiman, R. J., Decamps, H., 1997. The ecology of interfaces: riparian zones. Annual Review
of Ecology and Systematics 28, 621-658.
Neter, J., Wasserman, W., Kutner, M.H., 1985. Applied linear statistical models: Regression,
analysis of variance, and experimental designs. Irwin Pub, Homewood, IL.
Noon, B.R., Murphy, D.D., Beissinger, S.R., Shaffer, M.L., Dellasala, D., 2003.
Conservation planning for US National Forests: conducting comprehensive
biodiversity assessments. BioScience 53, 1217-1220.

147

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Norse, E.A., Rosenbaum, K.L. Wilcowe, B.A., Wilcox, B.A., Romme, W.H., Johnston,
D.W., Stout, M.L., 1986. Conserving biological diversity in our national forests. The
Wilderness Society, Washington, DC.
Noss, R.F., 1990. Indicators for monitoring biodiversity: A hierarchical approach.
Conservation Biology 4, 355-364.
Noss, R.F., 1996. Ecosystems as conservation targets. Trends in Ecology and Evolution 11,
351.
Noss, R.F., 1999. Assessing and monitoring forest biodiversity: a suggested framework and
indicators. Forest Ecology and Management 115, 135-146.
Noss, R.F., Cooperrider, A.Y., 1994. Saving Nature’s Legacy. Island Press, Washington, DC.
Obbard, M.E., 1987. Red squirrel. In: Novak, M., Baker, J.A., Obbard, M.E., Mallock, A.M.
(Eds.) Wild furbearer management and conservation in North America. Ontario
Trappers Association, Ontario Ministry of Natural Resources, Queen’s Printer,
Toronto, ON, pp. 264-281.
O'Farrell, M.J., Gannon, W.L., 1999. A comparison of acoustic versus capture techniques for
the inventory of bats. Journal of Mammology 80, 24-30.
Olden, J.D., Jackson, D.A., 2002. Illuminating the “ black box’’: Understanding variable
contributions in artificial neural networks. Ecological Modelling 154, 135-150.
Onal, H., 1997. Trade-off between structural diversity and economic objectives in forest
management. American Journal of Agricultural Economics 79, 1001-1012.
Ottaviani, D., Lasinio, G.J., Boitani, L., 2004. Two statistical methods to validate habitat
suitability models using presence-only data. Ecological Modelling 179, 417-443.
Oxbrough, A.G., Gittings, T., O’Halloran, J., Giller, P.S., Smith, G.F., 2005. Structural
indicators of spider communities across the forest plantation cycle. Forest Ecology
and Management 212, 171-183.
PCI Geomatics Corporation, 2001. PCI Works Version 7.0. Richmond Hill, ON.
Pearce, J., Ferrier, S., 2000. Evaluating the predictive performance of habitat models
developed using logistic regression. Ecological Modelling 133, 225-245.
Poiani, K.A., Richter, B.D., Anderson, M.G., Richter, H.E., 2000. Biodiversity conservation
at multiple scales: Functional sites, landscapes, and networks. BioScience 50, 133146.

148

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Province of British Columbia, 1991. Resource Inventory Committee Standards. Available at
http://srmwww.gov.bc.ca/risc/standards.htm accessed December 2005.
Province of British Columbia, 1995. Biodiversity Guidebook (Forest practices code of
British Columbia). Ministry of Forests, Government of British Columbia. Available
at http://www.for.gov.bc.ca/tasb/legsregs/fpc/fpcguide/biodiv/biotoc.htm accessed
March 2006.
Province of British Columbia, 1998. Inventory Methods for small mammals: shrews, voles,
mice and rats. Standards for Components of British Columbia’s Biodiversity, No. 31.
Resources Inventory Committee. Available at
http://ilmbwww.gov.bc.ca/risc/pubs/tebiodiv/smallmammals/index.htm accessed
December 2006.
Province of British Columbia, 1999a. Landscape Unit Planning Guide. Ministry of Forests,
Forest Practices Branch Operations Division and Ministry of Environment, Lands and
Parks Resource Stewardship Branch. Victoria, BC.
Province of British Columbia, 1999b. Inventory Methods for forest and grassland songbirds.
Standards for Components of British Columbia’s Biodiversity, No. 15. Resources
Inventory Committee. Available at
http://ilmbwww.gov.bc.ca/risc/pubs/tebiodiv/songbird/index.htm accessed December
2006.
Province of British Columbia, 2000. Overview Aerial Survey Standards for British Columbia
and the Yukon. Resource Inventory Committee. Available at:
http://srmwww.gov.bc.ca/risc/pubs/teveg/foresthealth/index.htm accessed December
2005.
Province of British Columbia, 2002. Vegetation Resource Inventory (VRI) Update. Ministry
of Sustainable Resources Management. Victoria, BC.
Purvis, A., Hector, A., 2000. Getting the measure of biodiversity. Nature 405, 212-219.
Ralph, C.J., Geupel, G.R., Pyle, P., Martin, T.E., DeSante, D.F., 1993. Handbook of field
methods for monitoring landbirds. U.S. Department of Agriculture, Forest Service
General Technical Report PSW-GTR-144, Albany, CA.
Ralph, C.J., Sauer, J.R., Droege, S., 1995. Monitoring bird populations by point counts. U.S.
Department of Agriculture, Forest Service General Technical Report PSW-149,
Portland, OR.
Redak, R.A., 2000. Arthropods and multispecies habitat conservation plans: are we missing
something? Environmental Management Vol. 26, Supplement 1, S97-S107.

149

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Rempel, R.S., Carr, A.P., 2003. Patch Analyst extension for ArcView: version 3. Available at
http://flash.lakeheadu.ca/~rrempel/patch/index.html accessed March 2006.
Reyers, B., Van Jaarsveld, A.S., Kruger, M., 2000. Complementarity as a biodiversity
indicator strategy. Proceedings of the Royal Society of London B series 267, 505-513.
Reynolds, R.T., Scott, J.M., Nussbaum, R.A., 1980. A variable circular-plot method for
estimating bird numbers. Condor 82, 309-313.
Ricketts, T.H., Daily, G.C., Ehrlich, P.R., 2002. Does butterfly diversity predict moth
diversity? Testing a popular indicator taxon at local scales. Biological Conservation
103,361-370.
Robbins, C.R., 1981. Bird activity levels related to weather. In: Ralph, C. J., Scott, J. M.
(Eds.). Estimating numbers of terrestrial birds. Cooper Ornithological Society Studies
in Avian Biology 6, Allen Press, Lawrence, KS, pp. 301-310.
Root, K.V., Akgakaya, H.R., Ginzburg, L., 2003. A multispecies approach to ecological
valuation and conservation. Conservation Biology 17, 196-206.
Root, R.B., 1967. The niche exploitation pattern of the blue-gray gnatcatcher. Ecology 37,
317-350.
Rosenzweig, M.L., 2003. Win-Win Ecology: How the Earth's Species Can Survive in the
Midst Of Human Enterprise. Oxford University Press, New York, NY.
Rushton, S.P., Omerod, S.J., Kerby, G., 2004. Special profile: New paradigms for modelling
species-distributions? Journal of Applied Ecology 41, 193-200.
Rykiel, E.J., Jr., 1996. Testing ecological models: the meaning of validation. Ecological
Modelling 90, 229-244.
Saab, V., 1999. Importance of spatial scale to habitat use by breeding birds in riparian forests:
a hierarchical analysis. Ecological Applications 9, 135-151.
Sarakinos, H., Nicholls, A.O., Tubert, A., Aggarwal, A., Margules, C.R., Sarkar, S., 2001.
Area prioritization for biodiversity conservation in Quebec on the basis of species
distributions: a preliminary analysis. Biodiversity and Conservation 10, 1419-1472.
Sarkar, S., Margules, C., 2002. Operationalizing biodiversity for conservation planning.
Journal of Bioscience 27, 299-308.
Schulze,C.H., Waltert, M., Kessler, P.J.A., Pitopang, R., Shahabuddin, Veddeler, D.,
Muhlenberg, M., Gradstein, S.R., Leuschner, C., Steffan-Dewenter, I., Tschamtke, T.,
2004. Biodiversity indicator groups of tropical land-use systems: comparing plants,
birds and insects. Ecological Applications 14, 1321-1333.

150

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Scott, J.M., Heglund, P.J., Morrison, M.L. (Eds.), 2002. Predicting Species Occurrences:
Issues of Accuracy and Scale. Island Press, Washington, DC.
Scott, N.J., Jr., 1994. Complete species inventories. In: Heyer, W., Donnelly, M.,
McDiarmid, R., Hayek, L., Foster, M. (Eds.), Measuring and Monitoring Biological
Diversity: Standards and Methods for Amphibians, Smithsonian Institute Press,
Washington, DC, pp. 78-84.
Selker, H.P., Griffith, J.L., Patil, E., 1995. A comparison of performance of mathematical
predictive methods for medical diagnosis: identifying acute cardiac ischemia among
emergency department patients. Journal of Investigative Medicine 43,468-476.
Seoane, J., Carrascal, L.M., Alonso, C.L., Palomino, S., 2005. Species-specific traits to
prediction errors in bird habitat suitability modelling. Ecological Modelling 185, 299308.
Sergio, F., Newton, I., Marchesi, L., 2005. Conservation: Top predators and biodiversity
Nature 436, 192-192.
Shepard, R.N., 1962a. The analysis of proximities: multidimensional scaling with an
unknown distance function. I. Psycometrika 27, 125-139.
Shepard, R.N., 1962b. The analysis of proximities: multidimensional scaling with an
unknown distance function. II. Psycometrika 27, 219-246.
Shurin, J.B, Allen, E.G., 2001. Effects of competition, predation, dispersal on local
and regional species richness. American Naturalist 158, 624-637.
Simberloff, D., 1998. Flagships, umbrellas, and keystones: is single species management
passe in the landscape era? Biological Conservation 83, 247-257.
Simberloff, D., 1999. The role of science in the preservation of forest biodiversity. Forest
Ecology and Management 115, 101-111.
Soule, M. E., Terborgh, J., 1999. Continental conservation: scientific foundations of regional
reserve networks. Island Press, Washington, DC.
Spence, R.J., 2001. The new boreal forestry: adjusting timber management to accommodate
biodiversity. Trends in Ecology and Evolution 16, 591-593.
StataCorp., 2003. Stata Statistical Software: Release 8. StataCorp LP., College Station, TX.
StatSoft, Inc., 2003. Statistica (data analysis software system), version 6. StatSoft Inc., Tulsa,
OK.

151

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Staudhammer, C.L., LeMay, V.M., 2001. Introduction and evaluation of possible indices of
stand structural diversity. Canadian Journal of Forest Research 31,1105-1115.
Stelfox, J.B., 1995. Introduction. In: Relationships between stands age, stand structure, and
biodiversity in Aspen mixedwood forests in Alberta. Alberta Environmental Centre
and Canadian Forest Service, Edmonton, AB, pp. 1-12.
Stephens, P.A., Sutherland, W.J., 1999. Consequences of the Allee effect for behaviour,
ecology and conservation. Trends in Ecology and Evolution 14, 401-405.
Stone, J.N., Porter, J.L., 1998. What is forest stand structure and how to measure it?
Northwest Science 72, 25-26.
Sustainable Forestry Initiative, 2004. Sustainable forestry initiative standard: 2005-2009
standard. Available at: http://www.aboutsfb.org/generalPDFs/SFBStandard20052009.pdf accessed May 2006.
Tabachnik, B.G., Fidell, L.S., 2001. Using Multivariate Statistics. 4th ed. Allyn and Bacon,
Boston, MA.
Thomas, D. W., Bell, G. P., Fenton, B. B., 1987. Variation in echolocation call frequencies
recorded from North American vespertilionid bats: a cautionary note. Journal of
Mammalogy 68, 842-847.
Tyre, A.J., Possingham, H.P., Lindenmayer, D.B., 2001. Matching observed pattern with
model process: can territory occupancy provide information about life history
parameters. Ecological Applications 11, 1722-1737.
Tyre, A.J., Tenhumberg, B., Field, S.A., Niejalke, D., Parris, K., Possingham, H.P., 2003.
Improving precision and reducing bias in biological surveys by estimating false
negative error rates in presence-absence data. Ecological Applications 13, 1790-1801.
United Nations Environment Programme, 1992. Convention on Biological Diversity. Rio de
Janeiro, Brazil, 3-14 June. United Nations Department of Public Information, New
York, NY.
Van Home, B., 2002. Approaches to habitat modelling: the tensions between patterns and
process and between specificity and generality. In: Scott, J.M., Heglund, P.J.,
Morrison, M.L. (Eds.), Predicting Species Occurrences: Issues of Accuracy and
Scale. Island Press, Washington, DC, pp. 63-72.
Vaughan, I. P., Ormerod, S. J., 2005. The continuing challenges of testing species
distribution models. Journal of Applied Ecology 42, 720-730.

15 2

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Vayssieres, M.P., Plant, R.E., Allen-Diaz, B.H., 2000. Classification trees: an alternative
non-parametric approach for predicting species distributions. Journal of Vegetation
Science 11, 679-694.
Verbyla, D.L., Litaitis, J.A., 1989. Resampling methods for evaluation of classification
accuracy of wildlife habitat models. Environmental Management 13, 783-787.
Vos, C.C., Verboom, J., Opdam, P.F.M, Ter Braak, C.J.F., 2001. Toward ecologically scaled
landscape indices. The American Naturalist 183, 24-41.
Walters, C.J., 1986. Adaptive Management of Renewable Resources. Macmillan, New York,
NY.
Wiens, J.A., 1989. Spatial scaling in ecology. Functional Ecology 3, 385-397.
Willis, K.J., Whittaker, R.J., 2002. Species diversity: scale matters. Science 295, 1245-1248.
Wilson, D.E., Cole, F.R., Nichols, J.D., Rudran, R., Foster, M.S. (Eds.), 1996. Measuring and
Monitoring Biological Diversity, Standard Methods for Mammals. Smithsonian
Institution Press, Washington, DC.
Wilson, J.B., 1999. Guilds, functional types, and ecological groups. Oikos 86, 507-522.
World Commission of Forests and Sustainable Development, 1999. Our forests, our future:
report of the World Commission on Forests and Sustainable Development.
Cambridge University Press, New York, NY.
Worth A.P., Cronin, M.T.D., 2003. The use of discriminant analysis, logistic regression and
classification tree analysis in the development of classification models for human
health effects. Journal of Molecular Structure 622, 97-111.
Zabel, C.J., Anthony, R.G., (Eds.), 2003. Mammal Community Dynamics Management and
Conservation in the Coniferous Forests of Western North America. Cambridge
University Press, NY.
Zar, J.H., 1999. Biostatistical Analysis, Fourth Edition. Prentice-Hall Inc., Upper Saddle
River, NJ.
Zenner, E.K., Hibbs, D.E., 2000. A new method for modeling the heterogenity of forest
structure. Forest Ecology and Management 129, 75-87.
Zielinski, W.J., Kucera, T.E., 1995. American Marten, Fisher, Lynx and Wolverine: Survey
methods for their detection. USDA Forest Service, General Technical Report PSWGTR-157. Portland, OR.

153

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Zimmerman, G.M., Goetz, H., Mielke, Jr., P.W., 1985. Use of an improved statistical method
for group comparisons to study the effects of prairie fire. Ecology 66, 606-611.

154

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Appendix I. Explanation for electronic files and data CD.
This electronic appendix includes a menu structure that accesses a literature review
for all vertebrate species expected to occur in the Williams Lake Study Area, detection
summaries for each species by year and, for each species that had sufficient detections to
model, candidate variables used in models to predict presence and absence, model results,
and validation results. The literature review (Gillingham, 2003) was mostly complete at the
outset of this study, although I added and contributed to several species. The goal of this
literature review was to identify key structural and non-structural (biological) requirements
of species expected to occur in the study area. The literature review was used in the
development of a priori groupings of species based on structural associations, as part of a
parallel study (Gillingham and Psyllakis, 2004; see original lifeform groupings file; Figure
1.1). I used this literature review extensively to determine candidate models and ran all
subsequent analyses. It is included here for completeness.
To access the files, select the “index.html” file to load the menu structure. It may take
a minute to completely load in your browser. Once loaded, an expandable menu structure
will appear (Figure 1.1) that lists each species, in taxonomic order. The literature cited is also
available at any time by clicking the appropriate location in the header. Selecting the folder
for the species of interest expands the text of the heading and displays the available files
(Figure 1.2). Using the Red-naped Sapsucker (Sphyrapicus nuchalis) as an example, several
additional tables are presented that explain each available file for species that were modelled
in the order displayed in Figure 1.2: Files included in the “General” folder are: Literature
Review (Table 1.1), Detection Summary (Table 1.2), and Grouping (Table 1.3). Files included
in the “Models” folder include: Candidate Variables (Table 1.4), Model Selection Results
(Table 1.5), Best Model(s) (Table 1.6), and Validation (Table 1.7).

A p p e n d ix I
Click on th e folder* to expend end collapse them and on the te x t o f heading you can view species-specific html files. Because the number o f
detections a ffe c te d our a b ility to analyze each species, not all html files a re available fo r every species. To open the lite ra tu re c ite d in a sep a ra te
window click here.

Ei _J I n d i v i d u a l S p e c i e s
E l _ L ] A m p h to la n s

E

ffi

* iR e p t i l e s

3

_ L ) B ird s

3

_ jM a m m a ls

j j o r i g h a l L ife fb rm G ro u p in g s

Figure 1.1. Screen capture of the menu structure included in Appendix I accessed by
selecting the index file on the attached data CD. Literature cited can be accessed from this
location at any time by clicking the appropriate link in the header.

155

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

ED _ ) Vauxs Swift
El _ ) Black-chinned Hummingbird
E l | Annas Hummingbird
E l | Calliope Hummingbird
E l I Rufous Hummingbird
ED
Belted Kingfisher
El _ j Lewis Woodpecker
El _ j| Red-naped Sapsucker
ED General
|S [ Literature Review
H i) Detection Summary
IP!) Grouping
□

._*| Models
|S ) Candidate Variables

|S) Model Selection Results
|S | Best Model(s)
|S) Validation

El _ _ | Williamsons Sapsucker
El _ ] Red-breasted Sapsucker
El _ J Downy Woodpecker

Figure 1.2. Screen capture of a portion of the bird species expected to occur in the Williams
Lake Study Area. Clicking on a species will display available files. In the “General” folder,
files include the complete literature review, number of detections, and a priori grouping.
Under the “Model” folder, files include the listing of the candidate variables used in the
competing model sets, the results of model selection, the explanation of the coefficients for
the top models, and the temporal and spatial validation results of those top models. Not all
species will have model files as several species were not sufficiently detected to model.

156

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Table 1.1. Using Red-naped Sapsucker (Sphyrapicus nuchalis) as an example, information
included with the literature review for each species expected to occur in the Williams Lake
Study Area is included on the data CD.1
Red-naped Sapsucker (Sphyrapicus nuchalis)

F e e d in g R e q u ir e m e n ts
H a b ita ts

SS 3-6: 6H Primary cavity nester often adj. to water in live/dead trees
(esp. deciduous). PCN.
Tree boles in deciduous/mixed woodlands near water.
Upland and Riparian forests.

O th e r C ritic a l
R e q u ire m e n ts /C o m m e n ts

NA

P ro v in c ia l S t a t u s / E L P

R e g io n a l/L o c a l/P ro v in c ia l

Breeds in B.C.
(1) Clutch dates 6 M ay-16 June. Clutch size 2. Young dates 14 M ay-4
Aug.
(1) Arrives in the Interior in late Mar.; main spring movem ent occurs
through Apr.; autumn migration begins late Aug. - mid Sept.; a few birds
documented into Oct.
(1) N o Interior winter records, but 3 coastal.
(1) Widespread breeder across central southern and southeastern BC
north to Yoho National Park and through the Chilcotin-Cariboo basin;
rarely further north.
(1) W idely distrib. across south. BC E o f the Pacific and Cascade ranges,
N through Chilcotin-Cariboo basin and Nechako plateau; rarely to Nulki
L. in Nechako lowlands; wanders irreg. W o f Pacific and Cascade ranges
and has been reported from the Fraser 1

S p e c ia lis t o r G e n e ra lis t

Generalist.

I n v e n to r y L is t
G e n e ra l N o n s tr u c tu r a l
In fo rm a tio n

S p e c ie s M ig ra to ry
S p e c ie s R e s id e n t

B r e e d s in A re a

P rin c ip le
P re d a to rs /P a ra s ite s

(3) w easels on fledged young.

P rin c ip le P re y /F o o d

(5) insects
NA
(1) may nest in trees inhabited by Pileated woodpecker, Northern flicker,
European starling.

P rin c ip le C o m p e tito r s

L in k s to o th e r s p e c i e s
T e rrito ria l S iz e
(E x c lu s iv e /O v e rla p p in g )

G e n e ra l R e q u ire m e n ts
V e g e ta tio n T y p e

V e g e ta tio n S p e c ie s
P re fe rre d

NA
(1) sea- 1300m elevation; (4) trees infected with disease that have been
weakened inside and out: weak excavators.
(1)(4) deciduous and mixed woodlands where poplar and birch are
common (M cGillivray & Sem enchuck 1998 in (4)).
(1)(4) aspen groves in open P. pine forests/aspen fir parkland/logged
forests where decid. groves remain/ aspen groves in open
rangeland/birch groves/subalpine forest edges/residential gardens; nests
in decid. trees; T. aspen/birches/poplar/B. cottonwood/

C o v er T ype / L evel
N eeded

Food

NA
(5) cambium, fruit, berries, pine pitch often used instead o f deciduous
sap.

1 Literature cited included in the literature review can be found on the CD.

157

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Table 1.1. Literature review for the Red-naped Sapsucker (continued).
T re e D ia m e te r C la s s
(D B H )
C o u r s e W o o d y D e b ris
D e c a y C la s s
S te m

D e n s ity

S ta n d A g e /T re e A g e
B a sa l A rea
U n d e rs to ry

(1) 15-64cm nesting tree dbh; (6) >/= 38 cm roosting tree.
NA
NA
NA
NA
NA
NA

D e a d fa lls S tu m p s
S n a g s e tc

(6) snag dbh for both nesting and roosting >/= 38 cm.

W a te r (L a k e s , S tr e a m s ,
R ip a ria n )
H a b ita ts A v o id e d

NA
NA

T y p e , s iz e /d e p th o r
o th e r c h a ra c te rs
needed
W a te r R e q u ir e m e n t
Y e a r - r o u n d ( Y /N )

N est T ype

H a b ita t R e q u ir e m e n ts

NA
for breeding.
NA
(1) cavity nester; nest heights .5-22.9m; cavity entrance hole
diameters 10-17cm (3) first cavity in unexcavated trees at low height;
in subsequent years new nests are excavated above the last on the
same tree.
(1) edge o f woodlands adjacent to water bodies such as streams,
ponds, sloughs, lakes, road edges, logging slashes, transmission line
rights-of-way, mountain m eadows (3) trees susceptible to heart rot
(Keisker 1987 in (3)); preference for trees that show

S n a g s /C a v itie s
re q u ire d
F o re s t C o v e r ty p e
O th e r B re e d in g N e e d s
W in te r F o o d
W in te r C o v e r

NA
NA
NA
NA
NA

O th e r W in te r
R e q u ire m e n ts

M an ag em en t Issu e s

R e fe re n c e s U sed
(N u m b e r)

NA
(1) In BC the geographical and biological relationships between
nuchalis and ruber are complex; many areas where they meet, nest
side by side, and interbreed.
(1) Campbell et al. 1990 (18); (2) Keisker 1987 (1012); (3) Daily
1993 (1034); (4) McClelland & McClelland 2000 (1035); (5) Ehrlich
et al. 1998 (2037); (6) BC Ministry o f Forests 2001 (3719)

158

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Table 1.2. Example of the Detection Summary file for the Red-nap Sapsucker. This file
exists for all species detected in the Williams Lake Study Area from 2001-2004.
Detection summary for Red-naped Sapsucker (Sphyrapicus nuchalis)
Year
2001
2002
2003
2004
All

Total
Detections
41
99
137
84
Years

Plot/Transect
Detections
24
45
64
35
All

Table 1.3. Example of the lifeform assignment table for the Red-naped Sapsucker found on
the data CD. Lifeform groupings were part of a concurrent study and not discussed in this
dissertation. The information is included for completeness.
Lifeform assignment for Red-naped Sapsucker (Sphyrapicus nuchalis )
This species was assigned to 1 Lifeform grouping(s)
Lifeform 8.0: Standing dead, dying trees and forages in trees, forest openings or edges;
Requires dead or dying trees for nesting, denning, perching, or foraging' 'Includes
species that excavate their own cavities as well as those that use secondary cavities and
natural cavities'
A full description of each lifeform grouping and their member species can be found by
clicking on the ‘Original Lifeform Grouping’ at the bottom of the folder list.

159

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Table 1.4. Example of Candidate Variables for the Red-naped Sapsucker included on the
data CD. Candidate variables were selected and used to construct competing models for all
species that were detected in the Williams Lake Study Area and met prevalence criteria
(>10% and <90% occurrence).
Candidate Variables for examining presence/absence of Red-naped Sapsucker
(Sphyrapicus nuchalis)
Plot-Level Variables

Vegetation Resource
Inventory (VRI) Variables
Stand-Level Variables

Landscape-Level Variables

percent gap, percent at cover, deciduous stems per ha,
live basal area, dead basal area, large basal area,
distance from forest edge
aspen cover, stand structure class, crown closure,
leading species height, live basal area
percent gap, percent at cover, deciduous tree stems per
ha, dead basal area, live basal area, large tree basal
area, stand distance from forest edge
interspersion juxtaposition index 50 ha, mean patch
size 50 ha, edge density 50 ha, mean core area 50 ha,
proportion conifer 50 ha, proportion aspen shrub 50 ha

16 0

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Table 1 .5. Example of the Model Selection Result file included on the data CD for the Rednaped Sapsucker. Models with AIC < 2.0 are reported. Tables I.5A, I.5B, I.5C include the
results for models using plot, Vegetation Resource Inventory (VRI), and stand structural data
models, respectively.
Table 1.5A. Results from plot models for the Red-naped Sapsucker.
Model
K N ROC
LL
AICc
AAIC
wi
Evidence
_____________________________________________________________________Ratio
Percent canopy gap
4 228 0.753 -133.015 274.139 0.000 0.359 1.000
Distance to forest edge
Proportion aspen and
shrub (50 ha)*
Percent canopy gap
Distance to forest edge
Proportion aspen and
shrub (50 ha)*
Natural stump / ha

5 228

0.750 -132.761 275.703

1.564

0.164 2.186

Distance to forest edge
Proportion aspen and
shrub (50 ha)*

3 228

0.745 -134.827 275.707

1.568

0.164 2.190

Distance to forest edge
Proportion aspen and
shrub (50 ha)*
Natural stump / ha

4 228

0.743 -133.903 275.913

1.774

0.148 2.428

* arcsine transformation

161

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Table I.5B. Results from Vegetation Resource Inventory (VRI) models for the Red-naped
Sapsucker.
Note: stand structure classes 6, 8, 13 not used due to complete separation

K

N

ROC

AAIC

wi

0.603 -55.706

115.463 0.000

0.175

Evidence
Ratio
1.000

Sum of crown closure

2

83

Aspen cover

2

83

0.599 -56.052

116.153

0.691

0.124

1.413

Aspen cover
Sum of crown closure

3

83

0.687 -55.188

116.529

1.066

0.103

1.704

Aspen cover
Leading species height

3

83

0.655 -55.227

116.606

1.143

0.099

1.771

Live basal area

2

83

0.538 -56.389

116.827

1.365

0.089

1.979

Sum of crown closure
Leading species height

3

83

0.621 -55.521

117.195

1.732

0.074

2.378

Model

LL

AICc

Table I.5C. Results from stand-structure data models for the Red-naped Sapsucker.

K

N

Percent Aspen
Edge density (50 ha)

3

Edge density (50 ha)

2

Model

ROC

LL

AICc

AAIC

wi

97

0.716 -56.393 118.914 0.000

0.313

Evidence
Ratio
1.000

97

0.704 -57.487 119.017 0.102

0.297

1.052

162

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Table 1.6. Example of the Best Model file for the Red-naped Sapsucker. The P-coefficeints
and odds ratios are reported for each variable included in the top model (i.e., lowest AICc) for
plot (Table I.6A), Vegetation Resource Inventory (Table I.6B), and stand models(Table I.6C).
Table I.6A. Top logistic regression model that best discriminated between presence and not
detected at the plot sceale for the Red-naped Sapsucker.

Variable
Percent canopy gap

1.251

S.E.
0.637

Z
1.960

p-value
0.050

95% C.I.
Odds
Ratio Lower Upper
3.494 0.949
12.862

Distance to forest edge

-0.007

0.003

-2.700

0.007

0.993

0.988

0.997

Proportion aspen and
shrub (50 ha)*

3.413

1.296

2.630

0.008

30.363

2.232

413.075

Constant

-1.398

0.648

-2.160

0.031

P

* arcsine transformed

Table I.6B. Top logistic regression model that best discriminated between presence and not
detected at the stand scale using Vegetation Resource Inventory data for the Red-naped
Sapsucker.

95% C.I.
Variable_________ (3______S.E._____ z
Sum of crown
-0.019
0.015 -1.238
closure

p-value
0.216

1.430

0.147

Constant

0.986

1.450

Odds
Ratio
0.981

Lower
0.956

Upper
1.008

163

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Table I.6C. Top logistic regression model that best discriminated between present and not
detected at the stand scale using stand level data for the Red-naped Sapsucker.
95% C.I.
Variable
Percent Aspen
Edge density (50
ha)
Constant

0.041
0.012

S.E.
0.030
0.003

1.369
3.571

Pvalue
0.171
<0.000

-2.371

0.783

-3.030

0.002

P

z

Odds
Ratio
1.042
1.012

Lower Upper
0.981
1.106
1.005
1.019

Table 1.7. Example of Validation Files included on the data CD for the Red-naped Sapsucker.
Files are included for those species that had sufficient detections in 2004 at the spatial or
temporal validation plots to test models at either, or both, the plot and stand scale. For spatial
validation models, P -coefficients, p-value, and odds ratio are reported for each variable
included in the model, as well as the ROC values for the model constructed with
development data and its spatially validated counterpart. For temporal validation tables the pcoefficient for each variable is reported using a single year’s data for all years, including
2004, as well as ROC values for those models.
Table I.7A. Spatial validation of the best plot-level model for the Red-naped Sapsucker.

1.251

2001-2003
Odds
Pvalue
Ratio
0.050 3.494

Distance to
forest edge

-0.007

0.007

Proportion
aspen and
shrub (50 ha)*

3.413

0.008

Constant

-1.398

0.031

Variable
Percent
canopy gap

P

-1.496

2004
Odds
Pvalue Ratio
0.418 0.224

0.993

-0.011

0.014 0.989

30.363

1.131

0.482

-0.345

0.731

ROC

P
0.753

ROC
0.737

3.100

16 4

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Table I.7B. Temporal validation of the best plot-level model for the Red-naped Sapsucker.

P
Variable
Percent canopy gap

2001-03
1.251

2001
2.096

2002
0.456

2003
0.558

2004
-1.236

Distance to forest
edge

-0.007

-0.005

-0.013

-0.007

-0.011

Proportion aspen and
shrub (50 ha)*

3.413

1.195

2.995

2.058

5.501

Constant

-1.398

-3.441

-1.713

-1.325

-1.232

ROC

0.753

0.692

0.748

0.692

0.773

Table I.7C. Spatial validation of the best stand-level model for the Red-naped Sapsucker.

2004
Odds
Pvalue Ratio
0.542 9.052

0.041

2001-2003
Odds
Pvalue
Ratio
0.171 1.042

0.716 2.203

Edge density
(50 ha)

0.012

<0.000

1.012

0.011

0.072

Constant

-2.371

0.002

-3.332

0.005

Variable
Percent
Aspen

P

ROC

P

1.011

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

ROC
0.752

Table I.7D. Temporal validation of the best stand-level model for the Red-naped Sapsucker.

Variable
Percent Aspen
Edge density (50 ha)

B
2001-03
0.041
0.012
-2.371

0.037
0.007

2002
0.080

2003
-0.006

2004
-0.072

0.014
-4.421

0.014
-3.188

0.006
-1.574

0.761

0.726

0.702

2.731

Constant
ROC

2001

0.716

0.659

Literature Cited
Gillingham, M.P., 2003. Lifeforms, structure and biodiversity: species literature review.
March 2003. 651 pp. Available at http://web.unbc.ca/biodiversity/ accessed March
2006.
Gillingham, M.P., Psyllakis, J.M., 2004. Lifeforms: structure and biodiversity. 27 April 2004.
120 pp. Available at http://web.unbc.ca/biodiversity/ accessed March 2006.

166

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Appendix II. List of vertebrate species (200 species), with associated number of detections,
detected in the Williams Lake Study Area from 2001-2004. Species modelled are in bold text.
Species
Reptiles

Amphibians

Detections
all years

Common Garter Snake
Western Terrestrial Garter
Snake

Thamnophis sirtalis
Thamnophis elegans

2

Columbia Spotted Frog
Long-toed Salamander

Rana luteiventris
Ambystoma
macrodactylum
Bufo boreas
Rana sylvatica

8
7

151
47

Empidonax alnorum
Recurvirostra americana
Botaurus lentiginosus
Fulica americana
Corvus brachyrhynchos
Falco sparverius
Anthus rubescens
Setophaga ruticilla
Turdus migratorius
Anas americana
Haliaeetus leucocephalus
Strix varia
Bucephala islandica
Ceryle alcyon
Chlidonias niger
Picoides arcticus

86
1
2
203
39
3
1
53
1133
65
5
48
38
1
113
54

Poecile atricapillus
Dendroica striata
Dendragapus obscurus
Anas discors
Bombycilla garrulus
Larus Philadelphia
Poecile hudsonicus
Aegolius funereus
Euphagus cyanocephalus
Certhia americana
Molothrus ater
Bucephala albeola
Stellula calliope
Branta canadensis

805
4
2
52
6
1
11
2
3
366
220
238
4
305

Western Toad
Wood Frog
Birds

Scientific Name

Alder Flycatcher
American Avocet
American Bittern
American Coot
American Crow
American Kestrel
American Pipit
American Redstart
American Robin
American Widgeon
Bald Eagle
Barred Owl
Barrow's Goldeneye
Belted Kingfisher
Black Tem
Black-backed
Woodpecker
Black-capped Chickadee
Blackpoll Warbler
Blue Grouse
Blue-winged Teal
Bohemian Waxwing
Bonaparte's Gull
Boreal Chickadee
Boreal Owl
Brewer's Blackbird
Brown Creeper
Brown-headed Cowbird
Bufflehead
Calliope Hummingbird
Canada Goose

3

167

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Species
Canvasback
Cassin's Vireo
Cedar Waxwing
Chipping Sparrow*
Cinnamon Teal
Clay-colored Sparrow
Common Goldeneye
Common Loon
Common Merganser
Common Nighthawk
Common Raven
Common Redpoll
Common Snipe
Common Yellowthroat
Cooper's Hawk
Dark-eyed Junco*
Downy Woodpecker
Dusky Flycatcher
Eastern Kingbird
European Starling
Evening Grosbeak*
Gadwall
Golden Eagle
Golden-crowned Kinglet
Golden-crowned Sparrow
Gray Catbird
Gray Jay
Great Blue Heron
Great Gray Owl
Great Homed Owl
Greater Scaup
Greater Yellowlegs
Green-winged Teal
Hairy Woodpecker
Hammond's Flycatcher
Hermit Thrush
Herring Gull
Hooded Merganser
Homed Grebe
Killdeer
Least Flycatcher
Least Sandpiper
Lesser Scaup
Lesser Yellowlegs
Lincoln's Sparrow

Scientific Name
Aythya valisineria
Vireo cassinii
Bombycilla cedrorum
Spizella passerina
Anas cyanoptera
Spizella pallida
Bucephala clangula
Gavia immer
Mergus merganser
Chordeiles minor
Corvus corax
Carduelis flammea
Gallinago gallinago
Geothlypis trichas
Accipiter cooperii
Junco hyemalis
Picoides pubescens
Empidonax oberholseri
Tyrannus tyrannus
Stumus vulgaris
Coccothraustes vespertinus
Anas strepera
Aquila chrysaetos
Regulus satrapa
Zonotrichia atricapilla
Dumetella carolinensis
Perisoreus canadensis
Ardea herodias
Strix nebulosa
Bubo virginianus
Aythya marila
Tringa melanoleuca
Anas crecca
Picoides villosus
Empidonax hammondii
Catharus guttatus
Larus argentatus
Lophodytes cucullatus
Podiceps auritus
Charadrius vociferus
Empidonax minimus
Calidris minutilla
Aythya affinis
Tringia flavipes
Melospiza lincolnii

Detections
all years
1
1190
31
1650
26
28
6
174
2
13
358
17
109
178
2
2403
69
172
11
2
1209
16
1
734
2
3
329
1
9
30
4
46
75
239
147
1009
2
13
2
82
213
1
195
7
20
16 8

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Species
Long-eared Owl
MacGillivray's Warbler
Magnolia Warbler
Mallard
Marsh Wren
Merlin
Mountain Bluebird
Mountain Chickadee
Nashville Warbler
Northern Flicker
Northern Goshawk
Northern Harrier
Northern Pintail
Northern Pygmy Owl
Northern Rough-winged
Swallow
Northern Saw-whet Owl
Northern Shoveler
Northern Waterthrush
Olive-sided Flycatcher
Orange-crowned Warbler
Osprey
Pacific Slope Flycatcher
Pied-billed Grebe
Pileated Woodpecker
Pine Grosbeak
Pine Siskin*
Purple Finch
Red Crossbill
Red-breasted Merganser
Red-breasted Nuthatch*
Red-eyed Vireo
Redhead
Red-naped Sapsucker
Red-necked Grebe
Red-tailed Hawk
Red-winged Blackbird
Ring-billed Gull
Ring-necked Duck
Ruby-crowned Kinglet*
Ruddy Duck
Ruffed Grouse
Rufous Hummingbird
Rusty Blackbird
Sandhill Crane

Scientific Name

Detections
all years

Asio otus
Oporornis tolmiei
Dendroica magnolia
Anas platyrhynchos
Cistothorus palustris
Falco columbarius
Sialia currucoides
Poecile gambeli
Vermivora ruficapilla
Colaptes auratus
Accipiter gentilis
Circus cyaneus
Anas acuta
Glaucidium gnoma
Stelgidopteryx serripennis

2
92
17
312
308
4
8
1267
10
553
12
5
3
20
4

Aegolius acadicus
Anas clypeata
Seiurus noveboracensis
Contopus borealis
Vermivora celata
Pandion haliaetus
Empidonax difftcilis
Podilymbus podiceps
Dryocopus pileatus
Pinicola enucleator
Carduelis pinus
Carpodacus purpureus
Loxia curvirostra
Mergus serrator
Sitta canadensis
Vireo olivaceus
Aythya americana
Sphyrapicus nuchalis
Podiceps grisegena
Buteo jamaicensis
Agelaius phoeniceus
Larus delawarensis
Aythya collaris
Regulus calendula
Oxyura jamaicensis
Bonasa umbellus
Selasphorus rufus
Euphagus carolinus
Grus canadensis

29
7
145
488
689
1
2
106
384
18
863
4
262
5
1868
51
14
361
2
88
286
1
132
2021
145
350
9
1
279
16 9

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Species
Scientific Name
Detections
_________________________________________________________________ all years
Savannah Sparrow
Passerculus sandwichensis
55
Say's Phoebe
Sayornis saya
6
Semipalmated Plover
Charadrius semipalmatus
1
Semipalmated Sandpiper
Calidris pusilla
5
Sharp-shinned Hawk
Accipiter striatus
3
Sharp-tailed Grouse
Tympanuchus phasianellus
3
Short-billed Dowitcher
Limnodromus griseus
3
Solitary Sandpiper
Tringa solitaria
58
Song Sparrow
Melospiza melodia
113
Sora
Porzana Carolina
109
Spotted Sandpiper
Actitis macularia
27
Spruce Grouse
Dendragapus canadensis
35
Swainson's Thrush*
Catharus ustulatus
1959
Tennessee Warbler
Vermivora peregrina
3
Three-toed Woodpecker
Picoides tridactylus
102
Townsend's Solitaire
Myadestes townsendi
248
Townsend's Warbler
Dendroica townsendi
135
Tree Swallow
Tachycineta bicolor
157
Varied Thrush
Ixoreus naevius
19
Veery
Catharus fuscescens
26
Vesper Sparrow
Pooecetes gramineus
9
Violet-green Swallow
Tachycineta thalassina
16
Warbling Vireo
Vireo gilvus
655
Western Kingbird
Tyrannus verticalis
2
Western Tanager
Piranga ludoviciana
1111
Western Wood-Pewee
Contopus sordidulus
504
White-crowned Sparrow
Zonotrichia leucophrys
19
White-winged Crossbill
Loxia leucoptera
12
Willow Flycatcher
Empidonax traillii
204
Wilson's Phalarope
Phalaropus tricolor
26
Wilson's Warbler
Wilsonia pusilla
150
Winter Wren
Troglodytes troglodytes
103
Wood Duck
Aix sponsa
2
Yellow Warbler
Dendroica petechia
98
Yellow-headed Blackbird
Xanthocephalus
168
xanthocephalus
Yellow-rumped Warbler*
Dendroica coronata
1863
Mammals

Beaver
Big Brown Bat
Black Bear
Bushy-tailed Woodrat
Columbian Ground Squirrel
Common Shrew
Cougar

Castor canadensis
Eptesicus fuscus
Ursus americanus
Neotoma cinerea
Spermophilus columbianus
Sorex cinereus
Puma concolor

6
6
451
3
4
4
10
170

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Species

Scientific Name

Canis latrans
Coyote
Deer Mouse
Peromyscus maniculatus
Dusky Shrew
Sorex monticolus
Ermine
Mustela erminea
Fisher
Martes pennanti
Grey Wolf
Canis lupus
Grizzly Bear
Ursus arctos
Heather Vole
Phenacomys intermedius
Hoary Bat
Lasiurus cinereus
Little Brown Myotis
Myotis lucifugus
Long Legged Myotis
Myotis volans
Mustela frenata
Long-tailed Weasel
Lynx
Lynx canadensis
Marten
Martes americana
Meadow Jumping Mouse
Zapus hudsonius
Meadow Vole
Microtus pennsylvanicus
Moose
Alces alces
Mule Deer*
Odocoileus hemionus
Muskrat
Ondatra zibethicus
Northern Bog Lemming
Synaptomys borealis
Northern Flying Squirrel
Glaucomys sabrinus
Porcupine
Erethizon dorsatum
Red Fox
Vulpes vulpes
Red Squirrel*
Tamiasciurus hudsonicus
Silver-haired Bat
Lasionycteris noctivagans
Lepus americanus
Snowshoe Hare
Southern Red-backed
Clethrionomys gapperi
Vole
Striped Skunk
Mephitis mephitis
Western Jumping Mouse
Zapus princeps
Western Long-eared
Myotis evotis
Myotis
Yellow-bellied Marmot
Marmota flaviventris
Tamias amoenus
Yellow-pine Chipmunk
* species not modelled because of commonness

Detections
all years
160
455
1
29
7
10
1
2
11
21
3
78
42
10
2
97
876
2255
46
3
24
8
27
2449
2
2218
2602
8
2
163
1
317

171

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Appendix III. List of vertebrate species used in non-metric multidimensional scaling
ordination analysis (n = 66). Species were detected in the Williams Lake Study Area from
2001-2004 and ordination analysis was used in conjunction with cluster analysis to determine
plots with similar species co-occurrence. Species detected at less than 5% of plots were
omitted as were species that had high levels of spatial uncertainty associated with their
detection (e.g., rarely detected not in flight, calls that travel long distances).

Species

Scientific Name

Amphibians

Western Toad

Bufo boreas

Birds

Alder Flycatcher
American Crow
American Redstart
American Robin
Black-backed Woodpecker
Black-capped Chickadee
Brown Creeper
Brown-headed Cowbird
Cassin's Vireo
Chipping Sparrow
Clay-colored Sparrow
Common Raven
Common Yellowthroat
Dark-eyed Junco
Downy Woodpecker
Dusky Flycatcher
Golden-crowned Kinglet
Gray Jay
Hairy Woodpecker
Hammond's Flycatcher
Hermit Thrush
Least Flycatcher
MacGillivray's Warbler
Mountain Chickadee
Northern Flicker
Northern Waterthrush
Olive-sided Flycatcher
Orange-crowned Warbler
Pileated Woodpecker
Pine Siskin
Red-breasted Nuthatch
Red-eyed Vireo
Red-naped Sapsucker
Red-tailed Hawk

Empidonax alnorum
Corvus brachyrhynchos
Setophaga ruticilla
Turdus migratorius
Picoides arcticus
Poecile atricapillus
Certhia americana
Molothrus ater
Vireo cassinii
Spizella passerine
Spizella pallida
Corvus corax
Geothlypis trichas
Junco hyemalis
Picoides pubescens
Empidonax oberholseri
Regulus satrapa
Perisoreus canadensis
Picoides villosus
Empidonax hammondii
Catharus guttatus
Empidonax minimus
Oporomis tolmiei
Poecile gambeli
Colaptes auratus
Seiurus noveboracensis
Contopus borealis
Vermivora celata
Dryocopus pileatus
Carduelis pinus
Sitta canadensis
Vireo olivaceus
Sphyrapicus nuchalis
Buteo jamaicensis

Total
detections all
years
151
86
39
53
1133
54
805
366
220
1190
1650
28
358
178
2403
69
172
734
329
239
147
1009
213
92
1267
553
145
488
689
384
863
1868
51
361
88

17 2

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Ruby-crowned Kinglet
Ruffed Grouse
Song Sparrow
Spruce Grouse
Swainson's Thrush
Three-toed Woodpecker
Townsend's Solitaire
Townsend's Warbler
Veery
Warbling Vireo
Western Tanager
Western Wood-Pewee
Willow Flycatcher
Wilson's Warbler
Winter Wren
Yellow Warbler
Yellow-rumped Warbler

Regulus calendula
Bonasa umbellus
Melospiza melodia
Dendragapus canadensis
Catharus ustulatus
Picoides tridactylus
Myadestes townsendi
Dendroica townsendi
Catharus fuscescens
Vireo gilvus
Piranga ludoviciana
Contopus sordidulus
Empidonax traillii
Wilsonia pusilla
Troglodytes troglodytes
Dendroica petechia
Dendroica coronata

Total
detections all
years
2021
350
113
35
1959
102
248
135
26
655
1111
504
204
150
103
98
1863

Black Bear
Coyote
Deer Mouse
Ermine
Long-tailed Weasel
Lynx
Meadow Vole
Moose
Mule Deer
Red Fox
Red Squirrel
Snowshoe Hare
Southern Red-backed Vole
Yellow-pine Chipmunk

Ursus americanus
Canis latrans
Peromyscus maniculatus
Mustela erminea
Mustela frenata
Lynx canadensis
Microtus pennsylvanicus
Alces alces
Odocoileus hemionus
Vulpes vulpes
Tamiasciurus hudsonicus
Lepus americanus
Clethrionomys gapperi
Tamias amoenus

451
160
455
29
78
42
97
876
2255
27
2449
2218
2602
317

Species

Mammals

Scientific Name

173

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Appendix IV. Comparison of cluster analysis results for groupings of plots based on
vertebrate-species co-occurrence (X, lettered groups) and structural characteristics (Y,
numbered groups). Structure groups with the highest number of plots overlapping with
species groups are in bold text (Y). Overlap for groups based on each method was minimal
(range = 16.7% - 50.0%).

Plot
PI
P17
P18
P19
P22
P38
P39
P40
P41
P44
P45
P47
P48
P62
P74
P77
P78
P106
P107
P120
P121
P124
P125
P127
P135
P136
P137
P139
P140
P169
P170
P179
PI 27A
P127B
P2
P3
P4
P5
P70
P71
PI 12
PI 13

A 1 B 2 C 3 D 4
X Y
X
X
X
X
X
X
X
X
Y
X Y
X
Y
X
Y
X
X
X
Y
X
Y
X
Y
X
X
Y
X
Y
X
Y
X
Y
X
X
Y
X
X
Y
X
X
Y
X
Y
X
X
X Y
X
Y
X
Y
X Y
Y
X
Y X
Y X
X
X
Y
X
X

Cluster Groups
5 F 6 G 7 H 8 I

9 J

10K11

Y
Y
Y
Y
Y

Y
Y
Y
Y
Y

Y
Y

174

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Cluster Groups
4 E 5 F
6 G 7 H 8 I
Y
Y
Y
Y
Y

Plot
A 1 B 2 C 3 D
X
PI 17
X
P153
X
P156
X
P157
X
P158
X
Y
P187
X
Y
P192
X
P209
P178A
Y X
P57A
X
X
P6
Y
X
P7
X
P9
X Y
P ll
Y X
P13
X
P16
X
P30
X
P31
Y
X
P43
Y
X
P46
Y
X
P58
X
P59
X
P76
P104
X
X
P108
Y
X
P109
X
P110
Y
X
PI 18
X
PI 19
X
P122
Y
X
P128
X
Y
P129
X
P130
Y X
P131
P138
X
X
P150
X
P151
X
P155
X
P159
X
P190
Y
Y X
P210
P110A
X
X
Y
P114A
P118A
X
X
P8
Y
X
P42
Y
X
P72
Y
Y X
P86
Y X
P97

9 J 1 0 K 1 1

Y

Y
Y

Y
Y
Y

175

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Plot
A IB
P126
P188
P193
P10
P12
P14
P25
P53
P80
P82
P96
PI 14
P134
P141
P142
P145
P149
P171
P172
Y
P173
P178
P184
P191
Y
P199
P223
P224
P236
P243
P141A
P15
P54
P55
P56
P57
P67
P68
P87
P90
P92
P174
Y
P176
P177
P181
PI83
P202
P203
P206
P208
P222

2 C
Y
Y

Cluster Groups
3 D 4 E
5 F
6 G 7 H
X
X
X
Y X
Y
X
Y
X
Y
X
Y
X
X Y
Y
X
Y
X
X Y
Y X
Y
X
X
X
X
Y
X
X
X Y
Y X
Y
X
X
Y
X
X
Y
X
Y
X
X
X
X Y
X Y
X Y
X Y
X Y
Y
X
Y
X
Y
X
Y
X
Y
X
X
Y
X
Y
X
X
X Y
Y
X
X
X
X
X Y

81
Y

9 J

10K11

Y
Y
Y

Y
Y
Y

Y
Y
Y
Y

17 6

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Plot
A IB
P237
P238
P240
P241
P152A
P56B
P20
P21
P32
P33
P34
Y
P35
Y
P36
P37
P166
P23
P24
P27
P28
P49
P50
P60
P61
P63
P64
P65
P79
P81
P101
P102
P103
P105
P146
P147
P148
P160
P161
P180
P195
P207
P180A
P180B
P26
P88
P89
P91
P93
P94
P95

2 C

3 D 4 E
Y

Y
Y

Y

Y
Y

Y
Y
Y

Y

Y
Y

Y

Cluster Groups
5 F
6 G 7 H 8 I
9 J 10K11
X
Y
X
X
Y
X
Y
X Y
X Y
X
X
X
Y
Y
X
X
X
X
Y
Y
X
X
Y X
Y X
X Y
X
Y
X
Y X
X
X Y
X Y
X Y
X
X
X
X Y
Y
X
Y
X
X
X Y
X
Y
X
Y
X
Y
X
Y
X
X
X Y
X
Y
Y
X
Y
X
X
Y
X
Y
X
X
Y
X
Y
Y
X
177

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Plot
P143
P144
P185
P196
P201
P204
P205
P212
P230
P231
P232
P233
P239
P242
P50A
P90A
P29
P51
P52
P69
P132
P133
P152
P154
P186
P198
P211
P220
P221
P234
P235
P56A
P83
P84
P85
Pi l l

A

IB

2 C

3 D 4 E

Cluster Groups
5 F
6 G 7 H

Y
Y
Y
Y

Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y

81

9 J
X Y
X
X
X
X
X
X
X
X
X
X
Y X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
Y X
X

Y
Y
Y
Y

10K11
Y
Y
Y
Y

Y
Y
Y

X
X
X
X

178

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Appendix V. Indicator species analysis to identify vertebrate species most closely associated
with the 11 groups identified with cluster analysis of plots based on species co-occurrence in
the Williams Lake Study Area from 2001-2004.
We used indicator species analysis (Dufrene and Legendre, 1997) to determine
species most associated with each of the 11 plot groups identified with cluster analysis (Table
V.l). Indicator species analysis describes how well each species separates among groups by
comparing the occurrence of each species within each group (McCune and Grace, 2002). A
perfect match occurs when the species only occurs in one group, never others. The p-value is
calculated based on the proportion of randomized trials with indicator values equal to or
exceeding the observed value (McCune and Grace, 2002) and tests the hypothesis of no
differences among groups. The table is provided only as supplemental information as we
made no attempt to interpret relationships.

Table V .l. Species included in each group as determined by indicator species analysis.
Groups were defined by cluster analysis of species co-occurrence for plots in the Williams
Lake Study Area from 2001-2004.
group
1
1
1
1
1

p-value
0.03
0.01
0.02
0.02
0.01

1
1

0.02
0.02

Black Bear

Latin
Canis latrans
Vireo cassinii
Catharus guttatus
Lepus americanus
Lynx canadensis
Tamiasciurus
hudsonicus
Ursus americanus

American Crow

Corvus brachyrhynchos

2

0.23

Gray Jay
Spruce Grouse
Western Tanager
Willow Flycatcher

Perisoreus canadensis
Dendragapus canadensis
Piranga ludoviciana
Empidonax traillii

3
3
3
3

0.01
0.05
0.08
0.04

Chipping Sparrow

Spizella passerine

4

0.01

Common Raven
Mountain Chickadee
Pileated Woodpecker

Corvus corax
Poecile gambeli
Dryocopus pileatus

5
5
5

0.06
0.05
0.09

Species
Coyote
Cassin's Vireo
Hermit Thrush
Snowshoe Hare
Lynx
Red Squirrel

179

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Table V .l. Species included in each group as determined by indicator species analysis
(continued).

Veery
Western Wood-Pewee

Latin
Empidonax alnorum
Turdus migratorius
Poecile atricapillus
Molothrus ater
Bufo boreas
Picoides pubescens
Empidonax oberholseri
Picoides villosus
Empidonax minimus
Microtus pennsylvanicus
Mustela frenata
Colaptes auratus
Vermivora celata
Contopus borealis
Sphyrapicus nuchalis
Buteo jamaicensis
Bonasa umbellus
Catharus fuseescens
Contopus sordidulus

Pine Siskin
Red-breasted Nuthatch
Ruby-crowned Kinglet
Warbling Vireo

Carduelis pinus
Sitta canadensis
Regulus calendula
Vireo gilvus

7
7
7
7

0.01
0.16
0.02
0.02

Clay-colored Sparrow
Dark-eyed Junco
MacGillivray's Warbler
Yellow-pine Chipmunk
Townsend's Solitaire
Yellow-ramped Warbler

Spizella pallida
Junco hyemalis
Oporornis tolmiei
Tamias amoenus
Myadestes townsendi
Dendroica coronata

8
8
8
8
8
8

0.29
0.01
0.42
0.01
0.03
0.08

Hammond's Flycatcher
Northern Waterthrash
Red-eyed Vireo
Wilson's Warbler
Yellow Warbler

Empidonax hammondii
Seiurus noveboracensis
Vireo olivaceus
Wilsonia pusilla
Dendroica petechia

9
9
9
9
9

0.22
0.02
0.02
0.04
0.11

Species
Alder Flycatcher
American Robin
Black-capped Chickadee
Brown-headed Cowbird
Western Toad
Downy Woodpecker
Dusky Flycatcher
Hairy Woodpecker
Least Flycatcher
Meadow Vole
Long-tailed Weasel
Northern Flicker
Orange-crowned Warbler
Olive-sided Flycatcher
Red-naped Sapsucker
Red-tailed Hawk
Ruffed Grouse

group
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6

p-value
0.03
0.01
0.01
0.02
0.05
0.01
0.03
0.02
0.01
0.05
0.04
0.01
0.08
0.01
0.02
0.03
0.03
0.08
0.01

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.

Table V .l. Species included in each group as determined by indicator species analysis
(continued).
Species
Moose
American Redstart
Common Yellowthroat
Deer Mouse
Song Sparrow
Townsend's Warbler

Latin
Alces alces
Setophaga ruticilla
Geothlypis trichas
Peromyscus maniculatus
Melospiza melodia
Dendroica townsendi

group
10
10
10
10
10
10

/7-value
0.12
0.01
0.03
0.03
0.04
0.01

Black-backed Woodpecker
Brown Creeper
Southern Red-backed Vole
Golden-crowned Kinglet
Ermine
Mule Deer
Swainson's Thrush
Three-toed Woodpecker
Red Fox
Winter Wren

Picoides arcticus
Certhia americana
Clethrionomys gapperi
Regulus satrapa
Mustela erminea
Odocoileus hemionus
Catharus ustulatus
Picoides tridactylus
Vulpes vulpes
Troglodytes troglodytes

11
11
11
11
11
11
11
11
11
11

0.02
0.01
0.01
0.01
0.01
0.01
0.01
0.01
0.01
0.01

L iterature Cited
Dufrene, M., Legendre, P., 1997. Species assemblages and indicator species: the need for
flexible asymmetrical approach. Ecological Monographs 67, 345-366.
McCune, B., Grace, J.B., 2002. Analysis of Ecological Communities. MjM Software
Design, Gleneden Beach, OR.

R ep ro d u ced with p erm ission o f th e copyright ow ner. Further reproduction prohibited w ithout perm ission.