dynamique spatio-temporelle des mammifères hivernant dans ... · estimated effects of current and...
TRANSCRIPT
DYNAMIQUE SPATIO-TEMPORELLE DES MAMMIFÈRES HIVERNANT DANS UNE FORÊT BORÉALE DE L’EST DU
CANADA
Thèse de doctorat
TOSHINORI KAWAGUCHI
Doctorat en sciences forestières Philosophiae doctor (Ph.D.)
Québec, Canada
© Toshinori KAWAGUCHI, 2015
iii
RÉSUMÉ
La sélection de l'habitat par les espèces animales est rarement abordée par des études à long
terme. Basé sur 11 années de pistage sur la neige, j’ai examiné 1) s’il est possible d’élaborer
un indice de population fiable avec des dénombrements de pistes comparés à des ventes de
peaux de mustélidés, l'écureuil roux et la marte, 2) si la sélection de l'habitat du lièvre est
influencée par la densité de l’espèce, 3) si la profondeur de la neige exerce une influence sur
l'utilisation de l’habitat du lièvre, 4) et si l’association spatiale entre la martre et le lièvre est
réduite lorsque l’abondance de prédateurs concurrents, le lynx du Canada et le renard roux,
augmente.
Chaque année, 91,3 km ± 28,9 km (moyenne ± SD) de transects ont été parcourus. Pour
le premier objectif, des modèles linéaires généralisés du nombre de pistes de chaque espèce
ont été développés, en fonction de l'effet de l'année (variable catégorique) et des descripteurs
de la végétation. Les estimations des effets de l'année étaient étroitement associées avec les
ventes des peaux d'écureuil roux et de belettes. Le nombre moyen de pistes par effort
d’échantillonnage étaient associés avec les ventes de peaux de martre. La fréquentation de
jeunes peuplements (20-40 ans) était influencée par l’indice de population de lièvres durant
l'année précédente. À l’intérieur d’un hiver, le lièvre était davantage associé à feuillage au-
dessus de 2 m (données LiDAR) à mesure que la neige devenait plus profonde. Finalement,
la relation de causalité entre le lièvre, la martre, l'écureuil roux, le renard roux et le lynx a été
déterminée par l'analyse de piste (path analysis). L’association spatiale entre les lièvres et la
martre diminuait lorsque l’abondance de lynx dans l'année précédente était élevée.
Cette étude démontre l’importance de la prise en compte de la dynamique écosystémique
à long terme tel que le climat et la dynamique de la population, et de l’espèce focale, lors de
l’étude de la sélection de l’habitat. Elle incite à la prudence dans les projections à long terme
basées sur des approches simples telles que les indices de qualité des habitats. Dans un
contexte d’aménagement forestier, il est probable que les changements à court terme et à long
terme dans la végétation et l’enneigement, suite aux pratiques forestières et aux changements
climatiques, auront des effets complexes sur la répartition spatiale des mammifères
hivernants.
v
ABSTRACT
Habitat selection by animals has rarely been the focus of long term studies. Based on 11
years study of snow tracking, I investigated whether 1) population indices derived from snow
tracking agreed with pelt sales in marten, red squirrel and weasels, 2) habitat selection by
snowshoe hare is influenced by conspecific density, 3) snow depth influenced habitat use
pattern of snowshoe hare, 4) spatial association between marten and hare is reduced when
other hare predators, lynx and fox, are more abundant.
Each year, 91.3km ± 28.9 km (mean ± SD) of transects were surveyed. For the first
objective, generalized linear models were used for track count of each species as function of
year effect (categorical variable) and vegetation variables. Estimates of year effects agreed
strongly with pelt sales of red squirrel and weasels. Mean track counts by sampling effort
agreed with marten pelt sales. Hare track counts in young (20-40y) forest stands declined
with an increase of conspecific density with one year lag. Hare track counts were increasingly
associated to stands with high foliage density above 2m (measured with LiDAR), as snow
became deeper in the course of winter. Finally, path analyses of the causal relationship
between spatial distributions of hare, marten, red squirrel, red fox and Canada lynx suggested
that the hare-marten spatial association declined when lynx abundance in the previous year
was high.
This thesis underlines the importance of accounting for long term ecosystem dynamics
such as population and climate, including those of the focal species, in the study of habitat
selection. It raises questions about the validity of long-term projections based on simple
approaches such as habitat suitability indices. In a forest management context, short- and
long-term changes in the vegetation and snow cover, following forest management and
climate change, will have complex effect on wintering mammal spatial distribution.
vii
TABLE OF CONTENTS
RÉSUMÉ ............................................................................................................................... iii
ABSTRACT ............................................................................................................................ v
TABLE OF CONTENTS ...................................................................................................... vii
LIST OF TABLES ............................................................................................................... xi
LIST OF FIGURES ............................................................................................................. xiii
ACKNOWLEDGMENTS ................................................................................................ xvii
AVANT-PROPOS .............................................................................................................. xix
GENERAL INTRODUCTION ............................................................................................ 1
Ecosystem management and limits of short term studies ............................................. 2
Habitat selection ................................................................................................................ 3
Interspecific interactions and habitat selection .............................................................. 4
Winter and seasonal aspects of habitat selection ........................................................... 5
Objectives .......................................................................................................................... 7
STUDY SITE ......................................................................................................................... 9
Climate condition .............................................................................................................. 9
Plant species composition ................................................................................................. 9
Mammal species composition .......................................................................................... 9
Natural disturbance ........................................................................................................ 10
Ecosystem management ................................................................................................. 10
GENERAL METHODS ..................................................................................................... 19
Strengths .......................................................................................................................... 19
Weaknesses ...................................................................................................................... 20
Sampling design .............................................................................................................. 20
Survey conditions ............................................................................................................ 21
Statistical procedures for adjusting potential bias in track counts ............................ 21
CHAPTER 1 - SNOW TRACKING AND TRAPPING HARVEST AS RELIABLE SOURCES FOR INFERRING ABUNDANCE: A 9-YEAR COMPARISON .............. 27
Abstract ............................................................................................................................ 28
Introduction ..................................................................................................................... 29
Field-Site Description ..................................................................................................... 32
Methods ............................................................................................................................ 33
Results .............................................................................................................................. 37
Discussion ........................................................................................................................ 38
Acknowledgments ........................................................................................................... 40
CHAPTER 2 – INFLUENCES OF CURRENT AND RECENT CONSPECIFIC DENSITY ON HABITAT SELECTION OF SNOWSHOE HARE .............................. 49
Abstract ............................................................................................................................ 50
viii
Introduction .................................................................................................................... 51
Methods ........................................................................................................................... 52
Results ............................................................................................................................. 55
Discussion ........................................................................................................................ 56
Acknowledgements ......................................................................................................... 58
CHAPTER 3 – VARIATION OF SNOW DEPTH AFFECTS THE SPATIAL DISTRIBUTION OF SNOWSHOE HARE ..................................................................... 63
Abstract ........................................................................................................................... 64
Introduction .................................................................................................................... 65
Methods ........................................................................................................................... 66
Study site ...................................................................................................................... 66
Assessing snowshoe hare spatial distribution .............................................................. 67
Understory cover, stand height and foliage density ..................................................... 67
Statistical analysis ........................................................................................................ 69
Results ............................................................................................................................. 70
Discussion ........................................................................................................................ 70
Acknowledgements ......................................................................................................... 72
CHAPTER 4 – WINTER SPATIOTEMPORAL DYNAMICS OF A BOREAL PREDATOR-PREY COMPLEX ...................................................................................... 81
Abstract ........................................................................................................................... 82
Introduction .................................................................................................................... 84
Methods ........................................................................................................................... 87
Study area ..................................................................................................................... 87
Snow tracking ............................................................................................................... 87
Estimation of population indices .................................................................................. 88
Exploratory path analysis ............................................................................................ 89
Dynamics of spatial association ................................................................................... 90
Results ............................................................................................................................. 90
Discussion ........................................................................................................................ 92
Acknowledgements ......................................................................................................... 94
Appendix ....................................................................................................................... 103
Appendix 1. The best causal graphs for each year during study period, 2004-2014. 103
Appendix 2. Result of d-separation test representing all d-separation claims and its probability of independence. ...................................................................................... 107
Appendix 3. Estimated path coefficients of edges in the best graphs for each year. . 112
GENERAL CONCLUSION ............................................................................................ 117
The challenge for developping a reliable population index ...................................... 117
Feedback effect of density on habitat selection of snowshoe hare ........................... 118
ix
Snow depth influenced habitat use of hare ................................................................. 118
Interactions in spatial distribution among mammals ................................................ 119
Caveats ........................................................................................................................... 120
Management implications ............................................................................................ 121
Long term studies and snow tracking ......................................................................... 122
LITERATURES CITED .................................................................................................. 124
xi
LIST OF TABLES
GENERAL METHODS
Table Method 1. Sampling effort for snow tracking at the Montmorency Forest, southern Quebec, Canada, 2004-2014. ........................................................................................ 25
CHAPTER 1
Table 1. 1. Sampling effort for snow tracking and track counts for American Marten, American Red Squirrel, and weasels at the Montmorency Forest, southern Quebec, Canada, 2004-2012. ...................................................................................................... 46
Table 1. 2. Pearson correlations between pelt sales, and different population indices that were based on snow-tracking over the three species (n = 9): a) American Marten, b) American Red Squirrel, and c) weasels, in southern Quebec, Canada, 2004-2012. Year effect GLM indicates estimates of year effect from a Generalized Linear Model. .......................... 47
CHAPTER 2
Table 2.1. Sampling effort for snow-tracking of snowshoe hare (Lepus americanus) in the Montmorency Forest, southern Quebec (Canada), 2004-2014. .................................... 59
Table 2.2. Estimated effects of current and lag density (previous winter) on habitat selection of snowshoe hares in the Montmorency Forest, Québec, 2004-2014 (n = 10). Estimates are shown for models including either current or lagged effects of density. Positive estimates indicate a greater association at higher density. Adjusted R2 values can be negative, because unlike raw R2, they are penalized by the number of parameters. .... 60
CHAPTER 3
Table 3. 1. Sampling effort for snow-tracking for snowshoe hare (Lepus americanus) at the Montmorency Forest in southern Quebec, Canada, 2012 -2014. ................................. 77
Table 3. 2. List of models for testing hypothesis regarding effect of interaction between LiDAR derived tree height, penetration rate, regenerating forest and snow depth on track counts of hare. X indicates a corresponding variable included into the model. ........... 78
Table 3. 3. Model comparison among candidate models for habitat use by snowshoe hare in the Montmorency Forest, Quebec, 2012-2014. The variable YEAR was treated as a categorical variable. The w indicates Akaike weight. .................................................. 79
Table 3. 4. Estimated effects of snow depth on habitat use by snowshoe hare in the Montmorency Forest, Quebec, 2012-2014: a) hare response mean LiDAR penetration rate (n = 336) and b) hare response to proportion of understory cover (n = 180) under different snow depth. .................................................................................................... 80
CHAPTER 4
Table 4. 1. Sampling effort for snow tracking and track counts for snowshoe hare (Lepus americanus), American marten (Martes americana), red squirrel (Tamiasciurus
xii
hudsonicus), and Canadian lynx (Lynx canadensis) and red fox (Vulpes vulpes) at the Montmorency Forest, southern Quebec, Canada, 2004-2014. ................................... 100
Table 4. 2. Model fit for dynamics of edges in spatial distribution between Hare and Marten in the Montmorency Forest, Québec, 2004-2014. Candidate models include variables: Hare-marten path coefficients (Hare-Marten), hare-marten path coefficients in the previous year (Hare-Martent-1), marten population index (Marten population), hare-fox path coefficients (Hare-Fox), and hare-lynx path coefficient (Hare-Lynx). .............. 101
Table 4. 3. Model fit for dynamics of edges in spatial distribution between marten and squirrel in the Montmorency Forest, Québec, 2004-2014. Candidate models include variables: Hare-lynx path coefficients (Hare-Lynx), hare population index (Hare population), hare-fox path coefficients (Hare-Fox). ....................................................................... 102
Table 4. 4. Result of d-separation test representing all d-separation claims and its probability of independence. ......................................................................................................... 107
Table 4. 5. Estimated path coefficients of edges in the best graphs for each year. ............ 112
xiii
LIST OF FIGURES
STUDY SITE
Figure Site 1. Inter-annual dyhnamics in weather condition in winter (January to March), 2004-2014: a) Snow depth and b) Temperature. Black line indicates mean value. Blue line indicates either seasonal maximum or minimum value depending on variable of interst. Grey error bar indicates standard deviation. ..................................................... 12
Figure Site 2. Vegetation maps of the study site. The map was created based on vegetation map in March 2014. ...................................................................................................... 13
Figure Site 3. Distribution of clear-cut performed during the study period, 2004-2014. Different colors are corresponding to year of clear-cut performed. ............................. 14
Figure Site 4. Inter-annual variations in proportion of area of each habitat types in Foret Montmorecy, southern Quebec, 2004-2014. ................................................................ 15
GENERAL METHOD
Figure Method 1. Spatial distribution of transects in Forêt Montmorency. ......................... 23
Figure Method 2. Schematic representation of an off-road snow-tracking transects. .......... 24
Photo 1. Effect of snow depth on understory cover represented by photographie. The photo represents that small tree was buried by deep snow. ............................................................ 16
Photo 2. Graphical representation of four types of winter habitats in the study site, the Montmorency Forest, Québec: Regenerating forest (top left), young forest (top right), mature forest (bottom left), old forest (bottom right). .................................................. 17
CHAPTER 1
Figure 1.1. Graphical representation of spatial location of snow tracking site (the Montmorency Forest) and trapping area of Furbearer Management Unit (UGAF) 39, southern Quebec, Canada, 2004 - 2012. Black area indicates the location of snow tracking sites. Gray area indicates the location of trapping area. ................................. 41
Figure 1. 2. Inter-annual dynamics of winter precipitation and winter temperature in the study sites, the Montmorency Forest and the Laurentides Wildlife Reserves (UGAF 39), southern Quebec, Canada, 2004-2012: a) Winter temperature (oC), b) winter precipitation (mm). The data for 2006 was not available. ............................................ 42
Figure 1. 3. Spatial distribution of sampling transects in the Montmorency Forest, southern Quebec, Canada, 2004-2012. Black lines indicate off-trail transects and gray lines indicate either roads or trails. ........................................................................................ 43
Figure 1. 4. Comparison of population trends between snow tracking and pelt sales across three taxa: a) American Marten, b) American Red Squirrel and c) weasels, southern Quebec, Canada, 2004-2012. Two population indices are presented: left) Year effect of a Generalized Linear Model (Year effect GLM), right) tracks/exposure time. Black lines represent pelt sales and gray lines represent population indices of year effect GLM (right) or tracks/exposure time (left). Vertical bars represent standard errors. ............. 44
xiv
CHAPTER 2
Figure 2. 1 Estimated population index of snowshoe hare over 11 years from 2004 to 2014 in the study site. The index was developed from coefficients of the year effect estimated from generalized estimating equations (GEE). Vertical bars indicate standard errors. 61
Figure 2.2. Association of hares with 0-20y forest stands explained by a) immediate effect (Dt) only and b) with immediate and time-lag effects (Dt + Dt-1) of the density index over 10 years (2005 - 2014) in the study site (n = 10). Points with standard error bars indicate model coefficients. Dashed lines indicate 95 % confidence bands of the fitted regression line values. .................................................................................................. 62
CHAPTER 3
Figure 3. 1. Satellite snow depth over distribution of snowshoe hare (Lepus americanus). Satellite data was obtained from Canadian Meteorological Centre (Brown and Brasnett 2010). The date of measurement for the map was 1 March in 2012. The resolution was 24km x 24km. The location of the study site was represented by a black star. The histogram showed frequency distribution of snow depth over hare distribution. Snow depth at the study site was 85.1cm on this date. .......................................................... 74
Figure 3. 2. Vegetation map and sampling location in the study site, 2012-2014. The vegetation map was produced by using the one in 2012, ............................................. 75
Figure 3. 3. Effect of interaction between snow depth and vegetation structure on habitat use by hares at the Montmorency Forest, Canada, 2012-2014: A) hare response to LiDAR penetration rate, B) hare response to mean understory cover under different snow depth. High penetration rate values indicate low foliage density above 3m from ground. ..... 76
CHAPTER 4
Figure 4. 1. Population dynamics of five species over 11 years, 2004-2014: a) Snowshoe hare, b) red squirrel, c) American marten, d) Lynx and e) red fox. Gray error bars indicate standard errors. ............................................................................................................. 96
Figure 4. 2. Summary of path analysis results linking spatial distributions of predator, prey and vegetation attributes, 2004-2014. Thickness of line is proportional to the number of years with evidence for an edge. Red colored edges indicate positive coefficients and blue colored edges indicate negative coefficiens. Grey colored edges indicate that path coefficient were either positive or negative depending on study year. ........................ 97
Figure 4. 3. Dynamics of edges in spatial distributions between hare and marten, 2004 -2014 (n = 10). The graphs represent a) relationship between hare-marten spatial association in the current winter and the one in the previous winter and b) hare-marten spatial association in the current winter and lynx population index in the previous winter. Points with standard error bars indicate path coefficients. Dashed lines indicate 95 % confidence bands of the fitted regression line values. .................................................. 98
Figure 4. 4. Dynamics of edges in spatial distributions between squirrel and marten, 2004 -2014 (n = 10): a) Relationship between squirrel-marten spatial association and hare population index, b) Relationship between squirrel-marten spatial association and hare-marten association. Points with standard error bars indicate path coefficients. Dashed lines indicate 95 % confidence bands of the fitted regression line values. .................. 99
xv
Figure 4. 5. The best causal graphs for each year during study period, 2004-2014. Solid line indicates significant path (p < 0.05). Dashed line indicates non-significant path (p < 0.05). Red line indicates positive effect from a variable to the other and blue lines indicate negative. ........................................................................................................ 106
xvii
ACKNOWLEDGMENTS
I am indebted to a great number of recherchers and volunteers who have helped me
progress in this PhD project. Their collaboration and contribution made this thesis possible.
Financial support for this project was provided by a scholarship to T. Kawaguchi from
the “Leadership and Sustainable development Scholarship Program” of Laval University as
well as Foundation F.-K.-Morrow, and by a Natural Sciences and Engineering Research
Council of Canada (NSERC) Discovery grant to A. Desrochers.
I thank the Ministère des Forêts, de la Faune et des Parcs for providing us with pelt sales
data, trapping effort data and geographical information for chapter 1. Especially, I appreciate
H. Bastien for providing me useful comments and advice for the manuscript in chapter 1.
We are grateful to the 29 skilled field workers who contributed to the collection of snow
tracking data. Among 29 skilled workers, I would like to thank especially M. Lapointe for
her great contribution to data collection under harsh winter environment. I also thank C.
Villeneuve for teaching me how to drive a snowmobile safely and for his great contribution
to data collection.
I appreciate my collegues for providing me useful comments on my project and for
having discussions on ecological and statistical issues. I thank J. Marchal for providing me
useful advice on R programming. I have frequently discussed statistical issues with him
during lunch time. The discussion made me think the issues further and deeper. And I thank
J. Faure Lacroix for helping me improve my french writing and pronouciation in french
presentation in Colloque 3 and other opportunities. F. Fabianek introduced me to boreal forest
in the Montmorency Forest in 10 days of his field work. The field work was useful for me to
xviii
capture characteristics of boreal forest and to develop the initial idea of my thesis. I thank S.
Beaudoin for her assistance in correcting in my French during PhD program. I thank C. Roy,
S. Renard and N. S. Baker for their comments on the earlier version of this thesis.
I appreciate D. Fortin, L. Bélanger and C. Samson for their assistance in the design of the
study and W.F.J. Parsons in Centre d’Ètude de Forêt (CEF) for his assistance in linguistic
corrections. My special thanks was given to C. Samson for his contribution to this thesis from
the beginning to the end. His comments and questions often led me to think further and deeper.
I also appreciate I. D. Thompson and M. Mazerolle, members of jury for my defense, for
providing me with thoughtful feedbacks on the thesis.
Lastly, I would like to give my gratitude to my director, André Desrochers, for having
provided me with useful advice and comments on this thesis for this 4 years. He had been
working on data collection and initiation of snow tracking project from 2000 before I started
the PhD program. Again, I apprecite his great amount of contribution to the project.
xix
AVANT-PROPOS
Cette thèse de doctorat inclut quatre chapitres. Je suis l’auteur principal de tous les
chapitres, aussi que l’introduction générale et la conclusion générale. Toutes les analyses
statistiques ont été éffectués par moi. Mon directeur de recherche, André Desrochers, a
largement contribué à l’élaboration de cette thèse. De plus, Héloïse Bastien (Ministère des
Forêts, de la Faune et des Parcs, en cours) est coauteur du premier chapitre.
Le prémier chapitre a été accepté par la revue scientifique Northeastern Naturalist.
1
GENERAL INTRODUCTION
Ecological long term studies are often essential to detect and understand ecological
processes and patterns including population dynamics and spatial distribution of animal
species (Lindenmayer et al. 2012). Ecological patterns requiring longitudinal studies
typically include, but are not limited to, responses to climate change (Brown and Braaten
1998) and population dynamics (Fryxell et al. 1999). In the Canadian boreal forest, fur returns
were instrumental in revealing the 10-year population cycle in several mammals including
snowshoe hare (Lepus americanus) and lynx (Lynx canadensis) (Elton and Nicholson 1942).
A classical long term study in Yukon showed that population dynamics of snowshoe hare are
driven primarily by a predator, the Canada lynx (Krebs et al. 2001a). In the case of habitat
preference of snowshoe hare, monitoring habitat preferences for eight years let researchers
realize that at low population densities, hares used mature pine forest less frequently than
dense immature pine forest. However, at peak density, they used mature pine forest as
frequently as immature pine forest (Mowat 2003). Long term work in Finland empirically
demonstrated mesopredator release in a hare-fox-lynx relationship, in which the top predator
(lynx, Lynx lynx) suppressed abundance of mesopredator (red fox, Vulpes vulpes), indirectly
leading to prey population increase (hare, Lepus timidus) (Elmhagen et al. 2010).
An additional reason for the need to conduct longitudinal ecological studies is the
presence of time lags in processes such as species redistribution following landscape changes
(Metzger et al. 2009), local extinction after deforestation (Brooks et al. 1999), species
invasion (Crooks 2005), and population dynamics (Fryxell et al. 1991, Erb et al. 2001). Time
lags can be caused by long processing times following the perception of a stimulus (Brooks
et al. 1999), intervening processes between two processes of interest (Magnuson 1990), or
2
feedbacks (Framstad et al. 1997). Thus, accounting for recent ecological states is often
required to understand what causes current ecological phenomena.
Ecosystem management and limits of short term studies
In the boreal forest, forestry practices under the context of ecosystem management (defined
in Gauthier et al. 2008) attempt to emulate habitat and landscape pattern produced by natural
disturbances (Long 2009) and attempts to manage forest within natural range of variability
(Keane et al. 2009). On the other hand, the Convention on Biological Diversity (CBD)
mentioned that at least 17 per cent of terrestrial are expected to be conserved by 2020 as well
as prevention of species extiction. Under this context, a large number of studies have
investigated wildlife-habitat relationships to conserve a focal species via protecting their
preferred habitat and to establish a management ‘baseline’. Habitat relationship studies in
this context typically 1) evaluate effect of forestry practices on animal distribution (e.g.,
Ferron et al. 1998, de Bellefeuille et al. 2001, St-Laurent et al. 2008), 2) examine if forestry
practices emulate dynamics of habitat use along a stand age gradient observed in natural
disturbance (e.g., Allard-Duchêne et al. 2014) and 3) describe habitat use patterns along stand
age gradient under natural disturbance (e.g., Hodson et al. 2011). Most of these studies
substitute space for time, i.e. use a cross-sectional approach, to infer long-term relationships.
The results derived from these studies have been, and remain, important to establish a basic
understanding of habitat use, an end-result of habitat selection, and patterns of population
dynamics. However, since wildlife-habitat relationships have been shown to vary with
wildlife population density fluctuating over time (examples above), thus the studies might
potentially introduce confounding effects due to conditions prevailing at the time of those
studies.
3
Habitat selection
Habitat selection is a decision-making process whereby individuals preferentially use, or
occupy, a non-random set of available habitat (Morris 2003), which ecologists postulate as a
tool to maximize fitness. The process is influenced by various factors including conspecific
density (Fretwell and Lucas 1970), weather (Reid et al. 2012), interference or exploitation
competition (Morris et al. 2000) and predation risk (Hildén 1965, Laundré et al. 2010,
Thomson et al. 2006).
Under the assumption of ideal free distribution (Fretwell and Lucas 1970), animals move
freely among habitats of different quality, in accordance to what is expected to maximize
their fitness. Fitness per capita declines with increase of conspecific density due to
intraspecific competition. According to the isodar model (Morris 2003), individuals use the
best habitat at low density. However, as density increases, individuals are exposed to
intensive intraspecific competition. As a consequence, fitness in the best habitat can decrease
to the point where fitness in an alternative habitat is equal or higher to that in the best habitat.
Eventually, this leads individuals to move from high-density (best) to lower-density
(alternative) habitat. The effect of conspecific population density on habitat selection has
been documented for various taxa, including mammals (e.g., fat sand rat Psammomys obesus,
Shenbrot 2004; white-footed mouse Peromyscus maniculatus, Morris 1996) and birds (e.g.,
brown-headed cowbird Molothrus ater, Jensen and Cully 2005) taxa (Dreisig 1995, Haché et
al. 2012).
To this date, effects of density on wildlife-habitat relationships have been investigated in
cross-sectional approaches (Morris 2003, Hodson et al. 2010). However, given the facts that
population dynamics of animals synchronized over space (Liebhold et al. 2004), variation
4
captured in cross-sectional approach might not be as large as the one captured in longtitudinal
studies, potentially leading to poor performance of space-for-time substitution.
In addition, time lags are generally expected, and sometimes observed in species
redistribution due to lagged perception of stimuli or long processing time (Brooks et al. 1999).
Since habitat selection involves responses to stimuli, the effect of density on habitat selection
could be lagged. Despite their likely occurrence, time lag effects of density on habitat
selection per se have not been investigated.
Interspecific interactions and habitat selection
Habitat selection by preys and predators each can be influenced by interspecific
interactions modulated by prey abundance, predation risk, and intra-guild competition (Lima
2002, Gorini et al. 2012). Predation risk often leads prey to maintain anti-predator responses
to prior presence of predators, for example moose (Alces alces) – grey wolf (Canis lupus)
(Latombe et al. 2014), elk (Cervus canadensis)-wolf (Fortin et al. 2005).
In a prey-predator interaction, habitat selection by predators often translates into
occupying areas with high prey abundance, as has been shown in the lynx - snowshoe hare
case (Keim et al. 2011).
When several prey species are available, the strength of the spatial association between a
predator and a given prey species can decrease when the abundance of alternative prey
species increases (prey switching; Murdoch et al. 1975). Prey switching is exemplified with
wolves that concentrate on areas used by deer (Odocoileus virginianus) during winter, but
switch to beavers (Castor canadensis) during summer, when the latter become available
(Latham et al. 2013).
5
Habitat selection can be influenced by competition for shared resources by other species
(Morris et al. 2000, Morris 2003). For example, population density of collared lemmings
(Dicrostonyx groenlandicus) in their preferred habitat can decline with increase in density of
competitor, brown lemmings (Lemmus trimucronatus), in the same habitat (Morris et al.
2000). Competition among predators having dietary overlap often results in spatial
segregation among species as shown in the marten (Martes americana) – fisher (Pekania
pennant, formerly Martes pennanti) relationship (Fisher et al. 2013).
North American boreal forests typically host a multi-species predator-prey system
including snowshoe hare, American marten, red squirrel (Tamiasciurus hudsonicus), lynx
and red fox. To exemplify how the complexity of this system influnece habitat selection of a
species, we consider marten. Marten has been long thought to be an old-growth specialist
species (Buskirk and Powell 1994) but now known to use a wider range of habitats including
young and mature forest (Potvin et al. 2000). A deeper understanding of marten habitat
selection can be attributed to the spatial distribution of their prey, snowshoe hare and red
squirrel (Powell et al. 2003). Since those prey species have been confirmed to appear in
mature forest (40yr-) (Hodson 2011, Allard-Duchêne et al. 2014), association between
mature or old forest and marten could be due to association between habitat and those prey.
Still, as observed in a case of prey switching and competition, the strength of the spatial
association between marten and its prey itself could be subject to change due to abundance
of alternative prey and competitors.
Winter and seasonal aspects of habitat selection
The boreal forest is characterized by its harsh winter environment with extreme low
temperatures and deep snow cover (Brandt et al. 2013). Of course, the main challenges
6
encountered in winter by animals are lower food availability and high homeostasis demands.
But recent climate change (Pauli et al. 2013) may add to the existing challenges, for example,
by creating a mismatch between molt to white pelage by hares and the timing of snowfall
(Mills et al. 2013). Lower snow accumulation may also reduce survival of small mammals
such as voles which are dependent on subnivean space (Pauli et al. 2013). However, winter
is also challenging for field work, and therefore a relatively small number of studies have
been conducted during this season (Campbell et al. 2005).
Spatial distribution in summer might be different from the one in winter because
availability of required resource might be subject to vary due to weather, especially in
landscapes where snow depth regularly exceeds 1 m. Variable snow accumulation affects
accessibility to resources including food and thermal cover for both herbivore and carnivore
(Halpin and Bissonette 1988, Morrison et al. 2003). When small trees are covered by deep
snow, availability of understory cover may decline to the point where food (White et al. 2009)
and anti-predator cover (Litvaitis et al. 1985) become sparse enough to elicit changes in
habitat selection.
As described above, habitat selection can be considered as a complex process involving
multiple stimuli and potentially including lags in response to stimuli, all with potential
consequences on the spatio-temporal dynamics of mammals. For example, snowshoe hare –
habitat relationship is known to vary with conspecific density, but this reponse could be
lagged. In addition, preferred habitat of hare might vary with climate, particulary snow depth.
Given an association between prey and predators, spatial dynamics of prey would greatly
influence the dynamics of predator. Long term studies, enabling to capture large varitation
7
and allowing the investigation of lagged responses, would contribute to address those
uncertainty prevailing in species-habitat relationship studies.
Objectives
The general objective of my thesis is to gain a better understanding of spatio-temporal
dynamics of wintering mammals in an eastern Canadian boreal forest under varying
abundance, trophic and competitive interactions with accounting for potential lagged
response to these factors, based on a 11 years of snow-tracking. My thesis contains four
specific objectives, each corresponding to a chapter.
The first thesis objective (Chapter 1) was to design and evaluate different population
indices developed from snow tracking data by comparing with pelt sales data for three
mammal species: American marten, red squirrel and weasels. If the two indices describe
population changes accurately over time, I predict that they will be highly correlated, after
accounting for sources of error described above.
My second objective (chapter 2) was to investigate influences of current and recent
conspecific density on habitat selection of snowshoe hare. In this chapter, I predicted that at
high population densities, snowshoe hare distribution expands into less preferred habitat, thus
weakening the association between hares and most preferred habitat. I predicted that the the
shift in distribution would be stronger in the winter following high density rather than in the
current winter, i.e. will be lagged.
My third objective (Chapter 3) was to examine whether variation of snow depth affects
the spatio-temporal distribution of snowshoe hare. I predicted that with increasing snow depth,
snowshoe hares will less frequently use sites dominated by saplings and lower trees.
8
My fourth objective (Chapter 4) was to explore the spatio-temporal dynamics of five
mammal species: snowshoe hare – red squirrel – marten – lynx – red fox, in a temporally-
changing habitat. I tested whether the strength of the marten-hare spatial association is
affected by presence of lynx and fox, competitors, and abundance of alternative prey, red
squirrel.
9
STUDY SITE
Geographical location and topography
Snow tracking was performed in the Forêt Montmorency, about 80 km north of Quebec City
(47° 20’N, 71° 10’W). The area of this study site is 66 km2. Altitude ranges from 650 m to
1000 m and maximum degree of slope was 40o.
Climate condition
From 1999 to 2011, annual mean temperature was 0.3 °C. Annual precipitation was 1417 mm
(33 % as snow). Maximum snow depth at weather station in the Montmorency Forest ranged
from 62 to 146 cm in 2004-2014 (Figure Site 1a). Mean daily temperature ranged from -15
oC to -9 oC (Figure Site 1b). Snow accumulation in this site was high enough to bury small
trees (Photograph 2).
Plant species composition
Balsam fir (Abies balsamea) dominates second-growth mature forest stands. Black
spruce (Picea mariana), white or paper birch (Betula papyrifera), trembling aspen (Populus
tremuloides), and white spruce (P. glauca) are also common. Recent (less than 5-yr-old)
clear-cuts are generally colonized by red raspberry (Rubus idaeus), balsam fir, and white
birch (de Bellefeuille et al. 2001). Dominance of balsam fir was confirmed by the prism
survey conducted in 2012, which showed that 77% of trees captured in the survey was balsam
fir and only 9% of tree was white birch.
Mammal species composition
A lot of mammal species are commonly observed in the study site: snowshoe hare,
American marten, American red squirrel, Canadian lynx, red fox, grey wolf, weasels (mostly
10
Mustela erminea based on known species distribution), moose, beaver, river otter, and
porcupine (Bouliane et al. 2015). Mountain lion was rarely observed. There was no sighting
of coyote (Canis latrans) and bob cat (Lynx rufus). Small mammlas such as meadow vole
(Microtus pennsylvanicus) and southern red-backed vole (Clethrionomys gapperi) are also
present.
Natural disturbance
Outbreak of spruce budwarm and windthrow were main natural disturbance in the site,
which were driving sources of developing mosaic forest.
Ecosystem management
The principal strategy to protect biodiversity is to “manage natural mosaic forest within
variabily of natural primitive forest” (Bélanger 2001). Specifically, the management goal is
to be achieved through three practices: 1) employ practices maintaining natural regeneration
process, 2) emulate natural disturbances to conserve mosaic forest, 3) developping modality
to conserve ecological value.
The main forestry practice in this area is clearcut with protection of soils and regeneration
(CPRS; Bélanger et al. 1991), resulting in an ever changing landscape (Figure 1 and Figure2).
A dense road is present, with ca. 150 km of roads, i.e. over 2 km/km2. The forest has been
managed to perserve 20% of regenerating forest (1-20yr) and 20 % of youg forest (20-to-
40yr) and 20% of mature forest (40-80yr), 20 % of older forest with irregular management
(e.g., all types of partial cutting) in a landscape at a scale of 10km2. In addition, in a
management unit of 10-15km2, similar proportion of different size of harvest patches (0.5ha-
to-30ha, 10ha-to-30ha, and 30ha-to-100ha) were created. As a consequence of these
11
management, the landscape in this study site was mosaic forest and contained a large amount
of small patches. During the study period, average of actual proportion of each of habitat
type were 26.4 ± 3.5 % (mean ± standard deviation), 23.8 ± 2.9 %, and 49.7 % ± 1.1 for
regenerating forest, young forest and mature forest respectively. Targeted proportion of each
plant species (balsam fir, white spruce, black spruce, white birch) are 60, 14, 3, 10-to-25 %
separately.
In addition, the study area is also managed for other uses such as recreational and research
activity. During winter, some of roads and trails has been used for cross-country ski and
snowshoe hiking in the study years.
12
Figure Site 1. Inter-annual dyhnamics in weather condition in winter (January to March), 2004-2014: a) Snow depth and b) Temperature. Black line indicates mean value. Blue line indicates either seasonal maximum or minimum value depending on variable of interst. Grey error bar indicates standard deviation.
13
Figure Site 2. Vegetation maps of the study site. The map was created based on vegetation map in March 2014.
14
Figure Site 3. Distribution of clear-cut performed during the study period, 2004-2014. Different colors are corresponding to year of clear-cut performed.
15
Figure Site 4. Inter-annual variations in proportion of area of each habitat types in Foret Montmorecy, southern Quebec, 2004-2014.
16
Photo 1. Effect of snow depth on understory cover represented by photographie. The photo represents that small tree was buried by deep snow.
17
Photo 2. Graphical representation of four types of winter habitats in the study site, the Montmorency Forest, Québec: Regenerating forest (top left), young forest (top right), mature forest (bottom left), old forest (bottom right).
19
GENERAL METHODS
(While the following chapters provide description about the field method, this section
provides more detatiled descriptions on the field work used in the entire thesis.)
Most of this thesis was based on snow tracking, a method commonly used in northern
countries (e.g., Canada: Silva et al. 2009; Finland: Sulkava 2007; Norway: Pedersen et al.
2010). Snow tracking can be achieved in two ways, 1) searching for tracks (used here), and
2) tracking at bait stations (not used here) (Halfpenny, 1995).
Snow tracking is non-invasive and is used to monitor changes in relative population size
(Raphael 1994) as well as to determine habitat selection for many mammalian species.
Population trend derived from snow tracking surevey was corresponding to the one from
mark-repcature method. Snow tracking has been used to conduct reliable field surveys of
American marten, fisher, lynx and wolverine (Halfpenny 1995). Snow tracking and the
classic pellet count technique can show similar patterns in habitat use for the snowshoe hare
(Litvaitis et al. 1985).
Strengths
Snow tracking is often easy to implement, less expensive in comparison with other
techniques such as live trapping. In addition, this technique has no bias associated with bait.
Compared to other non invasive methods (e.g., camera trap), snow tracking can survey
extensive areas, which is potentially important in detecting rare animals and also in
investigating animals with large home range size (> 50 ha). For several mammalian species
active in winter, snow tracking has higher detection probability (Gompper et al. 2006). Given
larger variation in seasonal snow depth (approximately 50cm to 150cm) as observed in the
study area (Montmorency Forest), snow tracking would provide more stable detection rate
20
comparing with camera traps which are installed at a fixed height and are potentially buried
in deep snow.
Weaknesses
The main weakness of snow tracking is that it does not provide a direct estimate of
number of animals, given the lack of individual identification. Furthermore, the detection and
count of animal tracks is easily affected by weather such as strong wind, snow fall, and snow
quality (Gompper 2006). Snow tracking is more labor-intensive than camera traps, which
may explain why several government agencies that used to employ snow tracking as
monitoring method now use camera traps (Claude Samson, pers. comm.). Despite these
caveats, snow tracking offers a high potential for long-term, extensive studies, because of its
simplicity and relatively low cost.
Sampling design
Snow-tracking has been conducted at Forêt Montmorency each winter (date range: 20
December – 15 April) since March 2000, initially using a haphazard selection of transects
(2000-2003), followed by a more systematic design (2004-present), consisting of 67 sample
units placed on a 1-km square grid (Figure Method 1). Each grid point was numbered and
sorted randomly at the beginning of a field season to determine the temporal sampling order
within the season. In order to cover larger area and visit a site regulaly, off-road transects
sampled in the previous 2 years were removed from the pool of points available for sampling.
Off-road transects were 2 km long, except for truncated transects occurring near the border
of the study area (Figure Method 2). Additionally, some off-road transects were erased from
the sampling area because observers cannot access them securely. Road and trail transects
covers 150 km of roads and 40 km of trails. However, the actual length of transect sampled
21
depended on snow condition, weather, time of day, and personnel available (Table Method
1). Each year, snow tracking surveys are conducted from late December to late March.
Survey conditions
Snow quality and climate can affect the detection rate of animal tracks. Thus, we
conduct snow tracking when no strong wind (faster than 20m/s) and no snow fall has occurred
in approximately the last 24 h before field work. Each transect is surveyed only once each
winter. At the start of each transect, observers entered snow conditions on a GPS and start
looking for tracks by snowshoe, skis, by foot or by snowmobile at speed slower than 20 km/h.
When a track is detected, observers obtained its coordinate with a GPS and measured stride
length, width as well as footprint width and depth to help identify the species in case of doubt.
All six species studied here leave tracks that are usually easy to detect and identify, and track
misidentification is assumed to be negligible.
Personnel who participated in snow tracking included experienced field workers and
volunteers. To avoid variation in quality and capacity of observers in detection and
identifying tracks, two days of field training was required for inexperienced participants, and
volunteers were required to measure and photograph tracks of equivocal origin. Thus, this
entire thesis assumed that the quality and capacity of observers were not different among
years and within years. Number of observators per a sampling event ranged from one to two
in most sampling events, depending on difficulty of terrain (for security) and experience.
Statistical procedures for adjusting potential bias in track counts
Since snow tracking was performed on both trail and off-trail transects, surveying on
trails and roads might detect more tracks than on off-trail transect for certain species. For
22
example, red fox and wolf have been confirmed to use road as movement corridors (Coffin
2007). In addition, road has no canopy cover and lateral cover. For mammal species which
require these vegetation cover, road might not be used as much as off-trail transect, resulting
in lower track detection rate on road. Exposure time and temperature are often factors
affecting detection rate of tracks (Roy et al. 2010, Bois et al. 2012).
The following chapters integrated these factors into statistical models as fixed effects to
account for these biases in track presence or counts. While N-mixture models are often used
to account for detection probability (MacKenzie et al. 2002), this thesis did not use them
because the aim of this thesis did not require to estimate true occupancy but simply assess
whether and how track count and environmental conditions, including other wildlife,
covaried while accounting for possible bias.
23
Figure Method 1. Spatial distribution of transects in Forêt Montmorency.
24
Figure Method 2. Schematic representation of an off-road snow-tracking transects.
25
Table Method 1. Sampling effort for snow tracking at the Montmorency Forest, southern Quebec, Canada, 2004-2014.
Year
Road and trail Off-trail
Length (km)
Number of observers
Sampling events (days)
Length (km)
Number of observers
Sampling events (days)
2004 109.5 4 11 44.4 4 30 2005 110.9 6 18 20.2 2 13 2006 187.5 3 21 47.5 3 21 2007 125.6 3 14 30.8 3 14 2008 146.7 5 17 47.3 4 19 2009 162.8 4 19 26.4 3 13 2010 151.4 10 13 29.6 5 10 2011 143.8 4 10 24.7 2 9 2012 135.9 2 13 69 6 21 2013 124.4 5 9 27.3 2 11 2014 124.6 5 12 22.9 2 9 Total 1523.1 51 157 390.2 36 170 Mean 138.5 4.6 14.3 35.5 3.3 15.5 SD 23.3 2.1 3.9 14.9 1.3 6.6
27
CHAPTER 1 - SNOW TRACKING AND TRAPPING HARVEST AS
RELIABLE SOURCES FOR INFERRING ABUNDANCE: A 9-YEAR
COMPARISON
Toshinori Kawaguchi1,*, André Desrochers1 and Héloïse Bastien2
1Centre d’Étude de la Forêt, Université Laval, 2405, rue de la Terrasse, Québec (Québec),
G1V 0A6, Canada. 2Direction de la Gestion de la Faune de la Capitale-Nationale et de la
Chaudière-Appalaches, Ministère des Forêts, de la Faune et des Parcs, 1300, rue du
Blizzard, local 100, Québec (Québec) G2K 0G9, Canada.
This chapter was accepted by the journal Northeastern Naturalist on 3 November 2015.
Therefore, the format in this chapter was adjusted to the guidline of the journal.
28
Abstract
Trapping harvest and snow tracking are frequently used to infer population dynamics, yet
there have been few evaluations of those indices. We developed population indices for Martes
americana Turton (American Marten), Mustela spp. (Weasels) and Tamiasciurus hudsonicus
Erxleben (American Red Squirrel) from 9 years of snow tracking data in Eastern Canada.
Population indices were mean track counts per unit effort and derived from a Generalized
Linear Model (GLM) of track counts as a function of year and covariates including forest age.
Mean track counts had significant correlation with American Marten and Weasels pelt sales.
Year effects in GLM were correlated with American Red Squirrel and Weasels pelt sales.
Both methods are in agreement therefore they likely are replicating population dynamics of
the selected species.
29
Introduction
Monitoring animal populations is a key to understanding the state of ecosystems, their
functions and their responses to anthropogenic and natural disturbances (Lindenmayer et al.
2012). Direct estimates of population densities by live trapping or mark-recapture is very
labor intensive and expensive (Gese 2001). Thus, ecologists often resort to population indices,
such as those derived from snow tracking (Pellikka et al. 2005), trapping harvest (Roberts
and Crimmins 2010), observation reports by hunters (Simard et al. 2012) and scat surveys
(Krebs et al. 2001a, Mowat and Slough 2003).
Snow tracking is non-invasive (Halfpenny et al. 1995) and is frequently used to estimate
relative abundance of wintering mammals in North America (Mowat and Slough 2003) and
Europe (Pellikka et al. 2005). Assuming that the number of tracks is proportional to
population size, some studies use tracking counts per unit effort to infer relative abundances
of mammals such as Lepus americanus Erxleben (Snowshoe Hare), Tamiasciurus hudsonicus
Erxleben (American Red Squirrel) (Jensen et al. 2012), Martes americana Turton (American
Marten) (Krebs 2011), and Mustela frenata Lichtenstein (Long-tailed Weasel) (Fitzgerald
1977). Snow tracking is ease of use (Halfpenny et al. 1995) and has higher detection rates
than other non-invasive techniques such as camera traps and track plates (Gompper et al.
2006). Also, snow tracking does not require the baits or attractants that are employed in other
population monitoring techniques (Raphael 1994).
However, there may be significant noise and bias in population indices obtained from
snow tracking data. Track counts can be affected by weather conditions such as strong winds
and recent snow falls (Raphael 1994, Gompper et al. 2006), and can vary with the activity
level of animals. For example, American Marten activity decreases at extremely low
30
temperatures in American Marten (Thompson and Colgan 1994) and weasels (Robitaille and
Baron 1987). Presence of prey can influence track counts of the predator as shown in Lynx
canadensis Kerr (Canada Lynx) (Keim et al. 2011). Observer bias and miss identification can
be biases.
In many provincial and state agencies, trapping harvest in the form of total catch and
catch per unit effort is commonly used to infer relative population sizes and trends of fur-
bearing species (Douglas and Strickland 1987), particularly American Marten and Pekania
pennanti (formaly Martes pennanti ) Erxleben (Fisher) (Jensen et al. 2012). In the United
States, harvest surveys are frequently used to monitor Lynx rufus Schreber (Bobcat)
population status (Roberts and Crimmins 2010). In Quebec, Canada, the number of pelts that
are sold is the main and sometimes the only index of population size, in the case of Weasels
composed mostly (probably more than 95% based on known species distribution) of Mustela
erminea Linnaeus (Short-Tail Weasel) and Long-tailed Weasel, Mephitis mephitis Schreber
(Striped Skunk), and Ondatra zibethicus Linnaeus (Muskrat). For Canada Lynx, American
Marten and Fisher, abundances can be inferred from the logbooks that trappers are required
to maintain. Those logbooks record the numbers of trapped animals for each species, together
with trapping effort. Numbers of pelts sold can be obtained from fur transaction reports.
Trapping harvest is mainly affected trapping effort (Fortin and Cantin 2004), which in turn
is influenced by several ecological and socioeconomic factors, such as food abundance (Ryan
et al. 2004), market prices (Weinstein 1977), government quotas (Smith et al. 1984),
landscape changes (Raphael 1994) and other disturbances over time (Raphael 1994), and
temperature (Kapfer and Potts 2012).
31
In addition to problems that are specific to each of these indirect methods, temporal
changes in sampling locations may introduce additional bias to the index. Moreover, even if
field surveys are conducted in fixed locations each year, successional patterns in vegetation
will affect the response of different species to the sampled plots (Anderson 2001). As is the
case with any model, those that are used to develop population indices should optimize the
combination of simplicity and accuracy.
Few studies have evaluated the reliability of population indices that have been obtained
through those indirect methods. In a five-year study in Ontario, Thompson and Colgan (1987)
reported different population trends from trapping harvest and snow tracking. Thompson et
al. (1989) again found no correlation between track counts and trapping harvest, but they did
find a significant correlation between track counts and live trapping data. The authors
suggested that track counts correctly described population changes over 5 years.
To evaluate population indices that were based on snow tracking and pelt sales, we
examined the correlation between the two techniques with data from three mammal species:
American Marten, weasels, and American Red Squirrel. While two indices were not direct
measure of population size, we hypothesized that if two indices describe population changes
accurately over time, they are highly correlated, after accounting for sources of error
described above. We predicted that if one or both of the indices failed to describe annual
population dynamics, correlation coefficient between two indices should not be different
from zero for the species concerned.
32
Field-Site Description
We conducted snow tracking surveys in the Montmorency Forest, a 66 km2 area, about
80 km north of Quebec City (47° 20’N, 71° 10’W), Canada (Fig. 1. 1). Trapping and hunting
for the three species were not allowed in this area. Most of the study area was originally
clearcut between 1941 and 1945 and is now managed with a combination of clear-cuts and
selective cuts. The resulting forest is composed of mature (more than 40-y-old), mid-
succession (21- to 40-y-old) and regenerating (less than 20-y-old) forest stands, which
comprise 55 %, 25 % and 20 % of the study area at the time of the study, respectively.
Locations of different-aged stands shift with time due to continuing timber harvest and forest
stand succession; mean stand age remained stable throughout the study period (42.97 ± 1.67
years, range 0 – 113 y). A dense road network is present, with about 150 km of dirt roads.
During winter, several roads were groomed by machinery for cross-country ski trails.
Elevation ranges from 650 m to 1000 m. From 1999 to 2011, annual mean temperature was
0.3 °C. Annual precipitation was 1417 mm (33 % as snow). Maximum snow depth at weather
station in the Montmorency Forest ranged from 62 to 146 cm in 1999–2011 (Vigeant-
Langlois and Desrochers 2011). Abies balsamea (Linnaeus) Miller (Balsam Fir) dominated
second-growth mature forest stands. Picea mariana (Miller) Britton, Sterns and Poggenburg
(Black Spruce), Betula papyrifera Marshall (White Birch), Populus tremuloides Michaux
(Trembling Aspen), and P. glauca (Moench) Voss (White Spruce) were also common. Recent
(less than 5 y) clear-cuts were generally colonized by Rubus idaeus Linnaeus (Red
Raspberry), balsam fir, and white birch (de Bellefeuille et al. 2001).
We obtained pelt sales data from fur transaction reports of Furbearer Management Unit
(French acronym, UGAF) 39, a 7934 km2 area (Fig. 1.1). In this area, Balsam Fir and Black
33
Spruce were dominant plant species. White Birch, Trembling Aspen, Yellow Birch and
maples were also common (Dussault et al. 2006). Similar to the Montmorency Forest, timber
harvest was conducted in UGAF 39, resulting in a heterogeneous stand mosaic (Dussault et
al. 2006). From 2004 to 2012, mean daily winter temperature ranged from -16 to -9 oC and
total winter precipitation ranged from 165 to 265mm (Environment Canada 2014; Fig. 1. 2).
In summary, the main difference between snow tracking and trapping areas was the
presence of trapping in the latter. Thus, furbearers were possibly more abundant in the snow
tracking area than in the trapping area, potentially resulting in different movement patterns
due to higher density. But we assumed that those two locations exhibited similar population
trends. We based our assumption on the facts that two areas were geographically overlapped
and had similar forest composition and forest management, in addition to the fact that wildlife
populations within 100 km have been shown to vary synchronously in various taxa including
mammals (Liebhold et al. 2004).
Methods
Snow tracking
We conducted snow tracking each winter (20 December - 31 March) from 2004 to 2012.
We counted tracks along a subset of a network including about 150 km of roads, 40 km of
trails, and 60 km of off-trail straight line transects (Fig. 1.3). None of the roads that were
surveyed had been snowplowed. Transect length depended upon snow condition, current
weather, time of day, and personnel availability (Table 1.1). Off-trail transects were randomly
selected from a systematic grid covering the entire study area at the beginning of each year.
However, in the selection process, we removed transects that had been surveyed in the
previous 2 years. For each year, we surveyed selected transects only once to cover larger area
34
as much as possible with a GPS. As a result, we surveyed an average of 91.3km ± 28.9 km
(mean ± SD) of transects each year (Table 1.1). Our surveys on transects were performed
within a 24 to 72 hours after the last snowfall exceeded 3 cm. We recorded all tracks found
within 2 m of either side of the transect lines into a GPS receiver and identified the species,
based on track pattern and size. Conspecific tracks that were within 3 m of a recorded track
were ignored. Temperature and snow depth were measured at an Environment Canada
weather station at the Montmorency Forest.
We georeferenced track and transect data with ArcGIS (Version 10.1, ESRI, 2012) and
then split transects into 200 m fixed-length segments, totaling 4123 200-m segments for the
entire study (Table 1.1). We counted tracks on each transect segment, and generated buffers
with a radius of 100 m around each segment. We selected this fixed-length as a compromise
that limits the number of zeros in the data while retaining a sufficiently large number of
sampling units. Within each buffer, we calculated the mean age of forest stands (weighted by
area), slope (the difference between minimum and maximum elevation), variance of age and
mean elevation. Mean forest stand age was calculated as follows, with age in years and w in
hectares for each forest stand in a buffer:
𝑀𝑒𝑎𝑛 𝑎𝑔𝑒 = ∑ 𝑤1𝑎𝑔𝑒1+𝑤2𝑎𝑔𝑒2+⋯+ 𝑤𝑖𝑎𝑔𝑒𝑖 𝑘=1
𝑖
∑ 𝑤1+ 𝑤2+⋯ + 𝑤𝑖𝑘=1𝑖
(1)
Since buffers occasionally included roads, rivers and lakes, we also calculated the
percentage of vegetation cover inside each buffer.
Population indices from snow tracking data
We assumed that track count is a function of population size, animal activity level,
exposure time of transect since the last disturbance. We developed two population indices.
35
The first was mean track count per 200-m transect segment per hours of exposure since the
last snowfall. This index has the advantage of simplicity, but ratios raise statistical issues
including spurious correlation (Atchley et al. 1976). Thus, we used a second index based on
a generalized linear model (GLM) with a negative binomial distribution and log link by using
package MASS in the R software (Venables and Ripley 2002). We used negative binomial
distribution instead of Poisson due to large number of zero count. In the GLMs, track counts
were function of year as a categorical variable, and combinations of the following covariates:
mean stand age, variance of stand age, mean temperature since last disturbance, exposure
time since last disturbance (snow or wind), slope, mean elevation, proportion of vegetated
area and transect type (road or off-trail). Transect type was included into the model to account
for differences in length of off-trail transect surveyed among study years. Year effect
represents a population index (or ‘anomaly’) by comparing mean counts in a given year with
that in the reference year (2004). Thus, year effect estimates reflected differences in mean
track counts between 2004 and other years, after accounting for the effect of covariates on
track counts described above. We were aware that GLMs do not account for spatial
autocorrelation which likely occurred in track counts. However, average Moran’s I for
American Marten, American Red Squirrel and weasels over study years were respectively
0.044 (range = 0.015-0.080), 0.062 (range = 0.007-0.119) and 0.024 (range = 0.006-0.040).
These values suggest weak spatial autocorrelation in track counts after accounting for
covariates. More importantly, we considered years, not transects, as sampling units for
statistical inference on the comparison of time series. Thus we do not consider spatial
autocorrelation an issue.
36
Pelt sales
We used pelt sales data as a proxy for trapping harvest, based on fur transaction reports,
which were obtained from Quebec’s Ministère des Forêts, de la Faune et des Parcs. Pelt sales
data were composed mainly of trapping harvest and of number of animals hunted. There were
112 trappers in the area who used body-gripping traps (Model 120, 160 or 220). Trapping
season was from 18 October to 1 March of the following year, and there was no harvest limit
for the three species that were studied. However, for American Marten, the ministry asked
trappers to stop trapping American Marten when they caught more females than males late
in the trapping season. We obtained data of trapping effort (number of traps x number of
nights) for American Marten from trappers’ mandatory reports and daily logbooks of the
government. Data for American Red Squirrel and weasels were not available. Average
trapping effort for American Marten from 2004 to 2012 was 70187 [range: 52046 - 87756]
traps x night. Average number of trapping logbooks submitted was 72 [range: 58 -83].
Population indices from pelt sales data
Various confounding factors such as temperature and trapping effort can influence both pelt
sales and track counts. We calculated Pearson product-moment correlations (r) between raw
pelt sales and four factors: trapping effort, daily mean temperature (January to March) (oC),
winter total precipitation (mm) and pelt price in the previous year adjusted for 2012
(Canadian dollars) in the UGAF 39. Only factors which had significant correlation with pelt
sales were accounted in testing correlation between two indices.Testing correlation between
two indices
In order to test if two indices agree, we calculated Pearson correlation between pelt sales
and each of the annual snow tracking indices. If pelt sales have significant correlation with
any of potential external factors such as trapping effort, winter temperature and winter
37
precipitation, we also modeled pelt sales as a function of the significant external factors and
one of population index derived from snow tracking data. When estimated parameter of an
annual tracking index was significantly different from zero, we deemed two indices agreed.
The procedure was replicated for each of population index based on snow tracking. The entire
statistical analysis was performed in the software R (R Development Core Team 2013).
Results
Annual track counts were 139 ± 83 (mean ± SD) [range: 66 - 329] for American Marten,
543 ± 723 (mean ± SD) [range: 63-2061] for American Red Squirrel, and 120 ± 174 (mean
± SD) [range: 11 - 575] for weasels. Mean annual pelt sales for American Red Squirrel,
weasels, and American Marten were 184 [range: 74- 319], 361 [242 - 636], and 921 [629-
1334] pelts respectively. None of the factors among trapping effort, winter temperature,
winter precipitation and pelt price in the previous year were significantly correlated with pelt
sales (-0.64 < r < 0.44, P > 0.05).
Mean of track counts hr-1 200 m-1 for American Red Squirrel, weasels, and American
Marten were 0.029 [0.004 –0.131], 0.006 [0.007- 0.027], and 0.008 [0.006 – 0.016]
respectively. There was strong agreement between mean track counts per unit effort and pelt
sales in American Marten and weasels (Fig. 1.4; Table 1.2, but the agreement was not
statistically significant in the case of squirrel (Fig. 1.4; Table 1.2).
Year effect estimates counts 200 m-1 (compared with 2004) in the best generalized linear
model ranged from -1.76 to 2.34 for American Red Squirrel, -1.54 to 1.91 for weasels, and -
0.61 to 0.81 for American Marten. Year effect estimates agreed strongly with pelt sales of
weasels and American Red Squirrel (Fig. 1.4; Table 1.2), but the agreement was not
38
statistically significant in the case of American Marten (Table 1.2). Agreement between
GLM-based snow-tracking indices and pelt sales was the highest in weasels, followed by
American Red Squirrel and American Marten (Table 1.2).
Discussion
Track counts data were generally in strong agreement with pelt sales over the nine years
of this study, which indicated that both methods capture a real signal in the three taxa that
were studied. Our results differed from those of Thompson et al. (1989), who found no
correlation between track counts and trapping harvest in Short-tailed Weasel, Vulpes vulpes
Linnaeus (Red Fox), or Lynx. Agreement between track counts and pelt sales is remarkable
especially because they were derived from completely different methods, from different sets
of locations, with contrasting trapping effort. Thus, both data sets likely captured population
phenomena occurring over the entire study area.
Nevertheless, agreement between indices may arise from common biases, such as weather
and food availability effects on exploratory behavior and movements. Thus, it could be
argued that the variation in the population indices had little to do with actual population size.
However, there was little confounding effect of trapping effort, mean daily temperatures or
precipitation (snow) on indices, because none of these factors were strongly correlated with
indices. Food-related biases may have occurred, though. Jensen et al. (2011) reported that
success rate of American Marten harvest was lower in mast year of Fagus grandifolia Ehrhart
(American Beech) and Sorbus aucuparia Linnaeus (Mountain Ash) than in year of mast
failure. When food is abundant, American Marten could be less likely to be attracted to bait
associated with trapping devices, possibly leading to underestimates of the relative
abundance. High food abundance might decrease track counts. High food abundance was
39
reported to decrease daily movement length of Mustela nivalis Linnaeus (Least Weasel)
(Klemola et al. 1999), potentially leading to fewer track counts.
American Marten track counts showed a lower degree of agreement with pelt sales than
did weasels. Lower agreement could have arisen from a policy which was implemented by
the Quebec government and the policy recommended that trappers stop trapping adults (H.
Bastien, personal communication). However, we failed to identify when the policy was
implemented. In Michigan, trapping harvest limits greatly impacted the number of fisher
harvested (Hiller et al. 2011).
American Red Squirrel track counts also exhibited a lower degree of agreement with pelt
sales than did weasels. Sales of American Red Squirrel pelts might reflect population sizes
poorly because not all pelts were sold, given that American Red Squirrel is often used as bait
for other furbearers (H. Bastien, personal communication). Furthermore, American Red
Squirrel pelts are considered of no significant commercial value by local trappers. The price
of a American Red Squirrel pelt varied between 0.65 and 1.44 Canadian dollars (CAD) from
2004 to 2012 (where 1 CAD ~ 1 USD), which was much less than the price of a weasels pelt
(range: 2.24 CAD – 8.87 CAD) or a American Marten pelt (range: 44.88 CAD – 121.71
CAD). Thus, the motivation for capture contrasted strongly between species.
Population indices are not substitutes for true estimates of abundance. The indices used
in this study were not compared with population trends obtained from direct measures of
abundance, but given the high correlations obtained, at least in the case of American Marten
and weasels, our study adds to existing support for the use of either snow tracking or pelt
sales. With those indices, one should be able to infer relative inter- annual population
40
changes of the furbearer species studied, and possibly others such as Red Fox and Canada
Lynx. Therefore, use of these indices would be useful to investigate impact of local forestry
and wildlife management on population dynamics of furbearer species.
Acknowledgments
Financial support for this project was provided by a scholarship to T. Kawaguchi from
the “Leadership and Sustainable development Scholarship Program” of Laval University, by
a scholarship to T. Kawaguchi from the Foundation F.-K.-Morrow and by a Natural Sciences
and Engineering Research Council of Canada (NSERC) grant to A. Desrochers. We are
grateful to the 29 skilled field workers who contributed to the collection of snow tracking
data, and to the Ministère des Forêts, de la Faune et des Parcs for providing us with pelt sales
data, trapping effort data and geographical information. We thank D. Fortin, L. Bélanger and
C. Samson for their assistance in the design of the study and W.F.J. Parsons in Centre d’Ètude
de Forêt (CEF) for his assistance in linguistic corrections. Lastly, we thank the anonymous
reviewers for their thoughtful comments and suggestions to improve our manuscript.
41
Figure 1.1. Graphical representation of spatial location of snow tracking site (the Montmorency Forest) and trapping area of Furbearer Management Unit (UGAF) 39, southern Quebec, Canada, 2004 - 2012. Black area indicates the location of snow tracking sites. Gray area indicates the location of trapping area.
42
Figure 1. 2. Inter-annual dynamics of winter precipitation and winter temperature in the study
sites, the Montmorency Forest and the Laurentides Wildlife Reserves (UGAF 39), southern
Quebec, Canada, 2004-2012: a) Winter temperature (oC), b) winter precipitation (mm). The
data for 2006 was not available.
43
Figure 1. 3. Spatial distribution of sampling transects in the Montmorency Forest, southern
Quebec, Canada, 2004-2012. Black lines indicate off-trail transects and gray lines indicate
either roads or trails.
44
Figure 1. 4. Comparison of population trends between snow tracking and pelt sales across three taxa: a) American Marten, b) American Red Squirrel and c) weasels, southern Quebec, Canada, 2004-2012. Two population indices are presented: left) Year effect of a Generalized Linear Model (Year effect GLM), right) tracks/exposure time. Black lines represent pelt sales
45
and gray lines represent population indices of year effect GLM (right) or tracks/exposure time (left). Vertical bars represent standard errors.
46
Table 1. 1. Sampling effort for snow tracking and track counts for American Marten, American Red Squirrel, and weasels at the Montmorency Forest, southern Quebec, Canada, 2004-2012.
Year On
road/trails (km)
Off-trail (km)
Sampling events (days)
200m segment
(n)
American Marten (count)
American Red
Squirrel (count)
weasels (count)
2004 33 17 14 252 70 133 36 2005 60 5 14 325 66 187 11 2006 51 23 12 370 88 63 58 2007 71 16 14 435 140 2061 130 2008 97 23 12 602 329 441 575 2009 58 12 11 346 98 78 76 2010 91 12 8 518 106 251 35 2011 110 13 8 619 163 175 89 2012 91 39 19 656 194 1497 71 Total 662 160 112 4123 1254 4886 1081
Average 74 18 12.4 458 139 543 120 Standard deviation
25 10 3.4 146 83 723 174
47
Table 1. 2. Pearson correlations between pelt sales, and different population indices that were based on snow-tracking over the three species (n = 9): a) American Marten, b) American Red Squirrel, and c) weasels, in southern Quebec, Canada, 2004-2012. Year effect GLM indicates estimates of year effect from a Generalized Linear Model.
Population index r P a) American Marten Mean of count/exposure hours 0.71 0.032
Year effect GLM 0.55 0.12 b) American Red Squirrel Mean of count/exposure hours 0.57 0.1
Year effect GLM 0.77 0.02 c) weasels Mean of count/exposure hours 0.87 0.002
Year effect GLM 0.85 0.004
49
CHAPTER 2 – INFLUENCES OF CURRENT AND RECENT
CONSPECIFIC DENSITY ON HABITAT SELECTION OF SNOWSHOE
HARE
TOSHINORI KAWAGUCHI, Centre d’étude de la forêt, and Département des sciences
du bois et de la forêt, Université Laval, 2405, rue de la Terrasse, Québec (Québec), G1V
0A6, Canada
ANDRÉ DESROCHERS, Centre d’étude de la forêt, and Département de science du bois
et de la forêt, Université Laval, 2405, rue de la Terrasse, Québec (Québec), G1V 0A6,
Canada
Key word: habitat selection, density-dependency, time lag, ideal free distribution, Lepus
americanus
50
Abstract
Ideal free distribution theory predicts that conspecific density affects habitat selection, but
possible lags in response to the density remain poorly documented. Snowshoe hare (Lepus
americanus) are known for their marked inter-annual variation in population density,
possibly leading to strong density-dependent effects on habitat selection. Based on 11 years
of snow tracking in southern Quebec (Canada), we investigated whether snowshoe hare
habitat selection exhibits instantaneous and delayed responses to its population density. We
measured the relationship between track counts in 100 m transect segments and the
proportion of forest stands of different age classes within 50 m of the transect. We developed
an index of population density with generalized estimating equations (GEE) looking at track
counts in response to year as a categorical variable. Snowshoe hares spread significantly from
preferred forest stands when density in the previous winter was high. To our knowledge, no
previous empirical studies have documented a lagged response to population density in
habitat selection. Time lags offer a possible explanation for deviations, which have appeared
in empirical studies of density-dependent habitat selection, from the ideal free distribution.
51
Introduction
Habitat selection is a process involving responses to a large variety of stimuli, such as
vegetation structure and composition, predation risk (Hildén 1965), weather (Reid et al. 2012),
and conspecific population density (Fretwell and Lucas 1970). The effect of conspecific
population density on habitat selection has been documented for various taxa, including
mammals (e.g., fat sand rat Psammomys obesus, (Shenbrot 2004); white-footed mouse
Peromyscus maniculatus, (Morris 1996); domestic sheep Ovis aries, (Mobæk et al. 2009)),
birds (e.g., brown-headed cowbird Molothrus ater; Jensen and Cully 2005), and fish (e.g.,
brown trout Salmo trutta; Ayllón et al. 2013). Under the ideal free distribution model
(Fretwell and Lucas 1970), animals move freely and rapidly among habitats of different
quality in order to maximize their fitness (Morris 2003). According to Morris’ (2003) isodar
model, fitness decreases with increases in density, which may lead individuals to move from
high-density to lower-density habitat.
Time lags are often observed in processes such as species redistribution following
landscape changes (Metzger et al. 2009), local extinction after deforestation (Brooks et al.
1999), species invasion (Crooks 2005), and population dynamics (Fryxell et al. 1991, Erb et
al. 2001). Time lags can be caused by: 1) long processing times following the perception of
a stimulus (Brooks et al. 1999), 2) intervening processes between two processes of interest
(Magnuson 1990), and 3) feedbacks (Framstad et al. 1997).
While several empirical studies have documented the ideal free distribution (Dreisig 1995,
Haché et al. 2012), deviations from ideal free model have been reported. Those deviations
have been interpreted as resulting from limited perceptual constrains or despotic behavior
(Oro 2008). However, to our knowledge, none has examined whether responses to density
52
are lagged, which may account for departures from IFD. Yet the perception of a decrease in
fitness or an increase in density, or that of available nearby habitats (Rosenzweig 1981), is
unlikely to be an instantaneous process.
Here, we documented population dynamics of snowshoe hare (Lepus americanus) over
11 winters, evaluated whether habitat selection of hare was affected by density, and tested
whether responses to density are delayed. More specifically, we measured the strength of
association between hares and forest stands of different age classes to identify the most and
least preferred habitats, based on winter track counts. Snowshoe hare was used as a model
species because of its strong population fluctuations (Krebs 2001, 2011) and reported
deviations from an ideal free distribution (Morris 2005). We predicted that at high population
densities, snowshoe hare distribution expands into less preferred habitat, thus weakening the
association between hares and most preferred habitat. Because we expect lags in snowshoe
hare responses, we predicted that the lagged shift in distribution would be stronger in the
winter following high density rather than in the current winter.
Methods
We conducted snow tracking surveys at the Montmorency Forest, a 66 km2 boreal forest
about 80 km north of Quebec City (47°20’N, 71°10’W), Canada. Most of the study area was
originally clear-cut between 1941 and 1945 and is now managed with a combination of clear-
cuts and selective cuts. The resulting forest is composed of mature (more than 40-y-old), mid-
succession (21- to 40-y-old) and regenerating (less than 20-y-old) forest stands, which
comprise ca. 55 %, 25 % and 20 % of the study area, respectively. Locations of different-
aged stands shift with time due to continuing timber harvest and forest stand succession:
mean stand age remained stable throughout the study period (43.32 ± 1.98 years; mean ± SD,
53
range 0 – 114 y). A dense road network is present, with about 150 km of roads, i.e., over 2
km/km2. During winter, several roads were groomed by machinery for cross-country ski trails.
Elevation ranges from 650 m to 1000 m. From 1999 to 2011, annual mean temperature was
0.3 °C. Annual precipitation was 1417 mm (33 % as snow). Maximum snow depth at weather
station in the study site ranged from 62 to 146 cm in 1999–2011 (Vigeant-Langlois and
Desrochers 2011). In our study area, balsam fir (Abies balsamea) dominates second-growth
mature forest stands. Black spruce (Picea mariana), white or paper birch (Betula papyrifera),
trembling aspen (Populus tremuloides), and white spruce (P. glauca) are also common.
Recent (less than 5-y-old) clear-cuts are generally colonized by red raspberry (Rubus idaeus),
balsam fir, and white birch (de Bellefeuille et al. 2001).
We conducted snow tracking each winter (20 December - 31 March) from 2004 to 2014.
We counted tracks along a subset of a network that included about 150 km of roads, 40 km
of trails, and 60 km of straight line transects inside forest stands. None of the roads that were
surveyed had been snow-plowed. Transect length depended upon snow condition, weather,
time of day, and personnel availability (Table 2.1). Each year, we surveyed selected transects
only once to cover larger area as much as possible, and surveyed tracks along 94.73 km ±
30.57 km (mean ± SD) of transects (Table 2.1). We surveyed transects within a 24 to 72 hours
after the last snowfall exceeded 3 cm. We recorded all hare tracks that fell within 2 m of
either side of the transect lines into a GPS receiver. Hare tracks that were within 3 m of a
recorded track were ignored. Any of single track, a trail and a network of hare track was
recorded as one track.
We geo-referenced track and transect data with ArcGIS (Version 10.1, ESRI 2012) and
split transects into 100 m segments, totaling 10436 100-m segments for the entire study
54
(Table 2.1). We counted tracks along each transect segment, and generated buffers with a
radius of 50 m. Winter home range size of snowshoe hare averages 2 ha (Beaudoin et al.
2004), thus we considered 2 ha of resulting buffer size as an sufficient size that reflects spatial
scale of hare. Within each buffer, we calculated the mean age of forest stands (weighted by
area), slope (the difference between minimum and maximum elevation), mean elevation, and
the proportions of the area occupied by 4 habitat types, based on forest stand age (0-20 y, 20-
40 y, 40-80 y). Older forest stands (older than 80 y) were rare and not included in the analyses.
Since buffers occasionally included roads, rivers and lakes, we also calculated the percentage
of vegetated area inside each buffer.
For indexing population density, we used model coefficients for year as a categorical
effect in a Generalized Estimating Equation (GEE) with a negative binomial distribution and
log link by using package geeM in the R software (McDaniel and Henderson 2015). GEE
was used to account for local spatial autocorrelation. In GEE, we used transect unit before
segmetation as a cluster. The model included the following covariates: hours of exposure
since last disturbance, mean stand age, squared mean age, slope, mean elevation, mean
temperature in the previous 24 hours, month as a categorical variable and year as a categorical
variable. The squared term was added because habitat use pattern showed a peak around 40y
stand age (Hodsons et al. 2011), thus resulting in better fit. Month was included into the
model to account for potential decline of hare population over winter (Kielland et al. 2010).
We validated this approach with trapping success data that had been obtained from the same
region; annual population indices that had been derived from snow tracking were highly
correlated with trapping success (Kawaguchi et al. in press).
55
To investigate our initial prediction, a GEE was used with a negative binomial distribution
and log link. In GEE, track counts were a function of hours of exposure since last disturbance,
proportion of each habitat type (regenerating forest: 0-20 y, young forest: 20-40 y, mature
forest: 40-80 y), slope, mean elevation, transect type (forest vs road), percentage of vegetated
area within the buffer and month (categorical variable). Our categorization of habitat types
was based on hare habitat use patterns in which hare appears at high density in 20-40 y forest,
also to avoid mixing preferred and unpreferred habitats in a category.
We developed a model for each habitat type and each year, and we used model estimates
for the effect of the proportion of each habitat as measures of habitat preference. To compare
immediate and delayed effects of snowshoe hare population density on habitat preference,
we used linear models for the preference as a function of population density in current (Dt)
and previous (Dt-1) winters separately and combined. The model integrated weight for each
observation which were calculated as following:
wi = (1/SEi)/(1/SE1 + 1/SE2 + … + 1/SEk)
Where wi is a weight for measured habitat preference i, SE is a standard error of estimated
coefficient of habtiat variable. Since we used lagged effect, we excluded habitat preference
in 2004 from the analysis. A model with the highest adjusted R2 was considered as the best
model. All statistical analyses were conducted in the R statistical environment (R
Development Core Team 2014).
Results
We found 14240 snowshoe hare tracks in total and track counts/km varied from 4.9 to
23.7 (Table 2.1).
56
The association between track counts and the proportion of 0- to 20-y-old stands was
negative each year (range of model estimates: [-0.011, -0.0001]), suggesting that this habitat
was the least preferred. In contrast, the relationship between track counts and the proportion
of 20- to 40-y-old stands was positive each year (range: [0.001, 0.012]), suggesting that 20-
to 40-y-old stands were most preferred. Relationships between track counts and the
proportion of older stands were either positive or negative, depending upon the year.
The lag model for response to 20- to 40-y-old performed best among the candidate models.
The lagged effect of density was significantly negative, suggesting that hare more frequently
used 20-to-40-y-old forest in response to higher density in the previous winter (Table 2.2;
Figure 2.2). In other habitat types (0- to -20 y forest and 40-80-y-forest), neither immediate
nor lagged effect were significant (Table 2.2; Figure 2.2) while lagged effect in the other
habitat types showed positive.
Discussion
Snowshoe hares wintering at the Montmorency Forest appeared to respond spatially to
their population density with a lag of one year. The models including delayed effects
explained the dynamic associations of hares with most preferred habitats much better than
the model with immediate responses to population density. In contrast, population density
poorly explained variation in hare responses to forest stands that were 40-to-80 y and 0-to-
20y.
The preference for forest stands that were 20- to 40-y-old was consistent with other
studies (Thompson et al. 1989, Hodson et al. 2011). The avoidance of hares of 0- to 20-y-old
forest stands was also consistent with past studies (Thompson et al. 1989, Potvin et al. 1999).
57
Also, avoidance could be explained by the fact that these stands were mostly covered by snow,
thus offering few opportunities for foraging as well as high predation risk.
The signs of the model estimates in 20-40 y forest stand associations can be explained by
an overflow of individuals from the preferred habitat, in response to changes in population
density. We interpret the lagged response of hares as a ‘buffer effect’ (Brown 1969, Gill et
al. 2001). The apparent overflow of hares from high-density habitat was delayed. Presence
of delayed effect would be due to delays in perception of stimuli (increased density) or that
of nearby available habitat.
We expected a lagged shift of snowshoe hare use into the less preferred habitat (0-to-20y
and 40-to-60y stand), thus positive coefficients of the lagged effect in those habitats. Contrary
to our expectation, none of lagged effect in these habitat appeared non significant while the
sign of coeffcients showed positive. This pattern could be attributed to higher mortality in
low-density habitat, potentially resulting in offsetting increased habitat use by dispersing hare.
As younger forest stands were more open and often partly covered by a thick snow layer
(Horstkotte and Roturier 2013), hares in this habitat would be more vulnerable to predators
such as Canada lynx (Lynx canadensis). Immediate and deferred costs of dispersal are known
to lower survival rates (Stamps et al. 2005). Thus, dispersed hare might have low survival
rate.
Numerous empirical studies on density-dependent habitat selection have been performed;
however, most of these studies showed at least one case in which animals did not follow the
ideal free distribution (Morris and MacEachern 2010). For example, in a study on snowshoe
hares in northwestern Ontario, Morris (2005) indicated that they exhibited subtle density-
58
dependent habitat selection, with large residual variation. In an Idaho study, local hare
extinction was not mitigated by greater densities in neighboring areas (Thornton et al. 2012b).
As demonstrated here, using a time-lag effect can significantly improve the explanatory
power of density-dependent habitat selection models.
Acknowledgements
Financial support for this project was provided by the “Leadership and Sustainable
Development Scholarship Program” of Laval University to T. Kawaguchi, by a scholarship
to T. Kawaguchi from the Foundation F.-K.-Morrow and by a Natural Sciences and
Engineering Research Council of Canada (NSERC) grant to A. Desrochers. We are grateful
to the 29 skilled field workers who contributed to the collection of snow tracking data, and
the Montmorency Forest for logistical support. We thank D. Fortin, L. Bélanger and C.
Samson for their assistance in the design of the study, and W.F.J. Parsons in Centre for Forest
Research (CFR) for linguistic corrections.
59
Table 2.1. Sampling effort for snow-tracking of snowshoe hare (Lepus americanus) in the Montmorency Forest, southern Quebec (Canada), 2004-2014.
Year km sampled 200 m
segment (n)
Track count On
road/trails Off-trail
2004 34 19 527 597 2005 61 6 671 390 2006 52 23 751 425 2007 72 17 890 943 2008 99 24 1234 1684 2009 59 12 715 344 2010 93 12 1055 834 2011 112 14 1263 1937 2012 93 42 1352 2896 2013 113 20 1325 3147 2014 52 13 653 1043 Total 840 202 10436 14240 Mean 76 18 949 1295 SD 27 9 306 996
60
Table 2.2. Estimated effects of current and lag density (previous winter) on habitat selection of snowshoe hares in the Montmorency Forest, Québec, 2004-2014 (n = 10). Estimates are shown for models including either current or lagged effects of density. Positive estimates indicate a greater association at higher density. Adjusted R2 values can be negative, because unlike raw R2, they are penalized by the number of parameters.
Stand age
Adjusted R2 Model estimates ( ± s.e)
Current Lag Current + lag Current density (Dt) Lag density (Dt-1)
0-20 y 0.13 0.09 0.04 0.0039 ± 0.003, P = 0.16 0.0038 ± 0.003, P = 0.21 20-40 y 0.39 0.47 0.46 -0.0034 ± 0.0013, P = 0.03 -0.0034 ± 0.0011, P = 0.02 40-80y -0.11 -0.08 -0.23 0.0003 ± 0.001, P = 0.82 0.0008 ± 0.001, P = 0.58
61
Figure 2. 1 Estimated population index of snowshoe hare over 11 years from 2004 to 2014 in the study site. The index was developed from coefficients of the year effect estimated from generalized estimating equations (GEE). Vertical bars indicate standard errors.
62
Figure 2.2. Association of hares with 0-20y forest stands explained by a) immediate effect (Dt) only and b) with immediate and time-lag effects (Dt + Dt-1) of the density index over 10 years (2005 - 2014) in the study site (n = 10). Points with standard error bars indicate model coefficients. Dashed lines indicate 95 % confidence bands of the fitted regression line values.
63
CHAPTER 3 – VARIATION OF SNOW DEPTH AFFECTS THE
SPATIAL DISTRIBUTION OF SNOWSHOE HARE
TOSHINORI KAWAGUCHI,1 Centre d’étude de la forêt, Université Laval, 2405, rue de
la Terrasse, Québec (Québec), G1V 0A6, Canada
ANDRÉ DESROCHERS, Centre d’étude de la forêt, Université Laval, 2405, rue de la
Terrasse, Québec (Québec), G1V 0A6, Canada
Key words: Snow depth, Habitat use, Snowshoe hare, LiDAR
64
Abstract
Snow accumulation changes access to vegetation by herbivores. Snowshoe hare (Lepus
americanus), a keystone species in the boreal forest, is known to most frequently use young
forest stands during winter. However, snow accumulation greatly varies in time and space,
possibly affecting habitat use by snowshoe hare, especially in regions where snow
accumulation is high. We measured shifts in habitat use by snowshoe hare as a function of
snow depth at the Montmorency Forest, Quebec, Canada. We surveyed on 67 km of transects
over 3 winters, found 2239 hare tracks and measured snow depth in 336 locations. We
analyzed track counts as a response to foliage density above 2 m indexed by penetration rate
obtained from LiDAR imagery, snow depth and the interaction between snow depth and
foliage density as explanatory variables. Snowshoe hares were less frequently found in sites
with high foliage density at the middle height but were found more frequently in sites with
high foliage density when snow accumulation increased. We concluded that snow depth
dynamics may introduce significant uncertainty in spatial distribution models for the species,
and possibly its interactions with predators.
65
Introduction
Snow depth greatly varies seasonally, inter-annually and geographically (Brown and
Braaten 1998). Snow depths in boreal forests in eastern North America often exceed 1m, and
occasionally 2m (Brown and Bransnett. 2010). However, snow depth varies greatly at the
forest stand scale, particularly in response to tree height and canopy cover (Horstkotte and
Roturier 2013). Snow depth often plays a significant role for wildlife including overwinter
survival, dispersal (Kielland et al. 2010, Campbell et al. 2005) and the supply of subnivean
space (Korslund and Steen 2006), daily movements and activity levels (Fuller 1991, Murray
and Boutin 1991), as well as access to food and thermal cover (Wolff 1980, Halpin and
Bissonette 1988, Morrison et al. 2003). If small trees, herbaceous vegetation and shrubs are
buried by deep snow, availability of understory cover may change. However, deep snow may
allow small herbivores to access to browse at higher levels.
Snow depth is often predicted to vary in response to climate change (Campbell et al.
2005). As a consequence, small herbivores such as the snowshoe hare (Lepus americanus)
should respond to changes in snow depth. Hare is strongly dependent on understory cover
(Litvaitis et al. 1985), mostly found in 20 to 40 y old boreal forests in eastern North America
(Thompson et al. 1989, Hodson et al. 2011).
Approximately 10 % of the geographical range of snowshoe hare (Figure 3.1) experiences
more than 1m of snow depth on a regular basis, and may therefore exhibit greater temporal
variation in the spatial distribution of small herbivores, both within years, as snow
accumulates, and among years. An early study, without detailed statistical analysis argued
that habitat shifts from summer to winter (Wolff 1980), but we know of no quatitative studies
examining spatial dynamics of small herbivores in response to snow depth during winter.
66
Here, we investigated whether snow depth influenced habitat use by snowshoe hare in a
boreal forest of southern Quebec, Canada. We predicted that with increasing snow depth,
snowshoe hares should less frequently use sites dominated by understory cover. Deeper snow
should provide access to vegetation above 2m, thus we predicted that hares should be more
strongly associated with higher density of foliage above 2m in high snow depth.
Methods
Study site
We conducted snow tracking surveys at the Montmorency Forest, a 66 km2 boreal forest
about 80 km north of Quebec City (47° 20’N, 71° 10’W), Canada. Most of the study area was
originally clear-cut between 1941 and 1945 and is now managed with a combination of clear-
cuts and selective cuts. The resulting forest is composed of mature (more than 40-y-old), mid-
succession (21- to 40-y-old) and regenerating (less than 20-y-old) forest stands, which
comprise ca. 55 %, 25 % and 20 % of the study area at the time of the study, respectively.
Locations of different-aged stands shift with time due to continuing timber harvest and forest
stand succession: mean stand age remained stable throughout the study period (43.3 ± 2.0 y,
range 0 – ca. 120 y). A dense road network is present, with about 150 km of dirt roads, i.e.,
over 2 km/km2. During winter, several roads were groomed by machinery for cross-country
ski trails. Elevation ranges from 650 m to 1000 m. From 2012 to 2014, mean winter
temperature was ranged from -14.5 to -9.2 °C. Maximum snow depth at the weather station
of the study area ranged from 58 to 111 cm. The Montmorency Forest exhibits higher snow
depth than in most of snowshoe hare’s range (Figure 3.1). In our study area, balsam fir (Abies
balsamea) dominates second growth mature forest stands. Black spruce (Picea mariana),
67
white or paper birch (Betula papyrifera), aspen (Populus tremuloides) and white spruce (P.
glauca) are also common (de Bellefeuille et al. 2001).
Assessing snowshoe hare spatial distribution
We conducted snow tracking each winter (1 January - 31 March) from 2012 to 2014.
Each year we counted tracks along a subset of 60 km of straight line transects inside forest
stands. Transect length depended upon snow condition, weather, time of day, and personnel
availability (Table 3.1). Off-transects were randomly selected from a systematic grid
covering the entire study area (Figure 3.2). We surveyed selected transects only once, with a
GPS receiver. As a result, we surveyed an average of 22 km ± 12 km (mean ± SD) of transects
each year (Table 3.1). Our surveys on transects were performed within a 24 to 72 hours after
the last snowfall exceeding 3 cm. We recorded all hare tracks found within 2 m of either side
of the transect lines into a GPS receiver. Hare tracks that were within 3 m of a recorded track
were ignored. Any of single track, a trail and a network of hare track was recorded as one
track. During snow tracking, we measured snow depth at 100-m intervals by pushing a stick
down until it hit the ground, marking the level of the snow surface, and measuring the length
between the tip of the stick and snow level marker.
Understory cover, stand height and foliage density
We measured understory cover in 2011 and 2012 along line transects used for snow
tracking, from August to November, i.e., in the absence of snow. We defined understory
cover as a visual estimate of the proportion of ground covered by live herbaceous of shrub
and conifer vegetation lower than 1.5 m from the ground, within 2 m on each side of the line
transect, at 50 m intervals. Thus, resulting understory cover plots were rectangular, 4 m x 50
m.
68
We obtained stand heights with a Light Detection and Ranging (LiDAR) image. LiDAR
is a remote-sensing technology that can indirectly characterize forest structure including
forest biomass, foliage density, canopy structure and tree height (Lim et al. 2003, Wulder et
al. 2008). 95% LiDAR return height was used as a proxy for tree height in our statistical
analysis. And LiDAR penetration rate above 2m from ground was used as an indicator of
foliage density. Airborne LiDAR data were obtained in August 2011 by using an Optech
ALTM 3100 sensor at a pulse repetition rate of 100 kHz, laser wavelength of 1046 nm,
divergence of 0.25 mrad and scan rate between 46 Hz and 56 Hz (For more details, see Racine
et al. 2014). The LiDAR image resolution was 5 m.
Georeferencing snow tracking data
We georeferenced track and transect data with ArcGIS (Version 10.1, ESRI, 2012) and
split transects into 200 m segments, totaling 336 200-m segments for the entire study (Table
3. 1). We obtained track counts on each transect segment, and generated buffers with a radius
of 100 m. Since home range of snowshoe hare ranged from 2 ha to 10 ha (Beaudoin et al.
2004), we considered 7 to 8 ha of resulting buffer size as an sufficient size that reflects spatial
scale of hare. Within each buffer, we calculated slope (the difference between minimum and
maximum elevation), mean elevation, proportion of low trees (0-6m) grids, understory cover,
mean penetration rate above 2 m and mean snow depth. Since we did not measure understory
cover for 2013 and 2014, we used 0-6 m of mean tree height as a surrogate for abundance of
understory cover because of its higher correlation with measured understory cover than either
0-2 m or 0-4 m (n = 146, r = 0.48, 0.35, 0.14 for 0-6m tree, 0-4m tree, 0-2 tree respectively).
To account for roads, rivers and lakes occasionally included in buffers, we calculated the
proportion of vegetated area.
69
Statistical analysis
We modeled track counts as a response to the following covariates: exposure time,
penetration rate above 2 m, snow depth, understory cover, vegetated area, 0-6m tree grid,
year effect (categorical variable), slope, mean elevation and mean temperature. Year effect
was integrated to account for possible relative changes in hare population size among the
study years (Mowat et al. 2003). We used two models: a) penetration rate model and b)
understory cover model (Table 3.2). Each model corresponded to two predictions: a) a
negative interaction term between penetration rate and snow depth, indicative of hares
responding to deep snow by increasing their use of foliage at middle height, and b) the
interaction term between understory cover and snow depth will be significantly negative if
snowshoe hares reduce their use of sites with high understory cover when snow is deeper.
Using generalized linear models (GLMs) with a negative binomial distribution and log
link, we verified that the interaction model representing our principal hypothesises received
the highest support comparing with other candidate models which do not include the
interaction term (Table 3.3). We selected the best-supported models with an information-
theoretical approach, thus by Akaike Information Criterion (AIC) (Burnham and Anderson
2002).
In order to account for effect of local spatial correlation, we reran the best supported
model by using generalized estimating equations (GEE) procedure with a negative binomial
distribution and log link by using the package geeM (McDaniel and Henderson 2015). With
GEE, we used transect unit before segmetation as a cluster. Parameter estimation was based
on quasi-likelihood. Significance of parameters was determined by Wald tests using
estimated robust standard errors. For the model b), we used the dataset only from 2012. All
70
statistical analyses were conducted in the R statistical environment (R Development Core
Team 2015).
Results
We found 2239 tracks in total, and measured snow depth at 336 locations. Averaged snow
depth in the field ranged from 33.8cm to 186.5cm, depending on advancement of winter, and
location.
Among the candidates of penetration models, the highest support evidenced by the
highest Akaike weight was given to the model including the interaction between snow depth
and penetration rate, snow depth, proportion of low tree grid, year effect, slope, elevation and
exposure time (Table 3.3a). In the best model, the interaction term between penetration rate
and snow depth wassignificantly negative (Table 3.4a), i.e., hare track counts higher at foliage
density above 2m particularly with deeper snow (Figure 3.3a).
Among the candidate models for understory cover, the highest support was received by
the model including the interaction between snow depth and vegetation cover, exposure time,
snow depth and understory cover (Table 3.3b). In the best model, the interaction term
between understory cover and snow depth was significantly negative (Table 3.4b), i.e., hare
track counts became lower at sites with high understory cover particularly with deeper snow.
Discussion
The results were consistent with our predictions, namely that snowshoe hares respond to
forest structure differently depending on snow depth. Hares exhibited strong preference
toward sites dominated by understory cover under low snow depth. The stronger association
71
with understory cover was consistent with several studies from low snow depth sites and
snow-free season (Wolfe et al. 1982, Litvaitis et al. 1985). However, as snow accumulated,
lower access to understory vegetation reduced the relative use of sites dominated by
understory cover.
We interpret the strong association between hare track counts and foliage density above
2m from ground under the condition of high snow depth as a result of an “elevator effect”
giving access to browse at higher positions from ground. If foliage density above 2 m acted
simply as cover which was also an important determinant of snowshoe hare distribution
(Hodson et al. 2010a), no statistical interaction between the effects of foliage density and
snow depth would be expected.
Food availability could be a key element in explaining their habitat use pattern. It was
reported that hare put higher priority on vegetation cover than food (Hodges 1999). Hodson
et al. (2011) reported that hare population density was lower at sites with higher food
availability (deciduous twigs) than at sites with lower food availability. If food was constraint
on habitat selection of hare, it wouldn’t show lower hare density at higher food availability
site. Thus, food did not affect habitat selection of hare and food availability was unlikely to
work as a constraint in forming their spatial pattern.
Several studies have documented changes in habitat use of small mammals due to changes
in snow depth due to the significant roles of snow depth (e.g., Reid et al. 2012). The majority
of studies have focused on phenomena under the snow surface, thus looking at supply of
thermoregulation by snow rather than facilitated access to foliage and vegetation cover.
Consequently, little focus has been given to “elevator effect”. While a few studies have
72
described habiat shift of snowshoe hare from summer to winter (Wolff 1980), to our
knowledge, our study is the first example that quantitatively demonstrated likely result of
“elevator effect” by high snow depth.
Our results would have important implications to wildlife management under future
climate change. Brown and Braaten (1998) reported that snow depth had increasing trend
from 1946 to 1995 in east coast of Canada while there was decreasing trend in snow depth in
other regions of Canada. Variation of the trend was also observed in the United States
(Kunkel et al. 2009). Recent studies indicated that climate change possibly affect dynamics
of snow depth (Christensen et al. 2013). Since our study demonstrated the change in habitat
use by hares due to high snow depth, fluctuation in future climate possibly affects winter
behavior and distribution of hare and in turn affects distribution of associated predators
including Canadian lynx and American marten (Apps 1999, Powell et al. 2003). We
concluded that snow depth dynamics may introduce significant uncertainty in spatial
distribution models for the species, and its interactions with predators.
Acknowledgements
Financial support for this project was provided by the scholarship “Leadership and
Sustainable Development Scholarship Program” of Laval University to T. Kawaguchi, by a
scholarship to T. Kawaguchi from the Foundation F.-K.-Morrow and by a Natural Sciences
and Engineering Research Council of Canada (NSERC) grant to A. Desrochers. We are
grateful to the 9 skilled field workers who contributed to the collection of snow tracking data,
the Montmorency Forest for logistical support. We appreciated J-C. Ruel, J. Bégin, E.B.
73
Racine for providing us with LiDAR data. We thank D. Fortin, L. Bélanger and C. Samson
for their assistance in the design of the study, I.D. Thompson for his advice on the manuscript.
74
Figure 3. 1. Satellite snow depth over distribution of snowshoe hare (Lepus americanus). Satellite data was obtained from Canadian Meteorological Centre (Brown and Brasnett 2010). The date of measurement for the map was 1 March in 2012. The resolution was 24km x 24km. The location of the study site was represented by a black star. The histogram showed frequency distribution of snow depth over hare distribution. Snow depth at the study site was 85.1cm on this date.
75
Figure 3. 2. Vegetation map and sampling location in the study site, 2012-2014. The vegetation map was produced by using the one in 2012,
76
Figure 3. 3. Effect of interaction between snow depth and vegetation structure on habitat use by hares at the Montmorency Forest, Canada, 2012-2014: A) hare response to LiDAR penetration rate, B) hare response to mean understory cover under different snow depth. High penetration rate values indicate low foliage density above 3m from ground.
77
Table 3. 1. Sampling effort for snow-tracking for snowshoe hare (Lepus americanus) at the Montmorency Forest in southern Quebec, Canada, 2012 -2014.
Year Track count Km sampled 200m segments (n) 2012 1227 36 180 2013 792 18 93 2014 220 13 63 Total 2239 67 336 Mean 746 22 112 SD 505 12 61
78
Table 3. 2. List of models for testing hypothesis regarding effect of interaction between LiDAR derived tree height, penetration rate, regenerating forest and snow depth on track counts of hare. X indicates a corresponding variable included into the model.
Variable Models
Penetration rate Understory cover
Exposure time since disturbance X X Understory cover X 0-6m tree grids X Vegetated area X
LiDAR penetration rate X Year effect X Snow depth X X
Penetration rate x Snow depth X Understory cover x Snow depth X
79
Table 3. 3. Model comparison among candidate models for habitat use by snowshoe hare in the Montmorency Forest, Quebec, 2012-2014. The variable YEAR was treated as a categorical variable. The w indicates Akaike weight.
Model K AIC w a) LiDAR penetration rate Penetration rate + 0-6m tree + Vegetated area + Snow depth + YEAR + Temperature + Slope + Elevation + Snow depth x Penetration rate
11 1883.76 0.96
Penetration rate + 0-6m tree + Vegetated area + Snow depth + YEAR + Temperature + Slope + Elevation 10 1889.88 0.04
Penetration rate + Snow depth + Vegetated area + YEAR + Temperature 9 1911 <.01
Penetration rate + Snow depth + Vegetated area + YEAR 6 1915.92 <.01 Penetration rate + Snow depth + Vegetated area + YEAR + Temperature + Slope + Elevation 8 1918.66 <.01
b) Understory cover Exposure + Understory cover + Snow depth + Understory cover x Snow depth 4 848.08 0.83
Exposure + Understory cover + Snow depth + Slope + Elevation + Understory cover x Snow depth 6 851.59 0.14
Exposure + Understory cover + Snow depth 3 855.43 0.02
Exposure + Understory cover + Snow depth + Slope + Elevation 5 859.21 <.01
80
Table 3. 4. Estimated effects of snow depth on habitat use by snowshoe hare in the Montmorency Forest, Quebec, 2012-2014: a) hare response mean LiDAR penetration rate (n = 336) and b) hare response to proportion of understory cover (n = 180) under different snow depth.
Variable Value SE P a) LiDAR penetration rate
(Intercept) 3.0019 1.2815 0.02 exposure time 0.0013 0.0057 0.81
Penetration rate -0.0242 0.0246 0.33 0-6m tree 0.0415 0.0122 <0.01
Snow depth 0.0147 0.0047 <0.01 Vegetation area -0.0007 0.0079 0.93
YEAR 2013 0.2886 0.2365 0.22 YEAR 2014 -0.4065 0.2634 0.12
Slope 0.002 0.0079 0.8 Elevation -0.0018 0.001 0.07
Mean temperature 0.025 0.0179 0.16 Penetration rate x
Snow depth -0.0004 0.0001 <0.01
b) Understory cover
(Intercept) 1.0632 0.6902 0.12 Exposure time -0.0034 0.0074 0.65
Understory cover 0.0818 0.0268 <0.01 Snow depth 0.0105 0.005 0.03
Understory cover x Snow depth -0.0009 0.0002 <0.01
81
CHAPTER 4 – WINTER SPATIOTEMPORAL DYNAMICS OF A
BOREAL PREDATOR-PREY COMPLEX
TOSHINORI KAWAGUCHIa, ANDRÉ DESROCHERSa,
a Centre d’étude de la forêt, Université Laval, 2405, rue de la Terrasse, Québec (Québec),
G1V 0A6, Canada
82
Abstract
1. Habitat selection by mesopredators is subject to trade-offs between energy gain from prey,
competition and predation risk from larger predators. These trade-offs should lead to
variability in the spatial association between large predators, small predators, and their prey.
2. We explored spatio-temporal relationships among five mammal species including
American marten (Martes americana), snowshoe hare (Lepus americanus), red squirrel
(Tamiasciurus hudsonicus), red fox (Vulpes vulpes) and Canadian lynx (Lynx canadensis).
More specifically, we examined whether spatial association between marten, a mesopredator,
and hare was influenced by the presence of predators, fox and lynx, as well as fox-hare or
lynx-hare associations, based on 11 years of snow-tracking along 976 km of transects in
southern Quebec (Canada). We also examined whether martens responded spatially to high
squirrel abundance when larger predators were present or when hare-marten association was
weak.
3. With path analyses, we built directed acyclic graphs depicting spatial associations among
species each year separately. We modeled marten-hare path coefficients as a function of lynx-
hare coefficients, fox-hare coefficients, marten-hare coefficients in the previous winter,
marten, lynx, fox population indices in current and previous year. We modeled marten-
squirrel path coefficients as function of lynx-hare coefficients, fox-hare coefficients, marten-
hare coefficients and hare population index.
4. Spatial association between marten and hare was weak or nonexistent when lynx was
spatially associated with hare. Hare-marten association was best explained by lynx
population index in the previous year and was negatively correlated with the lynx population
index. The model for squirrel-marten association including lynx population index in the
previous winter and hare-marten association performed best among twelve candidate models.
83
5. Our first prediction was not rejected, thus suggesting that habitat selection by marten, a
mesocarnivore species, was influenced by larger predator abundance in the previous year.
Evidence for switching to alternative prey was slightly weak. Presence of larger carnivores
likely led to uncertainty into predicting spatial distribution of mesocarnivores based simply
on their prey distribution.
Key words: Martes Americana, Lepus americanus, prey switching, community ecology,
competition, habitat selection
84
Introduction
Habitat selection is a decision-making process whereby individuals preferentially use, or
occupy, a non-random set of available habitat (Morris 2003). The process involves responses
to a large variety of stimuli, such as vegetation structure and composition, predation risk
(Hildén 1965), weather (Reid et al. 2012, Crowther et al. 2014), conspecific population
density (Fretwell and Lucas 1970). Individuals are expected to use space to maximize their
fitness (Morris 2003). In a prey-predator interaction, habitat selection by predators often
translates into occupying areas with high prey abundance. For example in snow tracking
study, with snowshoe hare (Lepus americanus) as a main prey species, lynx (Lynx
canadensis) tracks are more likely to be found in areas with higher hare abundance indexed
by hare track count in snow tracking study (Bayne et al. 2008, Keim et al. 2011).
However, when several prey species are available, the strength of the spatial association
between a predator and a given prey species may decrease when the abundance of alternative
prey species increases (prey switching; Murdoch et al. 1975). Prey switching is exemplified
with wolves (Canis lupus) that concentrate on areas used by deer (Odocoileus virginianus)
during winter, but switch to beavers (Castor canadensis) during summer, when the latter
become available (Latham et al. 2013).
Habitat selection of a species can be influenced by competition for shared resources by
other species (Morris et al. 2000, Morris 2003). For example, density of collared lemmings
(Dicrostonyx groenlandicus) in their preferred habitat declined with increase in density of
competitor, brown lemmings (Lemmus trimucronatus), in the habitat (Morris et al. 2000).
Similarly, habitat selection by a predator species might be influenced by competition for
shared prey by another predator species. Competition among predators having dietary overlap
85
often results in spatial segregation among species as shown in the marten-fisher relationship
(Fisher et al. 2013).
Habitat selection by mesopredators might be subject to change due to predation risk from
larger predators in addition to interspecific competition. Predation by large species on
mesopredators, sometimes leads to a decline in the abundance of mesopredators (Borer et al.
2003; Polis et al. 1989). For example, in a system including gray fox (Urocyon
cinereoargenteus, 3-5kg), coyote (Canis latrans, 8-20kg) and bobcats (Felis rufus, 5-15kg),
these species shared food resources such as small mammals (lagomorphs and rodents). 58%
of gray fox mortality and 40% of bobcat mortality were caused by coyotes, and gray fox
abundance was negatively associated with coyote abundance (Fedriani et al. 2000).
Because of the potential influence of competitors and alternative prey, habitat selection
by mesopredators should result from a trade-off between energy gain, predation risk,
intraspecific competition and intra-guild competition (Lima 2002, Gorini et al. 2012). When
a larger predator species occupies area with high primary prey abundance, mesopredators
might not occupy the area due to predator avoidance (Fortin et al. 2005, Latombe et al. 2014)
and a concomitant reduction in foraging success (Brown and Kotler 2004, Andruskiw et al.
2008). Instead, they might switch to occupy area with high secondary prey abundance for
energy gain. Despite this possibility, it has not been addressed how mesopredators distribute
themselves when they face with requirement of energy gain, competition and energy gain
simultaneously.
North American boreal forests typically host a predator-prey system including snowshoe
hare, marten (Martes americana), squirrel (Tamiasciurus hudsonicus), lynx and fox (Vulpes
86
vulpes). Their abundance is known to fluctuate strongly (e.g., hare; Krebs et al. 2001a,
marten; Flynn and Schumacher 2009). Lynx, marten and fox prey on hare (Martin 1994,
Poole and Graf 1996, Krebs et al. 2001b, Krebs 2011) and squirrel (Apps 1999). Hare and
squirrel are considered as more important food item comparing to small mammals such as
red-backed vole because hare and squirrel composed 78.9% of total carories comsumed based
on minimum caloric estimate (Cumberland et al. 2001). Lynx and fox prey on marten (Clark
et al. 1987, Thompson 1994, Hearn 2007, Naughton 2012). Under such a system, we expect
marten habitat selection to reflect a balance between energy gain from hares and predation
risk from lynx. In particular, even though marten tends to select areas of higher hare
abundance (Vigeant-Langlois and Desrochers 2011), marten should be less associated to
areas with abundant hare when lynx is present or when lynx is spatially associated with hare
because of predator avoidance and reduction in foraging success. In such cases, marten may
increase its use of alternative prey such as squirrels.
Here, we explored spatio-temporal relationships among five mammal species including
American marten, snowshoe hare, red squirrel, red fox and Canadian lynx. More specifically,
we tested 1) whether spatial association between marten and hare decreased when spatial
association with hare by known predators, fox and lynx, was strong, 2) whether hare–marten
association was reduced under relative abundance of lynx or fox, 3) whether stronger hare-
lynx association or higher abundance of fox and lynx lead to stronger marten-squirrel
association. To account for lags in effect of conspecific density and predation risk on
behavioral response including habitat selection (Kawaguchi and Desrochers unpublished
[chapter 2], Magnuson 1990, Laundré et al. 2001), effects of abundance and spatial
associations between a paired variable in the previous winter were also examined.
87
Methods
Study area
We conducted snow tracking surveys in a 66 km2 boreal forest 80 km north of Quebec
City (47° 20’N, 71° 10’W). Most of the study area was originally clear-cut between 1941 and
1945 and is now managed with a combination of clearcuts and selective cuts. The resulting
forest is composed of mature (more than 40-y-old), mid-succession (21- to 40-y-old) and
regenerating (less than 20-y-old) forest stands comprising ca. 55%, 25% and 20% of the study
area, respectively. Locations of different-aged stands shift with time due to continuing timber
harvest and forest stand succession, with a mean stand age remaining stable throughout the
study period (43.0 ± 1.7 years, range 0 – 114 yr). A dense road network is present, with 2 km
of roads/km2. During winter, several roads were groomed by machinery for cross-country ski
trails. Altitude ranges from 650 m to 1000 m. From 1999 to 2011, annual mean temperature
was 0.3 °C. Annual rainfall was 1417 mm (33 % in snow). Maximum snow depth at weather
station ranged from 62 to 146 cm in 2004-2014. Balsam fir dominates second growth mature
forest stands. Black spruce (Picea mariana), white birch (Betula papyrifera), aspen (Populus
tremuloides) and white spruce (P. glauca) are also common. Recent (less than 5 y) clearcuts
are generally colonized by raspberry (Rubus idaeus), balsam fir, and white birch (de
Bellefeuille et al. 2001).
Snow tracking
We conducted snow tracking each winter (20 December - 31 March) from 2004 to 2014.
We counted tracks along a subset of a network including about 150 km of roads, 40 km of
trails, and 60 km of straight line transects. None of the roads that were surveyed had been
snowplowed. Transect length depended upon snow condition, current weather, time of day,
88
and personnel availability (Table 4.1). Off-trail transects were randomly selected from a
systematic grid covering the entire study area at the beginning of each year. However, in the
selection process, we removed transects that had been surveyed in the previous 2 years. For
each year, we surveyed selected transects only once to cover larger area as much as possible
with a GPS. As a result, we surveyed an average of 92.4 km ± 29.8 km (mean ± SD) of
transects each year (Table 4.1). Our surveys on transects were performed within a 24 to 72
hours after the last snowfall exceeded 3 cm. We recorded all tracks that fell within 2 m of
either side of the transect lines into a GPS receiver and identified the species, based on track
pattern and size. Conspecific tracks that were within 3 m of a recorded track were ignored.
Any of single track, a trail and a network of conspecific tracks was recorded as one track.
We georeferenced track and transect data to ArcGIS (Version 10.1, ESRI, 2012) and split
transects into 400m fixed-length segments, totaling 2461 400-m segments for the entire study
(Table 4.1). We counted tracks on each transect segment, and generated buffers with a radius
of 200 m. Within each buffer, we calculated the mean age of forest stands (weighted by area),
proportion of area of mature forest (40-80 y).
Estimation of population indices
Population indices (henceforth, ‘abundances’) of hare, squirrel, marten and fox were
estimated from the year effect in generalized linear models in which track counts were
function of mean stand age, elevation, slope, elevation, temperature and year effect as
categorical variable (Kawaguchi et al. in press). Abundance of lynx was estimated from the
mean track count divided by exposure hours because of excessive amount of zeros in lynx
track counts from 2009 to 2011, i.e. no lynx tracks found in 2009 and 2011.
89
Exploratory path analysis
To identify the most important directed acyclic graph (hereafter, DAG) representing
spatio-temporal relationships among the selected species, we conducted exploratory path
analysis through d-separation test (Shipley 1997, 2000), which allows non-Gaussian or non-
linear data. The variables used were hare track counts, squirrel track counts, marten track
presence (binary), lynx track presence, fox track presence, mean temperature, transect type
(road or straight-line) and proportion of mature forest (40-80 y) within a buffer. We used the
method by Shipley (1997) to look for evidence of causal links (undirected edges). We
produced all possible DAG by changing directions of edges with the constraint of being
biologically meaningful. We assumed 1) no directed edge from animal tracks to habitat, 3)
no directed edge with a positive regression coefficient from a predator to a prey and 4) no
directed edge with negative path coefficient from prey to predator. We applied d-separation
test for all DAGs and then calculated AIC values. A DAG with a lower AIC was considered
better (Shipley 2012). We repeated these procedures for each year. In the procedure for
determining presence of edges, we used different type I error risks (alpha = 0.05, 0.1, 0.20,
0.3, 0.4 or 0.5). To examine dependencies among paired variables (e.g., hare and lynx), we
used a Generalized Linear Model (GLM) for one variable (e.g, lynx) as a function of the other
(e.g, hare). Since switching response variables can change the p value, we switched response
variable and then ran GLM once again. Then, the lowest p value was used to determine the
independency. To account for effect of exposure time since the last disturbance (wind or
snow) on detetion rate and thus on track count and presence, exposure time was integrated
into GLM whenever track count or presence of mammal species was used as response
variable. Since our aim is not to estimate true occupancy, we did not use N-mixture model
(See Mackenzie et al. 2002) to account for detection rate but we integrated the variables as
90
fixed effect in GLMs. For hare and squirrel track counts, negative binomial distribution and
log link were applied by using the package MASS in the program R. Binomial distribution
and logit link were used for transect type, marten, lynx and fox. Based on the best DAG, path
coefficients were calculated by regressing a child variable which has incoming arrows for
parent variables having outgoing arrows. GLM was used to calculate the path coefficients.
We are aware of potential spatial autocorrelation among transects segments and thus,
Moran’s I was calculated by using model residuals to verify degree of spatial autocorrelation.
Dynamics of spatial association
Path coefficients in the best graphs were used as a proxy for strength of spatial association
between two species for each year. When a direct edge was not present, a path coefficient
was set to zero. By using multiple linear models, we separately examined the response of the
marten-hare path coefficients and the response of squirrel-marten path coefficients as a
function of different combinations of covariates: the marten-hare coefficient and squirrel-
marten in the previous year marten, lynx and fox population indices in the current winter as
well as ones in the previous winter, and lynx-hare, fox-hare path coefficients. We used
adjusted r squares to compare model performance.. The model fit was performed by using
the package stats in the program R. All statistical analyses were conducted in the R statistical
environment (R Development Core Team 2015).
Results
We found and georeferenced 13186, 1307, 4916, 954 and 311 tracks in total for hare, marten,
squirrel, fox and lynx respectively (Table 4.1). Population index of lynx experienced a large
decline from 2008 to 2009. Lynx abundance was negatively correlated with fox abundance
(n = 11, r = -0.63, P = 0.04), but we found no other correlation in the dynamics of paired
91
species (n = 11, P > 0.05). Average Moran’s I for marten, hare, squirrel, fox and lynx over
study years were respectively 0.03 [range: -0.005 – 0.05], 0.1 [range: 0.05 – 0.25], 0.05
[range: -0.003 – 0.09], 0.03 [range: -0.002 – 0.06] and 0.03 [range: -0.01 – 0.08]. These values
suggest weak spatial autocorrelation in track counts after accounting for covariates.
Eleven DAGs were obtained through exploratory path analyses (Appendix 1 to see the
best DAG each year, Appendix 2 for detailed result of d-separation test on the best causal
graphs). The best DAGs for each year differed in topology. Three hare-to-marten edges out
of five were significant (P < 0.05, Figure 4.2, Appendix 3 for presenting all the path
coefficients estimated). Mean path coefficients from hare to marten was 0.070, ranging from
0.031 to 0.113. One hare-to-lynx edge out of two was significant. Mean path coefficient from
hare to lynx was 0.125 [range: 0.034 – 0.215]. Two edges from hare to fox remained in the
best graphs. The mean path coefficient was 0.042 [range: 0.033 – 0.053]. There was no edge
between hare and marten when hare and lynx had direct positive path coefficient. Path
coefficients from lynx to marten were negative in three years [range: –2.01 to -0.802].
Positive edges from squirrel to marten appeared in 4 years and mean of the path coefficients
was 0.225 [range: 0.078 – 0.400] (Figure 4.2).
The model for marten-hare dynamics, including lynx abundance in the previous year and
hare-marten association in the previous winter, performed best among the candidate models
(Table 4.2). Estimated effects of both lynx abundance in the previous winter and hare-marten
association in the previous winter were negative (Figure 4.3) and the effect of lynx abundance
in the previous year was significant (Table 4.2).
92
The model for marten-squirrel dynamics including hare-marten path coefficient and lynx
abundance in the previous year, performed best among the candidate models (Table 4.3). The
models including hare abundance or squirrel abundance poorly performed. Estimated effect
of hare-marten path coefficient was negative and that of lynx abundance in the previous year
was also negative (Figure 4.4). None of them appeared to be significant while hare-marten
coefficients in the best model was nearly significant.
Discussion
The spatial association between one predator, marten, and one of its main prey, hare,
depended on the abundance of larger predators in the previous year. The spatial association
between marten and an alternative prey, red squirrel, was best explained by recent abundance
of larger predators and the hare-marten spatial association rather than abundance of larger
predators and predator spatial association with the main prey.
We predicted that when lynx or fox were more strongly associated with hare, marten
would be less associated with hare and also predicted that hare-marten spatial association
should be reduced with greater lynx or fox abundance. Our first prediction was partially true
for lynx, not for fox. Our results showed that marten was not spatially associated with hare
when lynx was associated with hare. However, the second prediction received most support
by the result in which the model with lynx abundance in the previous year performed best for
explaining hare-marten spatial association. This argument is also supported by the fact that
80% of hare-marten links were concentrated in years after no detection of lynx tracks. These
results support the hypothesis that competition for shared prey between lynx and marten
influenced the hare-marten relationship, thus habitat selection of mesopredators.
93
Why did lynx abundance in the previous winter exert a stronger influence on the hare-
marten relationship than current lynx abundance? Lagged responses in predator-prey
dynamics have been documented elsewhere. For example, Laundré et al. (2001) observed 1
year delays in behavioral change in elk after reintroduction of wolves in Yellowstone. We
can speculate that the higher importance of lagged effect of lynx population index was due
to delays in assessing predation risk from lynx by marten, but testing this hypothesis would
require a dedicated study.
Thr model performance for dynamics of hare-marten association was improved by
including hare-marten association in the previous winter. And its estimated effects showed
that higher spatial associations between hares and marten in the current winter tended to lead
to a lower association in the following winter. Possible explanation could be feedback effect
in which hare took avoidance behavior against marten in response to recent presence of
marten. Avoidance behavior against predator by prey in response to direct and indirect cue
was commonly observed in various taxa (e.g., moose-wolf; Latombe et al. 2014). When hare
was spatially associated with marten, hare would more frequently encounter with marten or
would find indirect cue such as tracks. The detection of predator cues by hare would lead
hares to spend less time in area with marten, potentially resulting in a weaker spatial
association.
The best model for squirrel-marten spatial association indicated that the association was
reduced with stronger hare-marten association and with higher lynx relative abundance in the
previous year. While switching to alternative prey due to decline in primary prey abundance
has been widely documented(e.g., Thompson and Colgan 1990, Randa et al. 2009), we did
not find significant switching to alternative prey due to decline of primary prey abundance
94
and the models with prey relative abundance performed poorly. Rather, stronger support was
given to the hypothesis that the switching to alternative prey was due to combined effect of
decline in spatial association of predator with primary prey and also relative abundance of
larger predator. However, these factors nearly became significant potentially because of
presence of other preys such as ruffed grouse (Bonasa umbellus) (Cumberland et al. 2001) in
the study area, evidenced by detection of grouse track during the field work. Since grouses
were also available for marten, marten might allocate certain proportion of their hunting effort
(e.g., attempt to be spatially associated) to location of grouse, resulting in ambiguours
switching behavior from hare to squirrel.
In summary, our hypothesis that habitat selection by a mesocarnivore, the American
marten, was greatly influenced by larger predator abundance in the previous year was not
rejected. And our study provided weak evidence for prey switching to alternative prey.
Previous studes on prey switching have demonstrated that prey-predator spatial relationship
were dynamics rather than stable. To our knowledge, our study is the first to demonstrate a
case that prey-predator relationship can be dynamics due to larger predator. Further work on
predicting the spatial distribution of carnivores via the distribution of their prey, an area of
current interest (Trainor et al. 2013, Trainor and Schmitz 2014), should not ignore the
possible role of competitors sharing similar prey.
Acknowledgements
Financial support for this project was provided by a scholarship to T. Kawaguchi from
the “Leadership and Sustainable development Scholarship Program” of Laval University, by
a scholarship to T. Kawaguchi from the Foundation F.-K.-Morrow and by a Natural Sciences
and Engineering Research Council of Canada (NSERC) grant to A. Desrochers. We are
95
grateful to the 29 skilled field workers who contributed to the collection of snow tracking
data. We thank D. Fortin, L. Bélanger and C. Samson for their assistance in the design of the
study and thank I.D. Thompson and M. Mazerolle for their advice on the manuscript.
96
Figure 4. 1. Population dynamics of five species over 11 years, 2004-2014: a) Snowshoe hare, b) red squirrel, c) American marten, d) Lynx and e) red fox. Gray error bars indicate standard errors.
97
Figure 4. 2. Summary of path analysis results linking spatial distributions of predator, prey and vegetation attributes, 2004-2014. Thickness of line is proportional to the number of years with evidence for an edge. Red colored edges indicate positive coefficients and blue colored edges indicate negative coefficiens. Grey colored edges indicate that path coefficient were either positive or negative depending on study year.
98
Figure 4. 3. Dynamics of edges in spatial distributions between hare and marten, 2004 -2014 (n = 10). The graphs represent a) relationship between hare-marten spatial association in the current winter and the one in the previous winter and b) hare-marten spatial association in the current winter and lynx population index in the previous winter. Points with standard error bars indicate path coefficients. Dashed lines indicate 95 % confidence bands of the fitted regression line values.
99
Figure 4. 4. Dynamics of edges in spatial distributions between squirrel and marten, 2004 -2014 (n = 10): a) Relationship between squirrel-marten spatial association and hare population index, b) Relationship between squirrel-marten spatial association and hare-marten association. Points with standard error bars indicate path coefficients. Dashed lines indicate 95 % confidence bands of the fitted regression line values.
100
Table 4. 1. Sampling effort for snow tracking and track counts for snowshoe hare (Lepus americanus), American marten (Martes americana), red squirrel (Tamiasciurus hudsonicus), and Canadian lynx (Lynx canadensis) and red fox (Vulpes vulpes) at the Montmorency Forest, southern Quebec, Canada, 2004-2014.
YEAR On
road/trails (km)
Off-trail (km)
400m segments
(n)
Hare (count)
Lynx (count)
Marten (count)
Squirrel (count)
Fox (count)
2004 32 15 121 547 38 65 137 10 2005 58 5 159 371 17 59 178 18 2006 50 23 183 424 52 85 63 50 2007 68 15 209 880 55 136 2009 50 2008 94 21 289 1502 89 314 427 64 2009 55 11 166 310 0 93 78 114 2010 88 12 252 787 1 98 229 121 2011 106 11 295 1788 0 149 168 155 2012 89 35 312 2578 31 183 1398 188 2013 107 19 320 3001 14 100 148 113 2014 50 12 155 998 14 25 81 71 Total 797 179 2461 13186 311 1307 4916 954
Average 72 16 224 1199 28 119 447 87 Standard deviation 25 8 72 917 28 78 644 56
101
Table 4. 2. Model fit for dynamics of edges in spatial distribution between Hare and Marten in the Montmorency Forest, Québec, 2004-2014. Candidate models include variables: Hare-marten path coefficients (Hare-Marten), hare-marten path coefficients in the previous year (Hare-Martent-1), marten population index (Marten population), hare-fox path coefficients (Hare-Fox), and hare-lynx path coefficient (Hare-Lynx).
Model n Adjusted R2 Variable Estimate SE P
Hare-Lynx 11 -0.04 Hare-Lynx -0.14 0.17 0.44 Hare-Fox 11 -0.1 Hare-Fox -0.13 0.38 0.74
Fox population 11 -0.05 Fox population 0.02 0.02 0.5 Lynx population 11 0.03 Lynx population -1.61 1.43 0.29 Hare-martent-1 10 -0.09 Hare-martent-1 -0.22 -0.36 0.55
Lynx population t-1 10 0.34 Lynx population t-1 -2.94 1.23 0.041 Fox populationt-1 10 0.34 Fox populationt-1 0.04 0.0172 0.044 Hare-Martent-1 +
Lynx population t-1 10 0.47 Hare-Martent-1 -0.49 0.29 0.13
Lynx population t-1 -3.46 1.18 0.02
102
Table 4. 3. Model fit for dynamics of edges in spatial distribution between marten and squirrel in the Montmorency Forest, Québec, 2004-2014. Candidate models include variables: Hare-lynx path coefficients (Hare-Lynx), hare population index (Hare population), hare-fox path coefficients (Hare-Fox).
Model n Adjusted R2 Variable Estimate SE P Hare-Marten 11 0.02 Hare-Marten -1.1 1.01 0.3 Hare-Lynx 11 -0.02 Hare-Lynx 0.14 0.16 0.39
Hare population 11 -0.02 Hare population -0.07 0.08 0.39 Squirrel population 11 0.11 Squirrel population -0.06 0.04 0.17
Lynx population 11 -0.05 Lynx population -3.17 4.57 0.51 Lynx population t-1 10 -0.06 Lynx population t-1 -3.43 4.75 0.49
Fox population 11 0.05 Fox population 0.08 0.06 0.25 Hare-Lynx +
Lynx population t-1 10 -0.06
Lynx population t-1 -4.05 6.32 0.43 Hare-Lynx 0.165 0.26 0.35
Hare population + Hare-Lynx 11 -0.12
Hare population -0.05 0.1 0.67 Hare-Lynx 0.09 0.2 0.66
Hare population + Hare-Marten 11 -0.06
Hare population -0.05 0.09 0.59 Hare-Marten -0.9 1.11 0.44
Hare-Marten + Squirrel abundance 11 0.2
Hare-Marten -1.31 0.91 0.19 Squirrel population -0.06 0.04 0.12
Hare-Marten + Lynx population t-1
11 0.32 Hare-Marten -2.69 1.15 0.052
Lynx population t-1 -11.06 5.01 0.06
103
Appendix
Appendix 1. The best causal graphs for each year during study period, 2004-2014.
104
105
106
Figure 4. 5. The best causal graphs for each year during study period, 2004-2014. Solid line indicates significant path (p < 0.05). Dashed line indicates non-significant path (p < 0.05). Red line indicates positive effect from a variable to the other and blue lines indicate negative.
107
Appendix 2. Result of d-separation test representing all d-separation claims and its
probability of independence.
Table 4. 4. Result of d-separation test representing all d-separation claims and its probability of independence.
Year Dsep Claim P Distribution Link
2004
Fox_||_Lynx|{Transect type,40-80y,Temperature,Squirrel,Marten,Exposure time} 0.72 Binomial Logit
Fox_||_Squirrel|{Transect type,Hare,40-80y,Temperature,Marten,Exposure time} 0.67 Binomial Logit
Fox_||_Hare|{Transect type,40-80y,Marten,Exposure time} 0.45 Binomial Logit
Hare_||_Lynx|{Transect type,40-80y,Temperature,Squirrel,Exposure time} 0.04 Negative
binomial Log
Hare_||_40-80y|{Transect type,Exposure time} 0.15 Negative binomial Log
Marten_||_Lynx|{Transect type,40-80y,Temperature,Squirrel,Hare,Exposure time} 0.87 Binomial Logit
Marten_||_40-80y|{Transect type,Hare,Temperature,Exposure time} 0.93 Binomial Logit
Marten_||_Squirrel|{Transect type,Hare,40-80y,Temperature,Exposure time} 0.66 Binomial Logit
Temperature_||_Hare|{Transect type,40-80y} 0.59 Gaussian Identity Temperature_||_Fox|{Transect type,40-80y,Marten} 0.95 Gaussian Identity Transect type_||_40-80y|{NA} 0.77 Binomial Logit Transect type_||_Fox|{40-80y,Marten} 0.66 Binomial Logit
2005
Fox_||_Transect type|{Temperature,40-80y,Marten,Exposure time} 0.37 Binomial Logit
Hare_||_Lynx|{Squirrel,Fox,Exposure time} 0.39 Negative binomial Log
Hare_||_Fox|{Squirrel,40-80y,Marten,Exposure time} 0.13 Negative binomial Log
Hare_||_Temperature|{Squirrel,Exposure time} 0.16 Negative binomial Log
40-80y_||_Hare|{Temperature,Transect type,Squirrel} 0.42 Gaussian Identity 40-80y_||_Marten|{Temperature,Transect type} 0.06 Gaussian Identity 40-80y_||_Lynx|{Temperature,Transect type,Squirrel,Fox} 0.1 Gaussian Identity
Lynx_||_Marten|{Temperature,Squirrel,Fox,Exposure time} 0.51 Binomial Logit
Lynx_||_Transect type|{Temperature,Squirrel,Fox,Exposure time} 0.26 Binomial Logit
Lynx_||_Temperature|{Squirrel,Fox,Exposure time} 0.06 Binomial Logit Marten_||_Transect type|{Temperature,Exposure time} 0.49 Binomial Logit Marten_||_Squirrel|{Temperature,40-80y,Exposure time} 0.1 Binomial Logit
108
Marten_||_Hare|{Squirrel,Temperature,Exposure time} 0.03 Binomial Logit Squirrel_||_Fox|{Temperature,40-80y,Marten,Exposure time} 0.53 Negative
binomial Log
Squirrel_||_Transect type|{Temperature,40-80y,Exposure time} 0.11 Negative
binomial Log
Temperature_||_Fox|{40-80y,Marten} 0.55 Gaussian Identity Transect type_||_Hare|{Temperature,Squirrel} 0.12 Binomial Logit
2006
Fox_||_Squirrel|{40-80y,Lynx,Exposure time} 0.79 Binomial Logit Fox_||_Temperature|{40-80y,Lynx,Exposure time} 0.35 Binomial Logit Fox_||_Hare|{Transect type,Squirrel,Lynx,Exposure time} 0.64 Binomial Logit
Hare_||_40-80y|{Transect type,Squirrel,Exposure time} 0.86 Negative binomial Log
Hare_||_Marten|{Transect type,Squirrel,40-80y,Lynx,Fox,Exposure time} 0.11 Negative
binomial Log
40-80y_||_Fox|{Lynx} 0.61 Gaussian Identity Lynx_||_Squirrel|{40-80y,Transect type,Hare,Exposure time} 0.9 Binomial Logit
Lynx_||_40-80y|{Transect type,Hare,Exposure time} 0.73 Binomial Logit Squirrel_||_Transect type|{40-80y,Temperature,Exposure time} 0.47 Negative
binomial Log
Squirrel_||_Temperature|{40-80y,Exposure time} 0.78 Negative binomial Log
Temperature_||_Marten|{40-80y,Transect type,Squirrel,Lynx,Fox} 0.8 Gaussian Identity
Temperature_||_Lynx|{40-80y,Transect type,Hare} 0.28 Gaussian Identity Temperature_||_Hare|{40-80y,Transect type,Squirrel} 0 Gaussian Identity Transect type_||_Fox|{40-80y,Temperature,Lynx} 0.95 Binomial Logit
2007
Fox_||_Lynx|{Temperature,Hare,Exposure time} 0.81 Binomial Logit Fox_||_Marten|{Temperature,Transect type,Lynx,Exposure time} 0.75 Binomial Logit
Fox_||_Squirrel|{Temperature,40-80y,Exposure time} 0.35 Binomial Logit Hare_||_40-80y|{Transect type,Squirrel,Fox,Exposure time} 0.31 Negative
binomial Log
Hare_||_Temperature|{Transect type,Squirrel,Fox,Exposure time} 0.39 Negative
binomial Log
40-80y_||_Lynx|{Temperature,Hare} 0.12 Gaussian Identity 40-80y_||_Fox|{NA} 0.18 Gaussian Identity Marten_||_Hare|{Transect type,Squirrel,Fox,Temperature,Lynx,Exposure time} 0.52 Binomial Logit
Marten_||_Squirrel|{Temperature,40-80y,Transect type,Lynx,Exposure time} 0.79 Binomial Logit
Marten_||_40-80y|{Temperature,Transect type,Lynx,Exposure time} 0.13 Binomial Logit
Squirrel_||_Lynx|{Temperature,40-80y,Hare,Exposure time} 0.46 Negative
binomial Log
Temperature_||_40-80y|{NA} 0.57 Gaussian Identity Temperature_||_Fox|{NA} 0.89 Gaussian Identity Transect type_||_Fox|{Temperature} 0.99 Binomial Logit Transect type_||_40-80y|{Temperature} 0.78 Binomial Logit
109
Transect type_||_Squirrel|{Temperature,40-80y} 0.51 Binomial Logit Transect type_||_Lynx|{Temperature,Hare} 0.14 Binomial Logit
2008
Fox_||_Transect type|{40-80y,Temperature,Marten,Hare,Exposure time} 0.27 Binomial Logit
Fox_||_Squirrel|{40-80y,Temperature,Marten,Hare,Exposure time} 0.82 Binomial Logit
Fox_||_Lynx|{Transect type,Temperature,40-80y,Marten,Hare,Exposure time} 0.35 Binomial Logit
Hare_||_Marten|{40-80y,Transect type,Temperature,Exposure time} 0.38 Negative
binomial Log
Hare_||_40-80y|{Transect type,Exposure time} 0.33 Negative binomial Log
Hare_||_Lynx|{Transect type,Temperature,Exposure time} 0.6 Negative
binomial Log
Lynx_||_40-80y|{Transect type,Temperature,Exposure time} 0.9 Binomial Logit
Marten_||_Lynx|{Transect type,Temperature,40-80y,Exposure time} 0.9 Binomial Logit
Marten_||_Squirrel|{40-80y,Transect type,Temperature,Hare,Exposure time} 0.1 Binomial Logit
Squirrel_||_Lynx|{Transect type,Temperature,40-80y,Hare,Exposure time} 0.05 Negative
binomial Log
Temperature_||_Hare|{Transect type} 0.66 Gaussian Identity Temperature_||_40-80y|{Transect type} 0.21 Gaussian Identity Transect type_||_Squirrel|{40-80y,Temperature,Hare} 0.68 Binomial Logit
2009
Fox_||_Marten|{Temperature,Transect type,Squirrel,Exposure time} 0.76 Binomial Logit
Hare_||_Fox|{Temperature,Transect type,Squirrel,Exposure time} 0.47 Negative
binomial Log
Hare_||_40-80y|{Temperature,Squirrel,Exposure time} 0.67 Negative binomial Log
40-80y_||_Marten|{Temperature,Transect type,Squirrel} 0.77 Gaussian Identity 40-80y_||_Fox|{Temperature,Transect type,Squirrel} 0.62 Gaussian Identity Marten_||_Hare|{Temperature,Transect type,Squirrel,Exposure time} 0.42 Binomial Logit
Squirrel_||_Temperature|{40-80y,Transect type,Exposure time} 0.75 Negative
binomial Log
Transect type_||_40-80y|{Temperature} 0.4 Binomial Logit Transect type_||_Hare|{Temperature,Squirrel} 0.35 Binomial Logit Transect type_||_Temperature|{NA} 0.77 Binomial Logit
2010
Fox_||_Marten|{Temperature,Transect type,Squirrel,Hare,Exposure time} 0.33 Binomial Logit
Fox_||_Hare|{Temperature,Transect type,40-80y,Squirrel,Exposure time} 0.28 Binomial Logit
Fox_||_Squirrel|{Temperature,Transect type,40-80y,Exposure time} 0.34 Binomial Logit
Fox_||_40-80y|{Temperature,Transect type,Exposure time} 0.97 Binomial Logit
40-80y_||_Marten|{Temperature,Transect type,Squirrel,Hare} 0.74 Gaussian Identity
110
Temperature_||_Marten|{Transect type,Squirrel,Hare} 0.93 Gaussian Identity Temperature_||_Hare|{Transect type,40-80y,Squirrel} 0.3 Gaussian Identity Transect type_||_40-80y|{Temperature} 0.19 Binomial Logit Transect type_||_Squirrel|{Temperature,40-80y} 0.75 Binomial Logit
2011
40-80y_||_Hare|{Transect type,Temperature,Squirrel,Fox} 0.08 Gaussian Identity
Marten_||_40-80y|{Transect type,Temperature,Squirrel,Fox,Exposure time} 0.79 Binomial Logit
Marten_||_Hare|{Temperature,Squirrel,Fox,Transect type,Exposure time} 0.25 Binomial Logit
Squirrel_||_Transect type|{40-80y,Exposure time} 0.7 Negative binomial Log
Temperature_||_Squirrel|{Transect type,40-80y} 0.01 Gaussian Identity Temperature_||_Fox|{Transect type,40-80y,Squirrel} 0.03 Gaussian Identity Transect type_||_Marten|{Temperature,Squirrel,Fox} 0.66 Binomial Logit
2012
Fox_||_Hare|{Transect type,40-80y,Temperature,Squirrel,Exposure time} 0.6 Binomial Logit
Fox_||_40-80y|{Transect type,Temperature,Squirrel,Exposure time} 0.38 Binomial Logit
40-80y_||_Lynx|{Temperature,Fox} 0.74 Gaussian Identity 40-80y_||_Transect type|{NA} 0.12 Gaussian Identity Lynx_||_Hare|{Transect type,40-80y,Temperature,Fox,Exposure time} 0.61 Binomial Logit
Marten_||_Squirrel|{40-80y,Temperature,Hare,Transect type,Fox,Lynx,Exposure time} 0.47 Binomial Logit
Squirrel_||_Lynx|{40-80y,Temperature,Hare,Fox,Exposure time} 0.12 Negative
binomial Log
Transect type_||_Squirrel|{40-80y,Temperature,Hare} 0.03 Binomial Logit Transect type_||_Lynx|{Temperature,Fox} 0.12 Binomial Logit
2013
Fox_||_Lynx|{40-80y,Squirrel,Transect type,Hare,Exposure time} 0.98 Binomial Logit
40-80y_||_Hare|{Temperature,Transect type,Squirrel,Lynx} 0.01 Gaussian Identity
Lynx_||_Transect type|{40-80y,Squirrel,Exposure time} 0.34 Binomial Logit Lynx_||_Marten|{40-80y,Squirrel,Temperature,Transect type,Hare,Exposure time} 0.32 Binomial Logit
Marten_||_Fox|{40-80y,Transect type,Hare,Temperature,Exposure time} 0.35 Binomial Logit
Marten_||_Squirrel|{40-80y,Transect type,Temperature,Hare,Exposure time} 0.14 Binomial Logit
Squirrel_||_Fox|{40-80y,Transect type,Hare,Exposure time} 0.51 Negative
binomial Log
Squirrel_||_Temperature|{40-80y,Transect type,Exposure time} 0.15 Negative
binomial Log
Temperature_||_Fox|{40-80y,Transect type,Hare} 0.41 Gaussian Identity Temperature_||_Lynx|{40-80y,Squirrel} 0.56 Gaussian Identity Transect type_||_Temperature|{NA} 0.85 Binomial Logit Transect type_||_40-80y|{Temperature} 0.33 Binomial Logit
2014 Fox_||_Marten|{Temperature,Transect type,Hare,40-80y,Lynx,Exposure time} 0.56 Binomial Logit
111
Fox_||_Hare|{Temperature,40-80y,Squirrel,Lynx,Exposure time} 0.58 Binomial Logit
Hare_||_Transect type|{Temperature,40-80y,Squirrel,Exposure time} 0.6 Negative
binomial Log
40-80y_||_Squirrel|{Temperature,Transect type} 0.68 Gaussian Identity 40-80y_||_Marten|{Temperature,Transect type,Hare} 0.63 Gaussian Identity Lynx_||_Hare|{Temperature,40-80y,Squirrel,Marten,Exposure time} 0.9 Binomial Logit
Lynx_||_40-80y|{Temperature,Marten,Exposure time} 0.87 Binomial Logit Marten_||_Squirrel|{Transect type,Temperature,Hare,Exposure time} 0.43 Binomial Logit
Squirrel_||_Fox|{Transect type,40-80y,Lynx,Exposure time} 0.44 Negative
binomial Log
Squirrel_||_Lynx|{Transect type,Marten,Exposure time} 0.36 Negative binomial Log
Squirrel_||_Temperature|{Transect type,Exposure time} 0.63 Negative binomial Log
Temperature_||_Fox|{40-80y,Lynx} 0.86 Gaussian Identity Temperature_||_Lynx|{Marten} 0.87 Gaussian Identity Transect type_||_Lynx|{Temperature,40-80y,Marten} 0.48 Binomial Logit Transect type_||_Fox|{Temperature,40-80y,Lynx} 0.32 Binomial Logit
112
Appendix 3. Estimated path coefficients of edges in the best graphs for each year.
Table 4. 5. Estimated path coefficients of edges in the best graphs for each year.
Year P (graph) Parents Child Estimate Standard error z P
2004 0.839
Transect type Hare 0.928 0.27 3.433 0.001 Exposure time Hare -0.002 0.009 -0.212 0.832
Hare Marten 0.049 0.034 1.453 0.146 Transect type Marten 1.067 0.482 2.216 0.027 Temperature Marten -0.044 0.037 -1.187 0.235
Exposure time Marten -0.008 0.016 -0.48 0.631 Hare Squirrel 0.072 0.018 3.933 <0.001
40-80y Squirrel 0.009 0.007 1.307 0.191 Transect type Squirrel -1.063 0.377 -2.823 0.005 Temperature Squirrel 0.107 0.032 3.36 0.001
Exposure time Squirrel 0.04 0.01 3.854 <0.001 Squirrel Lynx 0.323 0.134 2.412 0.016 40-80y Lynx -0.018 0.013 -1.34 0.18
Transect type Lynx -1.523 0.812 -1.876 0.061 Temperature Lynx 0.131 0.069 1.905 0.057
Exposure time Lynx 0.011 0.02 0.524 0.6 Marten Fox 1.513 0.913 1.657 0.098 40-80y Fox -0.036 0.023 -1.573 0.116
Exposure time Fox 0.052 0.034 1.567 0.117 40-80y Temperature -0.071 0.025 -2.881 0.005
Transect type Temperature -3.408 1.101 -3.096 0.002
2005 0.007
Squirrel Hare 0.244 0.046 5.325 <0.001 Exposure time Hare -0.018 0.01 -1.795 0.073 Temperature Marten 0.518 0.102 5.077 <0.001
Exposure time Marten 0.025 0.021 1.182 0.237 40-80y Squirrel 0.031 0.008 4.021 <0.001
Temperature Squirrel 0.216 0.031 6.93 <0.001 Exposure time Squirrel 0.019 0.014 1.415 0.157
Squirrel Lynx 0.225 0.081 2.788 0.005 Fox Lynx 1.458 0.769 1.895 0.058
Exposure time Lynx -0.021 0.027 -0.77 0.441 Marten Fox 1.041 0.63 1.653 0.098 40-80y Fox -0.021 0.014 -1.483 0.138
Exposure time Fox 0.03 0.024 1.259 0.208 Transect type 40-80y 15.092 6.384 2.364 0.019 Temperature 40-80y -0.511 0.258 -1.98 0.049
Temperature Transect type 0.23 0.076 3.026 0.002
2006 0.515
Squirrel Hare 0.254 0.165 1.539 0.124 Transect type Hare 0.769 0.298 2.586 0.01 Exposure time Hare -0.013 0.009 -1.427 0.154
Squirrel Marten 0.268 0.226 1.187 0.235
113
Lynx Marten -1.594 0.82 -1.943 0.052 Fox Marten -1.098 0.675 -1.628 0.103
40-80y Marten 0.026 0.011 2.341 0.019 Transect type Marten 1.376 0.43 3.202 0.001 Exposure time Marten -0.047 0.016 -2.959 0.003
40-80y Squirrel 0.032 0.01 3.187 0.001 Exposure time Squirrel -0.011 0.013 -0.905 0.365
Hare Lynx 0.215 0.055 3.942 <0.001 Transect type Lynx -1.835 0.735 -2.497 0.013 Exposure time Lynx -0.012 0.014 -0.807 0.42
Lynx Fox -1.098 0.768 -1.429 0.153 Exposure time Fox -0.009 0.013 -0.673 0.501
40-80y Transect type -0.02 0.009 -2.313 0.021
Temperature Transect type 0.127 0.047 2.721 0.007
40-80y Temperature 0.049 0.015 3.327 0.001
2007 0.688
Squirrel Hare 0.02 0.009 2.128 0.033 Fox Hare -0.4 0.289 -1.385 0.166
Transect type Hare 1.218 0.268 4.551 <0.001 Exposure time Hare -0.023 0.008 -2.791 0.005
Lynx Marten -0.802 0.552 -1.453 0.146 Transect type Marten 1.986 0.437 4.543 <0.001 Temperature Marten 0.089 0.03 2.975 0.003
Exposure time Marten 0.025 0.015 1.694 0.09 40-80y Squirrel 0.02 0.004 5.56 <0.001
Temperature Squirrel 0.053 0.012 4.325 <0.001 Exposure time Squirrel 0.004 0.007 0.532 0.595
Hare Lynx 0.034 0.025 1.392 0.164 Temperature Lynx -0.049 0.032 -1.538 0.124
Exposure time Lynx -0.148 0.052 -2.847 0.004
Temperature Transect type -0.128 0.036 -3.582 <0.001
2008 0.471
Transect type Hare 1.352 0.199 6.785 <0.001 Exposure time Hare 0.027 0.009 3.03 0.002
40-80y Marten 0.017 0.006 2.788 0.005 Transect type Marten 0.783 0.326 2.403 0.016 Temperature Marten 0.081 0.027 3.044 0.002
Exposure time Marten -0.003 0.014 -0.212 0.832 Hare Squirrel 0.098 0.014 6.858 <0.001
40-80y Squirrel 0.024 0.006 4.316 <0.001 Temperature Squirrel 0.049 0.025 1.959 0.05
Exposure time Squirrel 0.013 0.013 0.987 0.324 Transect type Lynx -0.784 0.507 -1.548 0.122 Temperature Lynx -0.045 0.033 -1.359 0.174
Exposure time Lynx -0.025 0.021 -1.188 0.235 Hare Fox 0.033 0.023 1.414 0.157
Marten Fox 1.065 0.416 2.561 0.01
114
40-80y Fox 0.02 0.01 2.063 0.039 Temperature Fox -0.143 0.043 -3.298 0.001
Exposure time Fox -0.002 0.02 -0.106 0.916
40-80y Transect type -0.008 0.007 -1.085 0.278
Transect type Temperature -1.658 0.728 -2.277 0.024
2009 0.944
Squirrel Hare 0.254 0.121 2.105 0.035 Temperature Hare -0.032 0.026 -1.242 0.214
Exposure time Hare -0.022 0.017 -1.292 0.196 Squirrel Marten 0.158 0.131 1.206 0.228
Transect type Marten 0.588 0.485 1.211 0.226 Temperature Marten 0.04 0.034 1.198 0.231
Exposure time Marten -0.05 0.023 -2.211 0.027 40-80y Squirrel 0.016 0.009 1.804 0.071
Transect type Squirrel 0.61 0.561 1.087 0.277 Exposure time Squirrel -0.026 0.024 -1.118 0.264
Squirrel Fox 0.294 0.155 1.896 0.058 Transect type Fox -1.078 0.708 -1.521 0.128 Temperature Fox 0.134 0.045 2.94 0.003
Exposure time Fox 0.001 0.023 0.041 0.967 Temperature 40-80y -0.414 0.283 -1.466 0.145
2010 0.73
Squirrel Hare 0.199 0.043 4.583 <0.001 40-80y Hare -0.007 0.005 -1.505 0.132
Transect type Hare 1.464 0.305 4.795 <0.001 Exposure time Hare -0.01 0.008 -1.359 0.174
Hare Marten 0.114 0.034 3.378 0.001 Squirrel Marten 0.078 0.065 1.203 0.229
Transect type Marten 1.307 0.478 2.735 0.006 Exposure time Marten -0.015 0.012 -1.235 0.217
40-80y Squirrel 0.023 0.007 3.078 0.002 Temperature Squirrel 0.054 0.062 0.869 0.385
Exposure time Squirrel 0.023 0.011 2.121 0.034 Transect type Fox -1.014 0.563 -1.801 0.072 Temperature Fox 0.052 0.068 0.758 0.448
Exposure time Fox -0.005 0.011 -0.43 0.667 Temperature 40-80y 1.259 0.537 2.345 0.02
Temperature Transect type -0.133 0.063 -2.117 0.034
2011 0.028
Squirrel Hare 0.075 0.046 1.642 0.101 Fox Hare -0.207 0.143 -1.446 0.148
Transect type Hare 0.597 0.212 2.823 0.005 Temperature Hare -0.031 0.012 -2.603 0.009
Exposure time Hare -0.041 0.006 -7.026 <0.001 Squirrel Marten 0.396 0.116 3.419 0.001
Fox Marten -0.982 0.343 -2.866 0.004 Temperature Marten 0.058 0.026 2.203 0.028
Exposure time Marten -0.028 0.013 -2.179 0.029 40-80y Squirrel 0.022 0.007 3.367 0.001
115
Exposure time Squirrel -0.035 0.012 -2.974 0.003 Squirrel Fox 0.138 0.095 1.447 0.148 40-80y Fox 0.013 0.007 1.954 0.051
Transect type Fox -1.35 0.57 -2.366 0.018 Exposure time Fox -0.046 0.013 -3.537 <0.001 Transect type 40-80y 8.211 3.995 2.055 0.041 Temperature 40-80y -0.445 0.219 -2.03 0.043 Transect type Temperature 1.689 1.061 1.593 0.112
2012 0.106
40-80y Hare -0.004 0.002 -1.737 0.082 Transect type Hare 0.711 0.119 5.958 <0.001 Temperature Hare -0.037 0.014 -2.588 0.01
Exposure time Hare 0.005 0.004 1.273 0.203 Hare Marten 0.061 0.019 3.262 0.001 Lynx Marten -2.005 1.163 -1.725 0.085 Fox Marten -0.42 0.326 -1.29 0.197
40-80y Marten -0.006 0.006 -0.882 0.378 Transect type Marten 0.74 0.308 2.4 0.016 Temperature Marten -0.056 0.038 -1.457 0.145
Exposure time Marten 0.015 0.01 1.444 0.149 Hare Squirrel 0.096 0.008 11.538 <0.001
40-80y Squirrel 0.025 0.004 6.852 <0.001 Temperature Squirrel 0.1 0.022 4.558 <0.001
Exposure time Squirrel 0.001 0.006 0.194 0.846 Fox Lynx 1.01 0.725 1.393 0.164
Temperature Lynx -0.37 0.1 -3.716 <0.001 Exposure time Lynx 0.061 0.028 2.196 0.028
Squirrel Fox 0.087 0.02 4.302 <0.001 Transect type Fox -1.001 0.361 -2.775 0.006 Temperature Fox -0.092 0.036 -2.559 0.01
Exposure time Fox 0.044 0.01 4.263 <0.001 40-80y Temperature -0.02 0.011 -1.881 0.061
Transect type Temperature -2.383 0.51 -4.671 <0.001
2013 0.184
Squirrel Hare 0.168 0.051 3.289 0.001 Lynx Hare -0.877 0.33 -2.658 0.008
Transect type Hare 0.679 0.158 4.298 <0.001 Temperature Hare -0.025 0.006 -3.99 <0.001
Exposure time Hare 0.009 0.006 1.664 0.096 Hare Marten 0.032 0.017 1.869 0.062
40-80y Marten 0.012 0.008 1.473 0.141 Transect type Marten 1.219 0.385 3.169 0.002 Temperature Marten -0.023 0.019 -1.217 0.224
Exposure time Marten 0.021 0.016 1.32 0.187 40-80y Squirrel 0.017 0.006 2.627 0.009
Transect type Squirrel 0.982 0.313 3.134 0.002 Exposure time Squirrel 0.041 0.013 3.222 0.001
Squirrel Lynx 0.274 0.2 1.371 0.17 40-80y Lynx -0.036 0.016 -2.211 0.027
Exposure time Lynx 0.015 0.034 0.435 0.664
116
Hare Fox 0.053 0.016 3.29 0.001 40-80y Fox -0.007 0.007 -1.004 0.315
Transect type Fox -0.991 0.496 -1.998 0.046 Exposure time Fox 0.037 0.015 2.462 0.014 Temperature 40-80y -0.748 0.117 -6.369 <0.001
2014 0.982
Squirrel Hare 0.149 0.065 2.293 0.022 40-80y Hare -0.01 0.005 -1.93 0.054
Temperature Hare 0.061 0.017 3.648 <0.001 Exposure time Hare 0.004 0.008 0.452 0.651
Hare Marten 0.096 0.042 2.255 0.024 Transect type Marten 1.148 0.69 1.665 0.096 Temperature Marten -0.13 0.064 -2.034 0.042
Exposure time Marten 0.023 0.036 0.648 0.517 Transect type Squirrel -1.25 0.59 -2.12 0.034 Exposure time Squirrel -0.001 0.017 -0.052 0.958
Marten Lynx 1.474 0.919 1.604 0.109 Exposure time Lynx -0.065 0.028 -2.283 0.022
Lynx Fox -1.503 1.132 -1.328 0.184 40-80y Fox 0.016 0.011 1.438 0.151
Exposure time Fox -0.043 0.017 -2.499 0.012 Temperature 40-80y 0.329 0.254 1.297 0.197
40-80y Transect type -0.017 0.014 -1.234 0.217
Temperature Transect type 0.155 0.045 3.407 0.001
117
GENERAL CONCLUSION
This thesis addressed spatio-temporal dynamics of wintering mammals, with a focus on
snowshoe hare and marten. Habitat selection by snowshoe hare was influenced by current
density and recent density and they tended to move to older habitat in deeper snow. The long
term data combined with multiple species addressed habitat selection by mesopredators as
the likely result of trade-offs between energy gain, predation avoidance, and competitor
abundance, leading to changes in habitat selection depending on community dynamics.
The challenge for developping a reliable population index
Chapter 1 played a key role in the thesis, by providing evidence for the reliability of track
counts as a proxy for species dynamics. In this chapter, I estimated year effect in the
regression from snow tracking data, and showed a good agreement with pelt sales in red
squirrel and weasels and mean track counts accounting for exposure time agreed with marten
pelt sales. Besides supporting the reliability of track counts, those results should be of interest
to furbearer managers, because they support the use of pelt sales as a population index.
Therefore, pelt sales can be useful to investigate the effect of forestry on population trends of
mammals. Pelt sales have been recorded over the Quebec province since 2003 (the Ministère
des Forêts, de la Faune et des Parcs: https://www.mffp.gouv.qc.ca/faune/statistiques/chasse-
piegeage.jsp).
Then, in chapter 4, the result indicated track counts of marten were not only influenced
by habitat but also by track counts of prey, snowshoe hare and squirrel, and of competitor,
lynx and fox. Therefore, estimated year effect could be higher in the year when the space
highly used by preys was intensively surveyed. Incorporating prey track counts into the
118
regression analysis might increase agreement with pelt sales due to removal of bias associated
with prey and competitor, potentially leading to obtain better popualtion index.
Feedback effect of density on habitat selection of snowshoe hare
In chapter 2, I addressed the possible time lag effect of conspecific density on habitat
selection. Immediate effects of density combined with time lagged effect best performed in
explaining use of regenerating and young forests. Habitat use of hare has been frequently
investigated (e.g., Thronton et al. 2012, Bois et al. 2012). Though, it rarely addressed effect
of conspecific density on habitat selection and rarely showed population trends around study
years. Therefore, it would be difficult to perform further investigation if current or recent
density can have impact on observed habitat use. It is recommended that, where possible,
habitat studies should show population trend of a species in interest to allow comparisons
among studies.
In contrast to younger forests, I was unable to explain the use of older forests by
conspecific density. The result from chapter 4 revealed that hare tracks were spatially
associated with squirrel tracks. Allard-Duchene et al. (2014) also showed that hare and
squirrel had similar responses to stand age after fire disturbance, in which two species had
peaks at almost the same age. Since the use of mature forests was mediated by squirrels, this
might reduce predictive power of ideal free distribution, resulting in poor performance of
ideal free distribution models for this habitat.
Snow depth influenced habitat use of hare
In chapter 3, I found that snowshoe hare more frequently used foliage in the middle height
when snow was deeper. Stand ages at the peak habitat use of hare were variable with winter
119
habitat use studies. Since maximum snow depth varied with regions, variation of snow depth
may explain variations in hare habitat use pattern. Forestry practices including thinning and
clear-cut are also known to influence snow depth (Horstkotte and Roturier 2013). Difference
in forestry practice among regions would additionaly explain the variations.
Predators are often spatially associated with prey (e.g., lynx – snowshoe hare; Keim et al.
2011) as shown in chapter 4. In the studies of prey-predator spatial distribution, prey location
was often assumed to be fixed (e.g., Latham et al. 2013) even though we do know that the
spatial distribution of prey is dynamic, even at short time scales, as shown in Chapter 3.
Besides, as shown in chapter 2, prey (snowshoe hares) expanded their distribution toward
less preferred habitat in years of high abundance. Monthly and yearly spatial dynamics are
therefore likely to cause significant uncertainty in prey-predator studies that assume
negligible or nonexistent prey location shifts at timescales considered. Chapter 3 also has
implications in the longer run, if regional climate change does influence regional snow
precipitation in decadal time scales (Christensen 2013). Not only decadal dynamics in snow
depth change snowshoe hare distribution, but also that associated predators, marten and lynx.
Interactions in spatial distribution among mammals
In chapter 4, I hypothesized that the spatial association between a mesocarnivore (marten)
and an herbivore (hare) tended to decrease with higher abundance of competitor, lynx, in the
previous year. Lowered spatial association appeared to lead marten towards areas with an
alternative prey, red squirrel.
Prey switching has been investigated as function of ratio of primary prey and secondary
prey. However, as observed in chapter 4, larger predator abundance might have contributed
120
more to explain prey switching indirectly by decreasing spatial relationship between
predation and primary prey, rather than prey abundance.
Caveats
In my thesis, I investigated density-dependent habitat selection at yearly scale, implying
that population size of hare was assumed to be constant within a winter. However, it is known
that winter population size of snowshoe hare declines along with time (Kielland et al. 2010).
Thus, the inter-season declines in population size might affect habitat selection.
An important assumption in this thesis is that observed behavioral decision making by
animals should lead to fitness gain or maintain fitness. However, in this thesis, I did not
address fitness components (e.g., survival and population growth rate). Thus, my general
question regarding fitness is that « Does observed behavioral response to stimuli contribute
to fitness gain? And the spatial distribution of animals as the end result of the response can
contribute to fitness gain at the whole population level? ». More specific questions will be:
1) Which habitat had the most contribution to fitness (local population growth rate) of
snowshoe hare?
2) Did high local abundance of snowshoe hare or red squirrel contribute to increasing
local population growth rate or maintaining local population of marten?
3) When marten switched to the alternative prey, was its population size declined or
stable?
Chapter 4 focused on spatial association between preys and predators because strong
spatial asspciation was assumed to lead to successful hunting and energy gain, possibly
resulting in fitness gain. Since spatial association does not necessary mean hunting success
121
(Lima and Dill 1990), this assumption would be supported by providing an evidence that
frequency of occurrence of a certain prey in predator diet is increased with spatial association.
Diet analysis (e.g., scat survey) would enable to verify that the frequency of a certain prey
(e.g., hare) in mesocarnivore diet is declined in presence of larger predatopr sharing the prey
item. Such a verification would give strong support to the result observed in chapter 4.
My thesis addressed the questions about fundamental ecology but did not address
application aspect. A question is “Does the knowledge obtained throughout the entire thesis
improve prediction of spatial distribution of a animal species?”. In order to examine
applicability of the knowledge to wildlife management, it is important to conduct snow
tracking in another year in the study site and then compare observed track counts in 2015
with track counts that the knowledge obtained throughout the thesis predicts.
Management implications
Studies on spatial relationship between a species and forest stand types should ideally
address population trends of the species during study period and the previous year of first
study year in order to avoid confounding effects that would otherwise lead to possibly biased
assessments of preferred habitat. Though a lot of studies have addressed effect of forestry
practices including site preparation and commercial thinning on animal distribution (e.g.,
Thornton et al. 2012a), their interpretations of results are likely to vary with population status
during study period, thus leading to changes in management implication to forestry.
Conserving different types of habitat (stand age) with similar proportion in a landscape
as in the study site would be favorable for mammal community, at least a part of community,
and would increase resilience of the community to climate change. The first reason for this
122
statement is due to the result that preferred habitat of snowshoe hare varied with snow depth.
Given future fluctuation of snow precipitation (either increase or decrease), preparing
different habitat in terms of stand age would enable hare to switch to different habitat
depending on fluctuating snow depth. Since abundance of squirrel indexed by track count
was high in mature forest (seen in other literatures Fisher and Wilkinson 2005), presence of
different habitat can enable different prey items to exist in a landscape, potentially offering
chance of prey switching to carnivores.
Conserving carnivores, for example marten, might require setting conservation areas to
include high prey abundance areas. Protection of a focal species by setting conservation areas
would be efficient when strong association between a focal species and its preferred habitat
is validated. Marten is often considered as an indicator species (Thompson 1991) and is
registered in endangered species act in New Foundland (The Newfoundland Marten
Recovery Team 2010) and is considered as a species dependent on mature and over mature
forest. In chapter 4, marten showed variable response to mature habitat (40-80yr) which is
also observed in other literature (Potvin et al. 2000). On the other hand, marten was more
frequently (frequency of direct links occurred) spatially associated with prey species. Setting
conservation area might require to entail preferred habitat of prey item of marten.
Long term studies and snow tracking
While long term studies have been acknowledged as important to reveal ecological
processes, their success is often hindered by limited funding (Nelson et al. 2008) which is
required to assure personnel availability. For example, a 40y-study on demography of
seabirds, which had been supported financially, was closed due to termination of funding for
the project leading to lack of personnels (Birkhead 2014). Snow tracking study would have
123
potentials as a long term study due to less cost and its simplicity which might allow to
integrate citizen participation into long term monitoring (Dickinson et al. 2012) for assuring
personnels. In addition to these advantages, this thesis based on snow tracking has reavealed
important ecological process and therefore snow tracking would also have high capacity to
address complex ecological issues such as time lag and interspecific interaction, leading to
further understanding how a part of biodiversity is maintained.
125
LITERATURES CITED
Allard-Duchêne, A., D. Pothier, A. Dupuch, and D. Fortin. 2014. Temporal changes in habitat
use by snowshoe hares and red squirrels during post-fire and post-logging forest
succession. Forest Ecology and Management 313:17–25.
Anderson, D. R. 2001. The need to get the basics right in wildlife field studies. Wildlife
Society Bulletin 29:1294–1297.
Andruskiw, M., J. M. Fryxell, I. D. Thompson, and J. A. Baker. 2008. Habitat-mediated
variation in predation risk by the American marten. Ecology 89:2273–2280.
Apps, C. D. 1999. Space-use, diet, demographics, and topographic association of lynx in the
southern Canadian Rocky Mountains: a study. Pages 351– 371 in L. F. A. Ruggiero,
editor. Ecology and conservation of lynx in the United States. University Press of
Cololado, Boulder, Cololado, USA.
Atchley, W. R., C. T. Gaskins, and D. Anderson. 1976. Statistical properties of ratios. I.
Empirical results. Systematic Zoology 25:137–148.
Ayllón, D., G. G. Nicola, I. Parra, B. Elvira, and A. Almodóvar. 2013. Intercohort density
dependence drives brown trout habitat selection. Acta Oecologica 46:1–9.
Beaudoin, C., M. Crête, J. Huot, P. Etcheverry, and S. D. Côté. 2004. Does predation risk
affect habitat use in snowshoe hares? Écoscience 11:370–378.
Bélanger, L. 2001. La forêt mosaïque comme stratégie de conservation de la biodiversité de
la sapinière boréale de l’Est: l’expérience de la Forêt Montmorency. Le Naturaliste
Canadien 125:18–25.
Birkhead, T. 2014. Stormy outlook for long-term ecology studies. Nature 514:405.
126
Bélanger, L., L. Bertrand, P. Bouliane, and L.-J. Lussier. 1991. Plan général d’aménagement
de la Forêt Montmorency. Page xvii + 215. Université Laval, Faculté de foresterie et
de géomatique, Sainte-Foy, QC, Canada.
De Bellefeuille, S., L. Bélanger, J. Huot, and A. Cimon. 2001. Clear-cutting and regeneration
practices in Quebec boreal balsam fir forest: effects on snowshoe hare. Canadian
Journal of Forest Research 31:41–51.
Borer, E. T., C. J. Briggs, W. W. Murdoch, and S. L. Swarbrick. 2003. Testing intraguild
predation theory in a field system: does numerical dominance shift along a gradient
of productivity? Ecology Letters 6:929–935.
Bouliane, J., L. Bélanger, H. Sansregret, and P. Pineault. 2015. Plan d’aménagement forestier
intégré tactique 2014-2019. Les presses de l’Université Laval, Québec, QC, Canada.
Brandt, J. P., M. D. Flannigan, D. G. Maynard, I. D. Thompson, and W. J. A. Volney. 2013.
An introduction to Canada’s boreal zone: ecosystem processes, health, sustainability,
and environmental issues. Environmental Reviews 21:207–226.
Brooks, T. M., S. L. Pimm, and J. O. Oyugi. 1999. Time lag between deforestation and bird
extinction in tropical forest fragments. Conservation Biology 13:1140–1150.
Brown, J. L. 1969. The buffer effect and productivity in tit populations. American Naturalist
103:347–354.
Brown, J. S., and B. P. Kotler. 2004. Hazardous duty pay and the foraging cost of predation.
Ecology Letters 7:999–1014.
Brown, R. D., and R. O. Braaten. 1998. Spatial and temporal variability of Canadian monthly
snow depths, 1946–1995. Atmosphere-Ocean 36:37–54.
127
Brown, R. D., and B. Brasnett. 2010. Canadian Meteorological Centre (CMC) Daily Snow
Depth Analysis Data. Environment Canada, 2010. USA: National Snow and Ice Data
Center.
Burnham, K. P., and D. R. Anderson. 2002. Model selection and inference: a practical
information-theoretic approach. 2nd edition. Springer-Verlag, New York, NY, USA.
Buskirk, S. W., and R. A. Powell. 1994. Habitat ecology of Fishers and American Martens.
Pages 283–296 in S. W. Buskirk, A. S. Harestad, M. G. Raphael, and R. A. Powell,
editors. Martens, Sables, and Fishers. Comstock, Ithaca, NY, USA.
Campbell, J. L., M. J. Mitchell, P. M. Groffman, L. M. Christenson, and J. P. Hardy. 2005.
Winter in northeastern North America: a critical period for ecological processes.
Frontiers in Ecology and the Environment 3:314–322.
Christensen, J. H., K. Krishna Kumar, E. Aldrian, S.-I. An, I. F. A. Cavalcanti, M. de Castro,
W. Dong, P. Goswami, A. Hall, J. K. Kanyanga, A. Kitoh, J. Kossin, N.-C. Lau, J.
Renwick, D. B. Stephenson, S.-P. Xie, and T. Zhou. 2013. Climate Phenomena and
their Relevance for Future Regional Climate Change. Pages 1217–1308 in T. F.
Stocker, D. Qin, G.-K. Plattner, M. Tignor, S. K. Allen, J. Boschung, A. Nauels, Y.
Xia, V. Bex, and P. M. Midgley, editors. Climate Change 2013: The Physical Science
Basis. Contribution of Working Group I to the Fifth Assessment Report of the
Intergovernmental Panel on Climate Change. Cambridge University Press,
Cambridge, United Kingdom and New York, NY, USA.
Clark, T. W., E. Anderson, C. Douglas, and M. Strickland. 1987. Martes americana.
Mammalian Species 289:1–8
Crooks, J. A. 2005. Lag times and exotic species: The ecology and management of biological
invasions in slow-motion. Ecoscience 12:316–329.
128
Crowther, M. S., D. Lunney, J. Lemon, E. Stalenberg, R. Wheeler, G. Madani, K. A. Ross,
and M. Ellis. 2014. Climate-mediated habitat selection in an arboreal folivore.
Ecography 37:336–343.
Cumberland, R. E., J. A. Dempsey, and G. J. Forbes. 2001. Should diet be based on biomass?
Importance of larger prey to the American marten. Wildlife Society Bulletin 29:1125–
1130.
Dickinson, J. L., J. Shirk, D. Bonter, R. Bonney, R. L. Crain, J. Martin, T. Phillips, and K.
Purcell. 2012. The current state of citizen science as a tool for ecological research and
public engagement. Frontiers in Ecology and the Environment 10:291–297.
Douglas, C. W., and M. A. Strickland. 1987. Fisher. Pages 511–529 in M. J. Novak, J.A.
Baker, M. E. Obbard, and B. Malloch, editors. Wildlife furbearer management and
conservation in North America. Ontario Trappers association, North Bay, Ontario,
Canada.
Dreisig, H. 1995. Ideal free distributions of nectar foraging bumblebees. Oikos 72:161–172.
Dussault, C., M. Poulin, R. Courtois, and J.-P. Ouellet. 2006. Temporal and spatial
distribution of moose-vehicle accidents in the Laurentides Wildlife Reserve, Quebec,
Canada. Wildlife Biology 12:415–425.
Elmhagen, B., G. Ludwig, S. P. Rushton, and H. Lindén. 2010. Top predators, mesopredators
and their prey: interference ecosystems along bioclimatic productivity gradients.
Journal of Animal Ecology 79:785–794.
Elton, C., and M. Nicholson. 1942. The Ten-Year Cycle in Numbers of the Lynx in Canada.
Journal of Animal Ecology 11:215–244.
Erb, J., M. S. Boyce, and N. C. Stenseth. 2001. Population dynamics of large and small
mammals. Oikos 92:3–12.
129
Fedriani, J. M., T. K. Fuller, R. M. Sauvajot, and E. C. York. 2000. Competition and
intraguild predation among three sympatric carnivores. Oecologia 125:258–270.
Ferron, J., F. Potvin, and C. Dussault. 1998. Short-term effects of logging on snowshoe hares
in the boreal forest. Canadian Journal of Forest Research 28:1335–1343.
Fisher, J. T., and L. Wilkinson. 2005. The response of mammals to forest fire and timber
harvest in the North American boreal forest. Mammal Review 35:51–81.
Fisher, J. T., B. Anholt, S. Bradbury, M. Wheatley, and J. P. Volpe. 2013. Spatial segregation
of sympatric marten and fishers: the influence of landscapes and species-scapes.
Ecography 36:240–248.
Fitzgerald, B. M. 1977. Weasel predation on a cyclic population of the montane vole
(Microtus montanus) in California. Journal of Animal Ecology 46:367–397.
Flynn, R. W., and T. V. Schumacher. 2009. Temporal changes in population dynamics of
American martens. Journal of Wildlife Management 73:1269–1281.
Fortin, C., and M. Cantin. 2004. Harvest status, reproduction and mortality in a population
of American martens in Quebec, Canada. Pages 221–234 in D. J. Harrison, A. K.
Fuller, and G. Proulx, editors. Martens and fishers (Martes) in human-altered
environments: An international perspective. Springer, New York, NY, USA.
Fortin, D., H. L. Beyer, M. S. Boyce, D. W. Smith, T. Duchesne, and J. S. Mao. 2005. Wolves
influence elk movements: behavior shapes a trophic cascade in Yellowstone National
Park. Ecology 86:1320–1330.
Framstad, E., N. C. Stenseth, O. N. Bjornstad, and W. Falck. 1997. Limit cycles in norwegian
lemmings: tensions between phase-dependence and density-dependence. Proceedings
of Biological Sciences 264:31–38.
130
Fretwell, S. D., and H. L. Lucas. 1970. On territorial behaviour and other factors influencing
habitat distribution in birds. I. Theoretical development. Acta Biotheoretica 19:16–
36.
Fryxell, J. M., J. B. Falls, E. A. Falls, R. J. Brooks, L. Dix, and M. A. Strickland. 1999.
Density dependence, prey dependence and population dynamics of martens in Ontario.
Ecology 80:1311–1321.
Fryxell, J. M., D. J. T. Hussell, A. B. Lambert, and P. C. Smith. 1991. Time lags and
population fluctuations in white-tailed deer. Journal of Wildlife Management 55:377–
385.
Fuller, T. K. 1991. Effect of snow depth on wolf activity and prey selection in north central
Minnesota. Canadian Journal of Zoology 69:283–287.
Gauthier, S., M.-A. Vaillancourt, A. Leduc, L. De Grandpré, D. Kneeshaw, H. Morin, P.
Drapeau, and Y. Bergeron. 2008. Aménagement écosystémique en forêt boréale.
Presses de l’Université du Québec, Québec, QC, Canada.
Gese, E. M. 2001. Monitoring of terrestrial carnivore populations. Pages 372–396 in J. L.
Gittleman, S. M. Funk, D. W. MacDonald, and R. K. Wayne, editors. Carnivore
conservation. Cambridge University Press & The Zoological Society of London,
Cambridge, UK.
Gill, J. A., K. Norris, P. M. Potts, T. G. Gunnarsson, P. W. Atkinson, and W. J. Sutherland.
2001. The buffer effect and large-scale population regulation in migratory birds.
Nature 412:436–438.
Gilroy, J. J., and W. J. Sutherland. 2007. Beyond ecological traps: perceptual errors and
undervalued resources. Trends in Ecology & Evolution 22:351–356.
131
Gompper, M. E., R. W. Kays, J. C. Ray, S. D. Lapoint, D. A. Bogan, and J. R. Cryan. 2006.
A comparison of noninvasive techniques to survey carnivore communities in
northeastern North America. Wildlife Society Bulletin 34:1142–1151.
Gorini, L., J. D. C. Linnell, R. May, M. Panzacchi, L. Boitani, M. Odden, and E. B. Nilsen.
2012. Habitat heterogeneity and mammalian predator–prey interactions. Mammal
Review 42:55–77.
Haché, S., M.-A. Villard, and E. M. Bayne. 2012. Experimental evidence for an ideal free
distribution in a breeding population of a territorial songbird. Ecology 94:861–869.
Halfpenny, J. C., R. W. Thompson, S. C. Morse, T. Holden, and P. Rezendes. 1995. Snow
tracking. Pages 91–124 in W. J. Zielinski and T. E. Kucera, editors. American marten,
fisher, lynx, and wolverine: survey methods for their detection. USDA Forest Service,
Pacific Southwest Research Station, Albany, CA, USA.
Halpin, M. A., and J. A. Bissonette. 1988. Influence of snow depth on prey availability and
habitat use by red fox. Canadian Journal of Zoology 66:587–592.
Hearn, B. J. 2007, May. Factors affecting habitat selection and population characteristics of
American marten (Martes americana atrata) in Newfound Land. The University of
Maine, Maine, The United States.
Hildén, O. 1965. Habitat selection in birds: review. Annales Zoologici Fennici 2:53–75.
Hiller, T. L., D. R. Etter, J. L. Belant, and A. J. Tyre. 2011. Factors affecting harvests of
fishers and American martens in northern Michigan. Journal of Wildlife Management
75:1399–1405.
Hodges, K. E. 1999. The Ecology of Snowshoe Hares in Northern Boreal Forests. Pages 117–
162 in L. F. Ruggiero, K. B. Aubry, S. W. Buskirk, G. M. Koehler, C. J. Krebs, K. S.
132
McKelvey, and J. R. Squires, editors. Ecology and conservation of lynx in the United
States. University Press of Cololado, Colorado, the United States.
Hodson, J., D. Fortin, and L. Bélanger. 2010a. Fine-scale disturbances shape space-use
patterns of a boreal forest herbivore. Journal of Mammalogy 91:607–619.
Hodson, J., D. Fortin, and L. Bélanger. 2011. Changes in relative abundance of snowshoe
hares (Lepus americanus) across a 265-year gradient of boreal forest succession.
Canadian Journal of Zoology 89:908–920.
Hodson, J., D. Fortin, M.-L. LeBlanc, and L. Bélanger. 2010b. An appraisal of the fitness
consequences of forest disturbance for wildlife using habitat selection theory.
Oecologia 164:73–86.
Horstkotte, T., and S. Roturier. 2013. Does forest stand structure impact the dynamics of
snow on winter grazing grounds of reindeer (Rangifer t. tarandus)? Forest Ecology
and Management 291:162–171.
Jensen, P. G., C. L. Demers, S. A. Mcnulty, W. J. Jakubas, and M. M. Humphries. 2012.
Marten and fisher responses to fluctuations in prey populations and mast crops in the
northern hardwood forest. Journal of Wildlife Management 76:489–502.
Jensen, W. E., and J. F. Cully. 2005. Density-dependent habitat selection by brown-headed
cowbirds (Molothrus ater) in tallgrass prairie. Oecologia 142:136–149.
Kapfer, P. M., and K. B. Potts. 2012. Socioeconomic and ecological correlates of bobcat
harvest in Minnesota. Journal of Wildlife Management 76:237–242.
Keane, R. E., P. F. Hessburg, P. B. Landres, and F. J. Swanson. 2009. The use of historical
range and variability (HRV) in landscape management. Forest Ecology and
Management 258:1025–1037.
133
Keim, J. L., P. D. DeWitt, and S. R. Lele. 2011. Predators choose prey over prey habitats:
evidence from a lynx-hare system. Ecological Applications 21:1011–1016.
Kielland, K., K. Olson, and E. Euskirchen. 2010. Demography of snowshoe hares in relation
to regional climate variability during a 10-year population cycle in interior Alaska.
Canadian Journal of Forest Research 40:1265–1272.
Klemola, T., E. Korpimäki, K. Norrdahl, M. Tanhuanpaa, and M. Koivula. 1999. Mobility
and habitat utilization of small mustelids in relation to cyclically fluctuating prey
abundance. Annales Zoologici Fennici 36:75–82.
Korslund, L., and H. Steen. 2006. Small rodent winter survival: snow conditions limit access
to food resources. Journal of Animal Ecology 75:156–166.
Krebs, C. J. 2011. Of lemmings and snowshoe hares: the ecology of northern Canada.
Proceedings of the Royal Society B: Biological Sciences 278:481–489.
Krebs, C. J., R. Boonstra, S. Boutin, and A. R. E. Sinclair. 2001a. What drives the 10-year
cycle of snowshoe hares? BioScience 51:25–35.
Krebs, C. J., S. A. Boutin, and R. Boonstra. 2001b. Ecosystem dynamics of the boreal forest:
the Kluane project. New York, NY: Oxford University Press.
Kunkel, K. E., M. Palecki, L. Ensor, K. G. Hubbard, D. Robinson, K. Redmond, and D.
Easterling. 2009. Trends in twentieth-Century U.S. snowfall using a quality-
controlled dataset. Journal of Atmospheric and Oceanic Technology 26:33–44.
Latham, A. D. M., M. C. Latham, K. H. Knopff, M. Hebblewhite, and S. Boutin. 2013.
Wolves, white-tailed deer, and beaver: implications of seasonal prey switching for
woodland caribou declines. Ecography 36:1276–1290.
134
Latombe, G., D. Fortin, and L. Parrott. 2014. Spatio-temporal dynamics in the response of
woodland caribou and moose to the passage of grey wolf. Journal of Animal Ecology
83:185–198.
Laundré, J. W., L. Hernández, and K. B. Altendorf. 2001. Wolves, elk, and bison:
reestablishing the “landscape of fear” in Yellowstone National Park, U.S.A. Canadian
Journal of Zoology 79:1401–1409.
Laundré, J. W., L. Hernández, and W. J. Ripple. 2010. The landscape of fear: ecological
implications of being afraid. Open Ecology Journal 3:1-7.
Liebhold, A., W. D. Koenig, and O. N. Bjørnstad. 2004. Spatial synchrony in population
dynamics. Annual Review of Ecology, Evolution, and Systematics 35:467–490.
Lima, S. L. 2002. Putting predators back into behavioral predator–prey interactions. Trends
in Ecology and Evolution 17:70–75.
Lima, S. L., and L. M. Dill. 1990. Behavioral decisions made under the risk of predation: a
review and prospectus. Canadian Journal of Zoology 68:619–640.
Lim, K., P. Treitz, M. Wulder, B. St-Onge, and M. Flood. 2003. LiDAR remote sensing of
forest structure. Progress in Physical Geography 27:88–106.
Lindenmayer, D. B., G. E. Likens, A. Andersen, D. Bowman, C. M. Bull, E. Burns, C. R.
Dickman, A. A. Hoffmann, D. A. Keith, M. J. Liddell, A. J. Lowe, D. J. Metcalfe, S.
R. Phinn, J. Russell-Smith, N. Thurgate, and G. M. Wardle. 2012. Value of long-term
ecological studies. Austral Ecology 37:745–757.
Litvaitis, J. A., J. A. Sherburne, and J. A. Bissonette. 1985. Influence of understory
characteristics on snowshoe hare habitat use and density. Journal of Wildlife
Management 49:866–873.
135
Long, J. N. 2009. Emulating natural disturbance regimes as a basis for forest management:
A North American view. Forest Ecology and Management 257:1868–1873.
MacKenzie, D. I., J. D. Nichols, G. B. Lachman, S. Droege, J. Andrew Royle, and C. A.
Langtimm. 2002. Estimating site occupancy rates when detection probabilities are
less than one. Ecology 83:2248–2255.
Magnuson, J. J. 1990. Long-term ecological research and the invisible present. BioScience
40:495–501.
Martin, S. K. 1994. Feeding ecology of American Martens and Fishers. Pages 297–315 in S.
W. Buskirk, A. S. Harestad, M. G. Raphael, and R. A. Powell, editors. Martens,
Sables, and Fishers. Comstock, Ithaca, NY, USA.
McDaniel, L. S., and N. Henderson. 2015. geeM: Solve Generalized Estimating Equations.
R package version 0.7.4.
Mills, L. S., M. Zimova, J. Oyler, S. Running, J. T. Abatzoglou, and P. M. Lukacs. 2013.
Camouflage mismatch in seasonal coat color due to decreased snow duration.
Proceedings of the National Academy of Sciences 110:7360–7365.
Mobæk, R., A. Mysterud, L. Egil Loe, Ø. Holand, and G. Austrheim. 2009. Density
dependent and temporal variability in habitat selection by a large herbivore; an
experimental approach. Oikos 118:209–218.
Morris, D. W. 1996. Temporal and spatial population dynamics among patches connected by
habitat selection. Oikos 75:207–219.
Morris, D. W. 2003. Toward an ecological synthesis: a case for habitat selection. Oecologia
136:1–13.
136
Morris, D. W. 2005. Habitat-dependent foraging in a classic predator-prey system: a fable
from snowshoe hares. Oikos 109:239–254.
Morris, D. W., D. L. Davidson, and C. J. Krebs. 2000. Measuring the ghost of competition:
Insights from density-dependent habitat selection on the co-existence and dynamics
of lemming. Evolutionary Ecology Research 2:41–67.
Morris, D. W., and J. T. MacEachern. 2010. Active density-dependent habitat selection in a
controlled population of small mammals. Ecology 91:3131–3137.
Morrison, S. F., G. J. Forbes, S. J. Young, and S. Lusk. 2003. Within-yard habitat use by
white-tailed deer at varying winter severity. Forest Ecology and Management
172:173–182.
Mowat, G., and B. Slough. 2003. Habitat preference of Canada lynx through a cycle in
snowshoe hare abundance. Canadian Journal of Zoology 81:1736–1745.
Murdoch, W. W., S. Avery, and M. E. B. Smyth. 1975. Switching in Predatory Fish. Ecology
56:1094–1105.
Murray, D. L., and S. Boutin. 1991. The influence of snow on lynx and coyote movements:
does morphology affect behavior? Oecologia 88:463–469.
Naughton, D. 2012. The natural history of canadian mammals. University of Toronto, Canada.
Nelson, M. P., R. O. Peterson, and J. A. Vucetich. 2008. The Isle Royale Wolf-Moose project:
50 years of challenge and insight. George Wright Society 25:98–113
Oro, D. 2008. Living in a ghetto within a local population: an emprical example of an ideal
despotic distribution. Ecology 89:838–846
Pauli, J. N., B. Zuckerberg, J. P. Whiteman, and W. Porter. 2013. The subnivium: a
deteriorating seasonal refugium. Frontiers in Ecology and the Environment 11:260–
267.
137
Pedersen, Å. Ø., R. A. Ims, N. G. Yoccoz, V. H. Hausner, and K. H. Juell. 2010. Scale-
dependent responses of predators and their prey to spruce plantations in subarctic
birch forests in winter. Ecoscience 17:123–136.
Pellikka, J., H. Rita, and H. Lindén. 2005. Monitoring wildlife richness - Finnish applications
based on wildlife triangle censuses. Annales Zoologici Fennici 42:123–134.
Polis, G. A., C. A. Myers, and R. D. Holt. 1989. The ecology and evolution of intraguild
predation: potential competitors that eat each other. Annual Review of Ecology and
Systematics 20:297–330.
Poole, K. G., and R. P. Graf. 1996. Winter diet of marten during a snowshoe hare decline.
Canadian Journal of Zoology 74:456–466.
Potvin, F., L. Bélanger, and K. Lowell. 2000. Marten habitat selection in a clearcut boreal
landscape. Conservation Biology 14:844–857.
Potvin, F., R. Courtois, and L. Bélanger. 1999. Short-term response of wildlife to clear-
cutting in Quebec Boreal forest : multiscale effects and management implications.
Canadian Journal of Forest Research 29:1120–1127.
Powell, R. A., S. W. Buskirk, and W. J. Zielinski. 2003. Fisher and marten. Pages 635–649
in G. A. Feldhamer, B. C. Hompson, and J. A. Chapman, editors. Wild mammals of
north america : Biology, management and conservation. Second edition. The Johns
Hopkins University Press, Baltimore, Maryland, USA.
Racine, E. B., N. C. Coops, B. St-Onge, and J. Bégin. 2014. Estimating forest stand age from
LiDAR-derived predictors and nearest neighbor imputation. Forest Science 60:128–
136.
138
Randa, L. A., D. M. Cooper, P. L. Meserve, and J. A. Yunger. 2009. Prey Switching of
Sympatric Canids in Response to Variable Prey Abundance. Journal of Mammalogy
90:594–603.
Raphael, M. G. 1994. Techniques for monitoring populations of Fishers and American
Martens. Pages 224–240 in S. W. Buskirk, A. S. Harestad, M. G. Raphael, and R. A.
Powell, editors. Martens, Sables, and Fishers. Comstock, Ithaca, NY, USA.
R Core Team. 2013. R: A language and environment for statistical computing. R Foundation
for Statistical Computing, Vienna, Austria.
Reid, D. G., F. Bilodeau, C. J. Krebs, G. Gauthier, A. J. Kenney, B. S. Gilbert, M. C.-Y.
Leung, D. Duchesne, and E. Hofer. 2012. Lemming winter habitat choice: a snow-
fencing experiment. Oecologia 168:935–946.
Roberts, N. M., and S. M. Crimmins. 2010. Bobcat population status and management in
north America: Evidence of large-scale population increase. Journal of Fish and
Wildlife Management 1:169–174.
Robertson, B. A., and R. L. Hutto. 2007. Is selectively harvested forest an ecological trap for
Olive-sided Flycatchers? Condor 109:109–121.
Rosenzweig, M. L. 1981. A theory of habitat selection. Ecology 62:327–335.
Roy, C., L. Imbeau, and M. J. Mazerolle. 2010. Transformation of abandoned farm fields into
coniferous plantations: Is there enough vegetation structure left to maintain winter
habitat of snowshoe hares? Canadian Journal of Zoology 88:579–588.
Ryan, C. W., J. C. Pack, W. K. Igo, J. C. Rieffenberger, and A. B. Billings. 2004. Relationship
of mast production to big-game harvests in West Virginia. Wildlife Society Bulletin
32:786–794.
139
Shenbrot, G. 2004. Habitat selection in a seasonally variable environment: test of the isodar
theory with the fat sand rat, Psammomys obesus, in the Negev Desert, Israel. Oikos
106:359–365.
Shipley, B. 1997. Exploratory path analysis with applications in ecology and evolution. The
American Naturalist 149:1113–1138.
Shipley, B. 2000. A new inferential test for path models based on directed acyclic graphs.
Structural Equation Modeling: A Multidisciplinary Journal 7:206–218.
Shipley, B. 2012. The AIC model selection method applied to path analytic models compared
using a d-separation test. Ecology 94:560–564.
Silva, M., K. Johnson, and S. Opps. 2009. Habitat use and home range size of red foxes in
Prince Edward Island (Canada) based on snow-tracking and radio-telemetry data.
Central European Journal of Biology 4:229–240
Simard, M. A., S. D. Côté, A. Gingras, and T. Coulson. 2012. Tests of density dependence
using indices of relative abundance in a deer population. Oikos 121:1351–1363.
Smith, L. M., I. L. Brisbin, and G. C. White. 1984. An evaluation of total trapline captures as
estimates of furbearer abundance. Journal of Wildlife Management 48:1452–1455.
Stamps, J. A., V. V. Krishnan, and M. L. Reid. 2005. Search costs and habitat selection by
dispersers. Ecology 86:510–518.
St-Laurent, M. H., M. Cusson, J. Ferron, and A. Caron. 2008. Use of residual forest by
snowshoe hare in a clear-cut boreal landscape. Northeastern Naturalist 15:497–514.
Sulkava, R. 2007. Snow tracking: a relevant method for estimating otter Lutra lutra
populations. Wildlife Biology 13:208–218.
The Newfoundland Marten Recovery Team. 2010. Recovery plan for the threatened
Newfoundland population of American marten (Martes americana atrata). Page 31.
140
Wildlife Division, Department of Environment and Conservation, Government of
Newfoundland and Labrador, Corner Brook, Canada.
Thompson, I. D. 1991. Could marten become the spotted owl of eastern Canada? Forestry
Chronicle 67:136–140.
Thompson, I. D. 1994. Marten populations in uncut and logged boreal forests in Ontario.
Journal of Wildlife Management 58:272–280.
Thompson, I. D., and P. W. Colgan. 1987. Numerical responses of martens to a food shortage
in northcentral Ontario. Journal of Wildlife Management 51:824–835.
Thompson, I. D., and P. W. Colgan. 1990. Prey choice by marten during a decline in prey
abundance. Oecologia 83:443–451
Thompson, I. D., and P. W. Colgan. 1994. Marten activity in uncut and logged boreal forests
in Ontario. Journal of Wildlife Management 58:280–288.
Thompson, I. D., I. J. Davidson, S. O’Donnell, and F. Brazeau. 1989. Use of track transects
to measure the relative occurrence of some boreal mammals in uncut forest and
regeneration stands. Canadian Journal of Zoology 67:1816–1823.
Thomson, R. L., J. T. Forsman, F. Sardà-Palomera, and M. Mönkkönen. 2006. Fear factor:
prey habitat selection and its consequences in a predation risk landscape. Ecography
29:507-514.
Thornton, D. H., A. J. Wirsing, J. D. Roth, and D. L. Murray. 2012a. Complex effects of site
preparation and harvest on snowshoe hare abundance across a patchy forest landscape.
Forest Ecology and Management 280:132-139.
Thornton, D. H., A. J. Wirsing, J. D. Roth, and D. L. Murray. 2012b. Habitat quality and
population density drive occupancy dynamics of snowshoe hare in variegated
landscapes. Ecography 36:610–621.
141
Trainor, A. M., and O. J. Schmitz. 2014. Infusing considerations of trophic dependencies into
species distribution modelling. Ecology Letters 17:1507–1517.
Trainor, A. M., O. J. Schmitz, J. Ivan, and T. M. Shenk. 2013. Enhancing species distribution
modeling by characterizing predator-prey interactions. Ecological Applications
24:204–216.
Venables, W. N., and B. D. Ripley. 2002. Modern applied statistics with S. Fourth. Springer,
New York, USA.
Vigeant-Langlois, C., and A. Desrochers. 2011. Movements of wintering American marten
(Martes americana): relative influences of prey activity and forest stand age.
Canadian Journal of Forest Research 41:2202–2208.
Weinstein, M. S. 1977. Hares, lynx, and trappers. The American Naturalist 111:806–808.
White, K. S., G. W. Pendleton, and E. Hood. 2009. Effects of snow on sitka black-tailed deer
browse availability and nutritional carrying capacity in southeastern Alaska. Journal
of Wildlife Management 73:481–487.
Witt, J. C., C. R. Webster, R. E. Froese, T. D. Drummer, and J. A. Vucetich. 2012. Scale-
dependent drivers of ungulate patch use along a temporal and spatial gradient of snow
depth. Canadian Journal of Zoology 90:972–983.
Wolfe, M. L., N. V. Debyle, C. S. Winchell, and T. R. McCabe. 1982. Snowshoe Hare Cover
Relationships in Northern Utah. Journal of Wildlife Management 46:662–670.
Wulder, M. A., C. W. Bater, N. C. Coops, T. Hilker, and J. C. White. 2008. The role of
LiDAR in sustainable forest management. Forestry Chronicle 84:807–826.