Habitat overlap of Canada lynx and sympatric mesopredators increases following cyclical reduction in primary prey

Abstract

In multi-predator systems, niche expansion may increase habitat overlap among species and the potential for competitive or predatory interactions. The strength of those interactions is likely mediated by cyclic changes in prey availability. In the northern boreal and subboreal forests of Alaska and Canada, snowshoe hare populations cycle over an 8–11-year period and that variation can strongly influence the distribution and population dynamics of carnivores. We used camera traps to investigate the habitat use and overlap of lynx and sympatric mesopredators (coyote, fisher, wolverine) during two contrasting periods of hare abundance. Given optimal foraging theory and the relatively short interval between changes in prey abundance, we predicted that habitat overlap would increase during a low in hare abundance when Canada lynx and other mesopredators hunt alternative prey in varied habitats. We found that lynx occurrences mirrored the decrease in hares, while the occurrences of sympatric carnivore species increased during the low in hare abundance. Habitat overlap of lynx with other sympatric carnivores increased at a time of prey scarcity. Predator populations in subboreal forests may be in a dynamic state of habitat overlap dependent on cyclic prey abundance. Our results highlight the importance of long-term data and the consideration of natural cycles for interpreting broader scale questions about environmental disturbance and climate change.

Introduction

When resources are limiting, optimal foraging theory predicts that a species will expand the breadth of habitat and diet to maximize the rate of net energy intake^1,2. In a multi-predator system, niche expansion could increase habitat overlap among species and the potential for competitive or predatory interactions^3,4. Niche overlap among co-existing species may vary in response to fluctuating resources. When resources are abundant, each species may respond opportunistically to the same resource but with lessened competition. Alternatively, niche overlap and competitive interactions will increase following a decline in shared resources⁵. In systems where prey populations cycle with large amplitudes and over short durations, populations may be in a dynamic state of niche overlap dependent on prey-predator densities at a particular stage in a cycle.

In the northern boreal forests of Alaska and Canada, snowshoe hare (Lepus americanus) populations cycle over an 8–11-year period that results in contrasting years of prey abundance for the carnivore community^6–8. Canada lynx (Lynx canadensis; hereafter lynx), in particular, cycle asynchronously with snowshoe hare and are considered a specialist on this prey species^6–8. A time lag in the numerical response of predators to hare abundance can result in increased depensatory predation⁹. A combination of functional and numerical responses can contribute to density-dependent cycles¹⁰. Red squirrels (Tamiascurus hudsonicus) are often the second most dominant prey item in lynx diet, typically contributing more during lows in snowshoe hare abundance^11–13.

Coyote (Canis latrans) can also demonstrate asynchronous cycling dynamics in response to changes in the abundance of hare in the boreal forest^14,15. Although less pronounced, fisher (Pekania pennanti) populations in Ontario have shown a delayed and positive numerical response to increases in hare abundance¹⁶. Unlike lynx, coyotes and fishers are prey generalists that may be better able to adapt to low hare numbers by supplementing their diet with other prey (i.e., small mammals^13,16;). Large fluctuations in the abundance of hare will affect different predators to varying degrees, and likely will have strong influences on community dynamics.

Generalist predators are thought to have a stabilizing effect on predator-prey cycles because they feed on a diversity of prey and are less likely to over-exploit a single prey species¹⁰. The influence of generalist predators on lynx populations, especially during cyclic lows in hare numbers, is unclear. Previous studies have focused on potential coyote and lynx interactions by measuring numerical, functional, and behavioral responses to changes in hare abundance of each species individually^8,9,13,17. At a large spatial scale (129,000 km²), coyote activity can influence the distribution of lynx¹⁸. Coyotes and fishers can be both exploitative and interference competitors as well as occasionally preying on lynx^13,19,20. In Maine, predation was the leading cause of mortality for Canada lynx (28%) with fisher being the cause of 77% of predation events (14 of 18)¹⁹. Although wolverine (Gulo gulo) typically occupy a different ecological niche than lynx, the prey of the two predators overlaps; snowshoe hare can be a large component of wolverine diet [²¹, Robitaille et al. Unpubl. Data]. Wolverine may also be a dominant interference competitor for lynx²². In central British Columbia (BC), lynx are sympatric with coyotes, fisher, and wolverine, yet no studies have investigated the influence of cyclic prey abundance on the habitat overlap of lynx with this assemblage of predators.

Variability in spatial and temporal resources can influence the dynamics of niche differentiation and coexistence among species²³. Natural cycles in hare populations provide an opportunity to investigate the influence of short-term food abundance and scarcity on the habitat overlap of sympatric predators. We used camera traps to investigate the habitat overlap of lynx and sympatric mesopredators during two contrasting periods of hare abundance. We related co-occurring images of pairs of predators (lynx, coyote, fisher, wolverine) within a camera grid cell to spatial and temporal variation in habitat, prey, and forest disturbance. Given optimal foraging theory and the relatively short interval between changes in prey abundance, we predicted that habitat overlap between pairs of predators would increase during a low in hare abundance. Increasing habitat overlap would be a function of Canada lynx and other mesopredators hunting a broader range of prey in a wider range of habitats (i.e., niche expansion). These results provide new insights on ecologically similar predators that must adjust their use of habitats in response to spatially and temporally dynamic prey.

Materials and methods

Study area

The research was conducted in and adjacent to the John Prince Research Forest (JPRF) in north-central BC, Canada, which is co-managed by the University of Northern British Columbia, Binche Whut’en, and Tl’azt’en First Nations (Fig. 1). The study area is ~ 390 km² and characterized by rolling terrain with low mountains (700 m to 1500 m above sea level). The region represents the northern extent of contiguous Douglas-fir (Pseudotsuga menziesii var. glauca) forests in the interior of BC and is dominated by the Sub-Boreal Spruce biogeoclimatic zone²⁴. The study area is the focus of a long-term monitoring program investigating the influences of climate and landscape change.

Fig. 1.

Canada lynx study area with camera locations, streams and lakes, and canopy cover (3–10 m) in John Prince Research Forest, British Columbia, Canada.

The study area has experienced a wide variety of logging activities over the past 75 years and contains a mosaic of old and young forest (continuum from new harvest to old growth > 250 years old) with interspersed deciduous stands. From 2001 to 2017, the timber harvest in central and southern BC was increased to salvage dead and dying lodgepole pine (Pinus contorta) trees affected by a mountain pine beetle (Dendroctonus ponderosae; MPB) epidemic. This resulted in fundamental changes to the age distribution of forests across large portions of the province²⁵. The JPRF harvests approximately 20,000 m³ on an annual basis. More intensive industrial forest harvest occurred in the study area outside the Research Forest boundaries in the 15 years before this study (2000–2015). Although salvage harvest continued, relatively fewer stands were actively harvested in between the two data collection periods (e.g. 2015–2016 to 2020–2022; Li et al., in review).

American marten (Martes americana), Canada lynx, short-tailed weasel (Mustela erminea), American mink (Neogale vison), river otter (Lontra canadensis), fisher, wolverine, coyote, and red fox (Vulpes vulpes) are small- to medium-size carnivores present throughout the study area. Grizzly and black bears, wolves, and cougars are large carnivores in the study area. However, both bear species hibernate during the study period, and wolves and cougars occur at very low densities (JPRF, unpublished data). American beaver (Castor canadensis), muskrat (Ondatra zibethicus), snowshoe hare, red squirrels, flying squirrels (Glaucomys sabrinus), ruffed grouse (Bonasa umbellus), deer mice (Peromyscus maniculatus), and voles (Clethrionomys gapperi, Myodes spp. and Microtus spp.) are some of the most common prey species. There has not been any significant commercial trapping in the study area for almost two decades (JPRF, unpublished data).

Field data collection

We deployed 66 camera traps (Bushnell Trophy Trail Cameras models 119467 and 11947; Bushnell Outdoor Products, Missouri, USA) on a hexagonal grid (2.5 km apart; 5.41 km²) from January–March in 2015 (23 Jan–03 Apr) and 2016 (01 Feb–10 Apr) for a total of 20 weeks. We used camera data from 01 February–11 April in both 2020 and 2021. The hexagonal grid was randomly placed on the study area and cameras set near the center of each 5.41-km² cell. In 2020–2021, trail cameras (Browning Dark Ops HD Pro Trail Cameras model BTC-6HDP; Browning, Utah, USA) were set at the same sites as 2015–2016. Both camera types (Bushnell and Browning) had a 0.2s trigger speed.

Although some camera traps were set near roads and trails, none were set directly along these linear features. We used a consistent distance between camera traps and scent posts as well as camera trap height on a tree to standardize detection distance, camera angle, and field of view. At each site, a scent post was established 2.5–3 m directly in front of a camera between 0.5 and 1 m above the ground. Each scent post consisted of a small diameter log (< 15 cm) secured with one end above the ground (45 cm) and pointed directly at the camera. A local commercial lure containing beaver castor and catnip oil was placed at the end of the log. A small piece of beaver meat (~ 5 cm diameter) was hung by wire directly above the end of the log (~ 60 cm) to serve as an additional attractant. The use of bait and lure in our study was designed as a small, short-distance attractant to ensure animals near cameras were detected. We used a relatively weak scent (beaver castor and catnip oil) set low to the ground. Bait was small and consumed quickly in a single visit by most carnivores. For this reason, bait primarily served as an additional scent lure before and after consumption that encouraged animals already in the vicinity of the site to move into camera view²⁶.

Cameras were checked and the lure and bait were added every two weeks. In 2015–2016, cameras were set to take 30 s of video with a 1 s delay between video-recordings. In 2020–2021, cameras were set to take 10 s of video with a 1 s delay between video-recordings. These schedules allowed for nearly continuous recording of the time an animal was in view.

Although the exact timing of a peak and trough in hare populations was uncertain, the 2015–2016 and 2020–2021 data sets represented a time of high and low hare abundance, respectively. Occurrence rates and abundance estimates (N-mixture models) of snowshoe hares at camera traps decreased 73% and 42% between the two winter periods, respectively²⁶. This trend was supported by a 40% decrease in hare densities measured at pellet plots established 3 years after the estimated hare peak (2018–2021; John Prince Research Forest, unpublished data).

Habitat covariates

The study area had a forest inventory derived from high-density light detection and ranging (LiDAR; 8–10 pulses/m²) data obtained in August and September 2015. Little forest harvesting and no forest fires occurred between periods, so we assumed that the 2015 LiDAR data was representative for both study periods. Independent covariates were represented by three broad categories that included forest cover and structure, disturbance, and prey abundance (Table 1). Cover and structure, indices of forested habitat, and prey abundance are common ecological factors that can influence the distribution of Canada lynx and other mesopredators e.g^27,28. Landscape disturbance can also influence the distribution of Canada lynx e.g^18,29.

Table 1.

Variables used in the development of habitat models (multinomial logistic regression) for Canada Lynx and sympatric mesopredators (wolverine, coyote, fisher) at remote camera sites in central British Columbia, Canada, 2015–16 and 2020–21.

Parameter	Description	Variable type
cover (0–3 m)	Average canopy cover 0–3 m (LiDAR, 50-m radius or stand scale)	Continuous
cover (3–10 m)	Average canopy cover 3–10 m (LiDAR, 50-m radius or stand scale)	Continuous
cover (10 + m)	Average canopy cover > 10 m (LiDAR, 50-m radius or stand scale)	Continuous
canopy height	Average canopy height (LiDAR, 50-m radius)	Continuous
coarse woody debris	Volume coarse woody debris (field data; 3, 50-m transects)	Continuous
conf/decid	Count of trees; conifer > 75% conifer, mixed/decid ≤ 75% conifer	Categorical
dist.riparian	Distance to stream and lake edge (m)	Continuous
dist.edge	Distance to forest stand age (m)	Continuous
prop. forest age (< 20 yrs)	Proportion of area (1000-m radius) with recent cutblocks (< 20 yrs)	Continuous
edge density	Edge density (1000-m radius)	Continuous
hare	Occurrence rate per site for hare (# days detected/survey days)	Continuous
red squirrel	Occurrence rate per site for squirrel (# days detected/survey days)	Continuous

Variables representing forest cover and structure from LiDAR data included canopy closure at 3 vertical layers (bottom [0–3 m], mid [3–10 m], and top [> 10 m]) and canopy height measured within a 50-m radius of camera trap locations and at the stand scale²⁶. The 50-m distance was designed to measure habitat at the camera trap location but also represent the characteristics of the larger forest stand. In addition, we also explored the use of cover covariates calculated across the entire stand that contained the camera trap. We ran all models at both the 50-m and stand polygon scale. The same models containing cover measured at the 50-m scale consistently ranked higher than models measured at the stand polygon scale. Based on this comparison, we used the 50-m buffer for our analyses but also reported a model with cover measured at the stand scale. Because LiDAR data were measured during the leaf-on season, we used a coarse correction factor for the winter based on cover data collected using the digital Canopeo Application (www.canopeo.com; Alamo Software Foundation) and a literature search^30,31. Measurements of canopy cover were taken in 4-cardinal directions at 3 m from the camera location and averaged during the leaf-on and leaf-off seasons. We applied a correction factor of −25% and − 52% change to the leaf-off season for sites with few (26–50%) and a large (51–100%) proportion of deciduous trees, respectively.

We used field measurements to assess tree type (conifer leading > 75% conifer or mixed/deciduous ≤ 75% conifer) and volume of coarse woody debris. Field data for tree type were based on a count of trees (> 12.5 cm dbh) within 4 plots (11.28 m at camera location and three 3.99-m plots at the end of 50-m transects). Volume of coarse woody debris (> 7.5 cm dbh) was measured along three 50-m transects spaced equally and radiating out from the camera location. We used GIS data to measure the distance of each camera to riparian areas, and forest edge, as well as proportion of area (1000-m radius) containing recent cutblocks (< 20 years in age), and edge density. Forest disturbance varied little across camera traps at the 50-m scale²⁶. For this reason, we measured disturbance variables (edge density, proportion of buffer with cut blocks < 20 years old) to a distance of 1000 m from the camera. We quantified the relative abundance of hare and red squirrel as the proportion of days that an animal was recorded at a camera within a monitoring period (# of days with a prey occurrence/active camera survey days). Red squirrels and hares, generally in North America as well as specifically in our study area, display contrasting habitat use. Snowshoe hare are strongly associated with regenerating forests characterised by dense understory e.g^32,33. Although red squirrels will occupy young or recently disturbed forest stands, they generally occur at higher densities in mature forest, particularly conifer-dominated stands that provide protection from predators, necessary microclimates for food caching (middens), and greater food availability^34,35.

Data analysis

We used multinomial logistic regression to investigate habitat characteristics that influenced the habitat overlap of lynx and sympatric mesopredators (fisher, wolverine, coyote) during winter. We considered the application of occupancy-based co-occurrence models³⁶, but that approach was unsuitable given the survey effort and timing (continuous monitoring throughout season of interest), camera spacing, and the use of bait and lure that were designed to provide a high likelihood of detection^37–39. In addition, home ranges of all mesopredators in this study were large enough to encompass multiple camera stations, violating the assumption of spatial independence for occupancy models^36,40,41. Occupancy analyses are most appropriate for closed sampling units where a true detection probability can be measured naively or relative to some set of environmental or temporal covariates. Most camera studies, including ours, cannot meet the assumption of closure leading to nebulous interpretations of the detection component of the occupancy result that could lead to bias, especially for vagile species^42–45. Multinomial logistic regression can be used when the dependent variable is more than two unranked categories. In these models, all outcomes were compared to a base category (i.e., 0).

We used two measures of activity at a camera trap to assess habitat overlap among lynx and other sympatric mesopredators. First, we used a single occurrence approach where we divided each winter into two ecologically meaningful time periods, late winter and mid-winter, centered on the peak in lynx breeding activity⁴⁶. Longer days, warmer temperatures, and changing snow conditions also differentiate late winter from mid-winter. This resulted in 4 survey sessions (two sessions for each winter) for each cyclic period (hare high and low abundance). The response variable for each session was the occurrence (1) or non-occurrence (0) of a lynx at a camera trap regardless of the number of occurrences during a survey session.

Second, we used a measure of lynx prevalence at camera traps, no/low versus high/moderate prevalence, as a relative measure of lynx activity during each ecological period (high vs. low hare abundance). For the dependent variable in the prevalence models, we first summed all daily occurrences (independent visit = 24 h) at a site during each biological period (winter low hare abundance, winter high hare abundance). We then calculated the occurrence rate (#occurrence days/#operational camera days) for each site. We used the 50th percentile of the occurrence rate to identify camera sites that differed according to two categories in relative use: no/low (0) and high/moderate (1) use. For our analyses, the binary response variable provided an ecologically plausible measure of lynx prevalence and habitat overlap likely more representative of common and widespread species. For the winter low season, a “0” represented an occurrence rate of < 1.4 lynx occurrences/site/day. For winter high, a “0” equated to < 2.9 occurrences/site/day. For all other sympatric carnivores, the 50th percentile reflected a true binary outcome (e.g., 0 = 0 fisher occurrences/site/day).

There were four possible outcomes in each multinomial regression model set (e.g., lynx occurrence or lynx prevalence) according to the occurrence or relative prevalence of lynx and the occurrence of a sympatric mesopredator (fisher, wolverine, coyote) at a camera site (Fig. 2). For example, when modelling the habitat association between lynx and fisher, the base outcome (0) represented sites with the no (occurrence models) or relatively low use by lynx (prevalence models) and no fisher. The contrast outcomes included sites with the occurrence or relatively moderate/high use by lynx with no fisher (1), sites with no or low use by lynx and fisher was present (2), and sites with the occurrence or moderate/high use by lynx and fisher was present (3; Fig. 2). We fit models for all pairs of potential competitors of lynx (i.e., lynx-fisher, lynx-wolverine, and lynx-coyote). All two-species models (e.g., lynx-fisher) were fit during two contrasting periods of prey abundance within the hare cycle (hare low and hare high).

Fig. 2.

Example model development and outcomes (multinomial logistic regression) investigating the habitat overlap of Canada lynx and sympatric mesopredators (coyote, fisher, and wolverine) at camera traps during two contrasting periods of prey abundance in central British Columbia, Canada, 2015–2016 (high hare) and 2020–2021 (low hare).

We used an Information Theoretic Model Comparison approach to assess a set of 12 multinomial regression models for each model set (e.g., lynx occurrence or lynx prevalence model sets) to explain habitat associations and differentiations between lynx and other sympatric mesopredators (⁴⁷; Table 2). We used the difference in Akaike’s information criterion for small sample sizes (ΔAIC_c) and Akaike weights (AIC_cw) to rank and compare models. A model with a ΔAIC_c < 2 was considered to be equivalent to the model with the minimum score (47). When models had ΔAIC_c values that were nearly equivalent, we selected the model with the fewest parameters.

Table 2.

A priori candidate models (multinomial logistic regression) representing habitat overlap of Canada Lynx and sympatric mesopredators (wolverine, coyote, fisher) at remote camera sites in central British Columbia, Canada, 2015–2016 and 2020–2021.

Model Name	Model	K
Canopy Cover	cover (0–3 m) + cover (3–10 m) + cover (10 + m) + conf/decid	5
Riparian + Cover	dist. rip. + cover (0–3 m) + cover (3–10 m) + cover (10 + m)	5
Canopy Height + Ground cover	canopy height + cover (0–3 m) + cwd	4
Disturbance	prop. forest age (< 20 yrs) + edge density	3
Disturbance + Edge	prop. forest age (< 20 yrs) + dist. edge	3
Edge + Cover	dist. edge + cover (0–3 m) + cover (3–10 m) + cover (10 + m)	5
Stand cover (SC)	SC (0–3 m) + SC (3–10 m) + SC (10 + m) + conf/decid	5
Prey	hare + squirrel	3
Cover + Prey	cover (0–3 m) + cover (3–10 m) + cover (10 + m) + hare + squirrel	6
Prey + Disturbance	hare + squirrel + prop. forest age (< 20 yrs) + edge density	5
Disturbance + cover	cover (0–3 m) + cover (3–10 m) + cover (10 + m) + prop. forest age (< 20 yrs) + edge density	6
null	no independent covariates	1

We used the receiver operating characteristic and resulting area under the curve (AUC) to assess the predictive ability of each of the three categorical outcomes within the best multinomial regression model (e.g., lynx only, fisher only, fisher–lynx)⁴⁸. We used a onefold cross validation routine to withhold each record sequentially from the model building process and then calculated the independent probability of that withheld record being a species occurrence. We considered a model with an AUC score of 0.7 to 0.9 to be a useful application and a model with a score > 0.9 as highly accurate⁴⁹. We used the AUC scores to compare the relative predictive ability of the three outcomes for each two-species model set.

We used 95% confidence intervals to assess the strength of effect of each predictor covariate on the dependent variable. We used tolerance scores to assess variables within each model for excessive collinearity⁵⁰. All data analyses were performed using Stata (version 17.0; Statacorp, College Station, Texas).

Results

Camera trap occurrences

In 2015–2016 and 2020–2021, cameras functioned properly an average of 96.0% (SE = 0.9%; 8,866 camera-days) and 96.8% of potential camera days (SE = 0.7%; 8,947 camera-days), respectively. There was a 76% (732 to 176 occurrences) decrease in lynx occurrence rates between 2015–2016 and 2020–2021 during the winter (Table 3). There was a similar steep decrease in snowshoe hares (72.5%; 708 to 195 occurrences). Both fisher and wolverine occurrence rates increased from 2015–2016 to 2020–2021 (Table 3). Coyote occurrences were similar during the two hare abundance periods.

Table 3.

Total occurrences and occurrences per 100 camera-days (rate) for sympatric mesopredators and prey at camera traps during two contrasting periods of hare abundance in central British Columbia, Canada, 2015–2016 (high) and 2020–2021 (low).

2015–2016 (Hare High)			2020–2021 (Hare Low)
Species	Total	Rate		Total	Rate
Lynx 2015	303	6.89	Lynx 2020	105	2.36
Lynx 2016	429	9.60	Lynx 2021	71	1.58
Lynx Total	732	8.26	Lynx Total	176	1.97
Wolverine 2015	9	0.20	Wolverine 2020	25	0.56
Wolverine 2016	6	0.13	Wolverine 2021	70	1.55
Wolverine Total	15	0.17	Wolverine Total	95	1.06
Fisher 2015	1	0.02	Fisher 2020	29	0.65
Fisher 2016	1	0.02	Fisher 2021	23	0.51
Fisher Total	2	0.02	Fisher Total	52	0.58
Coyote 2015	8	0.18	Coyote 2020	13	0.29
Coyote 2016	18	0.40	Coyote 2021	9	0.20
Coyote Total	26	0.29	Coyote Total	22	0.25
Hare 2015	396	9.01	Hare 2020	129	2.90
Hare 2016	312	6.98	Hare 2021	66	1.47
Hare Total	708	7.99	Hare Total	195	2.18
Squirrel 2015	516	11.74	Squirrel 2020	480	10.81
Squirrel 2016	1288	28.82	Squirrel 2021	2111	46.86
Squirrel Total	1804	20.35	Squirrel Total	2591	28.96

Habitat overlap

In general, habitat overlap between lynx and sympatric mesopredators increased during a period of low hare abundance (2020–2021). This relationship was most pronounced for wolverine-lynx and coyote-lynx occurrence models and for wolverine-lynx and fisher-lynx prevalence models. For differences in AIC_c scores and weights, null models ranked the lowest for all species and time periods (ΔAIC_c >29.6, AIC_cw = < 0.001).

2015–2016 winter models

No occurrence or prevalence models predicting the habitat overlap of lynx and sympatric mesopredators had useful predictive accuracy during the hare high period (2015–2016; Tables 4 and 5). For the 2015–2016 winter season, the canopy cover and cover + disturbance models for both occurrence and prevalence best explained shared habitat use of lynx and coyote (ΔAIC_c < 2; Tables 4 and 5). The prevalence model best differentiated sites with relatively high use by lynx and no use by coyotes (AUC > 0.77). There was a clear top model for both occurrence and prevalence model sets (cover + disturbance) for the lynx-wolverine comparison (AIC_cw > 0.909; Tables 4 and 5). With only two total occurrences (one in each year), we were unable to run models for fisher during this survey period.

Table 4.

AIC_c scores and weights (w) for multinomial logistic regression models (Top 95%AIC_c weights) representing habitat overlap of Canada Lynx (occurrence response variable) and sympatric mesopredators (wolverine, fisher, and coyote) at camera sites in central British Columbia, Canada, 2015–2016 and 2020–2021. Area under the curve (AUC) and standard error (SE) for the receiver operating characteristic represents the predictive accuracy of each model. Bolded value = AUC ≥ 0.7.

HARE LOW WINTER
Model	K	AICc	ΔAICc	w	AUC1* (SE)	AUC2* (SE)	AUC3* (SE)
COYOTE - LYNX
Riparian + Cover	5	384.5	0	0.940	0.74 (0.03)	0.19 (0.06)	0.70 (0.12)
Cover and Prey	6	391.4	6.9	0.030	0.72 (0.03)	0.44 (0.06)	0.59 (0.14)
WOLVERINE - LYNX
Cover and Disturbance	6	540.1	0.0	0.898	0.72 (0.04)	0.75 (0.05)	0.82 (0.06)
Edge + Cover	5	546.0	5.9	0.048	0.70 (0.04)	0.73 (0.05)	0.83 (0.05)
Riparian + Cover	5	546.0	5.9	0.046	0.77 (0.04)	0.70 (0.05)	0.79 (0.06)
FISHER - LYNX
Riparian + Cover	5	473.4	0.0	0.905	0.74 (0.03)	0.63 (0.07)	0.23 (0.06)
Cover and Disturbance	6	479.2	5.8	0.051	0.72 (0.03)	0.62 (0.07)	0.52 (0.13)
HARE HIGH WINTER
Model	K	AIC c	ΔAIC c	w	AUC1 (SE)	AUC2 (SE)	AUC3 (SE)
COYOTE - LYNX
Cover and Disturbance	6	449.5	0.0	0.459	0.69 (0.03)	0.10 (0.10)	0.42 (0.10)
Site Cover	5	449.9	0.4	0.390	0.69 (0.03)	0.30 (0.25)	0.29 (0.06)
Prey and Disturbance	6	452.9	3.4	0.087	0.71 (0.03)	0.25 (0.21)	0.56 (0.08)
Stand Cover	5	455.1	5.6	0.029	0.68 (0.03)	0.34 (0.18)	0.59 (0.08)
WOLVERINE - LYNX
Cover and Disturbance	6	415.7	0.0	0.909	0.70 (0.03)	0.68 (0.16)	0.25 (0.09)
Prey and Disturbance	5	421.8	6.1	0.043	0.71 (0.03)	0.50 (0.17)	0.33 (0.09)

*AUC1 = lynx, AUC2 = sympatric mesopredator, AUC3 = lynx + mesopredator.

Table 5.

AIC_c scores and weights (w) for multinomial logistic regression models (Top 95%AIC_c weights) representing habitat overlap of Canada Lynx (prevalence response variable) and sympatric mesopredators (wolverine, fisher, and coyote) at camera sites in central British Columbia, Canada, 2015–2016 and 2020–2021. Area under the curve (AUC) and standard error (SE) for the receiver operating characteristic represents the predictive accuracy of each model. Bolded value = AUC ≥ 0.7.

HARE LOW WINTER
Model	K	AICc	ΔAICc	w	AUC1* (SE)	AUC2* (SE)	AUC3* (SE)
COYOTE - LYNX
Riparian + Cover	5	110.7	0.0	0.758	0.86 (0.05)	0.46 (0.18)	0.35 (0.15)
Cover and Disturbance	6	113.2	2.5	0.214	0.84 (0.05)	0.31 (0.13)	0.48 (0.17)
WOLVERINE - LYNX
Riparian + Cover	5	140.6	0.0	0.470	0.76 (0.08)	0.33 (0.09)	0.89 (0.06)
Cover and Disturbance	6	140.9	0.3	0.407	0.72 (0.09)	0.39 (0.10)	0.86 (0.07)
Canopy Height + Ground Cover	4	145.5	4.9	0.040	0.65 (0.10)	0.61 (0.10)	0.76 (0.08)
Edge + Cover	5	146.1	5.5	0.030	0.64 (0.10)	0.41 (0.10)	0.84 (0.07)
Site Cover	5	146.2	5.6	0.028	0.67 (0.10)	0.41 (0.10)	0.82 (0.08)
FISHER - LYNX
Riparian + Cover	5	145.2	0.0	0.408	0.79 (0.07)	0.36 (0.10)	0.82 (0.09)
Canopy Height + Ground Cover	4	145.5	0.3	0.344	0.68 (0.09)	0.62 (0.10)	0.74 (0.09)
Cover and Disturbance	6	147.4	2.2	0.135	0.72 (0.08)	0.47 (0.10)	0.79 (0.09)
Stand Cover	5	150.0	4.8	0.036	0.62 (0.09)	0.59 (0.10)	0.91 (0.05)
Site Cover	5	150.4	5.2	0.030	0.71 (0.08)	0.44 (0.10)	0.78 (0.09)
HARE HIGH WINTER
Model	K	AIC c	ΔAIC c	w	AUC1 (SE)	AUC2 (SE)	AUC3 (SE)
COYOTE - LYNX
Site Cover	5	138.5	0.0	0.5471	0.77 (0.07)	0.39 (0.12)	0.63 (0.11)
Cover and Disturbance	6	140.1	1.6	0.2511	0.75 (0.08)	0.48 (0.16)	0.60 (0.13)
Riparian + Cover	5	140.8	2.3	0.175	0.73 (0.07)	0.56 (0.16)	0.69 (0.09)
WOLVERINE - LYNX
Cover and Disturbance	6	103.8	0.0	0.984	0.88 (0.05)	0.48 (0.26)	0.19 (0.11)

*AUC1 = low lynx, AUC2 = sympatric mesopredator, AUC3 = low lynx + mesopredator.

In general, for the lynx-coyote and lynx-wolverine models, the occurrence and prevalence of lynx was associated with camera sites with more mid-strata cover (3–10 m), less top-strata cover (10 m +), and less recent harvest < 20 years old (Table 6). Those same forest conditions were related to the habitat use of lynx and coyotes when they occurred at the same camera sites during the 2015–2016 winter. There was no relationship between covariates in the top-ranked model and the shared habitat use of lynx and wolverine. The occurrence of wolverine at sites with no or low prevalence of use by lynx was associated with less mid-strata cover (3–10 m), more top-strata cover (10 m +), and more recent harvest < 20 years old (Table 6).

Table 6.

Coefficients (SE) for top-ranked models (occurrence and prevalence response variable) illustrating habitat overlap by Canada Lynx and sympatric mesopredators. Bolded values represent coefficients with 95% confidence intervals that did not overlap zero.

	Covariate	Lynx	Coyote	Coyote– lynx	Lynx	Wolverine	Wolverine– lynx	Lynx	Fisher	Fisher– lynx
OCCURRENCE
HARE HIGH	dist_riparian
	cover (0–3 m)	0.17 (0.23)	−0.91 (0.57)	−0.45 (0.38)	0.12 (0.23)	−0.44 (0.62)	−0.10 (0.47)
	cover (3–10 m)	0.63 (0.18)	0.77 (0.69)	0.52 (0.42)	0.57 (0.19)	−1.53 (0.56)	0.34 (0.39)
	cover (10 + m)	−0.37 (0.23)	−0.70 (0.40)	−0.59 (0.49)	−0.41 (0.23)	−0.73 (0.77)	−0.08 (0.34)
	age < 20	−0.32 (0.09)	−0.72 (0.40)	−0.38 (0.17)	−0.34 (0.09)	0.36 (0.15)	0.12 (0.23)
	edge density	0.26 (0.17)	−0.02 (0.77)	0.47 (0.36)	0.30 (0.17)	0.38 (0.32)	0.10 (0.50)
HARE LOW	dist_riparian	−0.56 (0.25)	−0.11 (0.43)	−1.5 (0.91)				−0.64 (0.21)	0.03 (0.20)	−0.41 (0.86)
	cover (0–3 m)	−0.42 (0.24)	−0.36 (0.50)	−0.55 (0.93)	−0.24 (0.27)	−0.23 (0.31)	−0.80 (0.41)	−0.27 (0.25)	1.03 (0.43)	−0.09 (0.52)
	cover (3–10 m)	1.02 (0.21)	−0.21 (0.53)	−0.87 (0.65)	1.05 (0.28)	1.10 (0.27)	1.33 (0.35)	0.91 (0.21)	−0.27 (0.27)	0.36 (0.37)
	cover (10 + m)	−1.13 (0.26)	−0.25 (0.43)	−1.14 (0.89)	−1.16 (0.31)	−0.74 (0.34)	−2.06 (0.50)	−1.08 (0.26)	0.86 (0.42)	−0.46 (0.67)
	age < 20				−0.19 (0.11)	−0.48 (0.17)	−0.56 (0.17)
	edge density				0.23 (0.21)	0.45 (0.22)	0.44 (0.26)
PREVALENCE
HARE HIGH	conifer	−1.41 (0.59)	−0.35 (0.58)	−1.77 (0.92)
	cover (0–3 m)	0.49 (0.58)	−0.63 (0.55)	−0.84 (0.54)	0.34 (0.59)	−0.67 (1.01)	−0.46 (0.65)
	cover (3–10 m)	2.33 (0.78)	0.72 (0.81)	2.98 (1.03)	1.27 (0.58)	−1.32 (0.91)	0.74 (0.54)
	cover (10 + m)	−1.30 (0.68)	0.3 (0.78)	−2.32 (0.82)	−1.81 (0.77)	−1.54 (1.32)	−0.79 (0.54)
	age < 20				−1.17 (0.46 )	0.47 (0.37)	0.05 (0.37)
	edge density				0.54 (0.38)	0.26 (0.53)	0.055 (0.61)
HARE LOW	dist_riparian	−1.54 (0.54)	−0.40 (0.43)	−1.80 (1.47)	−1.42 (0.56)	−0.40 (0.39)	−1.84 (0.67)	−1.16 (0.52)	0.09 (0.30)	−2.07 (1.11)
	cover (0–3 m)	−0.34 (0.77)	−0.088 (0.87)	−1.63 (1.10)	−1.34 (0.59)	−1.10 (0.62)	−0.87 (0.91)	−0.71 (0.56)	0.63 (0.61)	0.30 (1.00)
	cover (3–10 m)	2.69 (0.58)	−0.11 (0.76)	0.44 (0.82)	2.25 (0.57)	1.17 (0.58)	3.30 (0.70)	2.01 (0.56)	−0.29 (0.55)	2.12 (0.61)
	cover (10 + m)	−2.29 (0.77)	0.06 (0.77)	−2.02 (1.29)	−2.26 (0.60)	−0.97 (0.61)	−3.18 (1.06)	−2.03 (0.60)	0.53 (0.64)	−1.93 (1.02)

2020–2021 winter models

The top occurrence models for both lynx-wolverine and lynx-coyote had useful predictive accuracy (AUC > 0.70; Table 4) for the hare low period. In 2020–2021, the highest ranked prevalence model for all species contained covariates for distance to riparian edge and the LiDAR measures for mid- and top-strata canopy cover (Table 5). However, the second ranked model for both lynx-fisher and lynx-wolverine were nearly equivalent (ΔAIC_c < 2; Table 5). The occurrence or prevalence of lynx was positively associated with mid-strata cover (3–10 m) and cameras that were closer to riparian edge (Table 6). As with the 2015–2016 results, lynx were negatively associated with top-strata cover (10 m +). For lynx prevalence, there was shared habitat use by lynx and wolverine and fisher, but not coyotes (Table 5). For occurrence models, habitat use overlapped for lynx and coyote as well as lynx and wolverine (Table 4). Lynx and wolverine, fisher, and coyote overlapped in their use of camera sites with more stand cover from 3 to 10 m and avoided sites with greater amounts of cover that was > 10 m in the canopy (i.e., older or taller forest; Table 6). Riparian features were important for explaining overlap in habitat use between the mesopredators and lynx.

Discussion

Short-term fluctuations in prey abundance may drive fluctuations in habitat overlap and potential niche expansion and competitive interactions. Consistent with optimal foraging theory (i.e., niche expansion during prey scarcity^1,2), we observed differences in the habitat use and overlap of sympatric carnivores between the two hare abundance periods. Overlap in habitats used by lynx and sympatric mesopredators increased during the 2020–2021 winter, a time of prey scarcity. In general, habitat overlap between lynx and other mesopredators occurred in seral forests or riparian areas characterized by greater mid-level cover and less top-level cover, indicating increased shared use of these habitats during the hare low. During the hare high, lynx used similar habitats, but exhibited minimal overlap with other mesopredators.

This is the first study to use camera traps to document changes in habitat overlap between Canada lynx and sympatric carnivore species during the snowshoe hare cycle. A significant change in the mesopredator community occurred between two time periods separated by only four years. Lynx occurrences mirrored the decrease in hares, while the combined occurrences of wolverine, fisher, and coyote increased during the low in hare abundance likely contributing to the observed overlap between lynx and these mesopredators. Although a number of niche axes are related to resource competition, shared use of habitats among species is one key element in assessing the potential for niche overlap^51,52. Temporal variation in cyclic resource availability is an important component of niche overlap⁵² that was also part of our study.

We concentrated on hares and squirrels as focal prey species for two reasons. Snowshoe hare abundance is a well-established bottom-up driver of predator population dynamics in northern forest ecosystems^8,9,35, and during periods of low hare abundance, red squirrels commonly represent the second most important prey item in lynx diets^11–13. In addition, the relative occurrence of both species could be quantified spatially and temporally using camera traps. Although hare abundance was the only prey variable known to change significantly between sampling periods, we acknowledge that unmeasured factors, including fluctuations in mice and vole populations or the availability of ungulate carrion, may also have influenced mesopredator abundance.

The potential influence of bait on movement patterns and habitat use is an important consideration when designing, implementing, and interpreting camera trap studies^40,53,54. Although lure and bait can decrease the detection of some prey species, it often increases or has no effect on the detection of many carnivore species^55–57. A study of fisher found that any effects of bait on animal movements were eclipsed by habitat heterogeneity (Stewart et al. 2019), and that bait may increase detection probability improving ecological inferences. In another analysis in our study area, we found that time since bait and lure were added to a site was not a significant influence on the detection probability of lynx²⁶. Bait and lure may have influenced detection of animals in our study, but it is unlikely to compromise inferences and may have improved detection of focal carnivore species.

The extent of animal home ranges is another important consideration when interpreting data from camera trap studies. For example, when an individual home range overlaps several camera trap locations, occurrences could represent the movements of a single individual or multiple individuals. However, based on non-systematic data collected during the low in hare abundance (unique markings of wolverine and fisher), we know that a minimum of 4 wolverine and 6 fisher individuals were detected at camera trap locations during the winter. Given these minimum counts and a study area population estimate of 11 lynx²⁶, occurrence and habitat results were representative of multiple individuals.

Vegetation cover is an important element of habitat for mesopredators and their prey across the boreal forest⁵⁸. Similarly, we found that cover was important, but the direction and magnitude of effect varied according to canopy strata and the solitary use of camera sites by lynx or shared habitat use with coyote, wolverine, and fisher. Typically, mid- and top-level cover was most important, but ground cover (0–3 m) was represented in several lower-ranked models. Past studies of habitat use of lynx and hare did not have multi-strata measures of vegetation cover, potentially obscuring the relationships that were revealed by the LiDAR data used for our research [e.g^59–61. Prey models consistently ranked lower than cover models, but hares were often a positive influence on the use of sites by lynx. Although disturbance models generally ranked lower than cover models, the model that combined cover and disturbance covariates ranked first or second on multiple occasions. Generally, a greater proportion of cutblocks (< 20 years old) across a camera cell negatively influenced the use of habitat by the mesopredators in this study.

Riparian habitat was important to lynx, fisher and wolverine, but only during the low in hare abundance. Riparian edges can be positively associated with species diversity and abundance including small mammals^62–64. Riparian areas likely provide a dual function for mesopredators, providing not only hunting opportunities, but also serving as travel corridors that provide cover from competitors and predators^63–65. Riparian forest often represents the only mature trees following clearcut timber harvest⁶⁶. Thus, loss of upland forest through human or natural disturbance, as is occurring across much of central BC, could lead to more concentrated use of riparian forest and the potential for niche compression and competition among mesopredators.

Hare abundance was the only major ecological condition that changed between time periods. There was relatively little forest harvesting and no fire activity in or adjacent to the study area during the time of data collection. Also, the study area has not had any significant commercial trapping for almost two decades. Deep snow should give lynx a competitive advantage over other carnivores (except wolverine) and there was a difference in snow depth in February/March (Fort St James, B.C.; https://climate.weather.gc.ca/historical_data) between 2015–2016 ( = 23.1 cm, SE = 1.3) and 2020–2021 ( = 35.5 cm, SE = 1.3). However, snow was deeper during the hare low when there were less lynx and more sympatric carnivores and is an unlikely explanatory factor for habitat overlap in our study.

Short-term population cycles can influence fluctuations in habitat overlap as well as the identity and roles of dominant and sub-dominant species. A given adaptation may not provide a competitive advantage to a predator during all phases of a prey’s population cycle. Our research revealed two potential outcomes for lynx and competitive mesopredators. During the hare high period, lynx had a numerical and functional advantage because of their specialized ability to hunt hare in deep snow⁶⁷. Those advantages would be less during the hare low period when alternative prey become more important. Both camera trap and capture data revealed that lynx were in poor body condition and without kittens during that period (²⁶, John Prince Research Forest, Unpub. Data). Thus, the lessened advantages when hunting alternative prey, had apparent costs to the fitness of individual lynx and overall population productivity. During the hare low, the combination of both intra- and inter-specific competition may have a relatively stronger influence on lynx populations.

Cyclic fluctuations in prey can alter the abundance and distribution of predator communities^14,68,69. During the hare low, the need to hunt other prey resources may have increased the distribution and abundance of sympatric carnivores and led to increased habitat overlap with lynx in our study area. Similarly, Arctic fox (Vulpes lagopus) co-occurred more with sympatric carnivores when rodent abundance was low⁶⁹. Increased habitat overlap during times of prey scarcity may be a combination of prey and predator abundance as well as changes in the spatial ecology of the predators that are forced to hunt for alternative prey across a wider range of habitats.

The measure of lynx prevalence suggested that habitat overlap increased between lynx and fisher during the hare low (i.e., fisher occurred at sites with greater lynx activity), and between lynx and coyote based on lynx occurrence. For wolverine, the increase in habitat overlap increased during the hare low for measures of lynx occurrence and prevalence. Environmental factors influencing species abundance may differ from those that determine their distribution or occurrence⁷⁰. Although our response variable for relative use was not a measure of abundance, occurrence rates of lynx were strongly correlated to site abundance estimates from N-mixture models in our study area (r= 0.88–0.95²⁶;). These findings suggest that different processes may influence lynx occurrence and relative activity and its influences on habitat overlap with sympatric mesopredators.

In the boreal forest, coyote, lynx, and snowshoe hare often follow a similar pattern of cyclic abundance^14,15, and habitat overlap can be similar between coyote and lynx during both a low and high in the hare cycle¹⁴. The degree of overlap between lynx and coyotes is likely dependent on the availability of non-hare prey¹⁴. Coyotes may switch to hunting voles in years with a reduction in the abundance of hares and increased numbers of small mammals¹³. Prey switching by coyotes may partially explain the similar coyote occurrence rates between 2015–2016 and 2020–2021 when we observed relatively fewer lynx and hare during the winter.

There are several proposed mechanisms to explain the cycling dynamics of lynx and hare, including climate⁷¹, weather⁶, forest succession⁶, food availability and quality (^7[,72[,73), and predation^6,7,15. Although many of these factors may work synergistically, predation is likely the most important driver of the hare-lynx cycle⁷. The pattern and scale of the hare cycle has long been a function of large, undisturbed, and contiguous forests and it is unclear how a fragmented landscape, like in many parts of central BC, may affect the cyclic dynamics and interactions of hare, lynx, and other carnivores in the long-term. Our research highlights the importance of considering habitat overlap among heterospecifics in disturbed landscapes when assessing, monitoring, and managing cyclic populations.

Wiens (1977⁵, 1993⁷⁴) suggested that unless populations are in a resource-based equilibrium, short-term studies will only provide snapshots in time that may be challenging to interpret within the broader context of cyclic fluctuations and ecological relationships. Our data spans a 7-year period and represents contrasting ecological conditions. Nonetheless, there is uncertainty in the mechanisms underlying the observed patterns of lynx and their competitors. Clearly, longer term monitoring at large spatial scales is required as has been achieved in a small number of other study areas^35,75–77. Substantial changes in the mesopredator community occurred over a very short time, thus, a frequent monitoring interval is required. Without a basic understanding of co-occurrence and potential competition among sympatric carnivores, questions regarding climate change, forest disturbance, and mesopredator persistence will be challenging to interpret.

Author contributions

S. C., C. J., and D. H. conceived the ideas and designed methodology; S. C. and D. H. collected the data; S. C. led the data analyses and the writing of the manuscript. All authors contributed critically to the drafts and gave final approval for publication.

Data availability

The data that support the findings of this study are openly available in Borealis.ca at https://borealisdata.ca/privateurl.xhtml?token=4f4a2b9f-1080-420f-9d96-d97cc14a206d.

Declarations

Competing interests

The authors declare no competing interests.