Contingent strategies of risk management by snowshoe hares

I searched for consistency in risk management with controlled experiments on snowshoe hares foraging on jack-pine (Pinus banksiana) boughs (Morris 2005) during winter in northern Ontario, Canada. Predation risk for snowshoe hares is more intense in winter than during other seasons (Griffin et al. 2005). Hares forage singly, learn quickly to exploit resource patches, and travel among patches along well-used runways. Hares are a suitable model for studying risk management because they have easily recognized vigilant behaviours (Morris and Vijayan 2018) in response to high mortality from numerous mammalian and avian predators (Krebs et al. 2014) in a stressful landscape of fear (Boonstra et al. 1998; Sheriff et al. 2009). Fear in northern Ontario is induced by a variety of common predators including lynx (Lynx canadensis), coyotes (Canis latrans), wolves (C. lupus), red foxes (Vulpes vulpes fulva), fishers (Martes pennanti), pine martens (M. americana), mink (Neovison vison), and large owls (e.g., Strix nebulosi, S. varia, and Bubo virginianus) whose abundances vary through space and over time.

My design mimics that of Morris and Vijayan (2018) who quantified hare foraging on pine boughs placed in an open field versus thick alder habitats while wildlife cameras recorded the hares’ behaviour. The nutritional quality of jack-pine boughs, a favoured food of hares in winter (Bergeron and Tardif 1988), declines from the farthest tip towards the base (Palo et al. 1992; Hodson et al. 2010). Boughs thus represent a depletable resource patch with diminishing returns to browsers such as snowshoe hares. The diameter of the residual stem at the point of browse yields a reliable and repeatable estimate of the hares’ GUD (Morris 2005; Hodson et al. 2010; Morris and Vijayan 2018). The GUDs, in combination with time-synchronized photos, yield the basic data required to assess risk management and possible trade-offs between time allocation and vigilance.

Snowshoe hares living in 20-year-old cutovers and small overgrown farm fields regularly used several different runways to cross the sharp ecotone along an abandoned 1.5-km long logging road (= open-trail habitat) separating dense 4-m tall alder (Alnus viridis) and red pine (P. resinosa) “shrublands” growing on each side. I selected eight of these runways as experimental stations and placed one pair of identical foraging patches along the runway on each side of the ecotone between habitats (one patch in each habitat at each station, illustrated in Fig. S1). I manipulated risk in four different ways. (i) Distance: I subdivided the eight stations into two equal-sized groups representing four where the farthest distance from the “open” trail and adjoining dense habitat was between 3 and 5 m (near) versus the remaining four where the distance was between 7 and 15 m (far). (ii) Habitat: I evaluated hare foraging (GUDs) and vigilance (images recorded by remote cameras) in each foraging patch. (iii) Cover: I created four different risk treatments in the open-trail habitat by placing equal amounts of natural cover overhead of the patches (top), on the sides parallel to the hare runway (side), on both the side and top (both), or >5 m away from the patch (none). (iv) Moon phase: I assessed all three factors during the full moon (risky) and subsequent new moon (less risky). Snowshoe hares combat reduced winter survival during the full moon (Griffin et al. 2005) by occupying dense and putatively safer habitats (Gilbert and Boutin 1991; Gigliotti and Diefenbach 2018), but not in all studies (e.g., Griffin et al. 2005).

I pounded metal t-bar fence posts through the snow pack and into frozen earth in the open habitat during late January 2018. I located the four posts at the corners of a 1-m square centred at 1.5 m from the ecotone. I strung a 6 mm polypropylene rope from post to post at 1.25-m height and used it to suspend alder branches providing overhead cover. I drove a fifth post into the ground at the ecotone and used it to mount two time-synchronized covert motion-detecting cameras (Reconyx “RapidFire PC90”, Homen, Wisconsin, USA). I placed the cameras back-to-back with equal fields of view to obtain exclusive images of hares in each habitat (Morris and Vijayan 2018).

I cut 160 1.5-m-long terminal alder branches for use as cover treatments (20 branches at each station). I arranged the boughs as treatments by either interweaving them on top of the suspension rope, imbedding the butt ends securely into snow to create a 1-m line of equally spaced branches on opposing sides of the support frame, or by sticking them in the snow between 5 and 10 m distance from the foraging station (for the control none treatment). I placed all 20 boughs on top of the suspension rope for the top treatment, 10 boughs on top and 5 boughs on each side for the both treatment, and 10 boughs on opposing sides for the side treatment. I provide photographs of the none and side treatments in Figs. S2 and S3).

I also cut 200 50-cm-long terminal boughs from 10–15-year-old jack-pine stands (∼20 km from the experimental site) 2 d before initiating each moon-phase treatment (400 boughs total) and stored them in an unheated shed. I measured the diameter at the base of 320 boughs with a vernier caliper (0.01 mm accuracy; mean = 7.375 mm, SD = 0.86) after excluding damaged or abnormally large (>10 mm) or small boughs (<5 mm) and any with cones. I marked each measured bough individually with a small piece of numbered orange flagging tape that I secured to the base with a thumbtack. I randomized the boughs among paired foraging patches and created two foraging patches per station by firmly implanting (5–10 cm) the base of two boughs into snow in each habitat at 1.5-m distance from the cameras. I placed the boughs in the open habitat in the center of the 1-m square treatment plots. Snow completely obscured the flagging-tape labels from foraging hares. I placed the boughs on opposite sides of the hare runway (∼20 cm apart). All patches were in place by early afternoon before hares began to forage. I allowed hares to forage on the boughs overnight and collected the residual boughs well after foraging ceased in the morning of the next day. I replaced the boughs with new ones, took the foraged boughs to my field office, and measured the diameter of each one at the point of browse (again with 0.01 mm accuracy). On some occasions, hares consumed all vegetation distal to a bulge in the stem near the base that had a greater diameter than the bough’s basal diameter. I recoded the cut points of these 13 boughs as equal to the basal diameter. I confirmed that hares were the only browsers consuming the stems by identifying tracks in the snow (that also reveal the presence of predators), characteristic bite angles associated with hare foraging, and by searching for other species in each of the more than 25 000 photographs recorded by the cameras.

I standardized all open stations 1 d before each experimental period with the both treatment, and allowed hares to forage on identical patches with two unmarked boughs in each habitat. I checked to ensure that hares foraged at each station before randomizing the treatments. I collected all of the browsed boughs and replaced them with the individually marked experimental ones. I wore large snowshoes during all fieldwork to minimize compression of the snow pack.

It is possible, given the spatial scale of my experiments, that an individual hare could visit more than one station during a single night of foraging. The potential for this effect to influence results is small because, (i) previous research documented that different stations were exploited simultaneously by different hares (Morris and Vijayan 2018), (ii) hares with individually recognizable phenotypes used only a single station, and (iii) even if a hare did visit more than one station, the effect on GUDs would require that it was the final forager to visit each one, and the effect on vigilance would be devalued by numerous images of other hares.

I set the cameras to record consecutive photographs with a 5 s delay. I viewed each photograph at the end of the experiment and recorded hare images as belonging to one of five classes of mutually exclusive behaviours: alert—head-up posture with ears pointing forward; vigilant—ears pointing forward while standing upright on only the back legs or sitting with an elevated posture and rigid front legs; feeding—clipping pine boughs, head-down searching for pine needles and stems or with food in the forager’s mouth; not-feeding—no clear evidence that the hare had food in its mouth or was actively searching for food in the patch; and partial image—images in which the portion of the hare in the camera’s field of view did not allow an unambiguous assignment to other categories. I calculated a single variable for “vigilance” as the sum of alert plus vigilant images because both revealed highly observant and apprehensive nonfeeding hares. I tallied images with two or more hares separately and scored their respective behaviours. I also counted all images that recorded one hare chasing or jumping over another.

Although my camera data must be highly correlated with the amount of time that hares spend in the foraging patches, they cannot be used easily to calculate QHRs, such as those generated from best-fit harvest curves (Morris 2014) or solution of Holling’s disc equation (Holling 1965; Kotler and Brown 1990; Brown 1999). The number of images of feeding hares should nevertheless approximate the total time allocated to foraging, and the difference in the number of images of foraging hares in paired patches should provide a reliable estimate of differences in foraging time. I used these differences, and the mean diameter of the two boughs at point of browse (GUD), to test the underlying relationships between QHRs and GUDs that emerge from the trade-offs between vigilance and time allocation imbedded in eq. (1) (e.g., if the GUDs in two patches are equal, then the QHR will be lower in the patch with the most images of foraging hares (more time spent foraging for the same net harvest)). All animals were cared for in accordance with the guidelines of the Canadian Council on Animal Care and all methods were approved by Lakehead University’s Animal Care Committee.

I allocated the four cover treatments (top, side, both, none) to the eight stations in a randomized double Latin square that accommodated my equal numbers of stations categorized as near and far. Each station received a different treatment on each of four consecutive days. I initiated the experiment 1 d before the 31 January 2018 full moon and repeated the entire design starting on 13 February 2018 (2 d prior to the subsequent new moon). Mean minimum and maximum temperatures were lower during the full-moon experiment (−27.9 and −16.6 °C, respectively) than during the new-moon experiment (−14.5 and −2.0 °C, respectively; temperatures recorded 35 km away by Environment Canada station ZTB, data downloaded from climate.weather.gc.ca/climate_data/ on 19 March 2018).

I was concerned that snowfall during the first night of the experiment on 30 January 2018 might have biased hare feeding strategies so I extended the experiment 1 d by repeating the 30 January treatment on 3 February 2018. I replicated this 5 d design during the new moon, then excluded the first day of each moon phase from analysis. Doing so yielded identically balanced designs for both treatment and pretreatment foraging during the two moon phases. The repeated measures of GUDs and vigilance across days maximized my ability to infer the direct and interactive effects of habitat, treatments, and moon phase on risk management by snowshoe hares.

I analyzed vigilance with a repeated-measures mixed-effects model (dependent variable = square root of (number of alert plus vigilant images + 0.5); repeated measure = day; fixed effects = habitat, distance, moon phase, and treatment nested within habitat; random intercept = camera (to control for the possible effect of differences in cover that might obscure distant hares); covariate = square root of (total images of hares + 0.5); all fixed effects except treatment and the covariate fully crossed; denominator degrees of freedom (df) adjusted with Satterthwaite’s approximation; post hoc least significant difference (LSD) comparisons among fixed effects). I included all two- and three-way interactions (for all but treatment nested within habitat) in an initial model and used Akaike’s Information Criterion for finite sample sizes (AICc) to choose the most parsimonious of alternative models (lowest AICc, if ΔAICc < 2 I chose the model with fewest parameters, (Burnham and Anderson 2002)). I repeated the analysis and model selection using only data from the open-trail habitat so I could unambiguously assess the main effect of treatment and its potential interactions.

I used two similar repeated-measures mixed-effects models and AICc to assess differences in GUDs (dependent variable = mean cut-point diameter; fixed covariate = square root of (number of alert plus vigilant images + 0.5); random covariate = mean basal diameter of the two boughs; random effect = camera). I did not include an intercept for camera because there should not be a consistent difference in the ability of the cameras to record foraging within resource patches. I used the most parsimonious models from the four mixed-effects analyses as inferential tools to evaluate potential relationships between vigilance and GUDs.

I completed my analyses with paired t tests comparing the number of images with feeding hares between full- and new-moon phases and between near and far stations. I repeated the tests with Morris and Vijayan’s (2018) data on the number of hares foraging on pine boughs in alder and open habitats. I evaluated whether these differences corresponded with the trade-offs predicted by eq. (1).

All analyses were conducted with SPSS 25 and Minitab 18 software. I report statistical comparisons as either marginal means or differences in the marginal mean values ± their respective standard errors. I compared relative effect sizes of significant main effects by calculating the ratio of the variance accounted for by the effect to that of the effect plus its error variance (partial eta², ${\eta }_p^2$ ) from the ANOVA tables (using the worksheet provided by Lakens 2013; 90% confidence intervals from Uanhoro 2017). Although useful for comparing main effects within each model, the ${\eta }_p^2$ values that I report here are “accurate” only for a fully fixed-effects model. They should not be interpreted as general statements on the proportional variances accounted for by the mixed-effects models or those expected in other studies with different designs.

I illustrate the significant results of the most parsimonious models with boxplots of the models’ predicted values. Presentation of raw scores would misinterpret statistical significance because they do not include the marginal effects of other terms in the models.

Although we did not quantify numerous field signs of predators in 2010 (Morris and Vijayan 2018), predation risk was undoubtedly high. Tracks of mammalian predators were common and our wildlife cameras captured multiple images of lynx as well as a red fox. Additional evidence of high predation risk was revealed at two different foraging stations where hares were killed by avian predators immediately following the experiments (Morris and Vijayan 2018). Comparable signs of predators were virtually nonexistent in 2018. A single set of tracks revealed that a wolf had crossed the open trail while I was placing cameras prior to the experiment. I observed no other predator tracks during or after the experiment ended (zero mammalian predators), and no evidence of hare attacks (blood or talon or feather imprints in snow) from owls or hawks. Predation risk was clearly much lower in 2018 than in 2010.

The cameras recorded 25 692 images of hares in 25 421 photographs during the 8 d of the 2018 experiment. Of these, 15 064 (58.6% of all hares) revealed foraging hares and 1407 (0.055%) revealed alert or vigilant hares.

I recovered all 320 foraged pine boughs and their accompanying individualized labels. I was thus able to unambiguously assign the cut-point diameter of each bough with its basal diameter.

Risk management by vigilance and time allocation, as illustrated by snowshoe hares, is not nearly as complex as it could be. The marginal effects of habitat and distance on vigilance were only apparent in the context of the full versus new moon. Meanwhile, moon phase had no overall detectable direct or indirect effect on GUDs and the main effects of habitat and distance depended on their interaction. These results are a rather stark contrast to the significant differences in vigilance between habitats reported in an earlier study of snowshoe hares foraging pine boughs in open versus alder habitats in the same area (Morris and Vijayan 2018). Hares in that study were most vigilant in dense alder habitat, opposite to my results demonstrating least vigilance there, but only during the new moon; strategies of risk management thus change through time.

Morris and Vijayan (2018) attributed high vigilance in dense habitat as a trade-off between the ability to elude or escape from predators through the maze of stems and branches afforded by alder and reduced sight lines in the alder that limited visibility required to detect predators. My treatments were designed to test this hypothesis. I reasoned that hares would be most vigilant in the open when their sight lines were restricted by my palisades of alder branches. I included a control and two additional treatments (top and both) to the side treatment in an attempt to clarify how various types of cover alter vigilance and time allocation. Although treatment was included in the two analyses restricted to tests in the open-trail habitat, its only significant effect was on vigilance through a complex interaction with moon phase. If sight lines modulate vigilance, then why was vigilance “reversed” in 2018 and without clear effects of my treatments?

The explanation appears to rest with differences in risk between 2010 and 2018. The clear reduction in predator activity between years suggests that hares were more vigilant and trepidatious toward predators in 2010 than in 2018. Comparison of camera images between the two study periods confirms the prediction. Fully 20% (1043 of 5344) of the images from 2010 revealed alert or vigilant behaviour compared with only 5% (1407 of 25692) in 2018. The proportion of photographs with two or more hares (3% in 2010 vs. 1% in 2018), as well as that of chases (36 in 2010 vs. 14 in 2018), was also strikingly higher in 2010. It is thus reasonable to attribute the higher vigilance observed in 2010 to a higher risk of predation. That risk was exacerbated by heightened competition that reduced the hares’ foraging efficiency.

Relationships between risk and its management through trade-offs between vigilance and time allocation by snowshoe hares in 2010 might appear inconsistent with theory. Hares were significantly more vigilant in the alder than in the open habitat. Increased vigilance should normally reduce the efficiency of resource harvest (increase GUD). Mysteriously, there was no difference in GUDs between habitats. The mystery is resolved when one considers that risk of mortality in a habitat must also include what happens after the prey detects a predator. Alder is escape habitat for hares. They evade predators by zig-zagging through its dense tangle of branches. Improved sight lines in the open allow hares to more easily detect approaching predators. But the net effectiveness of their vigilance is reduced because the probability of capture during flight is higher there than in the alder. Hares should be more prone to fleeing and forage more apprehensively, and thus less efficiently, in the open habitat. My data support the hypothesis. The proportion of images with hares foraging in alder was less than in the open (Fig. 3c).

China et al. (2008) proposed a novel safety in numbers hypothesis (e.g., Treisman 1975, Bednekoff and Lima 1998) for desert gerbils that might also help to explain hare foraging. The gerbils prefer semi-stabilized dunes where winds redistribute food daily (Kotler et al. 1993). As population sizes increase, competition associated with high density in the rich semi-stabilized habitat causes individuals to shift habitat use towards poorer stabilized dunes. The resulting increase in density there dilutes predation risk for each individual. Reduced risk lowers the value of vigilance, yields more opportunity for foraging and other activities, and promotes additional immigration. Simultaneously, this so-called “risk pump” necessarily drains safety from the semi-stabilized dunes, promoting even more immigration. The system finally attains equilibrium at high density when the GUDs converge (but not vigilance).

Might winter-foraging snowshoe hares pump safety from one habitat to another? Perhaps. My observations of similar GUDs, but divergent vigilance between habitats, is consistent with the risk-pump hypothesis. Hares use vigilance to manage risk from both predators and competing hares. So in order for a risk pump to operate, the reduction in predation risk must outweigh all additional density-dependent costs associated with competition for shared resources.

If we truly understand the trade-offs between vigilance and time allocation, then we must be able to explain patterns of vigilance during the different moon phases in 2018. Two hypotheses are consistent with those patterns.

First, bright moonlight should improve the effectiveness of vigilance as prey scan their surroundings for threats. Hares in the open habitat confirmed this expectation during the full moon by reducing the number of vigilant scans, especially so at near stations where they could quickly dart into the safety of alder (Fig. 1d). Even so, net risk typically intensifies for nocturnally foraging prey as the moon waxes towards full (Gilbert and Boutin 1991; Kotler et al. 2010; Gigliotti and Diefenbach 2018), so hares were more vigilant at stations where traversing the open habitat took them far from safety (Fig. 1a) and made them more visible to predators. Second, the hares’ energetic state was almost certainly lower during the cold nights of the full moon relative to those during the new moon. If hares require more energy during cold temperatures, then both vigilance and QHRs should be reduced (eq. (1)). Neither hypothesis is exclusive of the other, both are likely, and both predict my observation that there were more images of hares foraging during the full than during the new moon (Fig. 3a).

My data allow one final test. There was no simple difference in vigilance between near and far stations, but GUDs were lower at near stations than at far ones. Hares at near stations should have a lower QHR revealed by fewer images of foraging hares. Once again, data agree with the prediction. There were fewer images of foraging hares at near than at far stations (Fig. 3b).

Critics might wonder whether patterns of vigilance and foraging were related to concomitant changes in predator and hare density. The experimental design and field observations suggest otherwise. Significant population changes were unlikely given the short 18 d duration of the study, and the number of mammalian predators was effectively zero during the entire experiment.

It thus appears that hares tune their foraging to match the effectiveness of vigilance at reducing risk in different contexts and conditions. Gigliotti and Diefenbach (2018) reached a similar conclusion: flexible behaviour allows hares to match risk avoidance with local conditions. Vigilance alone cannot ameliorate risk that effects changes in apprehension revealed by GUDs (Dall et al. 2001; Kotler et al. 2010). Interactions of foraging and vigilance with habitat, moon phase, and distance from escape habitat highlight a sophisticated risk-management strategy contingent on changes in risk and opportunity. That strategy is nevertheless completely consistent with Brown’s (1999) theory of optimal trade-offs between vigilance and time allocation.

The agreement between theory and observation in snowshoe hares highlights the importance of adaptive risk management in natural populations. Hares merge their demand for energy with patterns of vigilance and time allocation in a way that maximizes overwinter survival. Their foraging strategy must necessarily impinge on subsequent population dynamics (Hik 1995) and their numerical interactions with other species. Hare foraging behavior thus adds to a growing body of knowledge gained from clever experiments that documents risk management as a major structuring force in the spatial dynamics and coexistence of species (e.g., Kotler et al. 2010; Camp et al. 2017). Any study of prey foraging that fails to include an assessment of the trade-offs is likely to misrepresent risk management and its important role in the dynamics of populations and communities.