Behavioral Ecology Advance Access originally published online on December 6, 2007
Behavioral Ecology 2008 19(2):245-254; doi:10.1093/beheco/arm116
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
What best explains vigilance in elk: characteristics of prey, predators, or the environment?
Department of Ecology, Montana State University, 310 Lewis Hall, Bozeman, MT 59717, USA
Address correspondence to S. Liley. E-mail: stewmtb{at}comcast.net.
Received 23 February 2007; revised 10 October 2007; accepted 17 October 2007.
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMETHODSRESULTSDISCUSSIONFUNDINGREFERENCES |
|---|
We quantified the vigilance levels of elk (Cervus elaphus) preyed on by wolves (Canis lupus) in Yellowstone National Park, between January and May in 2005 and 2006, and used Akaike's information criterion to compare a set of 38 regression models for vigilance levels. These models combined up to 9 predictor variables of 3 types: characteristics of the prey group (herd size and composition), characteristics of the predator (wolf pack size, distance away, and the presence/absence of a kill), and characteristics of the local environment (distance to woodland edges, snow depth, and snow cover). The set of models spanned a range of complexity from simple univariate models to complex combinations with up to 3 variables of each type. Complex models incorporating the characteristics of the wolf pack, the structure of the elk herd, and the environmental conditions had higher information content than simple models. Although univariate models of vigilance detect significant relationships, they have low information content relative to multivariate models. These results show that elk assesses factors of several types when assessing risk and deciding how much time to allocate to vigilance. In particular, we found that all well-supported models of vigilance included several "prey" variables and several "predator" variables. This result highlights the need to consider information about predators when trying to explain variation in vigilance levels in prey.
Key words: antipredator behavior, Canis lupus, Cervus elaphus, elk, predation risk, vigilance, wolf.
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONFUNDINGREFERENCES |
|---|
To reduce the risk of predation, animals engage in a range of behavioral responses, including habitat shifts, changes in group size, reduced activity, and increased scanning rates for predator detection. Antipredator vigilance has been studied in great detail over the past quarter century (Elgar 1989
; Lima and Dill 1990; Caro 2005). By increasing the effort put into scanning the environment, animals are more likely to detect a predator at a distance and consequently prevent or avoid a direct attack (Caro 2005). Vigilance has been studied in a wide variety of mammals and birds, and these studies have identified many factors that influence scanning rates, including group size (Underwood 1982; Elgar 1989; Roberts 1996), age and sex (Hunter and Skinner 1998; Childress and Lung 2003; Winnie and Creel 2007), body condition (Winnie and Creel 2007), position within the group (Underwood 1982; Keys and Dugatkin 1990; Hunter and Skinner 1998; Stankowich 2003; di Blanco and Hirsch 2006), habitat type (Lima 1987; Scheel 1993; White and Berger 2001), time of day (Elgar 1989; Scheel 1993), and local environmental conditions (Elgar 1989; Lima and Dill 1990).
These studies clearly demonstrate that animals adjust their scanning rates in response to their own condition, the size, or type of group they occupy and in response to the local environment. We know substantially less about the ways animals respond to variation in the type of threat that they face. In most studies, behavioral responses are recorded in response to simulated risk or in response to the immediate presence or absence of a natural predator. This has been a productive approach, but it gives little scope to ask whether the strength of antipredator responses varies in response to aspects of risk such as the distance to predators, the size of the attacking group, or cues about the predators' state of satiation (as indicated by the presence or absence of a kill or perhaps simply by its behavior). It is likely that prey assess risk as a continuous variable, rather than dichotomizing risk into periods of safety and danger, as is assumed by some models (Lima and Bednekoff 1999), experiments (Sih and McCarthy 2002), and observational studies (Creel and Winnie 2005). To address this broad hypothesis, one must test whether variables associated with the predator predict the strength of antipredator responses by the prey.
For example, consider the question "Are prey more or less vigilant when a kill has been made recently?" A priori, it is difficult to know whether prey should regard a fresh carcass as an indication of safety or of risk because the answer depends on the time scale over which prey assess risks. The fact that an elk (Cervus elaphus) has been killed clearly indicates that a lethal risk exists. On the other hand, wolf (Canis lupus) packs in Yellowstone National Park (YNP) kill at intervals that average 2–3 days (Stahler et al. 2006). The presence of a fresh kill is a reliable indicator that a hunt is nonrandomly unlikely in the immediate future. Wolves, like other carnivores, sometimes make multiple kills and sometimes hunt opportunistically at kill sites, but this does not appear to be the norm. Do elk perceive a kill as a cue of danger because it reveals that wolves were not only in the area, but hunting actively, or do they recognize a brief period of safety (a period when wolves are satiated and not likely to hunt) and reduce their vigilance levels accordingly?
More broadly, despite the depth and breadth of information on factors that affect antipredator vigilance, we know relatively little about the relative importance (RI) of various factors that stimulate vigilance. These factors can be placed in 3 broad categories: 1) those related to the prey, including age, sex, condition, or group size, 2) those related to the environment, including habitat type, light conditions, and other factors that might impede escape or defense, such as snow depth or cover, 3) factors related to characteristics of the predator, such as distance to predators, predator group size, presence of a kill, and time since the kill was made. We know of no field study that has simultaneously considered the roles of environmental, predator, and prey effects on vigilance. Here, we made observations of elk on the northern range of YNP, where the primary predator of adult elk is the wolf. We used model selection with Akaike's information criterion (AICc) to compare a set of 38 regression models of vigilance in elk. We selected these models a priori to span a range of complexity and to vary in the degree to which the focus is put on environmental, predator, and prey effects or combinations of these effects. This approach allows us to test whether simple or complex models better predict vigilance and to identify whether environmental, predator, or prey variables are typically included in the best models.
|
METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONFUNDINGREFERENCES |
|---|
We observed elk and wolves in the northern range of YNP. The study area was a system of river valleys typified by grassy open areas at low elevation and coniferous woodlands on the slopes above. Elk, numbering from 12 000 to 14 000 animals, are the most numerous ungulate on the site (Smith et al. 2004
), followed by bison (Bison bison), numbering 600–700 animals (Smith et al. 2003). Moose (Alces alces), mule deer (Odocoileus hemionus), white-tailed deer (Odocoileus virginianous), and bighorn sheep (Ovis canadensis) also occupy the northern range but at much lower densities than elk or bison. Elk are the primary prey of wolves throughout YNP, compromising 88% of the 2347 documented wolf kills from 1995–2003 (Stahler et al. 2006), and wolves are the primary predator of adult elk, accounting for more than 90% of the predation on adults (Smith et al. 2004, Creel and Winnie 2005; Creel et al. 2007). The spatial structure of wolf territories and the daily movements of wolves within territories cause predation risk for elk to vary spatially on a scale of meters to kilometers and temporally on a scale of hours to days (Creel et al. 2005; Winnie and Creel 2007). In the northern range, wolves and elk are easily observed, without disturbance, at distances of a kilometer or more (Smith et al. 2004). Consequently, this system and site provide an excellent opportunity to record variation in antipredator vigilance in response to variation in predator, prey, and environmental characteristics.
Study area
We studied the factors that affect vigilance levels in elk during the winter months (January to May) of 2005 and 2006. Our 830-km2 study site was located in the northeast portion of YNP, WY, centered on the "northern range," which is defined by the seasonal migration of the elk herd (12 000–14 000 individuals) that occupies the area. Wolves were reintroduced into the northern range in the winters of 1995 and 1996. During our study, wolf numbers on the site varied from 84 wolves in 7 packs in 2005 to 54 wolves in 6 packs in 2006. The study area is dominated by large open valleys comprised of sage (Artemesia spp.) and grassland (Festuca spp. and Agropyron spp.) with riparian areas bordering the small creeks and the Yellowstone, Lamar, and Gardner rivers. The upper elevations and north-facing slopes are primarily coniferous forests (Pinus spp. and Pseudostuga spp.) with small intertwined meadows. Elevation varies from 1500 to 3400 m above sea level with 87% of the area falling between 1500 and 2400 m above sea level. The climate is described by long, cold, and snowy winters and short, cool summers. For a more detailed description of the study site, see Houston (1982).
Prey (elk) variables
We observed elk 3–5 days a week from early January to May in 2 years, from sunrise to sunset. Observations followed a stratified sampling design. We stratified the northern range by location (Blacktail Deer Creek Plateau, Hellroaing, Tower Junction, Slough Creek, and Lamar/Soda Butte). Each stratum was approximately the same size and received approximately equal observational effort. We made behavioral observations at distances of 0.5–2 km using a tripod-mounted x20–56 Nikon ED spotting scope. These distances were usually sufficient to avoid detectably affecting elk behavior. If it became apparent that the elk were focusing on the observer or other humans (any vigilance directed toward humans) or if the elk retreated from any people, we ceased observations and omitted all data from that elk herd.
We observed elk after first recording their distance from the nearest known wolves (see Predator (wolf) variables). On sighting an elk or elk herd, we recorded the following preliminary data with respect to the prey (also see Predator (wolf) variables and Environmental variables):
- 1) Herd size: For many species, vigilance declines with group size. We defined a herd with the criteria of Creel and Winnie (2005) and Winnie and Creel (2007), which follow the general recommendations of Stankowich (2003). The criteria identified herds with a mean interindividual distance of less than 5 body lengths, whereas herds were typically separated by more than a kilometer.
- 2) Herd composition: (Calves [young of the year], cows [female > 1 year old], bulls [adult male with at least one brow tine], and spikes [1-year old antlered male with no brow tines]). Previous research has shown that for elk elsewhere in the ecosystem, females with calves exhibit the highest vigilance levels followed by females without calves and then males (Childress and Lung 2003
; Winnie and Creel 2007).
- 3) Day of year (Julian date): As winter progresses, forage quantity and quality decreases, and elk body condition declines (Winnie and Creel 2007
). Thus, elk face stronger constraints on time and energy as winter progresses.
- 4) Position within the herd: Peripheral animals may be more vigilant than interior animals if they are at a greater risk of being attacked first by a predator (Hamilton 1971
; Underwood 1982; Fitzgibbon 1990; Keys and Dugatkin 1990; Hunter and Skinner 1998; Stankowich 2003; di Blanco and Hirsch 2006).
- 2) Herd composition: (Calves [young of the year], cows [female > 1 year old], bulls [adult male with at least one brow tine], and spikes [1-year old antlered male with no brow tines]). Previous research has shown that for elk elsewhere in the ecosystem, females with calves exhibit the highest vigilance levels followed by females without calves and then males (Childress and Lung 2003
We used instantaneous scan sampling to collect behavioral data. Instantaneous scan sampling provides unbiased data appropriate to estimate the percentage of time an animal spends engaged in various activities and allows more groups to be sampled in a day than other sampling designs (Altmann 1974; Martin and Bateson 1986). We scanned through elk herds (N = 202 herds) at 5-min intervals, for a minimum of 3 and a maximum of 8 scans, recording (with a hand-held voice recorder) the following information for each individual in the herd: sex and age (cow, calf, bull, or spike), position within the herd (peripheral or interior), and behavior (grazing, moving, vigilant, bedded, bedded vigilant, or other). We defined interior animals as those that a predator could not approach without encountering another animal in the herd. Conversely, peripheral animals were those that could be encountered by a predator before any other animals in the herd. This definition follows "layers of protection" method of Stankowich (2003), with one layer. We used behavioral categories established by prior research on elk responses to wolves (Winnie and Creel 2007), classifying behavior as follows: 1) grazing: animals that were standing with their head down collecting forage or with their heads up chewing forage; 2) moving: animals that were walking or running without feeding; 3) vigilant: animals that were standing, head erect, with ears cocked forward in the direction of gaze (the position of the head and ears is critical for this category); 4) bedded: animals that were lying down on their sternum or side, head not erect (often ruminating or apparently sleeping); 5) bedded vigilant: animals that were lying on their sternum with head erect, with ears cocked forward in the direction of gaze; and 6) other: all other less common activities such as grooming or sparring.
To avoid pseudoreplication, we collapsed all the scans from a herd into a single vigilance rate. Because separate individuals and scans are not fully dependent, this is a conservative approach, relative to a repeated-measures model with random effects.
Predator (wolf) variables
As mentioned above, many studies of antipredator behavior (including our own: Creel and Winnie 2005; Winnie and Creel 2007) consider only whether the predator is present or absent, which is likely to oversimplify the information perceived and used by prey in making antipredator responses. In this study, we recorded several variables about the predator to test how they related to vigilance levels in elk. Because at least one (and up to 7) wolf in every pack in the northern range wore a radio collar during this study (Doug Smith, personal communication [Yellowstone Wolf Project Leader]), we located wolves via VHF radiotelemetry. We then located the pack visually by skiing or hiking to high ground and using a tripod-mounted x20–56 Nikon ED spotting scope and watching the wolves for as long as was needed to determine the following variables. 1) The number of wolves in the pack. For some coursing predators, as pack size increases the probability of capturing and killing prey individual increases, therefore increasing the risk to a prey group (Creel and Creel 1995). However, we unaware of any studies that have tested how predator group size might affect antipredator behavior in prey, specifically vigilance levels. 2) Whether the wolves were at or near a recent kill (<24 h old). Fresh blood, carcasses or portions of them, and scavengers helped us to determine if a kill had been made recently. 3) The location of the pack (determined by Global Positioning System, range-finding binoculars, compass, and United States Geological Survey 1:50 000 maps). We used this location in combination with a subsequent location of elk herds to determine the straight-line distance between elk and the nearest known wolves. By comparing vigilance levels across a wide range of distances from wolves (minimum 0.2 km, maximum 5.0 km), we essentially allowed elk to tell us when they perceived predators to be "present." Five kilometers was the greatest distance we considered because as the wolf–elk distance increased, there was increased potential for failing to detect wolves within the radius, due to the quadratic increase in area.
Environmental variables
We measured 3 environmental variables that we hypothesized or knew to affect vigilance levels in elk. First, we recorded the distance from the herd to timber. Numerous studies have shown that habitat type and distance to protective cover affect vigilance levels in prey (Lima 1987; Scheel 1993; White and Berger 2001). Depending on the prey and predator species of concern results have varied. Distance to timber variable captures 2 important aspects of the environment, habitat type (open vs. closed), and how far an animal was from protective cover. We measured distance to timber by visually estimating the distance and classifying observations into one of 5 categories: 0 (herd was in the timber), 1–30 (away from timber), 31–100, 101–300, and 301+ m. This categorization has previously been useful in measuring effects of risk on patterns of aggregation (Creel and Winnie 2005), decision-making (Winnie et al. 2006), habitat selection (Creel et al. 2005), and behavior (Winnie and Creel 2007) for elk elsewhere in the Yellowstone ecosystem. Because distance to timber and habitat type are strongly collinear (we used the covariance matrix to examine collinearity for all pairs of variables considered), we could not include both variables in our analyses. We preferred distance to timber because of its important role in prior studies. The second variable of concern was percent snow cover. We visually estimated how much of the landscape (0–100% by 10%) was covered in snow within the viewshed of herd. As snow cover increases it also increases the difficulty of moving and foraging. Snow cover was often less than 100%, and elk tended to congregate on snow-free areas. Finally, we obtained daily data on snow–water equivalent (SWE) from the regional NRCS SNOTEL site at Canyon, YNP, WY (within the study area). SWE provided information on the depth and density of the snow. We quantified snow cover and density (SWE) because deeper and denser snow limit elk movement and foraging, and these effects might place constraints on the time or energy that can be directed to vigilance.
Statistical methods
To test the RI of predator, prey, and environmental variables in their effects on vigilance, we used AICc (with sample size correction, AICc) to compare 38 a priori regression models (Burnham and Anderson 2002). We used AICc for model selection because it identifies models on the basis of both fit and parsimony (complex models must explain more variance to obtain AICc scores equal to simpler models). Explicit comparison of multiple models is well suited to answering our 2 fundamental questions. First, what types of variables best explain vigilance? Second, do simple or complex models perform better? Models varied from simple (1 parameter models) to complex (maximum of 9 parameters). Conceptually, our models fell into 3 subsets. The first subset included the simplest models, which contained independent variables from one of the 3 types of variables (predator, prey, or environment) (total of 15 models). In the second subset, we developed models of intermediate complexity that included parameters about the predator and the prey herd (total of 15 models). The last subset contained the most complex models, which contained parameters about the predator, the elk herd, and the environment (total of 8 models).
The potential set of models was extensive (2047 models, without considering interactions). For model sets, these large, comparing models that include all possible subsets of independent variables have been criticized as data dredging (Burnham and Anderson 2002). We restricted our a priori model set to 38 models (Table 1) that addressed hypotheses developed through field observations, our past research, and other studies. Rather than building the model set by exclusion of models from the full set, we built the model set by inclusion. We began by including univariate models and the full model to include the extremes of the continuum of model complexity. We selected candidate models of intermediate complexity in general by identifying models that our prior research (Creel and Winnie 2005; Creel et al. 2005; Winnie et al. 2006; Winnie and Creel 2007) suggested would be strong and then adding or deleting single variables or pairs of variables to test whether these changes yielded an increase or decrease in information content. For an example of this approach, compare models 6, 9, and 19 in Table 1. We avoided a priori models that included pairs of independent variables that we suspected would be highly collinear. After identifying our set of candidate models, we examined the correlation matrix for the independent variables to confirm this (Table 2). One strength of model selection using information theory is that several models are compared, so that a pair of variables that may be collinear can be evaluated separately and together, giving clearer inference on the explanatory power of each than would be obtained from any single model. The fact that the full model was highly ranked (Table 1), and every independent variable appears in at least one of the top 3 models (all with 
AICc < 1.2) confirms that each independent variable has predictive value after inclusion of the others. We did not include interactions in the model set simply because the huge number of potential interactions was too complex to evaluate in an a priori manner. Because we were surprised by the results for models including the presence/absence of a kill, we examined this effect a posteriori as an exploratory analysis.
|
|
We assessed normality for all parameters by observing their distributions and further tested for deviations from model assumptions by observing residual plots. We arcsine transformed parameters that were proportions prior to analysis and back transformed in plots of results (Zar 1999
). We used the generalized linear models module of STATISTICA 6.0 (Statsoft Inc, Tulsa, OK) to fit models and made comparisons using AIC corrected for sample size (AICc) (Burnham and Anderson 2002). We assessed goodness of fit using the most complex model in the a priori set, (K = 9) and found no evidence for lack of fit (
2 = 202.000, degrees of freedom [df] = 189, 
=1.069). |
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONFUNDINGREFERENCES |
|---|
Complex models predicted vigilance levels better than simple models (Table 1). The best model (AICc weight = 0.199) included 2 parameters related to predators and 3 related to prey. The second best model, which had information content comparable to the best model (AICc weight = 0.153) was the second most complex model in the a priori set, with 3 parameters related to predators, 3 related to prey, and 2 related to the environment. Six models were within 2 AICc units of the top model. All these included parameters related to both predators and prey. Out of these 6 models, 2 included predator, prey, and environmental parameters. Models with the simplest structure performed much worse; the best simple model (with parameters of only one type—predator, prey, or environment) was more than 16 AICc units worse than the top model (Table 1). Models with a single independent variable were all more than 29 AICc units worse than the best model. Overall, Table 1 reveals that decisions about vigilance are simultaneously affected by information about the predator, the prey, and the situation in which the encounter occurs. Predator and prey variables were particularly important (Table 3).
|
Predator (wolf) variables
Characteristics of the predator were important in determining vigilance levels. In all models within 2 AICc units of the best model, there was at least one wolf parameter and most of these models contained more than one (average of 2.167; Table1). Comparing RI scores (Burnham and Anderson 2002
) for the wolf variables, distance from the elk herd to the wolves was the most important, with an RI score of 1 (the maximum) (Table 3). Vigilance levels were greatest when wolves were within 1 km of the elk herd (Figure 1, Table 4). Elk also showed elevated vigilance levels when wolves were between 1 and 3 km away. At distances of greater than 3 km, vigilance returned to apparently baseline levels. The number of wolves in the pack also played an important role in determining vigilance levels, with stronger responses to packs larger than 10 wolves (Tables 3 and 4). The greatest effect occurred when pack sizes were greater than10 wolves (Figure 2). At pack sizes less than 10 wolves, vigilance levels were lower and approximately constant. The presence or absence of a kill had the lowest RI score, in comparison to the other wolf variables (Table 3). Overall, the presence of a kill had only weak effects on vigilance (Table 4). However, we found an unanticipated but strong difference between all-bull groups and cow groups (cows, calves, and spikes) in their reactions to the presence of a kill (Figure 3, Table 4). Cow groups significantly reduced vigilance levels when a kill was present (df = 123, P < 0.001), whereas all-bull group significantly increased vigilance levels when a kill was present (df = 79, P = 0.003).
|
|
i)
|
|
Prey (elk) variables
Parameters describing characteristics about the elk herd also played an important role in determining vigilance levels. Herd composition had the highest RI (RI = 1) (Table 3). Groups including cows (composed of cows, calves, and spikes) had higher vigilance levels than all-bull groups (Figure 4, Table 4). Vigilance levels also increased when the proportion of calves to cows in the herd increased (Figure 5, Table 4). The proportions of cows and calves are not fully independent, so these 2 effects cannot be fully disentangled (Figures 4 and 5, see percentages within bars). Herd size had a RI score of 0.65. Group sizes between 10 and 20 were more vigilant than smaller or larger groups. Once group size exceeded 20 individuals, vigilance levels decreased in a linear fashion (Figure 6, Tables 3 and 4). Here again, the effect of herd size cannot be fully disentangled from the effect of herd composition. Small groups are often all bulls (Figure 6, see percentages within bars), and bulls are substantially less vigilant than cows (Winnie and Creel 2007
). The other 2 elk parameters (proportion peripheral and day of year) were less important in determining vigilance levels: both these variables had low RI scores of 0.062 (Table 1).
|
|
|
Environmental variables
Environmental parameters had a weaker effect, and none of the top models were dominated by environmental variables. However, environmental parameters did appear in some of the best models (Table 1), and 2 environmental variables had significant partial regressions on vigilance (Table 4). Distance to timber had the highest RI (RI = 0.332, compared with RI = 1 for the top predator and prey variables). The farther the elk herds were from protective cover the more vigilant they became. This increase was linear until groups were about 100 m from timber and then appeared to approach an asymptote (Figure 7, Table 4). Snow depth and density (SWE) had a RI value of 0.311 and appeared in 2 of 6 models within 2 AICc units of the best model (Tables 1and 3). Vigilance levels decreased as SWE increased (Table 4). Snow cover had the lowest RI score (0.168) of all the environmental parameters and appeared once in the models within 2 AICc units of the best model (Tables 1 and 3). Unlike SWE, as the percentage of the landscape that was covered in snow increased, elk vigilance levels tended to increase, though the effect was weak and not significant (Table 4).
|
Comparison of the 3 subsets of models
When comparing the performance of the models from each of the 3 subsets, models from subset 1 (simple models with independent variables from one of the 3 categories of variables: predator, prey, or environment) performed the worst (Table 1: bold table entries indicate the best model in each category). Significant effects on vigilance can be detected with these univariate models (Table 4), but their information content was low in comparison to more complex models. Models from subset 2 (more complex models containing parameters about predator and prey) performed the best, appearing in two-thirds of the top models (Table 1). Models from subset 3 (the most complex models containing parameters about predator, prey, and environment) did not perform as well as models from subset 2, but they did represent one-third of the best models (Table 1).
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONFUNDINGREFERENCES |
|---|
Our results show that elk are sensitive to a suite of variables when allocating time to antipredator vigilance. Vigilance levels are sensitive to characteristics of predators, the structure of the group that is threatened, and the conditions under which the threat arises. Complex models that incorporate all these features perform better than simpler models. No models that failed to incorporate aspects of both predator and prey had high information content. Including environmental effects improved the models, but environmental effects were less important than predator and prey variables.
The importance of predator variables in determining vigilance levels is one of our most interesting results. The model selection results show that vigilance responds to variation in the attributes of the wolf pack to which the elk is responding, including its size, distance away, and the presence of a kill, which is a potential cue about the likelihood that the wolves are actively hunting. In our study, elk vigilance levels negatively correlated with distance to wolves at distances up to 3 km (Figure 1). This result shows that elk in the northern range detect and respond to predators within a large area, as they do in the Gallatin Canyon, also in the Greater Yellowstone Ecosystem (Creel et al. 2005; Winnie and Creel 2007). The sensory capabilities of elk are not well studied (Toweill and Thomas 2002), but our results indicate that they are capable of detecting and responding to wolves that are not in the process of attacking and are up to several kilometers away. Whereas most studies of antipredator behavior focus on short-term reactions to real or simulated attacks by predators (Lima 1998), this result suggests that we must also consider the importance of less dramatic but more frequent responses to subtle spatiotemporal variation in risk. Given elk can detect and respond to predators that are several kilometers away and given the high density of wolves in the Yellowstone ecosystem, it is perhaps not surprising that the demographic costs of behavioral responses to risk are substantial (Creel et al. 2007).
Several studies of hunting success have found that as the number of individuals in the predator group increases so does the probability of making a kill (Creel and Creel 1995; Funston et al. 2001; but see Thurber and Peterson 1993; Schmidt and Mech 1997). Despite the apparent importance of predator group size from the perspective of the predator, we have been unable to find any studies that measured how prey respond to this increased threat. It appears that elk in the northern range identify larger pack sizes as a greater risk (Figure 2).
Of all the predator variables, kill presence/absence had the most surprising effects. A priori, we did not have a strong expectation of whether vigilance would increase or decrease in the presence of a fresh kill and whether a kill indicates safety or danger probably depends on the time scale of the decision. Cow groups appear to treat a recent kill as indicating a period of safety and significantly reduced their vigilance levels, whereas bull groups responded to recent kills by significantly increasing their vigilance levels. We do not have data to explain this difference, although differences in individual condition and vulnerability may be involved (Winnie and Creel 2007).
Herds with a high proportion of cows were more vigilant in the northern range, as has been shown for elk elsewhere in the Yellowstone ecosystem (Laundre et al. 2001; Childress and Lung 2003; Winnie and Creel 2007). In the Gallatin Canyon portion of the ecosystem, we have shown that bulls enter the winter in significantly worse body condition than cows, so energetic constraints do not allow bulls to allocate as much time to vigilance (Winnie and Creel 2007). Childress and Lung (2003) found that cows with calves were more vigilant than cows without calves and bulls. Calves are killed more often than expected by chance in YNP (Smith et al. 2004; Creel et al. 2007), creating an obvious selection pressure in favor of high vigilance in nursery herds. However, it is worth noting that this logic would also predict higher levels of vigilance for bulls than cows (opposite to our data, as just discussed) because bulls are also killed more often than expected by chance in YNP (Smith et al. 2004; Winnie and Creel 2007).
Elk, like many ungulates, are less vigilant as herd size increases (Underwood 1982; Elgar 1989; Roberts 1996). This decrease in individual vigilance aligns well with Pulliam's (1973) model of "corporate vigilance." Herd size is collinear with the proportion of the herd that occupies peripheral positions within a herd. (As herd size increases, more animals can occupy central positions.) Because of this collinearity, and because herd size effects are well established in theory (Elgar 1989; Lima and Dill 1990), for other species (Roberts 1996), and for elk (Creel and Winnie 2005; Winnie and Creel 2007), we included herd position in only 4 models.
Vigilance might be more tightly constrained as winter progresses because body condition declines through the winter (Winnie and Creel 2007). Despite this logic, models that included the day of winter fared relatively poorly, the best being 3.33 AICc units worse than the best model. As Winnie and Creel (2007) noted, "[day of winter] is problematic because it is correlated with temperature and snow depth and thus contributes to overdispersion in the models that also contain these variables." Because we considered snow depth (SWE) important and included it in many a priori models, we included day of winter in only 3 models. Post hoc analysis of the correlation matrix showed that day of year and SWE were highly correlated.
The top models, within 2 AICc units of the best model, always included parameters related to predators and to prey but included an environmental effect in only a third of the cases. This suggests that vigilance is not well predicted simply by knowing whether the environmental circumstances are generally risky or safe. Elk adjust vigilance more clearly in response to the group they are in and the type of immediate threat they are facing from wolves. Environmental variables acted more as modifiers rather than drivers of elk vigilance. Distance to timber, which is closely related to habitat type (open vs. closed: see Creel et al. 2005; Winnie et al. 2006), appeared in one-third of the top models (Table 1). Based on variation in vigilance, elk in this study system identified timber as a location of safety and open areas as places of danger. This result aligns well with previous results showing that elk adjust herd sizes in response to distance from timber (Creel and Winnie 2005) and that elk move into timbered locations when wolves are present (Creel et al. 2005).
As SWEs increased, vigilance levels decreased, whereas as percent snow cover increased vigilance levels increased (Table 4). SWE and snow cover initially seem likely to be correlated but were not strongly related in our data. SWE was greatest from late March to mid-April, and during this period the landscape was rarely covered in 100% snow (ca. 12% of the time). We cannot fully explain the difference between SWE and snow cover in their effects on vigilance. Post hoc, it seems possible that high snow cover increases risk and thus promotes vigilance, and deep and/or dense snow conditions make foraging and travel sufficiently difficult, and high SWE acts as a constraint on vigilance.
In summary, complex models explain patterns of vigilance in elk better than simple models. Although many individual variables play a significant role, univariate models had low information content in comparison to more complex models. From this pattern, we conclude that elk use information of several types when assessing risk and allocating time to vigilance. In particular, predator characteristics were as important as prey characteristics in determining vigilance levels. These results highlight the need to consider information about predators when testing hypotheses about vigilance levels in prey.
|
FUNDING |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONFUNDINGREFERENCES |
|---|
National Science Foundation (IBN-0238169).
|
ACKNOWLEDGEMENTS |
|---|
We thank John Winnie and David Christianson for many helpful discussions and Douglas Smith and the Yellowstone Wolf Project for valuable assistance with field observations.
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONFUNDINGREFERENCES |
|---|
Altmann J. Observational study of behavior: sampling methods. Behaviour (1974) 49:227–267.[Medline]
Burnham KP, Anderson DR. Model selection and multimodel inference: a practical information-theoretic approach, 2nd ed (2002) New York: Springer, Berlin Heidelberg.
Caro T. Antipredator defenses in birds and mammals. (2005) Chicago: The University of Chicago Press.
Childress MJ, Lung MA. Predation risk, gender and group size effect: does elk vigilance depend on the behaviour of conspecifics? Anim Behav (2003) 66:389–398.[CrossRef][Web of Science]
Creel S, Chrisianson D, Liley S, Winnie J Jr. Predation risk affects reproductive physiology and demography of elk. Science (2007) 315:960.
Creel S, Creel NM. Communal hunting and pack size in African wild dogs, Lycaon pictus. Anim Behav (1995) 50:1325–1339.[CrossRef][Web of Science]
Creel S, Jr Winnie J. Responses of elk herd size to fine-scale spatial and temporal variation in the risk of predation by wolves. Anim Behav (2005) 69:1181–1189.[CrossRef][Web of Science]
Creel S, Winnie J Jr, Maxwell B, Hamlin K, Creel M. Elk alter habitat selection as an antipredator response to wolves. Ecology (2005) 86:3387–3397.[CrossRef][Web of Science]
di Blanco Y, Hirsch BT. Determinants of vigilance behavior in the ring-tailed coati (Nasua nasua): the importance of within-group spatial position. Behav Ecol Sociobiol (2006) 61:173–182.[CrossRef][Web of Science]
Elgar MA. Predator vigilance and group size in mammals and birds: a critical review of the empirical evidence. Biol Rev Camb Philos Soc (1989) 64:13–33.[Medline]
Fitzgibbon CD. Why do hunting cheetahs prefer male gazelles? Anim Behav (1990) 40:837–845.[CrossRef][Web of Science]
Funston PJ, Mills MGL, Biggs HC. Factors affecting the hunting success of male and female lions in the Kruger National Park. J Zool (2001) 253:419–431.[CrossRef][Web of Science]
Hamilton WD. Geometry for the selfish herd. J Theor Biol (1971) 31:295–311.[CrossRef][Web of Science][Medline]
Houston DB. The northern Yellowstone elk: ecology and management. (1982) New York: Macmillan.
Hunter LTB, Skinner JD. Vigilance behaviour in African ungulates: the role of predation pressure. Behaviour (1998) 135:195–211.
Keys GC, Dugatkin LA. Flock size and position effects on vigilance, aggression, and prey capture in the European starling. Condor (1990) 92:151–159.
Laundre JW, Hernadez L, Altendorf KB. Wolves, elk, bison: reestablishing the "landscape of fear" in Yellowstone National Park, U.S.A. Can J Zool (2001) 79:1401–1409.[CrossRef]
Lima SL. Distance to cover, visual obstructions, and vigilance in house sparrows. Behaviour (1987) 102:231–238.
Lima SL. Non-lethal effects in the ecology of predator-prey interactions. Bioscience (1998) 48:81–107.
Lima SL, Bednekoff PA. Temporal variation in danger drives antipredator behavior: the predation risk allocation hypothesis. Am Nat (1999) 153:649–659.[CrossRef][Web of Science]
Lima SL, Dill LM. Behavioral decisions made under the risk of predation: a review and prospectus. Can J Zool (1990) 68:619–640.
Martin P, Bateson P. Measuring behaviour: an introductory guide. (1986) Cambridge: Cambridge University Press.
Pulliam HR. On the advantages of flocking. J Theor Biol (1973) 38:419–422.[CrossRef][Web of Science][Medline]
Roberts G. Why individual vigilance declines as group size increases. Anim Behav (1996) 51:1077–1086.[CrossRef][Web of Science]
Scheel D. Watching for lions in the grass: the usefulness of scanning and its effects during hunts. Anim Behav (1993) 46:695–704.[CrossRef][Web of Science]
Schmidt PA, Mech DL. Wolf pack size and food acquisition. Am Nat (1997) 150:513–517.[CrossRef][Web of Science]
Sih A, McCarthy TM. Prey response to pulses of risk and safety: testing the risk allocation hypothesis. Anim Behav (2002) 63:437–443.[CrossRef][Web of Science]
Smith DW, Drummer TD, Murphy KM, Guernsey DS, Evans SB. Winter prey selection and estimation of wolf kill rates in Yellowstone National Park, 1995-2000. J Wildl Manage (2004) 68:153–166.[CrossRef]
Smith DW, Peterson RO, Houston DB. Yellowstone after wolves. Bioscience (2003) 53:330–340.[CrossRef][Web of Science]
Stahler DR, Smith DW, Guernsey DS. Foraging and feeding ecology of the gray wolf (Canis lupus): lessons from Yellowstone National Park, Wyoming, USA. J Nutr (2006) 136:1923S–1926S.
Stankowich T. Marginal predation methodologies and the importance of predator preferences. Anim Behav (2003) 66:589–599.[CrossRef][Web of Science]
Thurber JM, Peterson RO. Effects of population density and pack size on the foraging ecology of gray wolves. J Mammal (1993) 74:879–889.[CrossRef][Web of Science]
Toweill DE, Thomas JW. North American elk: ecology and management, 1st ed. (2002) Washington: Smithsonian Institution Press.
Underwood R. Vigilance behaviour in grazing african antelopes. Behaviour (1982) 79:81–107.
White KS, Berger J. Antipredator strategies of Alaskan moose: are maternal trade-offs influenced by offspring activity? Can J Zool (2001) 79:2055–2062.[CrossRef]
Winnie J Jr, Chrisianson D, Creel S, Maxwell B. Elk decision-making rules are simplified in the presence of wolves. Behav Ecol Sociobiol (2006) 61:277–289.[CrossRef][Web of Science]
Winnie J Jr, Creel S. Sex-Specific behavioral responses of elk to spatial and temporal variation in the threat of wolf predation. Anim Behav (2007) 73:215–225.[CrossRef][Web of Science]
Zar JH. Biostatistical analysis. (1999) 4th ed. New Jersey (NJ): Prentice Hall.
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  S. Creel, J. A. Winnie Jr., and D. Christianson Glucocorticoid stress hormones and the effect of predation risk on elk reproduction PNAS, July 28, 2009; 106(30): 12388 - 12393. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||