SEARCH

SEARCH BY CITATION

Keywords:

  • behavioural response race;
  • ecology of fear;
  • GPS ;
  • habitat selection;
  • predation;
  • predator–prey spatial game;
  • space race;
  • step selection functions

Summary

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information
  1. Predators impact prey populations not only by consuming individuals, but also by altering their behaviours. These nonlethal effects can influence food web properties as much as lethal effects. The mechanisms of nonlethal effects include chronic and temporary anti-predator behaviours, the nature of which depends on the spatial dynamics of predators and the range over which prey perceive risk.
  2. The relation between chronic and ephemeral responses to risk determines predator–prey interactions, with consequences that can ripple across the food web. Nonetheless, few studies have quantified the spatio-temporal scales over which prey respond to predation threat, and how this response varies with habitat features.
  3. We evaluated the reaction of radio-collared caribou and moose to the passage of radio-collared wolves, by considering changes in movement characteristics during winter and summer. We used an optimization algorithm to identify the rate at which the impact of prior passage of wolves decreases over time and with the predator's distance.
  4. The spatial and temporal scales of anti-predator responses varied with prey species and season. Caribou and moose displayed four types of behaviour following the passage of wolves: lack of response, increased selection of safe land cover types, decreased selection of risky cover types and increased selection of food-rich forest stands. For example, moose increased their avoidance of open conifer stands with lichen in summer, which are selected by wolves in this season. Also in winter, caribou increased their selection of conifer stands with lichen for nearly 10 days following a wolf's passage. This stronger selection for food-rich patches could indicate that the recent passage of wolves informs caribou on the current predator distribution and reveals the rate at which this information become less reliable over time.
  5. Caribou and moose used anti-predator responses that combine both long- and short-term behavioural adjustments. The spatial game between wolves and their prey involves complex and nonlinear mechanisms that vary between species and seasons. A comprehensive assessment of risk effects on ecosystem dynamics thus requires the characterization of chronic and temporary anti-predator behaviours.

Introduction

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Anti-predator behaviours are commonly observed amongst prey populations. These behaviours can be costly, however, because they generally take away time from other potentially fitness-enhancing behaviours (Brown & Kotler 2004). For example, prey may have to avoid the richest resource patches when trading-off energy gains for safety (Lima & Dill 1990; Lima 1998). Bison (Bison bison L.) have been observed to decrease their selection of optimal food items in response to predation risk, resulting in lower rates of energy gains (Fortin & Fortin 2009). Prey also uses vigilance to avoid being surprised by a predator, a behaviour that tends to decrease foraging rate (Lendrem 1983; Brown 1999). In general, the level of anti-predator behaviour should increase with the risk of mortality, the prey's energy state, and its fitness, and decrease as its marginal value of energy rises (Brown 1999; Brown & Kotler 2004).

The co-evolution of predators and prey may result in prey developing chronic anti-predator responses (Schmitz, Krivan & Ovadia 2004). Because of the high costs of anti-predator behaviours, a specific chronic response may not always maximize fitness, and it may be beneficial for the prey to adjust its level of anti-predator behaviour based on its current perception of the threat level. A prey should exhibit its strongest anti-predator behaviour in brief and infrequent high-risk situations, whereas allocation of anti-predator effort to high-risk situations should decrease as they become more frequent or lengthy (Lima & Bednekoff 1999; Creel et al. 2008). Chronic and ephemeral behavioural responses should ultimately be dictated by factors such as the mobility of predators, their hunting mode (Schmitz, Krivan & Ovadia 2004) and the prey escape tactics (Wirsing, Cameron & Heithaus 2010). An optimal prey response may not simply use either chronic or ephemeral anti-predator behaviours, but may use a combination of both. Such a combination of anti-predator responses can then be used to characterize a prey's ‘landscape of fear’, in which ‘hills’ and ‘valleys’ are defined by the predation risk and related in particular to spatial patterns in habitat features (Laundré, Hernández & Ripple 2010). Given the possibility of chronic and ephemeral behavioural responses to risk, the landscape of fear could vary broadly, from static to highly dynamic according to the spatial and temporal scales of the prey's perception of risk. A comprehensive assessment of the impact of predators should therefore involve the quantification of a prey's perceptual range (or the range within which signs of predators can trigger a response), together with the time that the perceived changes in threat last (Lima & Dill 1990; Lima & Zollner 1996).

Several studies have examined the impact of the presence of predators at fine temporal and spatial scales. For example, elk tend to leave food-rich grasslands and moved into the protective cover of wooded areas when wolves were present in the same drainage on the same day (Creel et al. 2005). They also increase their movement rate when wolves are within 5 km from its location during the previous 4 h (Proffitt et al. 2009), a response that may vary with the presence of calves, group size and season (Liley & Creel 2008). Elk become more vigilant when wolves are present within 3 km from their location (Liley & Creel 2008). Likewise, zebra (Equus quagga B.) display intense vigilance when lions (Panthera leo L.) are within 2 km (Périquet et al. 2012), and individuals become less abundant on grasslands during the night if lions have also been present during the past 24 h (Fischhoff et al. 2007). In general, the presence of lions within a 2-km fixed radius during the previous 24 h influences habitat preferences of both grazers and browsers (Valeix et al. 2009). Most of these fine-scale studies (i) approximate perception of predator presence by a binary presence/absence of predator (ii) in a fixed radius and (iii) over a fixed temporal window around the prey locations. This coarse level of information on prey's assessment of predator presence might not suffice to evaluate the role of anti-predator behaviours on the spatial dynamics of prey and, more generally, on food web properties.

Here, we evaluated the behavioural response of forest-dwelling woodland caribou (Rangifer tarandus caribou L.) and moose (Alces alces L.) to the passage of grey wolves, in the Côte-Nord region of Québec during winter and summer. Forest-dwelling caribou is considered threatened across the Canadian boreal forest (COSEWIC 2006), and the conservation of this ecotype has a strong impact on forest management. Food resources are not considered as a limiting factor for these caribou, and, instead, their population dynamics would be driven by top-down forces (Courtois et al. 2007). Given the importance of risk effects on top-down systems, a better understanding of the perception of predator presence by caribou, and their reaction to it, can provide valuable information of high conservation value. The response of moose to predation risk can also have consequences on caribou conservation, because a high hunting success of wolves on moose can lead to an increase in the wolf population, which in turn can be detrimental to the caribou populations (Wittmer et al. 2007).

We assessed nonlethal effects of wolves on caribou and moose by means of using step selection functions (SSF, Fortin et al. 2005) integrating information on both animal movements and habitat selection, along with an index of predator presence. This indexes a discounting function of both time since and distance from passage. The discounting function was assessed using optimization techniques applied to the SSFs, which identify the temporal scales and spatial extents over which caribou and moose respond to the recent passage of wolves in summer and winter. Our results highlight the dynamic aspect of the predator–prey game, which characterized by a combination of chronic and ephemeral anti-predator responses.

Materials and methods

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Study Area

The study area was located in the Côte-Nord region of Québec, Canada. Black spruce (Picea mariana M.) and balsam fir (Abies balsamea L.) are the dominant tree species. Mean daily temperatures vary from −23 °C in January to 14 °C in July, whereas mean annual precipitation is 715 mm (Crête & Courtois 1997). Snow accumulates from early November to early June, reaching a peak >2 m in mid-March (Houle et al. 2010). The area is subject to logging activities. We used Landsat Thematic Mapper images taken in 2000 with a 25-m resolution grid covering nearly 72 000 km2(50°N to 52°N, 68°W to 71°W) updated every year with information from forestry companies to create maps of land cover types.

Telemetry Data and Biological Seasons

An aerial survey conducted over the study area in March 2007 estimated caribou density to be 1·9 individuals/100 km2 and moose density was 4·3 individuals/100 km2 (Gingras, Audy & Courtois 1989). Five wolf packs were also identified in the study area.

From 2005 to 2009, a total of 23 caribou, 15 moose and 11 wolves were fitted with radio-collars recording locations every 4 h. Amongst these individuals, we used GPS data from 12 caribou, 12 moose and 7 wolves, which only included animals whose 100% minimum convex polygons containing their locations overlapped with at least one wolf for caribou and moose and with at least one caribou or one moose for wolves. For each species, the year can be divided into multiple biological seasons, characterized by behavioural differences between successive periods (Basille et al. 2013). For caribou and moose, we differentiated winter (December 28th – April 15th for caribou; October 12th – May 6th for moose) and summer (May 28th – September 17th for caribou; June 10th – October 11th for moose). We therefore used wolves' locations corresponding to the union of these periods (October 12th – May 6th for winter and May 28th – October 11th for summer).

Assessment of the Prey's Reaction to Predator Presence

The impact of prior wolf presence on the movements of caribou and moose was analysed using SSFs (Fortin et al. 2005) in winter and in summer. SSFs establish the relationship between animal movements and habitat features by comparing observed and random steps. To estimate SSFs, paths of each prey were broken down into steps, which correspond to the straight-line segment between successive locations at 4-h intervals. Each observed step was then paired with 20 random steps having the same starting point, but differing in length and/or direction. The lengths and turning angles of random steps were drawn from the empirical distributions comprised of the steps of all individuals of a given prey species collected during the year. The characteristics of pairs of observed and random steps were compared with conditional logistic regression, with the resulting SSF taking the general form:

  • display math(eqn 1)

where w(x) represents the SSF score for the real or random step described by the vector x of variables xi, and βi is the coefficient for xi. SSFs have the advantage of assessing the influence that both factorial and continuous variables can collectively have on the odds that a particular step is travelled, given local habitat features and the individual's previous travel direction. Steps with a higher score have higher odds of being chosen by an animal. Additional details on SSF assessment are provided in Fortin et al. (2005) and Forester, Im & Rathouz (2009).

To assess the impact of predation on selection of land cover type, the landscape was divided into eight classes, namely open areas (including burned areas, rocks, heath with and without lichen, and water), closed-canopy mature conifer, open conifer with lichen, open conifer, mixed/deciduous, regenerating cut, recent cut and road. A factorial variable with eight classes representing the different land cover types (coverType, eqn (eqn 2)) was then included in the model, along with a term of interaction between land cover type at the destination point and a predator presence index at the origin of the step (coverType × Pred, eqn (eqn 2)). The predator presence index (Pred) is a decaying function of both time and distance (see section below for details on the derivation of Pred). The model thus allows for prey to decrease their response to predators with time and the distance between them. Our analysis accounts for the predator presence index at the origin of the step because we wanted to assess whether the passage of a wolf would influence where the prey will be located 4 h later (i.e. the end point location of the step). To assess the impact of predators on movement characteristics of prey, an interaction term relating movement speed (i.e. step length/4-h step duration) with Pred (speed × Pred, eqn (eqn 2)) was also considered. Because prey tend to avoid areas of high predator density (e.g. Mech 1977; Lewis & Murray 1993), we accounted for this aspect by adding a term characterizing the difference in predator presence index between the origin and the end point of the step (diffPred, eqn (eqn 2)). The model also included altitude (Altitude, eqn (eqn 2)). We used a spline function on speed, with the 1st, 2nd and 3rd quartiles for the knots locations (Spline, eqn (eqn 2)) to avoid potential bias associated with empirical sampling (Forester, Im & Rathouz 2009).

  • display math(eqn 2)

Spatio-Temporal Association Between Wolves and their Prey

To test for the potential response of caribou and moose to wolves in their vicinity, we identified the most recent passage of a wolf within 15 km of their locations. To do so, at each 4-h time step, each 25 m × 25 m grid cell in the land cover map was assigned the time and date of the most recent passage of a wolf. For each caribou and moose location, we then verified whether the path of a radio-collared wolf crossed any of 24 rays of length 15 km oriented every 15°, pivoting around the location. The distance Di to this path and the time Ti elapsed since the passage of the wolf was recorded for each ray i (= 1…24) (maximum of 10 days), as illustrated in Fig. S1 (Supporting information). If a ray crossed multiple wolf paths, only the most recent one was recorded. For every caribou and moose location, 24 combinations of time and distance {(T1, D1),(T2, D2),…,(T24, D24)} were therefore recorded. We considered spatio-temporal limits of 15 km and 10 days, which largely exceeded the previous investigation of the literature (cf. Creel et al. 2005; Proffitt et al. 2009).

Impact of Predator Presence

For each location of caribou and moose, we estimated the predator presence index for each ray = 1, 2,…, 24 as a function of T and D: Predi(Ti, Di) = Predi,T(Ti) × Predi,D(Di). Assuming that the perception of predator presence, and therefore the response to predator, decreases with time and distance, four different discounting functions were tested for each variable T or D: exponential discounting (eqn (eqn 3)), hyperbolic discounting (eqn (eqn 4)), Gaussian discounting (eqn (eqn 5)) and sigmoidal discounting with a coefficient of 100, considered as a threshold function (eqn (eqn 6)). The predator presence index at a given location corresponds to the maximum value over the 24 rays: Pred(T, D) = maxi Predi(Ti, Di).

  • display math(eqn 3)
  • display math(eqn 4)
  • display math(eqn 5)
  • display math(eqn 6)

Discounting Rate Estimation

For a particular combination {kT, kD} of discounting rates, the level of empirical support received by the SSF was assessed for each season using the quasi-likelihood under independence criterion (QIC; Pan 2001). The QIC accounts for nonindependence between subsequent observations by being calculated while also considering independent clusters of observations (Craiu, Duchesne & Fortin 2008).

For each combination of discounting functions, the combination {kT, kD} leading to the lowest QIC was found using the Nelder and Mead's optimization algorithm (Nelder & Mead 1965), a nonlinear optimization technique that can minimize an objective function in a many-dimensional space. Because this method can converge towards a local optimum, the estimation of {kT, kD} was performed 100 times for each of the 16 combinations of discounting functions, with a different random initialization for each replication, resulting in 1600 estimations for each season. To avoid biologically unrealistic results, only solutions ensuring that the value T5% (corresponding to PredT (T) = 0·05) ∈ [0, 10] days and that the value D5% (corresponding to PredD (D) = 0·05) ∈ [0, 10 000] metres were retained. The predator presence index then corresponded to the model yielding the lowest QIC value.

Statistical Analysis

Our interest was to determine the response of prey to the recent passage of a wolf. We therefore conducted our SSF analysis only on locations for which at least one 15 km ray intersected a wolf's path <10 days old, which corresponded to 19% (4563) of the locations in winter and 24% (5647) of the locations in summer for caribou, 30% (6950) of the locations in winter and 25% (3011) of the locations in summer for moose.

Because of the interaction terms (eqn (eqn 2)), all covariates are not independent when considering all the different land cover types, and the variance inflation factor (VIF) was >10 for some of them, depending on season. The interaction terms between land cover types and the predator presence index were therefore selected before applying the optimization algorithm for each season, in a backward stepwise fashion, so that the VIF would be <10 for all covariates, except for the splines, which are by definition colinear. In the end, the VIF was <10 for all covariates of our final models (Table S1, Supporting information), thereby allowing for valid statistical inference (Chatterjee, Hadi & Price 2000).

Because caribou and moose locations were measured every 4 h, successive steps were not independent. Robust standard errors of SSF parameters can be estimated using a robust sandwich estimate of the covariance matrix (Wei, Lin & Weissfeld 1989). This approach requires partitioning the data into clusters of autocorrelated steps, each cluster being independent from the others (Wei, Lin & Weissfeld 1989; Hardin & Hilbe 2003). Our focus on species interaction led to the removal of a number of locations, which naturally created independent clusters of consecutive steps. Autocorrelation and partial autocorrelation analyses of the deviance residuals showed that autocorrelation disappeared beyond lag 2 for all animals, for both seasons. We thus ensured that all clusters were independent by removing locations so that the last location of a given cluster was separated by at least two steps (8 h) from the first location of the next cluster. A total of 0·5% of the data were removed for caribou in winter and 0·3% of the data for caribou in summer and for moose in winter and summer. The clustering technic was further used to estimate the SSFs, while accounting for the nonindependence amongst individuals of a given species. To assess whether movements were independent amongst radio-collared individuals, we estimated the distance between simultaneous locations of all individuals. Individuals were considered independent if they were separated by more than 100 m (as Fortin et al. 2009). When two individuals were closer than 100 m, the two clusters to which these locations pertained were merged into a single cluster. Overall, we ended up with a total of 223 clusters in winter and 452 clusters in summer for caribou, and 159 clusters in winter and 109 clusters in summer for moose.

We used 5-fold cross-validation for case–control design to evaluate model robustness (see Fortin et al. 2009 for details). This approach yields an average (n = 100) Spearman rank correlation (rS) and its associated 95% confidence intervals. Robust models have high rS.

Probability that a Wolf Returns to a Previously Visited Area

We evaluated the probability that a radio-collared wolf returned to a previously visited area once it had left the area. To do so, we divided the areas occupied by the radio-collared wolf into quadrats of 180 m × 180 m, 500 m × 500 m and 1500 m × 1500 m, and for each wolf location, we recorded the time elapsed between the departure and the return to a given quadrat. If wolves did not return to the quadrat, we measured the time elapsed between the time of the last location within a quadrat and the end of the season, and these values were considered as censored data. We used a mixed-effects model of Cox proportional hazards (CPH) to assess how the risk that a wolf does not return to a quadrat varies over time, considering ‘individual’ as a mixed effect to take into account potential behavioural plasticity amongst animals (Natarajan & McCulloch 2004). CPH analysis uses a hazard function h(t) to model the relative ‘risk’ or ‘hazard’ h that an individual does not return to a quadrat. The probability of returning to a previously visited quadrat during the season is then Preturn(t) = 1 − h(t). Because quadrats are fixed, locations located near the edge of a quadrat are likely to result in successive jumps out and back into the quadrat for small displacements. To account for this aspect, we performed the same analysis by considering only the events for which individuals had travelled at least 1 km once they had left the quadrat, to assess the probability of return once the wolf is gone from the general area.

Results

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

For both caribou and moose, in every season, the optimization algorithm identified a function for the predator presence index leading to a model that explained the data better than a parsimonious model that did not consider prior wolf passage (Table 1).

Table 1. Combinations of discounting functions and corresponding parameters giving the lowest quasi-likelihood under independence criterion (QIC) over all simulations for winter and summer for caribou and moose in the Côte-Nord region of Québec, Canada. ∆QIC is the difference of QIC with a baseline model that did not consider wolf prior passage (with no interaction term nor diffPred) and w is the weight of the complete model, with respect to the baseline model. The half-life and 5% values are the values of time (or distance) for which Pred is equal to 0·5 and 0·05, respectively, if the distance (or time) is null
SpeciesSeasonTime parametersDistance parametersQICΔQIC w
CaribouWinterGaussianGaussian27032·7112·091·00
Half-life: 112 hHalf-life: 86 m
5%: 233 h5%: 179 m
SummerSigmoidalSigmoidal33826·6419·731·00
Threshold: 32 hThreshold: 4688 m
MooseWinterSigmoidalExponential41612·761·270·66
Threshold: 162 hHalf-life: 68 m
 5%: 296 m
SummerGaussianGaussian17160·989·430·99
Half-life: 114 hHalf-life: 333 m
5%: 236 h5%: 691 m

Response of Caribou to the Recent Passage of Wolves

Caribou responded differently to the recent passage of wolves during winter and summer. During winter, the impact of prior wolf passage on caribou movements decreased continuously with time and distance (Table 1, Fig. 1). Thus, the impact of wolves on caribou movements disappeared more rapidly as the distance from the wolf path increased (Fig 1a,b). Further, the impact on caribou movements gradually disappeared for close distances as the time since the wolf's passage increased (Fig 1c,d). In contrast, during summer, the impact of wolf passage on caribou movements disappeared abruptly after 1·5 days, but this impact extended over 4·7 km around the wolf's path (Table 1, Fig. 2), that is, at shorter temporal scale and larger spatial scale than in winter.

image

Figure 1. Selection coefficients of the land cover types for which 90% confidence intervals with (solid lines) and without (dotted lines) interaction with Pred excluded 0 for caribou in winter as a function of time for (a) D = 10 m and (b) D = 100 m and as a function of distance for (c) T = 1 day and (d) T = 5 days. Confidence intervals are presented in finer lines.

Download figure to PowerPoint

image

Figure 2. Selection coefficients of the land cover types for which 90% confidence intervals with (solid lines) and without (dotted lines) interaction with Pred excluded 0 for caribou in summer as (a) a function of time for D > 4·7 km m and (b) as a function of distance for T > 32 h. Confidence intervals are presented in finer lines.

Download figure to PowerPoint

During winter, in the absence of impact of prior wolf presence, caribou selected open conifer stands with lichen and avoided closed-canopy mature conifer stands, mixed and deciduous forests, recent cuts and roads, with respect to open conifer stands without lichen (Table 2), which is the most frequent land cover type. They also selected areas of relatively high altitude. Caribou became increasingly likely to end their step in open conifer stands with lichen following the recent passage of a wolf in their vicinity, while they became less likely to end their step in open areas (Table 2; Fig. 1a–d). The passage of a wolf thus creates a highly dynamic landscape of fear whose shape was linear following the wolf's path (Fig. 3), due to the rapid decaying of the Gaussian function for distance, but the rather long temporal scale (Table 1). The passage of a wolf in a given area temporarily increased differences of selection of the different land cover types in the close vicinity of the wolf's path, mainly by polarizing selection between the selection of stands comprised of open conifer with lichen and the avoidance of open areas (Fig 3a–d).

Table 2. Coefficients (β), standard error (SE) and 95% confidence intervals (CI) for the step selection functions of caribou in winter and summer. The reference land cover type was open-canopy mature conifer forest; diffPred is the difference of index of impact of predator presence (Pred) between the end and the origin of the step (a positive coefficient for diffPred means that the individual moves towards a higher index). Altitude is in metres and speed in metres per hour
VariableWinterSummer
β SE95% CI β SE95% CI
  1. Models were robust to cross-validation, with an observed rS of 0·94 (95% CI: 0·87–0·98) in winter and of 0·91 (0·90–0·99) in summer, which exceeded the rS of −0·01 (−0·42 to 0·39) in winter and of 0·02 (−0·47 to 0·52) in summer expected from randomized data.

  2. a

    Coefficients for which the 95% confidence intervals excluded zero.

  3. b

    Coefficients for which the 90% confidence intervals excluded zero.

Open area−0·040·14(−0·30 to 0·23)−0·860·11(−1·08 to −0·65) a
Closed-canopy mature conifer−0·210·08(−0·36 to −0·06) a−0·120·06(−0·23 to −0·01) a
Open conifer with lichen0·440·05(0·34 to 0·54) a0·120·06(−0·00 to 0·24)b
Mixed/Deciduous−0·160·08(−0·32 to 0·01)b−0·550·09(−0·72 to −0·38) a
Regenerating cuts   −2·090·56(−3·19 to −0·98) a
Recent cuts−1·740·37(−2·47 to −1·01) a−1·300·36(−2·01 to −0·60) a
Roads−1·300·65(−2·57 to −0·03) a−1·610·71(−3·01 to −0·22) a
Altitude (×102)0·460·12(0·23 to 0·69) a0·430·05(0·32 to 0·54) a
diffPred0·760·27(0·24 to 1·29) a−0·590·23(−1·04 to −0·15) a
Spline1 (×102)−2·571·06(−4·64 to −0·49) a0·950·43(0·11 to 1·79) a
Spline2 (×102)2·851·21(0·47 to 5·23) a−0·290·50(−1·26 to 0·68)
Spline3 (×102)−2·430·62(−3·65 to −1·22) a−0·310·26(−0·82 to 0·20)
Spline4 (×102)2·110·40(1·33 to 2·89) a−0·370·09(−0·55 to −0·20) a
Open area × Pred−37·5217·53(−71·89 to −3·16) a0·620·35(−0·06 to 1·31)b
Closed-canopy mature conifer × Pred−1·641·83(−5·22 to 1·95)0·240·17(−0·09 to 0·58)
Open conifer with lichen × Pred1·550·61(0·36 to 2·74)a0·360·25(−0·12 to 0·85)
Mixed/Deciduous × Pred−1·011·71(−4·37 to 2·35)0·690·38(−0·06 to 1·44)b
Speed × Pred (×103)   0·440·24(−0·02 to 0·90)b
image

Figure 3. Temporal changes in the relative probability of occurrence of caribou before (first row), during (second row) and after passage (third and fourth row) of a wolf, in a 20 × 20 km portion of the study area in the Côte-Nord region, Québec (Canada). The red line represents the wolf's path. The grey scale represents the relative probability of occurrence of a caribou in the landscape. Only coefficients for which 90% confidence intervals excluded 0 were taken into account.

Download figure to PowerPoint

During summer, in the absence of impact of prior wolf presence, caribou selected open conifer stands with lichen and avoided open areas, closed-canopy mature conifer stands, mixed and deciduous forests, and all anthropogenic disturbances, with respect to open conifer stands without lichen. They also selected areas of relatively high altitude (Table 2). Caribou reacted to prior wolf passage by moving away from predator's paths, as indicated by the negative coefficient for diffPred (Table 2). The probability of occurrence of caribou in open areas and mixed and deciduous forests was higher if a wolf had been present within 4·7 km, during the past 32 h (Table 1; Fig. 2). The passage of a wolf creates a landscape of fear with abrupt delimitations due to the combination of sigmoidal functions and changed dynamically at broader spatial and finer temporal scales in summer than in winter (Fig 3e–h). The passage of a wolf in a given area created a temporarily more uniform selection than in the absence of wolves, because open areas, closed-canopy conifer stands, and mixed and deciduous forests become selected in a similar fashion.

Response of Moose to the Recent Passage of Wolves

Moose responded differently to the recent passage of wolves during winter and summer. During winter, the impact of prior wolf passage on caribou movements disappeared abruptly after c. 7 days, but decreased continuously and rapidly with distance (Table 1). During summer, the impact of prior wolf passage on caribou movements decreased continuously with time and distance, at slightly at shorter temporal scale and larger spatial scale. Thus, the impact of wolves on moose movements disappeared more rapidly as the distance from the wolf path increased (Fig 4a,b). Further, the impact on moose movements gradually disappeared for close distances as the time since the wolf's passage increased (Fig 4c,d).

image

Figure 4. Selection coefficients of the land cover types for which 90% confidence intervals with (solid lines) and without (dotted lines) interaction with Pred excluded 0 for moose in summer as a function of time for (a) D = 10 m and (b) D = 400 m, and as a function of distance for (c) T = 1 day and (d) T = 8 days. Confidence intervals are presented in finer lines.

Download figure to PowerPoint

During winter, in absence of the impact of prior wolf presence, moose selected mixed and deciduous forests and avoided closed-canopy mature conifer and open conifer stands with lichen, relatively to open conifer stands without lichen. They also selected areas of relatively high altitude (Table 3). Moose increased their speed following the recent passage of a wolf in their vicinity and were more likely to move away than towards recent wolf's paths (see diffPred, Table 3). However, they did not modify their selection of land cover types. The passage of a wolf creates a highly dynamic landscape of fear, by temporarily decreasing their selection for the area around the wolf's path in a linear fashion (Fig 5a–d) because the predator presence index is modelled by an exponential discounting function decreasing rapidly for distance and a sigmoidal function with a threshold of c. 7 days for time (Table 1).

Table 3. Coefficients (β), standard error (SE) and 95% confidence intervals (CI) for the step selection functions of moose in winter and summer. The reference land cover type was open-canopy mature conifer forest; diffPred is the difference of index of impact of predator presence (Pred) between the end and the origin of the step (a positive coefficient for diffPred means that the individual moves towards a higher index). Altitude is in metres and speed in metres per hour
VariableWinterSummer
β SE95% CI β SE95% CI
  1. Models were robust to cross-validation, with an observed rS of 0·81 (95% CI: 0·69–0·91) in winter and of 0·95 (0·82–0·97) in summer, which exceeded the rS of – 0·19 (−0·85 to 0·43) in summer and of 0·03 (−0·30 to 0·49) in winter expected from randomized data.

  2. a

    Coefficients for which the 95% confidence intervals excluded zero.

  3. b

    Coefficients for which the 90% confidence intervals excluded zero.

Open area−0·040·11(−0·26 to 0·18)−0·280·09(−0·45 to −0·11) a
Closed-canopy mature conifer−0·100·04(−0·18 to −0·02)−0·190·07(−0·33 to −0·05) a
Open conifer with lichen−0·370·11(−0·58 to −0·16) a−0·340·14(−0·61 to −0·06) a
Mixed/Deciduous0·240·06(0·12 to 0·36) a0·130·07(−0·02 to 0·27) b
Regenerating cuts0·200·28(−0·36 to 0·76)−0·380·29(−0·94 to 0·18)
Recent cuts0·070·15(−0·22 to 0·36)−0·080·15(−0·38 to 0·22)
Roads   −0·820·87(−2·52 to 0·89)
Altitude (×102)0·230·12(0·00 to 0·46) a−0·040·10(−0·22 to 0·15)
diffPred1·210·45(0·33 to 2·09) a0·650·26(0·14 to 1·16) a
Spline1 (×102)−0·080·68(−1·41 to 1·25)7·110·64(5·85 to 8·37) a
Spline2 (×102)0·000·01(−0·01 to 0·02)−4·020·76(−5·51 to −2·53) a
Spline3 (×102)−0·010·00(−0·02 to −0·00) a−2·630·52(−3·65 to −1·62) a
Spline4 (×102)0·000·00(−0·00 to 0·01)−0·480·23(−0·92 to −0·04) a
Open area × Pred−0·050·32(−0·67 to 0·56)   
Closed-canopy mature conifer × Pred0·160·22(−0·27 to 0·58)0·090·36(−0·62 to 0·79)
Open conifer with lichen × Pred0·650·49(−0·31 to 1·61)−13·615·08(−23·56 to −3·65) a
Mixed/Deciduous × Pred   0·360·36(−0·36 to 1·07)
Recent cuts × Pred   0·490·42(−0·33 to 1·31)
Speed × Pred (×102)0·570·15(0·28 to 0·85) a−0·020·07(−0·17 to 0·13)
image

Figure 5. Temporal changes in the spatial relative probability of occurrence of moose before (first row), during (second row) and after passage (third and fourth row) of a wolf, in a 20 × 20 km portion of the study area in the Côte-Nord region, Québec (Canada). The red line represents the wolf's path. The grey scale represents the relative probability of occurrence of a moose in the landscape. Only coefficients for which 90% confidence intervals excluded 0 were taken into account.

Download figure to PowerPoint

During summer, in the absence of impact of prior wolf presence, moose selected mixed and deciduous forests and avoided open areas, closed-canopy mature conifer and open conifer stands with lichen, relative to open conifer stands without lichen (Table 3). They became less likely to end their step in open conifer stands with lichen as distance to and time since prior wolf passage decreased (Table 3; Fig. 4). Moose were also more likely to move towards than away from recent wolf's paths (see diffPred, Table 3). The passage of a wolf creates a highly dynamic landscape of fear, especially by temporarily increasing the avoidance of open conifer stands with lichen (Fig. 5e–h).

Probability that Wolves Return to a Given Area

For wolves, we found that the probability of returning to a quadrat was slightly higher in winter than in summer (Fig 6a,b). For the six individuals for which data were available for both winter and summer, we computed the area of the 95% minimum convex polygon, divided by the number of locations, for each season and for each year. For each individual, we then calculated the median of these values over the years, for winter and summer, and computed the ratio between these two medians. The ratio was <1 for four individuals (values = 0·35, 0·66, 0·45, 0·28), c. 1 for one individual (value = 1·06) and >1 for one individual (value = 13·30), meaning that wolves tend to range over larger areas in summer than in winter. Moreover, the ratio of the median of the step length in winter and in summer for each of these individuals was always <1 (values = 0·20, 0·08, 0·28, 1·02, 0·11, 0·04), meaning that wolves moved more in summer than in winter. These two factors can explain the observed result that wolves tend to come back less frequently to each area of their territories in summer than in winter when considering all 4-h steps. On the other hand, when we only considered events when wolves had left the area (i.e. leave the quadrat by more than 1 km), we observed the opposite situation (Fig 6c,d).

image

Figure 6. Probability Preturn(t) of wolves returning to a 180 m × 180 m, 500 m × 500 m and 1500 m × 1500 m quadrats with respect to time elapsed since last visit, for all locations (a) in winter and (b) in summer, and when individuals had left the quadrat by more than 1 km (c) in winter and (d) in summer. 95% confidence intervals are represented in grey.

Download figure to PowerPoint

Discussion

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

We applied optimization techniques to movement analysis to demonstrate that both caribou and moose respond to the passage of wolves by displaying ephemeral responses, the nature of which differs importantly depending on species and season. Our analysis, however, does not identify the exact distance and time during which prey is capable of detecting the passage of a wolf. Given the discrete nature of our field data (locations recorded at 4 h intervals), the cues triggering the response of caribou and moose might have been detected at much closer distances and finer temporal scales than reported by our analysis, especially in summer when they travel at a faster speed (Basille et al. 2013). What our analysis does reveal is that, once large herbivores acquire a signal of predator presence, their movement and habitat selection can be altered over rather large distances and long time-scales.

Both caribou and moose showed seasonal differences in their response to the passage of wolves. Prior passage of wolves had an impact on both species at longer temporal scales, but finer spatial scales, in winter than in summer. First, caribou and moose could roam over larger areas during the 4-h relocation interval in summer than in winter because of their faster travel speed in the absence of snow (Basille et al. 2013). Therefore, they should be able to detect the presence of wolves at farther distances in summer than winter during the relocation interval, especially given that wolves also tend to range over larger areas in summer than in winter. Second, the vulnerability of prey increases with snow depth (Wikenros et al. 2009), and caribou and moose might get fitness benefits by having more acute responses to predation risk in winter than in summer.

Seasonal differences in the influence of wolf were stronger on movement decisions of caribou than moose (Table 2). This difference is somewhat surprising given that wolves generally focus their hunt on moose than caribou (Hayes et al. 2000; Joly & Patterson 2003). The difference might reflect a higher risk of mortality for caribou than moose following an encounter with wolves (Mech & Boitani 2010). Alternatively, radio-collared moose might have crossed fewer wolf paths compared with collared caribou because moose travelled less than caribou during both seasons.

We detected four types of behaviours following the passage of a predator, within the distance and time resulting from the optimization algorithm: lack of response in habitat selection, increased selection of safe land cover types, decreased selection of risky land cover types, and increased selection of food-rich areas.

Both species displayed a lack of response for a number of land cover types. For example, caribou avoided closed-canopy mature conifer in winter and in summer, as observed by Courbin et al. (2009), regardless of the whereabouts of wolves. Moose selected mixed and deciduous forests, regardless of the season and of the changes in wolf spatial dynamics. Mixed and deciduous forests are food-rich areas for moose all year long (Sæther & Andersen 1990). Although these stands are also selected by wolves (Courbin et al. 2009), they are organized in small patches in the study area, a patchiness that might reduce the risk of co-occurrence of the two species, even if both species are making selective use of these stands. This is consistent with other studies suggesting that a complex habitat structure reduces the predation rate (Warfe & Barmuta 2004) and favours the stability of predator-prey systems (Hauzy et al. 2010).

Caribou also increased their selection of safe land cover types in response to prior passage of wolves. They generally avoided open areas in summer, probably because they are areas of low forage density. On the other hand, they increased their selection of this land cover type when wolves had been within 4·7 km during the past 1·5 days. This reaction might be somewhat similar to caribou selecting lakes in winter (Fortin et al. 2008), which is considered an anti-predator response that would allow them to detect predator from far distances. Given that wolves tend to select open areas in summer (Courbin et al. 2009), the adaptive value of this decision by caribou remains unclear. Predation risk is a function of the probability of encounter with a predator, the probability that an attack takes place given an encounter and the probability of dying from that attack (DeCesare 2012). To obtain a net reduction in predation risk from using open areas, the decrease in detection time would have to outweigh the potential increase in encounter probability with the predator.

Moose avoided open conifer stands with lichen in summer following the passage of a wolf. Open conifer stands with lichen are never a key land cover type for moose, presumably because they do not contain as much food as other land cover types (Crête & Courtois 1997) and because they are selected by wolves (Courbin et al. 2009). Accordingly, the avoidance of open conifer stands with lichen became even stronger in areas that wolves had visited recently. This behavioural response is consistent with an anti-predator response to temporary elevated risk. Moving away from predators and selecting land cover types avoided by them have obvious advantages for prey in terms of the predator–prey space race, because it should increase their chances of being spatially separated (Sih 2005).

Finally, caribou increased their selection of open conifer stands with lichen in winter following the passage of wolves and moved towards wolf paths. As we indicated previously, wolves select open conifer stands with lichen during this season (Courbin et al. 2009; Houle et al. 2010). The long-term response of caribou is to reduce their selection of those stands in areas of high wolf density (Labbé 2012). In this context, our analyses appear to indicate that caribou could benefit from information on the recent passage of wolves to assess the importance of the threat and then adjust their use of food-rich areas based on their assessment of the current risk. Predator inspection has been reported for a broad range of taxa (Dugatkin & Godin 1992; FitzGibbon 1994; Brown & Dreier 2002; Nocera, Taylor & Ratcliffe 2008), and although approaching the predator (or areas where a predator had been present) can increase the risk of an attack, it can also provide information that ultimately could yield a net fitness gain. An important piece of information appears to be whether or not predators are still in the area. We found that once wolves had left an area, the probability that they would return was relatively low for ∼5 days, at which point, the probability of wolves returning remained constant over time (Fig. 6c) and therefore relatively high and poorly predictable. The recent passages of wolves thus seem to inform caribou on the distribution of their predator, thereby providing them with an opportunity to lower their anti-predator behaviour by making greater use of the richest lichen patches. As the information becomes outdated, anti-predator behaviour becomes more acute again. A puzzling aspect of our findings is how prey learns in the first place that they can use an area because wolves have left the areas. Perhaps the observed avoidance of a given area needs to be frequently reinforced by cues of predator presence; otherwise, prey returns rapidly to the area. This hypothesis requires further investigation. Overall, we suggest that the decrease in the selection of lichen-rich patches by caribou in winter indicates that the lower level of selection for open conifer stands with lichen reflects a long-term trade-off between food acquisition and predation risk and might be seen as a chronic anti-predator response.

Few studies consider that prey and predators are both in motion in their analyses (Lima 2002; Sih 2005; Hammond, Luttbeg & Sih 2007). Our study accounts for such dynamic distributions, and as such, it offers a rare assessment of how a combination of short- and long-term anti-predator responses to predation risk drives the spatial dynamics of large herbivores. These adaptive responses translate into dynamic patterns of habitat selection and movements, which in turn yield highly heterogeneous landscapes of fear (Figs 3 and 5). The results have implications on food web properties. For example, trophic cascades can result from long-term avoidance of food-rich habitats that are subject to high predation risk (e.g. Schmitz, Krivan & Ovadia 2004; Fortin et al. 2005), a response that was also observed for caribou in winter (Labbé 2012). However, caribou increased their selection of open conifer stands with lichen for a short period of time when wolves had recently left the area, which should reduce the potential for a trophic cascade. Our study outlines the complexity of the temporal scales over which prey respond to the passage of a predator. These scales vary amongst species and between seasons for a given prey. Predictions of food web dynamics should benefit by accounting for such variations in anti-predator behaviours.

Acknowledgements

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

This study was funded by the Natural Sciences and Engineering Research Council of Canada (NSERC) – University Laval Industrial Research Chair in Silviculture and Wildlife. We thank M. Basille for his precious insights on this study. We also thank B. Patterson for sharing his knowledge on wolf behaviour.

References

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information
  • Basille, M., Fortin, D., Dussault, C., Ouellet, J.P. & Courtois, R. (2013) Ecologically based definition of seasons clarifies predator-prey interactions. Ecography, 36, 220229.
  • Brown, J. (1999) Vigilance, patch use and habitat selection: foraging under predation risk. Evolutionary Ecology Research, 1, 4971.
  • Brown, G. & Dreier, V. (2002) Predator inspection behaviour and attack cone avoidance in a characin fish: the effects of predator diet and prey experience. Animal Behaviour, 63, 11751181.
  • Brown, J. & Kotler, B. (2004) Hazardous duty pay and the foraging cost of predation. Ecology Letters, 7, 9991014.
  • Chatterjee, S., Hadi, A. & Price, B. (2000) The Use of Regression Analysis by Example. John Wiley and Sons, New York, USA.
  • COSEWIC (2006) COSEWIC assessment and update status report on the woodland caribou Rangifer tarandus caribou in Canada. Tech. rep, Committee on the Status of Endangered Wildlife in Canada, Ottawa.
  • Courbin, N., Fortin, D., Dussault, C. & Courtois, R. (2009) Landscape management for woodland caribou: the protection of forest blocks influences wolf-caribou co-occurrence. Landscape Ecology, 24, 13751388.
  • Courtois, R., Ouellet, J., Breton, L., Gingras, A. & Dussault, C. (2007) Effects of forest disturbance on density, space use, and mortality of woodland caribou. Ecoscience, 14, 491498.
  • Craiu, R., Duchesne, T. & Fortin, D. (2008) Inference methods for the conditional logistic regression model with longitudinal data. Biometrical Journal, 50, 97109.
  • Creel, S., Winnie, J. Jr, Maxwell, B., Hamlin, K. & Creel, M. (2005) Elk alter habitat selection as an antipredator response to wolves. Ecology, 86, 33873397.
  • Creel, S., Winnie, J. Jr, Christianson, D. & Liley, S. (2008) Time and space in general models of antipredator response: tests with wolves and elk. Animal Behaviour, 76, 11391146.
  • Crête, M. & Courtois, R. (1997) Limiting factors might obscure population regulation of moose (Cervidae: Alces alces) in unproductive boreal forests. Journal of Zoology, 242, 765781.
  • DeCesare, N. (2012) Separating spatial search and efficiency rates as components of predation risk. Proceedings of the Royal Society B: Biological Sciences, 279, 46264633.
  • Dugatkin, L. & Godin, J. (1992) Predator inspection, shoaling and foraging under predation hazard in the trinidadian guppy, poecilia reticulata. Environmental Biology of Fishes, 34, 265276.
  • Fischhoff, I., Sundaresan, S., Cordingley, J. & Rubenstein, D. (2007) Habitat use and movements of plains zebra (Equus burchelli) in response to predation danger from lions. Behavioral Ecology, 18, 725729.
  • FitzGibbon, C. (1994) The costs and benefits of predator inspection behaviour in thomson's gazelles. Behavioral Ecology and Sociobiology, 34, 139148.
  • Forester, J., Im, H. & Rathouz, P. (2009) Accounting for animal movement in estimation of resource selection functions: sampling and data analysis. Ecology, 90, 35543565.
  • Fortin, D. & Fortin, M. (2009) Group-size-dependent association between food profitability, predation risk and distribution of free-ranging bison. Animal Behaviour, 78, 887892.
  • Fortin, D., Beyer, H., Boyce, M., Smith, D., Duchesne, T. & Mao, J. (2005) Wolves influence elk movements: behavior shapes a trophic cascade in yellowstone national park. Ecology, 86, 13201330.
  • Fortin, D., Courtois, R., Etcheverry, P., Dussault, C. & Gingras, A. (2008) Winter selection of landscapes by woodland caribou: behavioural response to geographical gradients in habitat attributes. Journal of Applied Ecology, 45, 13921400.
  • Fortin, D., Fortin, M.E., Beyer, H.L., Duchesne, T., Courant, S. & Dancose, K. (2009) Group-size-mediated habitat selection and group fusion-fission dynamics of bison under predation risk. Ecology, 90, 24802490.
  • Gingras, A., Audy, A. & Courtois, R. (1989) Inventaire aérien de l'orignal dans la zone de chasse 19 à l'hiver 1987–1988. Ministère du Loisir, de la Chasse et de la Pêche. Direction de la gestion des espèces et des habitats, Québec.
  • Hammond, J., Luttbeg, B. & Sih, A. (2007) Predator and prey space use: dragonflies and tadpoles in an interactive game. Ecology, 88, 15251535.
  • Hardin, J. & Hilbe, J. (2003) General Estimating Equations. Chapman & Hall, Boca Raton, Florida.
  • Hauzy, C., Tully, T., Spataro, T., Paul, G. & Arditi, R. (2010) Spatial heterogeneity and functional response: an experiment in microcosms with varying obstacle densities. Oecologia, 163, 625636.
  • Hayes, R.D., Baer, A.M., Wotschikowsky, U. & Harestad, A.S. (2000) Kill rate by wolves on moose in the Yukon. Canadian Journal of Zoology, 78, 4959.
  • Houle, M., Fortin, D., Dussault, C., Courtois, R. & Ouellet, J. (2010) Cumulative effects of forestry on habitat use by gray wolf (Canis lupus) in the boreal forest. Landscape Ecology, 25, 419433.
  • Joly, D.O. & Patterson, B.R. (2003) Use of selection indices to model the functional response of predators. Ecology, 84, 16351639.
  • Labbé, M.C. (2012) Jeu prédateur-proie entre le caribou forestier et le loup gris:un effet saute-mouton spatialement structuré. Master's thesis, Département de biologie, Université Laval, Quebec, Canada.
  • Laundré, J., Hernández, L. & Ripple, W. (2010) The landscape of fear: ecological implications of being afraid. Open Ecology Journal, 3, 17.
  • Lendrem, D. (1983) Predation risk and vigilance in the blue tit (parus caeruleus). Behavioral Ecology and Sociobiology, 14, 913.
  • Lewis, M. & Murray, J. (1993) Modelling territoriality and wolf-deer interactions. Nature, 366, 738740.
  • Liley, S. & Creel, S. (2008) What best explains vigilance in elk: characteristics of prey, predators, or the environment? Behavioral Ecology, 19, 245.
  • Lima, S. (1998) Stress and decision making under the risk of predation: recent developments from behavioral, reproductive, and ecological perspectives. Advances in the Study of Behavior, 27, 215290.
  • Lima, S. (2002) Putting predators back into behavioral predator-prey interactions. Trends in Ecology & Evolution, 17, 7075.
  • Lima, S. & Bednekoff, P. (1999) Temporal variation in danger drives antipredator behavior: the predation risk allocation hypothesis. The American Naturalist, 153, 649659.
  • Lima, S. & Dill, L. (1990) Behavioral decisions made under the risk of predation: a review and prospectus. Canadian Journal of Zoology, 68, 619640.
  • Lima, S. & Zollner, P. (1996) Towards a behavioral ecology of ecological landscapes. Trends in Ecology & Evolution, 11, 131135.
  • Mech, L. (1977) Wolf-pack buffer zones as prey reservoirs. Science, 198, 320321.
  • Mech, L.D. & Boitani, L. (eds) (2010) Wolves: Behavior, Ecology, and Conservation. University of Chicago Press, Chicago.
  • Natarajan, R. & McCulloch, C. (2004) Modeling heterogeneity in nest survival data. Biometrics, 55, 553559.
  • Nelder, J.A. & Mead, R. (1965) A simplex algorithm for function minimization. Computer Journal, 7, 308313.
  • Nocera, J., Taylor, P. & Ratcliffe, L. (2008) Inspection of mob-calls as sources of predator information: response of migrant and resident birds in the neotropics. Behavioral Ecology and Sociobiology, 62, 17691777.
  • Pan, W. (2001) Akaike's information criterion in generalized estimating equations. Biometrics, 57, 120125.
  • Périquet, S., Todd-Jones, L., Valeix, M., Stapelkamp, B., Elliot, N., Wijers, M. et al. (2012) Influence of immediate predation risk by lions on the vigilance of prey of different body size. Behavioral Ecology, 23, 970976.
  • Proffitt, K., Grigg, J., Hamlin, K. & Garrott, R. (2009) Contrasting effects of wolves and human hunters on elk behavioral responses to predation risk. The Journal of Wildlife Management, 73, 345356.
  • Sæther, B. & Andersen, R. (1990) Resource limitation in a generalist herbivore, the moose Alces alces: ecological constraints on behavioural decisions. Canadian Journal of Zoology, 68, 993999.
  • Schmitz, O., Krivan, V. & Ovadia, O. (2004) Trophic cascades: the primacy of trait-mediated indirect interactions. Ecology Letters, 7, 153163.
  • Sih, A. (2005) Predator-prey space use as an emergent outcome of a behavioral response race. Ecology of Predator–Prey Interactions (eds P. Barbosa & I. Castellanos), pp. 240255. Oxford University Press, New York, New York, USA.
  • Valeix, M., Loveridge, A.J., Chamaille ́-Jammes, S., Davidson, Z., Murindagomo, F., Fritz, H. et al. (2009) Behavioral adjustments of african herbivores to predation risk by lions: spatiotemporal variations influence habitat use. Ecology, 90, 2330.
  • Warfe, D. & Barmuta, L. (2004) Habitat structural complexity mediates the foraging success of multiple predator species. Oecologia, 141, 171178.
  • Wei, L., Lin, D. & Weissfeld, L. (1989) Regression analysis of multivariate incomplete failure time data by modeling marginal distributions. Journal of the American Statistical Association, 84, 10651073.
  • Wikenros, C., Sand, H., Wabakken, P., Liberg, O. & Pedersen, H.C. (2009) Wolf predation on moose and roe deer: chase distances and outcome of encounters. Acta Theriologica, 54, 207218.
  • Wirsing, A., Cameron, K. & Heithaus, M. (2010) Spatial responses to predators vary with prey escape mode. Animal Behaviour, 79, 531537.
  • Wittmer, H., Mclellan, B., Serrouya, R. & Apps, C. (2007) Changes in landscape composition influence the decline of a threatened woodland caribou population. Ecology, 76, 568579.

Supporting Information

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information
FilenameFormatSizeDescription
jane12108-sup-0001-FigS1-TableS1.docWord document82K

Fig. S1. Recording of the distance from a previous path and of the time elapsed since the path was created.

Table S1. Variance Inflation Factors (VIF) for the different variables retained in the Step Selection Functions for each species for each season.

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.