Introduction
Large herbivores, like most prey species, make substantial investments in avoiding predation risk (Tolon et al., 2009). Risk of predation influences large herbivore habitat selection, grouping dynamics, and anti-predator behaviours (Hebblewhite et al., 2006; Christianson & Creel, 2010). As a result, large herbivore prey are often faced with making trade-offs in pursuing foraging opportunities while avoiding areas of high predation risk (Creel et al., 2005; Hebblewhite & Merrill, 2009).
Lima & Dill (1990) established a conceptual model of predation risk by identifying two fundamental components of Holling’s (1959) disk equation for the risk of a prey being killed per unit time:
P (death) = 1 - exp(-αd T) eqn 1
where α is the probability of encounter and d is the probability of death given an encounter during time (T). Their approach considers the two main stages of predation and highlights the conditional nature of mortality on attacks. However, it does not explicitly account for how predation risk may vary spatially.
Because spatial data for predator and prey are increasingly available, predation risk to prey has been related to a predator’s abundance, occurrence, intensity of use or resource selection (Theuerkauf & Rouys, 2008; Thaker et al., 2011; Moll et al., 2017). For example, White et al. (2009) related wolf (Canis lupus ) density as a metric of predation risk to changes in elk (Cervus canadensis ) nutrition in Yellowstone National Park during and after wolf recolonization, whereas predation risk from wolves and bears was estimated for a range of prey using RSFs (Gustine et al., 2006). Hebblewhite & Merrill (2007) combined these approaches by weighting RSFs of wolves by their spatial abundance to reflect the importance of the numeric response in predation risk. Although commonly used, these metrics ignore a key component of predation — the probability of death given an encounter (Lima & Dill, 1990). Encounters, however, are extremely difficult to observe directly, even indirectly (e.g ., Eriksen et al., 2009; Whittington et al., 2011). As a result, studies have estimated risk of mortality solely using prey kill sites (Smith et al., 2005). For example, Kauffman et al. (2007) compared wolf kill sites to random locations in Yellowstone National Park to identify landscape features associated with where elk might be killed if visited. Disadvantages in using kill sites is that they often are biased towards large prey that are more readily detected (Webb et al., 2008; Bacon et al., 2011), and adequate samples take considerable time to accumulate.
An alternative to kill sites is to combine spatial distributions of predators and contents of their scats. Scat contents reflective a predation event and the scat location indicates the broad-scale spatial overlap of predators and prey. A scat-based approach may be advantageous over telemetry-based kill sites because scats can be less invasive and more cost efficient when using scat-detection dogs (Wasser et al., 2004; Mumma et al., 2017), particularly in multi-predator communities. In this paper, we compare a scat-based approach to quantifying spatial predation risk from bears (Ursus arctos/U. americanus ), cougars (Puma concolor ), coyotes (C. latrans ), and wolves for a partially migratory elk population in the eastern slopes of the Rocky Mountains, Canada. We compared predictions of predator-scat occurrence combined with the relative probability of elk being in the scat to a model of predation risk from elk kill sites. We used this approach to assess expected differences in predation risk between migratory and resident elk on their summer ranges.