Risk from a top predator and forest structure influence scavenging by smaller carnivores

Abstract

Large carcasses often attract multiple carnivore species, so subordinate carnivores must weigh the reward of a profitable meal with the risk of being attacked by dominant carnivores. These risk–reward trade-offs are likely influenced by a variety of factors, including scale-dependent risk from dominant carnivores (e.g., short- vs. long-term risk) and the amount of carcass remaining. In the southeastern United States, human hunters provision a large amount of white-tailed deer carrion, which appears to be an important food source for coyotes (a novel top predator), but we know little about how coyotes influence the scavenging behavior of smaller carnivores. In this study, we evaluated the relative importance of risk from coyotes, vulture activity, forest structure, and remaining food on bobcat, gray fox, raccoon, and opossum scavenging by deploying 71 deer carcasses within a managed forest in South Carolina during January 2020 and 2021. We found that coyotes only had direct effects on bobcat behavior, suggesting that competition for carcasses was greatest between these two species. However, the relative importance of long- versus short-term risk from coyotes was dependent on the stage in the scavenging process. Effects from forest structure were also stage-dependent, where tree density and age were related to carcass discovery for bobcats and raccoons, while minimal understory cover facilitated bobcat, gray fox, and opossum scavenging, despite short-term risk from coyotes. Vulture activity appeared to serve as a cue for gray foxes to discover carcasses. Ultimately, we found that risk from coyotes had species-specific and context-dependent effects on smaller carnivore scavenging. This represents some of the first direct evidence of how coyotes alter smaller carnivore behavior in a region where coyotes are a novel top predator. However, forest structure (particularly understory cover) seemed to mediate risk from coyotes, highlighting how habitat can influence predator–predator interactions. Future research should also investigate these interactions during other times of the year and try to quantify how human-provisioned carcasses influence populations and communities.

INTRODUCTION

Aggression between carnivores is typically greatest when species are similarly sized, closely related, or consume the same foods (Ritchie & Johnson, 2009). Larger carnivores are often dominant, though other factors such as group size and weapons can be influential (Allen et al., 2016; Palomares & Caro, 1999). There is also mounting evidence that top-down effects on subordinate carnivores are not consistent across space and time (Haswell et al., 2017; Jachowski et al., 2020), highlighting a need to better understand the mechanisms that govern these interactions. Predator–prey theory suggests that risk and risk responses are scale-dependent; the risky places hypothesis predicts that prey species will avoid places with high long-term predation risk (i.e., habitat where predators often hunt; Laundré et al., 2001), whereas the risky times hypothesis predicts that prey behavior should be most altered when predators are nearby (Creel et al., 2008). Moreover, environmental factors such as vegetative cover or human activity can mediate prey risk, creating refuges or ecological traps (Berger, 2007; Weldon & Haddad, 2005). A better understanding of when predator–prey theory applies to predator–predator interactions is needed.

Many carnivores are facultative scavengers, though large carcasses can be “hotspots” of encounter risk among carnivores (Sivy et al., 2017). Scavenging was traditionally viewed by ecologists as a “free meal,” because it requires little handling time and can have large caloric benefits (Wilson & Wolkovich, 2011). However, recent research has shown that large carcasses can be fatal attractants for subordinate carnivores if dominant carnivores catch them there (Sivy et al., 2017). Although carcasses can be risky places, subordinate carnivores can likely mediate that risk by visiting carcasses during less risky times (Swanson et al., 2016). Carnivores have been shown to increase vigilance at carcasses in response to sympatric predators (Atwood & Gese, 2008; Kautz et al., 2021), but very few studies have directly tested how predation risk influences carrion consumption. Understanding how subordinate carnivores navigate this risk–reward trade-off is one important facet of understanding the ecological effects of carrion on the landscape (Ruprecht et al., 2021).

Carrion can become available through multiple means and is increasingly acknowledged as a substantial proportion of many carnivore species' diets (Prugh & Sivy, 2020). Large carnivores provision large herbivore carrion, as do humans, the latter often through hunting or vehicle collisions (Pereira et al., 2014; Wilson & Wolkovich, 2011). A global review found that 79 vertebrate species ate carrion from big game hunters (Mateo-Tomás et al., 2015), and in the United States alone, an estimated 625,000,000 kg of carrion is left in the field by hunters every year (Oro et al., 2013). This is at least comparable to the amount of carrion provisioned by large carnivores, which is estimated to be 42.2 kg km⁻² year⁻¹ (Prugh & Sivy, 2020) or ~415,000,000 kg year⁻¹ in an area the size of the United States.

In the southeastern United States, white-tailed deer hunting (Odocoileus virginianus; hereafter “deer”) is a common recreational activity during the fall, resulting in a large amount of carrion available to scavengers. In 2020, there were 197,893 deer harvested in South Carolina alone (2.4 deer km⁻²; South Carolina Department of Natural Resources, 2021). Indeed, most studies on coyote (Canis latrans) diet in the Southeast have reported a spike in deer consumption coincident with hunting season, often large enough to make deer the most consumed food item during the fall (e.g., Kelly et al., 2015; Schrecengost et al., 2008; Swingen et al., 2015; Ward et al., 2018). Coyotes likely acquire deer through multiple means, including finding hunter-killed (or crippled) deer before the hunter can, and scavenging from carcass dump sites or field-dressed deer. Importantly, coyotes have only become established across the Southeast in the last 30–50 years (Hody & Kays, 2018) and have been shown to be a substantial source of fawn mortality (Kilgo et al., 2019; Nelson et al., 2015). Therefore, although hunters generally perceive coyotes as an invasive, problematic species in the Southeast, they are likely supplementing coyote diets with a profitable, low-effort food source.

Apart from humans, coyotes have no predators across most of the eastern United States, making scavenging less risky relative to where they coexist with large carnivores (Ruprecht et al., 2021; Sivy et al., 2017). However, coyotes could be a source of risk for smaller carnivores intending to scavenge. Across their range, coyotes suppress and kill multiple fox species, and will also sometimes kill bobcats (Lynx rufus; Fedriani et al., 2000; Moehrenschlager et al., 2007; Nelson et al., 2007). However, coyotes and bobcats appear to share less niche space relative to coyotes and foxes, including metrics like dietary overlap, habitat use, and temporal activity (Rich et al., 2018; Thornton et al., 2004; Witczuk et al., 2015). Coyotes seem to suppress raccoon (Procyon lotor) populations in some contexts (Crooks & Soule, 1999; Sargeant et al., 1993), but not others (Gehrt & Prange, 2007). Since coyotes have become established in the southeastern United States, there has been relatively little research on their interactions with smaller carnivores. So far, a study in Mississippi found that gray foxes (Urocyon cinereoargenteus) avoided coyote core areas but overlapped considerably with full coyote home ranges (Chamberlain & Leopold, 2005), and a similar pattern was found between bobcats and coyotes in Florida (Thornton et al., 2004). This suggests that scavenging on carrion in places with high coyote activity would be most risky, while carrion in places with lower coyote activity would be less risky.

In this study, we evaluated support for multiple hypotheses that might influence the detection and consumption of deer carcasses by bobcats, gray foxes, raccoons, and opossums (Didelphis virginianus). We hypothesized that risk from a top predator (coyotes) would be important, but short-term risk and long-term risk might not have parallel effects. We also thought that vulture activity (i.e., flying) could serve as a cue for carnivores, especially given vultures often find carcasses quickly (Turner et al., 2017). Forest structure could influence the probability or speed in which carnivores detect carrion, through being correlated with carnivore abundance (Nelson et al., 2007) or their ability to detect visual and olfactory cues. Vegetative cover could influence perceived risk (Flores-Morales et al., 2018; Prugh & Sivy, 2020), which may be highest from coyotes in open habitat given their cursorial foraging strategy. Lastly, the amount of carrion remaining likely influences the time spent consuming carrion. We approached our analysis by decomposing the scavenging process into multiple steps: (1) how long it took for our focal species to arrive (objective 1); (2) how much time they spent at carcasses (objective 2); and (3) how much time they spent feeding (objective 3). For the most part, we used the same hypotheses for each step of the scavenging process (Table 1), so that we could quantify when each factor was important. By testing multiple hypotheses in a multipredator system, we were also able to determine whether coyotes have species-specific effects on smaller carnivores.

TABLE 1 Descriptions and significant results for seven variables we hypothesized would influence carnivore detection and consumption of deer carcasses in South Carolina, USA.

Hypothesis	Description	Time until arrival				Time spent at carcass				Percentage of time feeding
		B	GF	R	O	B	GF	R	O	B	O
Long-term coyote risk	An estimate of nearby coyote activity derived from a passive camera array (mean no. coyote photos/day)	↓	…	…	…	↑	…	…	…	…	…
Short-term coyote risk	Coyote activity at the carcass site (the unit varies depending on the analysis)	…	…	…	…	…	…	…	…	↓	…
Vulture activity	Vulture activity at the carcass site during crepuscular periods (no. photos/day)	…	↓	…	…	NA	NA	NA	NA	NA	NA
Tree stand age	The average circumference of the five nearest trees of the dominant type (cm)	↑	…	…	…	…	…	…	…	…	…
Tree density	The no. trees (of the dominant type) within a 5 m radius of the carcass (unitless integer)	…	…	↓	…	…	…	…	…	…	…
Understory cover	The average index from four Robel pole measurements (unitless, possible values = 1.0–16.0)	…	…	…	…	…	↓	…	…	↓	↓↑
Carcass remaining	The amount of carcass remaining (unitless, possible values = 1–5)	NA	NA	NA	NA	NA	NA	NA	NA	…	↑

Note: The symbol ↑ indicates a significant positive effect, ↓ indicates a significant negative effect, ↑↓ indicates an interaction, while no effect is indicated by an ellipsis (…). NA indicates that the variable was not used in those analyses. Abbreviations: B, bobcat; GF, gray fox; O, opossum; R, raccoon. METHODS Study area

Our study area was in the Piedmont region of western South Carolina, characterized by a humid subtropical climate, mild winters, and gently rolling hills (Griffith, 2010). Various Native American tribes lived in the region during the period of European colonization, likely including the Yuchi, Westo, Chickasaw, Apalachee, Yamassee, and Savanna peoples (Cobb & DePratter, 2012; Native Land, 2022). As late as the early 1700s, much of the land was a savannah community, including native grasses and bison (Bison bison; McCormick County Chamber of Commerce, 2021). During the 1800s, much of this land was converted to cotton fields and our study area still contains a plantation house where slaves lived and worked. The dominant land cover changed again in the 1900s, this time to forest, which mostly consisted of managed loblolly pine (Pinus taeda) plantations in various stages of succession. Hardwoods dominated in drainages, including white oak (Quercus alba), southern red oak (Q. falcata), and hickory (Carya spp.). Understory vegetation included blackberry (Rubus allegheniensis), muscadine (Vitus rotundifolia), and fennel (Eupatorium spp.). Pastures and fields were intermittent throughout the landscape, including food plots for game species. The majority of deer hunting occurred during rifle season in October, November, and December.

We deployed deer carcasses across ~61 km² (~15,000 acres) of private land in McCormick County, South Carolina (Figure 1). Potential scavengers included coyote, red fox (Vulpes vulpes), gray fox, domestic dog (Canis familiaris), bobcat, striped skunk (Mephitis mephitis), raccoon, opossum, wild pig (Sus scrofa), turkey vulture (Cathartes aura), black vulture (Coragyps atratus), American crow (Corvus brachyrhynchos), and birds of prey (Accipitridae spp.). Black bears (Ursus americanus) were rarely seen in the study area. The average temperature was 11°C (52°F) during our first survey period (January 2020) and 8°C (47°F) during our second survey period (January 2021). It rained 1–3 cm for three days in January 2020 and five days in January 2021.

FIGURE 1. Location of the study area in McCormick County, South Carolina, USA. Most of the land was managed by loblolly pine plantations in various stages of succession. The distribution of 71 carcasses that were deployed in January 2020 and January 2021 is shown, as well as the 90 passive cameras in an array, which were used to estimate long-term coyote risk at the carcass sites.

Study design

We stratified carcass deployment locations based on coyote activity. We primarily used data from an established camera array in our study area (Saldo et al., 2023) to estimate coyote activity (Appendix S1). The array consisted of ~90 unbaited cameras deployed on dirt roads, each ~1 km apart (Figure 1). For each camera array site, we counted the number of photos of coyotes from the year prior to carcass deployment and divided that number by the number of days each camera was active (photos per day). In order to capture a gradient of coyote activity, we evenly stratified each camera site into low, medium, and high activity using quantile breaks in ArcMAP 10.7 (ESRI, Redlands, CA, USA). We randomly chose an equal subset from each category, avoiding placing sites next to each other when possible. Once we arrived at the camera, we deployed each carcass 250 m away in a random direction, while ensuring the carcass was in the same habitat type as the camera.

Carcass deployment

In early January 2020 and 2021 (immediately after deer hunting season), we collected unwanted portions of deer carcasses (bones, guts, hides) from local deer processors. We chose this timing because we wanted to be representative of deer hunting season (October–December) while limiting the availability of other carcasses on the landscape. While a whole deer carcass is representative of some carrion, we wanted to simulate the leftovers from field-dressed deer or those left at carcass dump sites, while also standardizing the amount of carcass at each site. We placed 15–16 kg of deer carcass into a 1 × 0.5 m cage made of metal garden fencing and secured the cage to a tree with wire (Figure 2), or with 1–2 metal 1-cm diameter corkscrew anchors driven into the ground. We used garden fencing with 5 × 7.5 cm cells because we wanted scavengers to be able to feed from the carcass without being able to pull out large portions and feed off camera. We attempted to diversify the type of carcass in each cage (e.g., guts and bones), but 26 of 75 cages were only filled with a single type. We placed a motion-activated wildlife camera (Bushnell, Overland Park, KS, USA) ~5 m away and set it to take three photos per trigger with a 1-min delay between triggers (Figure 2). We left carcasses out for a minimum of 21 days, and most sites were deployed for at least 28 days. We deployed 35 sites in January 2020 and 40 sites in January 2021. However, we censored two sites from each year from all analyses due to various issues (i.e., burned by fire, batteries died <7 days after deployment); therefore, we used 71 sites in our analyses unless specified otherwise.

FIGURE 2. An example of how deer carcasses were deployed and monitored in South Carolina, USA, to quantify scavenging activity: (a) the 1 × 0.5 m (metal garden fence) cage containing 15–16 kg of deer carcass secured to a tree with wire and wire clips, (b) the placement of the camera relative to the carcass, and (c) a GPS-collared coyote feeding from the carcass. Photo credit: the authors.

Photo processing

We processed photos using Timelapse2 (Greenberg, 2021). For each photo, we recorded the species and the amount of carcass remaining relative to when it was deployed (>75%, 75%–50%, 50%–25%, <25%, bones and hide, or unknown/moved). For our focal species (bobcat, gray fox, raccoon, opossum), we also recorded their behavior (head up, head down not feeding, head down feeding, unsure). We defined “head up” as any photo where the top of the neck was parallel with or raised above the top of the back. For quality control, each photo received a certainty score (100% sure, fairly confident, or unsure), and all photos not tagged as 100% sure were checked by the first author.

Objective 1: Time until arrival at the carcass

We quantified how long it took for our focal species to arrive at the carcasses (if they showed up at all). We jointly modeled two response variables: (1) whether the species was detected, and (2) the time until the species arrived (defined as the time difference between the deployment date–time and first detection in hours) or the hours until the site became nonoperational. Although our modeling approach accounted for variation in deployment time, we still wanted our sites to be representative, so we censored any sites where the focal species was never detected and that were deployed for <15 days (species-dependent censored site range = 6–10).

We hypothesized that short-term and long-term risk from coyotes could influence time until arrival. For short-term coyote risk, we used the time from deployment until coyotes arrived. We chose to use time until arrival rather than some other metric from the carcass sites (e.g., total time spent by coyotes) because we wanted to test whether and when coyotes discovered sites influenced how long it took for our focal species to arrive. For long-term coyote risk, we used the camera array and an inverse-distance weighted approach to quantify coyote activity from the previous year. Specifically, we counted the number of coyote photos taken in May, June, October, and November of the previous year by each camera in the array and divided by the number of days the camera was operational (coyotes per day). We used data from these months due to photo processing constraints (over three million photos were collected throughout the study), and because we wanted to capture coyote activity during key biological periods throughout the year (May and June = pup rearing; October and November = dispersal). To decide which array cameras would represent which carcass sites, we used ArcMAP to visualize a 1-km buffer around each carcass site and used up to three of the closest array cameras within that buffer (Appendix S1: Figure S1). We distance-adjusted each coyote per day value by subtracting the distance (in kilometers) from the carcass site from one, then multiplied that distance by coyotes per day so that closer array cameras were adjusted less. We then averaged these adjusted values for each carcass site: $$$ \frac{{\overline{X}}^{coyotes}}{{\mathrm{day}}_i}\times {w}_i, $$$ where w_i = 1 minus the distance from camera site i to the carcass in kilometers.

We also hypothesized that vulture activity and forest structure could influence time until arrival. We counted the number of vulture photos taken at each carcass site between the hours of 4 pm and 9 am, which is when carnivores would likely be cuing into vulture activity, based on exploratory analyses of diel activity (85%–95% of focal species detections were during nocturnal or crepuscular hours). Given this time period only represents a portion of the day (~3 h) when the typically diurnal vultures were active, we tested whether crepuscular vulture activity was representative of total vulture activity. We used data from 2021 (38 sites) to test for a correlation between the subset of crepuscular photos we used and all vulture photos and found a moderate correlation between the two datasets (R² = 0.618). Accordingly, we proceeded with crepuscular vulture activity as representative of vulture activity at the carcass sites. We divided the number of vulture photos at each site by the days the site was operational to standardize between sites. For vegetative structure, we measured several variables while deploying the carcass. To approximate stand age, we measured the circumference (at 1.5 m off the ground) of the five closest trees of the type that dominated the stand (usually loblolly pines), then averaged these five values. For tree density, we counted the number of trees (of the dominant species) within a 5-m radius of the carcass. We only counted the dominant species because we wanted to capture the management phase the stand was in, and midstory hardwoods would have confounded this in some stands. For understory cover, we viewed a Robel pole (Robel et al., 1970) from 10 m away from the carcass in the four cardinal directions and averaged those four values.

We used time until event (Cox proportional hazard) models to test our hypotheses. We tested for collinearity between our four predictor variables and found no pair with Pearson's >|0.55|. We centered and scaled each predictor variable, then fit 13 models for each of our focal species (Appendix S1: Table S1). In addition to univariate and global models, we included subglobal models that captured combinations of variables we hypothesized could be important. For example, we included a “forest structure” model, with stand age, tree density, and understory cover. We also included a model with an interaction between stand age and vulture activity, because we thought that vulture activity might be particularly influential at younger sites (with no or little canopy). We used the survival package (Therneau & Grambsch, 2000) in R version 4.4.1 (R Core Team, 2021) to fit our models and check model assumptions. Raccoons had one model and opossums had two models that violated Cox proportional hazard assumptions, so we excluded those models.

We used an information theoretic approach (Akaike information criterion [AIC]; Burnham & Anderson, 2002) to determine relative support for our hypotheses. We ranked models using corrected Akaike information criterion (AIC_c) and the MuMIn package (Barton, 2009) and determined that any model within 2 ΔAIC_c of the top model (unless ranked below the null model) would be considered a competitive model (Burnham & Anderson, 2002). For competitive models, we determined variable significance if 95% CIs did not overlap zero. When there were multiple competitive models containing the same variable, we reported statistics from the top ranked model.

Objective 2: Time spent at the carcass

For this analysis, we calculated the total time our focal species spent at each site. We grouped photos of the same species at the same site into detection events if they were separated by <30 min (Ridout & Linkie, 2009), then recorded the duration of a detection event as the time between the first and last photo. We recorded events consisting of one trigger as 1 min in duration because of the 1-min delay between triggers. Our response variable was the sum of the duration of each species' detection events at each site divided by the days the site was operational. We hypothesized that several variables could influence time spent at the carcasses (Table 1). For long-term coyote activity, stand age, tree density, and understory cover, we used the same measures as previously described. For short-term coyote activity, we calculated the minutes spent by coyotes and divided them by the days the carcass site was operational.

We used generalized linear models followed by AIC model selection for this analysis. We tested for collinearity between our four predictor variables and found none with Pearson's >|0.52|. We compared the fit of different distributions on our global models with six combinations for each species: Poisson and two types of negative binomial models, along with either zero inflation or no zero inflation specified. We found that a negative binomial distribution was always best, but only raccoons needed the zero inflation term. Therefore, we carried these distributions into each species' model structure. In addition to the global model, we built nine other models for each of our focal species (Appendix S1: Table S2). We used a similar combination of models as objective 1, with the addition of an interaction between short-term coyote activity and understory cover, because we thought that risk from coyotes might be particularly high at sites with more cover. We included a logoffset term to account for the number of days each carcass site was operational. We checked model fit using the DHARMa package (Hartig, 2020) and used the same AIC_c and variable rules as previously described.

Objective 3: Percent of time spent feeding on the carcass

We calculated the percentage of time our focal species spent feeding during each detection event (see objective 2) by counting the number of photos where the animal was feeding with its head down and dividing by the total photos during that detection event. Therefore, our response variable was the proportion of photos where the animal was feeding during the detection event. We hypothesized that the same variables from objective 2 would be important, with the addition of the amount of carcass remaining (Table 1). We used the same measures of long-term coyote activity, stand age, tree density, and understory variables as previously described. We quantified short-term risk by summing the cumulative minutes spent by coyotes at that carcass site in the previous seven days (prior to that detection event). We chose seven days because we found that average coyote activity dropped incrementally from the day of discovery to day five, and we added two days to account for lingering scent. For the carcass remaining, we used the state of the carcass as described in Photo processing and converted this to an ordinal variable (5 = >75% remaining, 4 = 75%–50%, 3 = 50%–25%, 2 = <25%, 1 = bones and hide only). We excluded events where the carcass state was classified as “unknown/moved” (n = 1 for each species).

To test how these variables influence time spent feeding, we used a similar analytical approach to objective 2 (generalized linear models), with the difference being our observational unit was detection events rather than sites. We decided to not include gray foxes or raccoons in this analysis because they were only detected feeding in 3 of 30 and 3 of 35 detection events, respectively. None of our variables had pairwise correlations >|0.61|. We used a binomial distribution because our response variable had an upper limit of one. We fit 14 models for each species (Appendix S1: Table S3), which included a random effect of carcass site and weights for the number of photos in the detection event, but were otherwise similar to previous analyses in structure, function, AIC_c, and variable rules.

RESULTS

We collected useful data from 71 deer carcasses deployed across our study area in January 2020 and January 2021 (Figure 1). Vultures and coyotes were detected most frequently, followed by our focal carnivores and wild pigs (Appendix S1: Figure S2). We also detected domestic dogs, American crows, red-tailed hawks (Buteo jamaicensis), golden eagles (Aquila chrysaetos), and owls.

Objective 1: Time until arrival at the carcass

We detected coyotes at nearly all sites not censored from this analysis (91%), followed by opossums (46%), bobcats (40%), raccoons (26%), and gray foxes (19%; Figure 3). For the sites where they did arrive, coyotes were typically the first to arrive (average days until detection ± SE = 5.93 ± 0.57), followed closely by opossums (6.05 ± 1.11), whereas gray foxes (10.00 ± 2.25), raccoons (10.06 ± 1.63), and bobcats (10.23 ± 1.74) took longer to arrive at the carcasses.

FIGURE 3. Site accumulation curves for various scavenging vertebrates visiting carcass sites in South Carolina, USA. Day “0” are photos from the day the carcass was deployed. The vulture data were derived from 2021 data (n = 38 sites), while the rest of the species' data were derived from 2020 and 2021 (n = 61–66 sites depending on the species; see methods in objective 1).

Our bobcat analyses had three competitive models (Appendix S1: Table S1). Bobcats arrived faster at sites with greater long-term coyote activity (hazard ratio = 2.766) and when the tree stand was younger (hazard ratio = 0.637; Table 2). Our gray fox analysis had three competitive models (Appendix S1: Table S1), and gray foxes arrived faster at sites with more vulture activity (hazard ratio = 1.545; Table 2). Our raccoon analysis had four competitive models (Appendix S1: Table S1), and raccoons arrived faster at sites with greater tree density (hazard ratio = 1.361; Table 2). For our opossum analysis, there were three competitive models (Appendix S1: Table S1), yet 95% CIs overlapped zero.

TABLE 2 Estimates and 95% CIs from statistical analyses of bobcat, gray fox, raccoon, and opossum activity at deer carcass sites in South Carolina, USA.

Analysis	Species	Variable	Estimate	Lower 95% CI	Upper 95% CI	Hazard ratio
Time until arrival	Bobcat	Long-term coyote activity	1.018	0.353	1.682	2.766
		Stand age	−0.451	−0.837	−0.064	0.637
Time until arrival	Gray fox	Vulture activity	0.435	0.040	0.831	1.545
Time until arrival	Raccoon	Tree density	0.308	0.032	0.584	1.361
Time spent at carcass	Bobcat	Long-term coyote activity	0.370	0.124	0.616	NA
Time spent at carcass	Gray fox	Understory cover	−3.197	−5.397	−0.997	NA
Percentage of time spent feeding	Bobcat	Short-term coyote activity × understory cover	6.157	1.522	10.792	NA
		Short-term coyote activity	−4.199	−11.209	2.811	NA
		Understory cover	−4.835	−10.448	0.777	NA
Percentage of time spent feeding	Opossum	Carcass remaining × understory cover	−0.445	−0.780	−0.110	NA
		Carcass remaining	−0.012	−0.263	0.239	NA
		Understory cover	−0.256	−0.764	0.251	NA

Note: Only variables from competitive models for which 95% CIs did not overlap zero are shown. Note that effects from the time until arrival analyses are from hazard estimates, which, in this context, increase if sites are discovered quicker. Estimates for time spent at carcass and percentage of time spent feeding are linear model estimates. Objective 2: Time spent at the carcass

Totaled across all sites, we found that coyotes spent 4.3× more time (133.9 h) at the carcass sites compared with opossums (31.1 h), which spent much more time than bobcats (4.0 h), gray foxes (2.8 h), and raccoons (1.6 h; Appendix S1: Figure S2). Coyotes also spent the most minutes per operational day at the sites (8.4 ± 2.4 min; mean ± SE), followed by opossums (3.4 ± 0.8 min), gray foxes (1.0 ± 0.6 min), bobcats (0.7 ± 0.3 min), and raccoons (0.6 ± 0.3 min).

The bobcat and gray fox analyses were informative, while the raccoon and opossum analyses were not. For bobcats, there was one competitive model (Appendix S1: Table S2). Long-term coyote activity had a positive effect on bobcat activity, where bobcats spent 10 additional seconds at the carcasses for each additional photo per day of coyotes from nearby cameras (Figure 4). For the gray fox analysis, there were two competitive models related to habitat (Appendix S1: Table S2). Although gray foxes spent little time at the carcasses to begin with, they spent 10 less seconds at the carcasses for every two-unit increase in understory cover, though this effect was primarily present when cover was relatively low (Figure 4). The null model was the top model for the raccoon analysis (Appendix S1: Table S2). For opossums, there were four competitive models (Appendix S1: Table S2), yet all 95% CIs for their estimates overlapped zero.

FIGURE 4. Significant effects on time spent at deer carcasses by bobcats and gray foxes in South Carolina, USA. (a) Time spent by bobcats versus coyote activity at nearby cameras (long-term coyote risk), and (b) time spent by gray foxes versus understory cover.

Objective 3: Time spent feeding on the carcass

We recorded 695 coyote detection events, 150 opossum events, 56 bobcat events, 35 gray fox events, and 30 raccoon events. Bobcat detections lasted the longest on average (13.8 ± 3.8 min; mean ± SE), followed closely by opossum (13.1 ± 1.5 min) and coyote (11.6 ± 0.9 min). Gray fox (4.8 ± 2.0 min) and raccoon (3.1 ± 0.8 min) detections were shorter. Notably, many of the detection events for each species were 1 min long (our minimum possible event time length): 73% for raccoons, 59% for bobcats, 57% for gray foxes, 50% for coyotes, and 46% for opossums. Averaged across all sites, our focal species were detected feeding in <10% of the photos within a detection event; bobcats were feeding in 9.8% ± 3.5% of photos, the same as opossums (9.8% ± 3.5%), followed by coyotes (8.4% ± 1.7%), raccoons (5.8% ± 3.3%), and gray foxes (5.5% ± 4.4%).

For the percent of bobcat photos feeding analysis, the short-term coyote activity × understory interaction model was the only competitive model (Appendix S1: Table S3). Bobcat feeding was jointly influenced by negative effects from coyote activity and understory cover (Table 2, Figure 5). Coyote activity was most influential, as bobcat feeding was greatest at low levels of coyote activity and lowest at high levels of coyote activity. However, minimal understory cover facilitated bobcat feeding, even at high levels of coyote activity (Figure 5). For opossum feeding, there was also only one competitive model: carcass remaining × understory cover interaction model (Appendix S1: Table S3). Opossum feeding had a strong positive association with the amount of carcass remaining, but particularly when understory cover was high (Figure 5). However, when there was little carcass remaining, the effect of understory reversed and little understory was associated with more feeding.

FIGURE 5. Significant effects on bobcat and opossum feeding on deer carcasses in South Carolina, USA. (a) An interaction between time spent by coyotes at the carcass site within the last seven days and understory cover on bobcat feeding. (b) An interaction between the amount of carcass remaining and understory cover.

DISCUSSION

Our findings provide additional support for the notion that dominant carnivores have scale-dependent and species-specific effects on subordinate carnivore behavior (Jachowski et al., 2020), which are mediated by vegetation structure (Gigliotti et al., 2021). Of the four smaller carnivores we studied, only bobcats seemed to be influenced by coyotes, suggesting that competition is greatest between these two similarly sized species. Additionally, how coyotes and forest structure influenced smaller carnivore behavior was dependent on which step in the scavenging process we were assessing. The diversity of facultative scavengers we detected, combined with the high intensity of carcass use by coyotes across both space and time, suggests that carcasses act as point sources of competition (Prugh & Sivy, 2020). Ultimately, we show that coyotes have context-specific effects on smaller carnivore behavior and suggest behavioral strategies for how these smaller carnivores coexist with this novel large carnivore in our system.

How coyotes influenced scavenging

Bobcats arrived faster and spent more time at carcasses where long-term coyote risk was greater, perhaps because they aimed to claim and defend the carcass. Although coyotes and opossums spent more total time at the carcasses, bobcats had the greatest percentage of detection events >30 min in duration (18%), suggesting they were defending the carcass in these cases. Bobcats have been shown to engage with and “win” most direct encounters with coyotes at carcasses, even though they are a smaller species (Allen et al., 2016). We did record two instances where a bobcat clearly defended a carcass from a coyote, but there were not enough of these interactions to analyze. Another plausible explanation is that bobcat and coyote density were spatially correlated, potentially since both species share small mammals as important parts of their diet throughout the year (Koehler & Hornocker, 1991; Neale & Sacks, 2001). Indeed, dietary overlap is likely greatest in the winter when food availability is lowest (Cherry et al., 2016), meaning that carcasses would have been a valuable resource for both species. Therefore, increased dietary overlap may be the underlying explanation driving these relationships we found.

Short-term coyote risk reduced the time bobcats spent feeding, perhaps because putting their head down and feeding would reduce their ability to detect a coyote. Coyotes may have had similar effects on gray foxes and raccoons, as four of six of their total feeding events occurred when recent coyote activity was zero and the other two occurred when recent coyote activity was relatively low. Dominant carnivores in Alaska (wolves and wolverines; Canis lupus and Gulo gulo) had similar effects on subordinate carnivores (coyotes and red foxes), where the dominant species spent twice as much time feeding on carcasses (Klauder et al., 2021). In Montana, beta (but not alpha) coyotes reduced time feeding on carcasses in the presence of wolves (Atwood & Gese, 2008). Although coyotes apparently reduced bobcat feeding intensity, it is possible that bobcats were able to compensate for this loss of food by eating a similar amount of food over a longer time period. Perhaps this also explains why bobcats spent the most time at the carcasses per detection event.

Long-term and short-term coyote risk only had significant effects on bobcats, which suggests that competition for carcasses was greatest between these two species. Indeed, risk from coyotes best explained bobcat behavior at every step of the scavenging process, and our results suggest that a combination of fear and proactive defense may be driving these responses. While coyotes can kill bobcats (Fedriani et al., 2000), a recent meta-analysis reported that 74% of studies did not find evidence for interference competition between bobcats and coyotes, and it was less likely to be found in forested habitat (like ours; Dyck et al., 2022). Therefore, it seems that coyotes influence bobcat behavior, but the outcomes of direct interactions are context-dependent. In summary, we demonstrate the importance of investigating species-specific patterns in response to risk from a top predator through multiple measures, both in terms of predation risk (short-term and long-term risk) and through decomposing the multistep behavioral response to resources in the environment (e.g., detection, investigation, consumption).

How forest structure influenced scavenging and mediated risk

Forest stand structure (tree age and density) influenced bobcat and raccoon time until arrival. Bobcats arrived faster at carcasses within younger tree stands, which is likely where bobcats are more abundant in the Southeast (Chamberlain et al., 2003; Little et al., 2018). This also supports our hypothesis that prey availability is one of the mechanisms that explains bobcat time to arrival, because small mammals and rabbits (bobcat primary prey) were most abundant in young pine stands (unpublished survey data). Raccoons arrived faster at carcasses where tree density was greater, which may likewise be related to habitat selection. Although raccoon habitat selection is often scale and season dependent, in the winter, raccoons seem to select hardwoods and to some extent immature pines (which tended to have denser trees in our study area; Byrne & Chamberlain, 2011; Chamberlain et al., 2003). It is possible that risk from coyotes is one of the mechanisms driving selection of denser forests by raccoons, as there are more options for climbing to escape predators. Future studies should try to quantify a baseline measure of scavenger abundance or density near carcass sites, so that these effects can be disentangled from characteristics of the site that facilitate discovery.

Understory cover was an important factor in several analyses and seemed to mediate risk from coyotes in some cases. Although gray foxes spent relatively little time at carcasses to begin with, they spent more time at sites with very minimal understory cover. We posit that gray foxes were more comfortable at carcasses with longer sightlines, which would allow them to detect dominant carnivores more quickly. Of our focal species, gray foxes were likely at the highest risk of predation by coyotes, considering coyote-specific mortality is well documented (e.g., Fedriani et al., 2000) and we observed gray foxes killed by coyotes in our study area. This may also explain why we did not detect red foxes at our carcasses, despite them being present in our study area (Saldo et al., 2023): predation risk from coyotes may be even greater for red foxes (Levi & Wilmers, 2012). We also found that bobcats fed more when understory cover was minimal. Importantly, this effect interacted with short-term coyote risk, where minimal understory cover facilitated bobcat feeding, even when coyote risk was high. Similarly, understory cover interacted with the amount of carcass remaining for opossum feeding, yet here we found that minimal understory cover only facilitated feeding when there was little carcass remaining (and had the opposite effect when there was more carcass). Perhaps this suggests a risk–reward trade-off, where opossums are willing to engage in riskier behavior when there is a lot of food, but less willing when there is little food. Regardless, these findings again highlight the complex relationship between understory cover and subordinate carnivore behavior.

Past research has explored how vegetative structure can mediate predation risk. At the landscape scale, increased vegetative cover has been associated with facilitating coexistence between carnivores, presumably by reducing encounter rates with top predators (Janssen et al., 2007; Nelson et al., 2007). However, at smaller scales, reduced vegetative complexity has been shown to benefit subordinate carnivores. For example, cheetah survival was greater in areas of low vegetative complexity, partially because large predators could ambush them in more complex habitats (Gigliotti et al., 2020). In a scavenging context, an ideal carcass for a subordinate carnivore might be in open habitat (to better detect predators), but still relatively close to cover (for escape), which would parallel risk-sensitive forage selection by ungulates (Creel et al., 2008; Stears & Shrader, 2015). Collectively, we show that forest structure had divergent effects on subordinate carnivore scavenging depending on the context.

How vulture activity influenced time until arrival

Gray foxes arrived faster at sites with more vulture activity, which could mean gray foxes were cuing in on vulture activity. Although gray foxes were nocturnal (and vultures were diurnal), it is possible they were attentive to vulture activity during crepuscular periods. Vultures discovered sites quickly and were the most detected taxa (on average) during 26 of the 30 days we monitored carcasses. Therefore, even though gray foxes slowly discovered sites during the first two weeks after deployment, vulture activity would have likely remained high enough to serve as a cue throughout this time period. Vulture activity was also a top model for raccoon and opossum time until arrival (with the same effect direction), but 95% CIs overlapped zero. This suggests that these two species may have also been cuing into vulture activity, though not as strongly as gray foxes. Our results somewhat contrast with an experimental study, also in South Carolina, which found that carnivores did not arrive faster at sites where vultures were not excluded (Hill et al., 2018). However, other studies have found results in line with ours and suggested explanations. In addition to simply locating carcasses (large carnivores have been shown to cue into vulture activity; Houston, 1979), small carnivores may be cuing in because vultures can tear open carcasses for them (Cortés-Avizanda et al., 2012). Alternatively, given how quickly vultures can discover carcasses, their activity may serve as an indication of a fresh carcass, which could mean that a coyote (or another dominant carnivore) has yet to discover it. Thus, cuing into vultures may ultimately be another coexistence strategy for subordinate carnivores when sympatric with a dominant carnivore.

Limitations

We used a standardized approach to quantify scavenging during a single season, which leaves our results subject to at least two limitations. First, although field manipulations can be a powerful approach to understand complex systems, it can be challenging to know how representative the experimental conditions are (Smith et al., 2020). Providing and monitoring carcasses is an established method in scavenging research, though carcasses are not often placed in a metal cage like ours were. Scavengers seemed to have no trouble feeding from the carcass through the cage (as intended), though they may have fed more if the cage was not present. Carcass size has been shown to have a powerful effect on who scavenges and for how long (Moleón et al., 2015; Turner et al., 2017), so we wanted to avoid this confounding variable while reducing the ability of scavengers to move the carcass out of view. Second, scavenging frequency by facultative scavengers has been shown to vary throughout the year, typically peaking in winter when other foods are most limited (Pereira et al., 2014; Turner et al., 2017). Therefore, quantifying scavenging behavior primarily in January likely represented the period of highest potential competition. Future research should compare scavenging dynamics during other parts of the year when other foods are more abundant.

CONCLUSIONS AND IMPLICATIONS

We draw several conclusions from the results of our study likely of interest to ecologists. First, when considering the full scavenging process, forest structure was often more important for smaller carnivores scavenging than risk from coyotes. We showed that understory cover influenced both bobcat and opossum feeding, likely mediating the risk–reward trade-off associated with scavenging. We also found that gray foxes spent more time at carcasses with less understory cover and raccoons potentially selected dense forests as a proactive antipredator strategy, though future work should try to account for how habitat influences baseline focal species density. Second, although coyotes may influence the behavior of all of our focal species, we only found direct evidence that bobcats were cognizant of risk from coyotes. Indeed, risk from coyotes best predicted bobcat behavior at every step of the scavenging process, but not at the same risk scale. Specifically, short-term risk did not seem to be important until bobcats were actually feeding. Third, the relatively low frequency of feeding by all carnivores suggests that many detections represented investigatory behavior, amplifying the importance of measuring consumption in scavenging research. Moving forward, quantifying carnivore responses to risk using multiple measures will continue to add nuance to how carnivores weigh the risks and rewards of scavenging (Ruprecht et al., 2021). Collectively, our results support an emerging body of literature showing effects from top carnivores are complex and context-dependent (Haswell et al., 2017; Jachowski et al., 2020).

Coyotes have various ecological roles across their range, and we highlight how they change smaller carnivore behavior in a region where they are a novel top predator. Indeed, as coyotes have expanded their range, ecologists have been interested in the extent to which they can functionally replace wolves (Benson et al., 2017; Gompper, 2002). We show how short-term risk from coyotes in our study area reduced scavenging by a smaller carnivore, just like short-term risk from wolves reduced scavenging by coyotes in other areas (Atwood & Gese, 2008; Klauder et al., 2021). However, in addition to having negative effects, large carnivores can also benefit other carnivores (including coyotes) by provisioning them with carrion (Prugh & Sivy, 2020; Ruprecht et al., 2021). Yet coyotes would likely only serve this provisioning role in areas where they kill adult ungulates, which occurs in the northeastern portion of their newly expanded range (Benson et al., 2017), but is less common in the southeastern portion (except see Chitwood et al., 2014). Thus, coyotes would primarily have negative top-down effects on other carnivores in systems such as ours. In areas where coyotes do kill adult ungulates regularly, future studies should investigate the extent to which coyotes facilitate scavenging opportunities for smaller carnivores.

Perhaps human hunters in the eastern United States have somewhat filled the lost large carnivore carcass-provisioning role, though likely in a manner more temporally and spatially constrained (Wilmers et al., 2003). The speed of discovery and time spent by coyotes at our carcasses suggests that hunter (or vehicle collision) provisioned carcasses are a profitable and low-risk food for coyotes in areas without larger carnivores. Given human-food subsidies have been shown to change carnivore behavior and improve their fitness (Newsome et al., 2015; Oro et al., 2013), future studies should try to quantify how the effects from human-provisioned carcasses scale up to influence population dynamics and community assemblage.

ACKNOWLEDGMENTS

E. McDaniel, A. Jamison, E. Nowlin, M. Clark, S. Westwood, and V. Patch assisted in the field and with photo processing. Page's Deer Processing and Rock House Road Deer Processing provided deer carcasses. We appreciate Davis Land and Timber and National Deer Association for land access and field housing. We also thank the South Carolina Department of Natural Resources, (especially J. Cantrell and C. Ruth) for providing funding for this study. J. Kilgo, M. Muthersbaugh, C. Jachowski, and M. Childress provided conceptual guidance.

FUNDING INFORMATION

This project was supported by funding from the South Carolina Department of Natural Resources.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest.

DATA AVAILABILITY STATEMENT

Site-level and detection-level data (Jensen, 2023a, 2023b) are available from Figshare: https://doi.org/10.6084/m9.figshare.22337458 and https://doi.org/10.6084/m9.figshare.22337446, respectively.

ETHICS STATEMENT

We used GPS data from coyotes as part of the study design. All capture and handling procedures were permitted by Clemson University IACUC permit no. AUP 2018-031 and USDA Forest Service permit no. USFS 2018-031.