Land cover and space use influence coyote carnivory: evidence from stable-isotope analysis

Study area

Coyotes in our study area inhabited and traversed approximately 4,100 km² of private and public lands across Georgia (Burke, Columbia, Glascock, Jefferson, Lincoln, McDuffie, Richmond, Warren, and Wilkes Counties), and South Carolina (Abbeville, Aiken, Edgefield, Greenwood, Lexington, McCormick, and Saluda Counties) in the southeastern United States (Fig. 1). Coyotes captured in Georgia and South Carolina commonly moved between our respective study areas that straddled the Savannah River, and likely represented one population, leaving one distinct study area that we referred to as the Savannah River study area (SRA). The SRA had mild sub-tropical climate throughout the year. Summers were generally hot and humid with an average high temperature of 20 °C, while winters were mild with an average low temperature of 1 °C (Ward et al., 2018).

The SRA landscape contained a mix of forested, early successional, agricultural, and urban land covers. Forest cover was predominately loblolly (Pinus taeda) and shortleaf (Pinus echinata) plantations, oak (Quercus spp.)-hickory (Carya spp.)-pine (Pinus spp.) woodlands, and successional pine and hardwood forests containing oak, hickory, sweetgum (Liquidambar styraciflua), loblolly pine, and shortleaf pine. Agricultural lands were intermittent land cover and consisted mostly of pastures and crops such as cotton (Gossypium spp.), corn (Zea mays), tobacco (Nicotiana tabacum), soybeans (Glycine max), peanuts (Arachis hypogaea), and peaches (Prinus persica).

Using frequency of occurrence of prey in coyote scats, Ward et al. (2018) reported that the diets of resident coyotes in SRA consisted predominantly of white-tailed deer (Odocoileus virginianus; 42.8%), rabbit (Sylvilagus spp.; 28.6%), small mammals (22.3%), fruit (24.1%), and other items (13.3%) such as wild pig (Sus scrofa), insects, birds including northern bobwhite (Colinus virginianus) and wild turkey (Meleagris gallopavo), and anthropogenic food. The small mammal community in SRA included squirrels (Sciurus spp.), eastern woodrats (Neotoma floridana), hispid cotton rats (Sigmodon hispidus), mice (Peromyscus spp.), shrews (Blarina spp., Sorex spp.), voles (Microtus spp.), armadillos (Dasypus novemcinctus), and Virginia opossum (Didelphis virginiana). Fruit common to the SRA and available to coyotes included persimmon (Diospyros virginiana), blackberries (Rubus spp.), wild plums (Prunus spp.), wild grapes (Vitis spp.), black cherry (Prunus serotine), and peach.

Animal captures and sampling

To determine space use, we deployed GPS collars on coyotes during fall/winter of 2015–2017. We captured animals with foothold traps (Victor #3 Softcatch, Woodstream Corporation, Lititz, PA, USA; MB 550 or MB 450, Minnesota Trapline Products, Pennock, MN, USA) with offset or padded jaws. Coyotes were typically restrained with a catchpole, muzzle, and hobbles for processing. Although most coyotes were not anesthetized, some were chemically immobilized prior to processing using a Ketamine/Xylazine mixture administered intramuscularly at 0.8 mg/kg for Ketamine and 0.1 mg/kg for Xylazine when it seemed appropriated for the safety of animals and researchers. At the end of procedures, we administered coyotes with Yohimbine at 1.0 mg/kg to intramuscularly as an antagonist to Xylazine. We determined sex and weight of coyotes, and estimated ages of animals by body condition and tooth wear (Gipson et al., 2000). To estimate the ages of our study animals, we used April as the birth month for all coyotes because litters were born March–May (Gier, 1968; Knowlton, 1972). For example, if an animal was captured during January 2016 and presumed to be a pup we estimated the animal’s age as 9-months-old assuming it was likely born during April 2015. If the animal was presumed to have been born during April 2014, we estimated its age as 21-months-old. We then categorized coyotes ≥24 months old as adults, 12–24 months old as juveniles, and <12-months-old as pups. We also collected hair samples to determine diet breadth by clipping 10–20 guard hairs from the withers (i.e., area in between the shoulder blades) of coyotes. We stored hair samples in paper envelopes in a climate-controlled facility until the time of analysis.

Prior to release at capture sites, we fitted each coyote with a mortality-sensitive satellite collar (either G2110L Iridium LITE/GPS collar, Advanced Telemetry Systems, Isanti, MN, USA or Litetrack Iridium collar, Lotek Wireless Inc., New Market, Ontario, Canada) set to record locations at 4-h intervals. After release, individuals were monitored for the life of the collar (up to ~18 months) or until mortality. Previously captured individuals were not re-sampled in subsequent years. Our methodology was refined to ensure minimum stress, handling time, and injury to the captured individual, approved by the University of Georgia Institutional Animal Care and Use Committee (protocols A2014 08-025-R2 and A2015 05-004-A5), Georgia Department of Natural Resources (Scientific Collection Permit #29), South Carolina Department of Natural Resources, and Alabama Department of Conservation and Natural Resources, met guidelines recommended by the American Society of Mammalogists (Sikes & Animal Care and Use Committee of the American Society of Mammalogists, 2016), and reported in compliance with the ARRIVE guidelines (Kilkenny et al., 2010).

Space use and land cover classification

We used a binary classification of space use (resident or transient) by collared animals instead of a trinary classification (resident, bider, or transient; Morin & Kelly, 2017; Ellington, Muntz & Gehrt, 2020) to stay consistent with our previous work (Hinton, van Manen & Chamberlain, 2015; Ward et al., 2018; Webster et al., 2022; Youngmann et al., 2022). We only classified coyotes with ≥4 months of telemetry data as residents or transients (Hinton, van Manen & Chamberlain, 2015; Ward et al., 2018). We used fixed kernel density estimators to calculate coyote space use in R (R Core Team, 2023) using the “h-plugin” smoothing parameter in the “ks” package (Duong, 2007) to generate 95% isopleths. We classified coyotes as residents when they exhibited stable space use for ≥4 months and had home ranges ≤45 km² (Hinton, van Manen & Chamberlain, 2015). For juvenile and adult coyotes, we classified transients as animals with ranging areas >45 km² and unstable, shifting space use (i.e., low site fidelity). However, adult-sized pups (9–11-month-old animals) that were transients during fall/winter captures would have been younger pup-sized animals (0–6 months old) during the previous spring/summer and still dependent on care from breeding residents to have survived spring/summer. Consequently, we classified all pups as residents assuming displacement from natal territories likely occurred during the fall hunting and trapping seasons in which breeders were killed and packs disbanded (Hinton & Chamberlain, 2022). This method allowed for unambiguous classification of residents and transients, but also meant that we were unable to determine territorial status of juvenile and adult coyotes with <4 months of movement data due to mortality or collar failure.

Within 95% isopleths, we calculated the percent of agricultural land use, canopy cover, and density of roads to account for the effects of land cover on coyote diets. We determined agricultural land use using the 2016 United States Department of Agriculture (USDA) Forest Service National Land Cover Database (NLCD) by combing cropland (class 81) and pasture (class 82) into one land cover class, whereas mean percent canopy cover was acquired using the 2016 NLCD tree-canopy layer (Homer et al., 2015). To calculate percentages of agricultural land use and canopy cover, we used Quantum GIS (QGIS) version 3.32 to overlay resident home ranges and transient ranges against 30-m pixel digital NLCD and NLCD tree-canopy layers. Similarly, we used QGIS to estimate road densities by overlaying home ranges and transient ranges on road layers acquired from the United States Geological Survey (USGS) National Transportation Dataset (NTD) for Georgia (U.S. Geological Survey, 2024a) and South Carolina (U.S. Geological Survey, 2024b).

Stable isotope analysis

We prepared hair samples for stable isotope analysis from captured coyotes following the recommendations of Post et al. (2007) in which we did not extract lipids from hair samples because a preliminary analysis revealed that C/N values of all samples were <4. This preparation entailed first removing any debris or skin from hair using a sterile surface wipe and then using scissors to macerate hair. We measured 1–2 mg of samples and packed them into pre-combusted 5 × 9 mm tin capsules (Costech Analytical Technologies Inc., Valencia, CA, U.S.A.). Once homogenized, we measured a 1–2 mg sample and packed it into a tin capsule as above. In addition, we prepared duplicate samples for 10% of samples from each species as a quality control measure. We then sent prepared capsules to the University of Georgia Stable Isotope Ecology Laboratory (UGA-SIEL) in Athens, GA, USA for carbon and nitrogen isotope analysis by continuous flow isotope ratio mass spectrometry. Isotope ratio analysis involved transforming solid-phase samples to gas by rapid and complete flash combustion of sample material. The ionized combustion products were then mass analyzed by means of differing mass/charge ratios among the various isotopic species of CO₂ and N₂. The UGA-SIEL used standards in each analysis for QA/QC as well as calculating δ¹³C and δ¹⁵N (the difference between a sample and natural abundance standard for which Δ = zero; reported in parts per thousand) with air being the standard for nitrogen analyses and the Chicago PDB Marine Carbonate Standard (U.S. National Institute of Standards and Technology (NIST)), Gaithersburg, MD, U.S.A.) for carbon. The UGA-SIEL used NIST reference materials 8549, 8558, 8568, and 8569 to calibrate working standards in the laboratory. Measurement accuracy was within 0.2‰ for carbon and 0.3‰ for nitrogen.

We used a multi-faceted approach to determine differences in diet breadth between resident and transient coyotes and estimate diet composition based on sampled food resources. First, we used analysis of variance (ANOVA) and t-tests to explore differences in δ¹³C and δ¹⁵N values among coyote age, sex, and space use status. We also used ANOVAs to explore differences in δ¹³C and δ¹⁵N values among our landscape metrics. We then used generalized additive models (GAMs) to assess the effect of residency, age, and body mass of coyotes and the size and land cover composition of space used by coyotes on δ¹³C and δ¹⁵N values. We used GAMs rather than standard linear models because GAMs can more efficiently capture nonlinear variation. For our GAMs, we used a thin-plate spline shrinkage approach to modify the smoothing penalty for model selection purposes and restricted k to avoid overly complicated smooths. We included a binary variable in our GAMs to represent residency (resident = 1, transient = 0) whereas age (months), space use (km²), body mass (kg), road density (km/km²), canopy cover ( $\overline{x}$% cover of home and transient ranges), and agricultural cover ( $\overline{x}$% of home and transient ranges) were included as continuous variables. Because we classified all pups as residents, size of space used by coyotes was not correlated with residency (t₃₉ = 0.908, P = 0.369) and was treated as a continuum of localized-to-long distance ranging behaviors. For each isotope value (δ¹³C and δ¹⁵N), we fit the full model with shrinkage and retained all terms for inference (Marra & Wood, 2011). The effective degrees of freedom (edf) estimated from GAMs are a proxy for the degree of non-linearity relationships between predictor and response variables (Marra & Wood, 2011). Terms with 0 edfs have no effect on model fit, whereas non-significant terms that have positive edfs have small but non-zero effects.

To compare niche structure and overlap between resident and transient coyotes relative to potential food sources, we estimated isotopic niche space for both groups by plotting the isotope values for all coyote hair samples, calculating convex hulls around each group, and then calculating size-corrected standard Bayesian ellipses using package “SIBER” (Jackson et al., 2011). We then computed the total area of each convex hull (TA), sample size-corrected standard ellipse area (SEAc), and the proportion of overlap of standard ellipses between resident and transient coyotes. Because we detected an effect of land cover on stable isotope values (see Results), we also compared niche structure and overlap between coyotes on SRA and the Albemarle Peninsula of North Carolina (Arbogast, Hodge & Brenner-Coltrain, 2017). Diets, space use, and habitat selection by coyotes on the Albemarle Peninsula are well known (McVey et al., 2013; Hinton, van Manen & Chamberlain, 2015; Hinton et al., 2017) and our familiarity with both populations allows for an appropriate comparison to draw inferences about coyote ecology in the southeastern United States. When comparing SRA and Albemarle Peninsula coyotes, we included all SRA coyotes with stable isotopic values sampled during our study because meta-data (e.g., age, weight, space used, and months monitored) were not used in our SIBER analysis (Table S1).

Finally, we used stable isotope mixing models to estimate proportional contribution of food items to coyote diets. We targeted food items of coyotes in our study area consumed during spring/summer (Schrecengost et al., 2008; Ward et al., 2018) to correspond to the time period relevant to our hair samples (i.e., when hair would be metabolically active). These food items included fruit, insects, small mammals including Leporidae, game birds (northern bobwhite and wild turkey), white-tailed deer, and anthropogenic food. We collected potential prey samples opportunistically throughout spring and winter of 2019. We collected white-tailed deer hair samples from legally harvested or road killed individuals in Aiken County, South Carolina, and Oglethorpe and Athens-Clarke Counties, Georgia. Hispid cotton rats (Sigmodon hispidus) were opportunistically live trapped using Sherman traps (H.B. Sherman Traps, Inc. Tallahassee, FL, USA) in Athens-Clarke County, Georgia. Hair samples from all prey species were collected by clipping hairs from the withers (i.e., area between the shoulder blades) of each animal. All animal handling procedures for rodents were approved by the University of Georgia Institutional Animal Care and Use Committee (protocol A2018 06-024). Blackberry and persimmon samples were collected on private farms in Oglethorpe County, GA. We stored samples in plastic storage bags, with hair samples stored at room temperature and fruit samples in a −20 °C freezer until analysis. We used the same methodology to homogenize, sample, and pack prey hair into a tin capsule as we did for coyote hair and then shipped capsules to UGA-SIEL for carbon and nitrogen isotope analysis. We homogenized fruit samples using an electric blade grinder. Because our collected prey samples did not encompass all potential prey types consumed by coyotes, we acquired additional isotope values for other potential prey and anthropogenic foods from the literature (Table S2). We relied on published isotopic data for food sources geographically proximate to our study area. However, when isotopic data were not available for food sources within SRA and surrounding areas, we used data from adjacent regions. For example, we used stable isotopic data for eastern cottontails (Sylvilagus floridanus) from the Edwards Plateau region of south-central Texas (Smith et al., 2014) because such data was not available for the southeastern United States. We then used the average carbon (δ¹³C) and nitrogen (δ¹⁵N) values for each group (i.e., fruit, insects, small mammals, game birds, white-tailed deer, and anthropogenic) to identify potential dietary endpoints (Table 1).

We estimated the proportional contributions of each food source to the composition of coyote diets using the R package “simmr” (Parnell et al., 2010) in which we used a Bayesian stable isotope mixing model. The mixing model used Markov chain Monte Carlo (MCMC) methods and accounted for error in estimates of fractionation factors and variation in elemental concentrations of carbon and nitrogen in food sources that could bias model output (Phillips & Koch, 2002). We assessed model convergence using Gelman diagnostics within package ‘simmr’ in which convergence diagnostic values fell between 0.01 of 1.00, indicating our models converged. A primary challenge in using stable isotope mixing models is that food sources in many study systems may have isotopic signatures that overlap in isotope space, preventing mixing models from discriminating among food sources (Phillips et al., 2014). We used correlations from matrix plots of food source proportions to evaluate if mixing models had difficulty separating food sources. Low correlations (0.00–0.39) between sources indicated that the stable isotope mixing models performed well, whereas high correlations (0.60–1.00) indicate when models struggling to differentiate between food sources (Parnell et al., 2010).

Prior to analysis, we corrected for changes in atmospheric CO₂ over the last ~150 years (i.e., the Suess effect; Keeling, 1979). Based on core records, we applied a time dependent correction of −0.022‰ per year (Francey et al., 1999). All samples were corrected to reflect 2019 values. Additionally, for fruit, insects, game birds, and anthropogenic food, we applied diet-to-hair trophic discrimination factors (TDFs) of 2.6‰ for δ¹³C and 3.4‰ for δ¹⁵N (Roth & Hobson, 2000; Derbridge, Krausman & Darimont, 2012). For small mammals and white-tailed deer, we used a TDF of 1.6‰ for δ¹³C assuming that the hair of prey would be 1‰ lower than the TDF for prey tissues (e.g., muscle and other organs) that would have been assimilated by coyotes (Newsome et al., 2015). We did not adjust TDFs for δ¹⁵N, because tissue-specific TDFs were not reported for δ¹⁵N (Caut, Angulo & Courchamp, 2009; Newsome et al., 2015). Finally, we used an error estimate (SD) of 0.5‰ for δ¹³C and δ¹⁵N in our mixing models (Newsome et al., 2015). We conducted all statistical analyses in Program R (R Core Team, 2023).

Carbon isotope values

Canopy cover was the only informative predictor in our δ¹³C GAM in which ¹³C enrichment in SRA coyotes declined with increasing canopy cover. This was likely a consequence of predominance of C₃ plants in forested, closed-canopy habitats, while more open habitats, such as grasslands, woodlands, and agricultural cover, contain C₄ plants, which have greater δ¹³C (O’Leary, 1988; Ehleringer, Cerling & Helliker, 1997). Additionally, depleted δ¹³C values in animals can be attributed to depletion of δ¹³C values in plants growing under the canopy of dense forest stands (Drucker et al., 2008; Bonafini et al., 2013; Sykut et al., 2021). The depletion of δ¹³C in plants under closed canopies results from conditions of reduced sunlight, poor ventilation, and greater humidity when compared to those in open habitats (Teiszen, 1991; Sykut et al., 2021). Approximately 43% (18 of 41) of coyotes included in our GAM inhabited the Thomas-Augusta area and Savannah River Site, the areas with the greatest canopy cover, and exhibited lower δ¹³C values than did coyotes in the Edgefield-Aiken area. Therefore, when SRA coyotes consume co-occurring prey species and fruiting plants in areas of increasing canopy cover, they likely experience δ¹³C depletion resulting from environmental factors influencing isotopic variation at the base of their food chain.

Elevated δ¹³C values in coyote tissue are generally accepted as an indicator of the prevalence of anthropogenic food sources in coyote diets, especially in urban environments (Newsome et al., 2010; Murray et al., 2015; Larson et al., 2020; Sugden et al., 2021; Gámez et al., 2022). For example, Larson et al. (2020) observed ¹³C enrichment in coyotes along a rural-to-urban gradient of southern California and attributed enrichment to increased consumption of anthropogenic foods and domestic cats (Felis catus) by coyotes in suburban and urban areas. Several studies used a threshold δ¹³C value of −20.5‰ to differentiate natural from anthropogenic resources in diets of coyotes (Newsome et al., 2015; Larson et al., 2020) and other wildlife (Nicholson & Cove, 2022). However, a −20.5‰ threshold would indicate that anthropogenic foods were prevalent in 38% of SRA coyote diets (22 of 58 coyotes; Table S1) despite the results of our stable isotopic mixing models herein and a previous scat analysis (Ward et al., 2018) strongly contradicting prevalent use of anthropogenic foods by SRA coyotes. In fact, Kays & Feranec (2011) reported a mean δ¹³C value of −20.6‰ (range = −21.6‰ to −18.9‰) for captive canids that were fed a diet of scrap beef and kibbled dog food which is similar to δ¹³C values in SRA coyotes (mean δ¹³C = −20.9‰; range = −23.5‰ to −16.2‰; Table S1). Additionally, δ¹³C values reported in a recent study for 14 of 16 coyotes (Fig. 4C) and all 15 red wolves (Canis rufus) sampled from the Albemarle Peninsula of North Carolina (Arbogast, Hodge & Brenner-Coltrain, 2017) were above the −20.5‰ threshold despite little evidence of coyotes and red wolves consuming anthropogenic foods (McVey et al., 2013; Hinton et al., 2017). The exception was the Milltail red wolf pack on Alligator River National Wildlife Refuge that neighbored a county landfill, though their diet comprised mostly wild prey (Dellinger et al., 2011). Space used by coyotes on the Albemarle Peninsula consisted of approximately 17% forest cover (Hinton, van Manen & Chamberlain, 2015) which was significantly less canopy cover than we observed in SRA. Consequently, elevated δ¹³C values in coyotes may reflect the broader interplay of canopy cover and other landscape variables, such as agriculture and landscaping, that affect local food web dynamics rather than excessive use of anthropogenic foods by coyotes. Canopy cover can vary considerably across urban land uses such as nature preserves, recreational areas, golf courses, residential and industrial subdivisions, and low- and high-density residential areas (Porter, Forschner & Blair, 2001; Gallo et al., 2017; Ossola et al., 2019) that can affect flora and fauna diversity and facilitate carbon enrichment in food webs along rural-to-urban gradients (Golubiewski, 2006; Larson et al., 2020; Rankovic et al., 2020; Trammell et al., 2020).

Nitrogen isotope values

High δ¹⁵N values in coyotes reflects consumption of protein-rich animal prey and increases with carnivory and trophic level (Reid & Koch, 2017; Gámez et al., 2022). Elevated δ¹⁵N values in SRA coyotes was correlated with low road density, intermediate agricultural cover, and reduced space use. Road density was lowest in areas used by coyotes in the Edgefield-Aiken area where agriculture cover was also the greatest. Considering these two factors and the effect of reduced space use by coyotes, we suggest that carnivory in coyotes likely increases when space use becomes more localized, especially in areas with low human presence and with some agricultural cultivation. Indeed, this appears to support previous findings that consumption of white-tailed deer by SRA coyotes was positively correlated with smaller home ranges (Ward et al., 2018) and coyotes were observed avoiding roads and increasing their use of forest-agriculture edges during spring/summer (Youngmann et al., 2022). Furthermore, SRA coyotes did not rely on roads for acquiring white-tailed deer via scavenging (Youngmann et al., 2022) and previous research indicated roads on the Savannah River Site did not increase carrion use by coyotes (Hill et al., 2018). However, coyotes were identified as the primary mammalian scavenger of wild pig carcasses on the Savannah River Site (Turner et al., 2017) with pigs observed in approximately 8% of coyote scats (see Table 2 in Schrecengost et al., 2008). Although most consumption of wild pigs by coyotes on the Savannah River Site occurred after pig control efforts, small hooves were found in scats indicating some predation on wild pigs could not be ruled out (Schrecengost et al., 2008). Therefore, elevated δ¹⁵N values in SRA coyotes likely resulted from carnivory and some scavenging of wild prey rather than consumption of anthropogenic foods.

Furthermore, coyotes on the Albemarle Peninsula in North Carolina exhibited greater δ¹⁵N values than did SRA coyotes (11.2‰ vs 7.0‰; Fig. 4C). Agricultural cover comprised approximately 50% of space used by coyotes on the Albemarle Peninsula (Hinton, van Manen & Chamberlain, 2015) whereas space used by SRA coyotes consisted of nearly 23% agricultural cover. There are a number of potential explanations for how agricultural land cover can influence δ¹⁵N values in coyotes. Firstly, primary prey of coyotes on the Albemarle Peninsula and SRA consisted of white-tailed deer, lagomorphs, and rodents (Hinton et al., 2017; Ward et al., 2018) and prey fed on agricultural crops which can increase δ¹³C (e.g., corn) and δ¹⁵N (e.g., winter wheat, soybeans, and peanuts). The Albemarle Peninsula was more intensively used for row crops (corn, soybean, and winter wheat) than was SRA in which two of the three primary crops were C₃ plants. Secondly, fruit comprised a greater proportion of coyote diets in SRA (~24% of food items in scats; Ward et al., 2018) than in diets of those inhabiting the Albemarle Peninsula (<3% of food items in scats; McVey et al., 2013; Hinton et al., 2017). The substantial differences in agricultural cover and fruit used by coyotes in SRA and Albemarle Peninsula likely attributed to the significant differences in the δ¹⁵N values of both populations in which variation in agricultural cover affected the degree of carnivory in coyote populations.

Coyote carnivory and diet

Coyote diets during spring/summer exhibited significant variability among individuals with mammalian prey and game birds contributing more to resident diets than to transient diets. Although transients appeared to compensate for this by increasing their consumption of fruit and insects, mammalian prey and game birds still contributed to 55.7% of their diet. Current literature detailing the use of white-tailed deer by coyotes throughout the southeastern United States consists of estimates varying between 5% and 68%, via scavenging and predation, particularly during spring/summer when neonates were available (Schrecengost et al., 2008; Kilgo et al., 2012; Nelson et al., 2015; Hinton et al., 2017; Ward et al., 2018; Hinton, Rountree & Chamberlain, 2021; Youngmann et al., 2023). Because our samples reflect diets of the previous spring/summer when coyote hairs were last metabolically active (Castelló, 2018), our estimated use of white-tailed deer by coyotes likely included their consumption of adults and fawns (Kilgo et al., 2012; Nelson et al., 2015; Ward et al., 2018). For example, Ward et al. (2018) reported use of adult white-tailed deer and fawns by resident coyotes in SRA and Alabama during spring/summer months ranged between 13–30% and 11–33%, respectively, with total use of deer ranging between 36–62%. Additionally, recent broad-scale research on coyote diets in the southeastern United States could not identify scavenging behaviors as an explanation for consistent year-round use of white-tailed deer (Cherry et al., 2016; Hinton et al., 2017; Ward et al., 2018; Youngmann et al., 2022).

Surprisingly, game birds such as northern bobwhite and wild turkey made significant contributions to the diets of resident and transient coyotes. Although studies typically report game birds as incidental prey for coyotes (Henke, 2002; Staller et al., 2005; Schrecengost et al., 2008; Melville et al., 2015; McVey et al., 2013), there is some evidence that coyotes may use game birds as a food source in SRA. A recent study using metabarcoding analysis reported that wild turkeys made up 9.2% of prey detected in coyote scats of SRA during spring (Youngmann et al., 2023). Therefore, it is possible that Ward et al. (2018) underestimated the contributions of game birds in SRA coyote diets; however, they used packs rather than scats as the unit of replication in which consumption of avian prey ranged from 0–14% for packs during spring/summer. Considerable variation in use of game birds by coyote packs (i.e., zero-inflated data) may be responsible for contradictory results produced among studies. Nevertheless, we detected a moderate negative correlation between the contributions of game birds and white-tailed deer in resident (r = −0.491) and transient (r = −0.430) diets indicating that when individuals consumed more game birds they consumed less deer and vice versa. Youngmann et al. (2023) observed a similar albeit weak negative correlation between the occurrence of wild turkey and white-tailed deer in coyote scats. Current management practices for game species involving supplemental feeding may cause some game bird populations to concentrate into small areas where they are more susceptible to predation leading to increased variance in game bird use by coyotes. Therefore, we suggest that the observed negative correlations between coyote consumption of white-tailed deer and wild turkeys by both studies indicates prey-switching behaviors in coyotes that permit resident coyotes to maintain stable home ranges (Ward et al., 2018).

Small mammals and fruit are commonly reported as substantial food items for coyotes throughout their geographic range (Jensen et al., 2022), although work in some regions of the southeastern United States reported that fruit was not an important food item (McVey et al., 2013; Hinton et al., 2017; Hinton, Rountree & Chamberlain, 2021). We observed less consumption of small mammals by coyotes but similar use of fruit than is typically reported by other studies. For example, our study observed less use of small mammals including lagomorphs by coyotes than did the range of mean spring/summer values reported for those prey types in a meta-analysis on coyote diets throughout North America (Jensen et al., 2022). We also observed a lesser proportion of resident coyote diets consisting of fruit than reported by Ward et al. (2018) though use of small mammals in our study was on par with those reported in their study. Regardless, our stable isotope mixing models herein and previous scat analysis suggest that the spring/summer diets of SRA coyotes consist of 15–30% small mammals and 10–25% fruit. Fruit seeds are commonly found in coyote scats by which coyotes may serve as an important seed disperser for dominant fleshy-fruited species (Silverstein, 2005; Roehm & Moran, 2013; Campbell & West, 2022; Draper et al., 2024). Indeed, fruits comprised a significant proportion of diets for transient coyotes and transients may play an important role in regional dispersal of seeds such as those of blackberries, persimmon, wild plums, and wild grapes more so than residents because of their wide-ranging movements (Gelmi-Candusso et al., 2023).

Finally, the contribution of anthropogenic foods that we observed for coyotes in SRA likely resulted from their use of agricultural crops and supplemental feed, as these two anthropogenic food sources are likely more widespread and available to coyotes than is garbage consisting of processed foods. For example, supplemental feeding of white-tailed deer is widely practiced by deer hunters throughout the United States including in SRA to optimize hunting success or enhance antler growth, fecundity, and survival of deer (Lambert & Demarais, 2001; Webb et al., 2008; Priesmeyer et al., 2012; Sorensen, van Beest & Brook, 2014; Haung et al., 2022). Corn is ubiquitous in supplemental feed for white-tailed deer and game birds and gives feed high δ¹³C values (Nájera-Hillman & Mandujano, 2013) that may be mistaken for processed anthropogenic food which also contains corn in the form of corn syrup and starch (Newsome et al., 2010; Sugden et al., 2021). Indeed, macroscopic remains of anthropogenic food such as undigested pasta, food wrappers, and other types of garbage were found in only 0.3% of all scats collected from coyotes in SRA (Ward et al., 2018) and foraging behaviors of residents indicated little use of roads and anthropogenic land cover that would expose coyotes to garbage and litter (Youngmann et al., 2022). We encourage future studies to consider the effects of agriculture, supplemental feeding of game species, and canopy cover on δ¹³C values in coyotes to better understand how coyotes and other canids exploit anthropogenic landscapes and resources. Additionally, in a similar manner that natural prey may be accounted for in diet studies, future study designs should consider surveying study areas and/or coyote territories to identify the distribution and density of predominant sources of anthropogenic foods that study animals are exposed to rather than assuming high δ¹³C values are a result of coyote reliance on anthropogenic foods.