Co‐occurrence of bobcats, coyotes, and ocelots in Texas

Abstract Interspecific competition among carnivores has been linked to differences in behavior, morphology, and resource use. Insights into these interactions can enhance understanding of local ecological processes that can have impacts on the recovery of endangered species, such as the ocelot (Leopardus pardalis). Ocelots, bobcats (Lynx rufus), and coyotes (Canis latrans) share a small geographic range overlap from South Texas to south‐central Mexico but relationships among the three are poorly understood. From May 2011 to March 2018, we conducted a camera trap study to examine co‐occurrence patterns among ocelots, bobcats, and coyotes on the East Foundation's El Sauz Ranch in South Texas. We used a novel multiseason extension to multispecies occupancy models with ≥2 interacting species to conduct an exploratory analysis to examine interspecific interactions and examine the potential effects of patch‐level and landscape‐level metrics relative to the occurrence of these carnivores. We found strong evidence of seasonal mutual coexistence among all three species and observed a species‐specific seasonal trend in detection. Seasonal coexistence patterns were also explained by increasing distance from a high‐speed roadway. However, these results have important ecological implications for planning ocelot recovery in the rangelands of South Texas. This study suggests a coexistence among ocelots, bobcats, and coyotes under the environmental conditions on the El Sauz Ranch. Further research would provide a better understanding of the ecological mechanisms that facilitate coexistence within this community. As road networks in the region expand over the next few decades, large private working ranches will be needed to provide important habitat for ocelots and other carnivore species.

Mexico but relationships among the three are poorly understood. From May 2011 to March 2018, we conducted a camera trap study to examine co-occurrence patterns among ocelots, bobcats, and coyotes on the East Foundation's El Sauz Ranch in South Texas. We used a novel multiseason extension to multispecies occupancy models with ≥2 interacting species to conduct an exploratory analysis to examine interspecific interactions and examine the potential effects of patch-level and landscape-level metrics relative to the occurrence of these carnivores. We found strong evidence of seasonal mutual coexistence among all three species and observed a species-specific seasonal trend in detection. Seasonal coexistence patterns were also explained by increasing distance from a high-speed roadway. However, these results have important ecological implications for planning ocelot recovery in the rangelands of South Texas. This study suggests a coexistence among ocelots, bobcats, and coyotes under the environmental conditions on the El Sauz Ranch. Further research would provide a better understanding of the ecological mechanisms that facilitate coexistence within this community. As road networks in the region expand over the next few decades, large private working ranches will be needed to provide important habitat for ocelots and other carnivore species.

K E Y W O R D S
Canis latrans, co-occurrence, Leopardus pardalis, log-linear modeling, Lynx rufus, multispecies occupancy models and predation and the reciprocal effects can promote or limit potential coexistence functions between different species (Davis et al., 2018;Santos et al., 2019). Taxa within Carnivora have been widely studied, given their role affecting prey populations, and subsequent habitat structure, and ecological integrity (Nagy-Reis, Nichols, Chiarello, Ribeiro, & Setz, 2017). Examining the co-occurrence patterns of carnivores can help identify the underlying factors affecting local species distributions, ecological functions, and partitioning of resources (Rosenzweig, 1966;Schoener, 1974;Davis et al., 2011;Davis et al., 2018).
Two or more similar-sized species that share similar niches cannot coexist without one species being excluded from the community (Di Bitetti et al., 2010). The causative mechanism can be interference competition, where one species is directly antagonistic toward another and exploitative competition, where indirect interactions between species occur for a shared resource (Lesmeister, Nielsen, Schauber, & Hellgren, 2015). In North America, coyote (Canis latrans) exhibit interspecific competition and aggression toward sympatric canids (Randa & Yunger, 2006) and smaller mesocarnivores (Crooks & Soulé, 1999).
Unlike competitive exclusion or aggression, mutual occurrence of species is often facilitated by niche segregation (Davis et al., 2018;Di Bitetti et al., 2010;Santos et al., 2019). The ability for ≥2 species to coexist relies on differences in fitness and niche overlap, and these niches are fundamentally a function of interspecific interactions (Smith, Thomas, Levi, Wang, & Wilmers, 2018). In the case of niche segregation, these can help alleviate foraging competition and decrease potential negative effects of displacement by another species (Witczuk, Pagacz, Gliwicz, & Mills, 2005). In Belize, ocelot activity was correlated with areas of jaguar presence due to a shared preference for habitats (Davis et al., 2011). Davis et al. (2018 suggested that spatial coexistence between overlapping carnivores might be reduced through fine-scale partitioning of activity patterns. Further, spatial coexistence can also be facilitated by human impacts and landscape-scale features (Lesmeister et al., 2015;Smith et al., 2018).

Studies examining bobcat-ocelot interactions in Texas seem
to suggest both species mutually co-occur in the same areas, with spatial coexistence facilitated by fine-scale habitat partitioning.
Furthermore, Leonard (2016) found that ocelots and bobcats often shared overlapping 95% home ranges and were both associated with closed-canopy forests at the home range, with ocelots using dense canopies more than bobcats.
Using long-term camera trap monitoring, habitat metrics, and occupancy modeling (Rota et al., 2016), we can now study the interactions (i.e., avoidance or coexistence patterns) of such unique carnivore guilds and discern potential effects of habitat variables.
Such results can aid in explaining potential thresholds for occurrence, habitat use, and help guide management or recovery strategies (Crooks, 2002;Meek et al., 2014;Wang et al., 2019;Zemanova et al., 2017).
From 2011 to 2018, we conducted a camera trap study in South Texas to explore ocelot-bobcat-coyote interactions and potential effects of landscape-and patch-level metrics relative to the occurrence of the focal species ( Figure 2). This study is the first application of a novel and multiseason extension to the multispecies occupancy model (MSOM) of two or more interacting species developed by Rota et al. (2016) using a log-linear parameterization (MacKenzie et al., In Review). Due to the absence of predator control in the study area and surrounding ranches, we expected to observe a more natural dynamic between the species, free from man-made influences (e.g., hunting pressure). Based on previous studies, we defined three principal hypotheses for this study: (a) probability of ocelot and bobcat occurrence and detection will be negatively influenced by the presence/detection of coyotes, but ocelot and bobcat will exhibit positive co-occurrence values; (b) there will be season-specific variations in detectability and occurrence of each species; (c) ocelot occurrence will be positively linked to dense canopy cover, lower woody patch density and higher forest cover, lower edge density and farther from roads; (d) bobcat occurrence will be positively linked to areas with less dense woody patches but greater edge densities, mixed canopies and more forest cover, and farther from roads; and (e) coyote occurrence will be linked to less forest cover, greater edge and patch densities, farther from roads and open canopy cover.
El Sauz (113 km 2 ) is managed for cattle ranching and wildlife, land stewardship conservation and was located at the comingling of the Coastal Sand Plain, Lower Rio Grande Valley, and Laguna Madre Barrier Islands and Coastal Marshes eco-regions (Bailey & Cushwa, 1981 (Shindle & Tewes, 1998;Leslie, 2016).

| Noninvasive camera surveys
We conducted camera surveys on the El Sauz Ranch from 1 May 2011 to 31 March 2018, as a part of a long-term ocelot-monitoring project.
Camera grids (1 × 1 km) were designed based on a systematic, gridbased sampling method with one randomized sampling point (i.e., camera station) within each grid cell (Lombardi, Comer, Scognamillo, & Conway, 2017;Meek et al., 2014). Following United States Fish and Wildlife Service guidelines (Permit Number permit TE822908-0) for ocelot camera surveys, we maintained a minimum of 1 km spacing between adjacent camera stations. This distance was originally defined based on mean minimum distance moved by ocelots using historic telemetry data collected in the early 2000s on a nearby private ranch. Due to previous suggesting ocelots in the region are forest-interior species (Harveson, Tewes, Anderson, & Laack, 2004;Horne, Haines, Tewes, & Laack, 2009;Tewes, 1986), camera grid cells were established in the live oak-thornscrub forests located in southwestern (n = 13) and northwestern (n = 15) areas of the ranch.
At each sampling point, camera stations were in areas within or adjacent to patches of thornscrub or live oak. At each camera station, two Cuddeback ® Expert Scouting Cameras and Cuddeback ® X-Change Color cameras (Non-Typical Inc) were attached to trees or wooden stakes 0.5 m above the ground. Each camera faced each F I G U R E 1 Geographic ranges and areas of geographic overlap of ocelots (Leopardus pardalis), bobcats (Lynx rufus), and coyotes (Canis latrans) in the southern United States, Mexico, and Central America (IUCN, 2016) other and was offset 1-2 m (Lombardi et al., 2017;Satter et al., 2019) to individually identify ocelots for the long-term monitoring project identify individuals for the concurrent monitoring study. No bait or lure was used to avoid influencing the behavior of the focal species.

| Environmental variables
We quantified landscape-or patch-level metrics we believe likely influenced seasonal co-occurrence patterns. To examine whether the spatial structure of woody vegetation influenced seasonal co-occurrence patterns, we conducted a 1-m land cover classification of the study area using 2014, 1-m National Agriculture Imagery Program Digital Orthophoto Quarter Quadrangles (Texas Natural Resources System) in ERDAS IMAGINE (Hexagon Geospatial) based on four broad habitat categories: herbaceous (i.e., coastal prairie, herbaceous emergent wetlands, grasslands), water (i.e., lagunas and anthropogenic waterways), bare ground (inland dunes, caliche roads, and Texas Farm-to-Market 186 [paved road]), and woody cover (thornscrub, mesquite, and live oak forests and mottes) (Jensen, 2016;Mata et al., 2018). Using a Trimble ® Geo 7 Series Handheld Computer with 1 m precision or a Trimble Nomad ® 1050 Series Handheld Computer with GBSS 1 m precision (Trimble Navigation, Ltd), we collected 629 ground-truth points collected in June and September 2016. We accurately assessed our classification using a confusion matric until we attained an 85% threshold (Mata et al., 2018). Because camera stations were placed 1 km apart, we placed 500 m buffers (hereafter, sampling unit) around each station, to avoid potential spatial pseudoreplication among sampling units (Lombardi et al., 2017). Within each buffer, we used FRAGSTATS 4.2 to examine three landscape metrics: woody patch density (PD; # patches/100 ha), edge density (ED; m/ha), and percent landscape (PLAND; %) (Zemanova et al., 2017). Due to previous research linking the occurrence of these species with canopy cover and distance to paved roads (Cain, Tuovila, Hewitt, & Tewes, 2003;Haines, Tewes, & Laack, 2005;Hinton, Manen, & Chamberlain, 2015;Horne et al., 2009), we attempted to examine the effect of each using a representative measurement for each sampling unit. The distance (km) from each camera station to the roadway was measured. Due to the location of the high-speed roadway on the southern boundary of the ranch and the availability of larger forest patches farther from the road, we believe this variable may act as a proxy for greater availability of forested habitat for each species. Canopy cover was quantified using a Geographic Resource Solutions ® (Geographic Resource Solutions) convex densitometer at 5 m in four cardinal directions and at the center of the camera station and then averaged the five values for each station. Canopy cover estimates were categorized into three classes (open < 25%, mixed 25%-75%, and dense > 75%).

| Multiseason multispecies occupancy models of three species
In Program R 3.6.1 (R Core Team, 2019), we implemented a novel For this study, we defined a capture history containing 14 seasons with five monthly (4-week) survey occasions per season (i.e., each season was 20 weeks, with five surveys implemented a non-Markovian multiseason model, where the probability of occupancy is independent of the previous occupancy state of a unit, which allows for season-specific occupancy probabilities (MacKenzie et al., 2018). A non-Markovian model was assumed to reduce the number of parameters to estimate due to the statistically small size.
Two small sets of candidate models were considered based on biological relevant a priori hypotheses regarding the co-occurrence patterns of ocelots, bobcats, and coyotes. Each candidate model set was analyzed separately to examine both behavioral influences on co-occurrence and the potential effects of habitat metrics. The first set of candidate models (H 1 -H 5 , plus a null model) were based on five a priori hypotheses examining the influence of behavior on detection and occupancy (Table A1). We hypothesized that the likelihood of felid occurrence (ocelots and bobcats) will be negatively influenced by the presence of coyotes (Hunter, 2019;Neale & Sacks, 2001 the hypothesis that detectability of ocelots and bobcats was negatively affected by the presence of coyotes in each occasion, respectively. Model H 5 refers to the hypothesis that occupancy of all species varied seasonally, and detection was a function of a species-specific seasonal effect. A null model (H 6 ) with no species interaction or seasonal effects on occupancy and detection was also considered.
Our second set of candidate models (H 7 -H 11 ) examined the potential effects of landscape-and patch-level variables on occurrence of each species (Table A1). We did not consider models that failed to converge, as this may be a result of over-parameterization for the sample size, or it is just a bad likelihood function with multiple maxima. Due to the complexity of these models, we limited models that tested the effects of these variables to no more than two biologically relevant covariates. Canopy cover around each sampling unit was used as a categorical variable where mixed canopies were used as a reference level as it was the most dominant cover type in the study area. Based on Horne et al. (2009) and Andelt (1985), Model H 7 reflected the hypothesis that compared to mixed cover, ocelots are more likely to occur in dense canopies,  (Haines et al., 2005), bobcats (Cain et al., 2003), and coyotes (Hinton et al., 2015), we hypothesized that proximity to roads affected felids and coyotes (Model (3) bobcats and coyotes will be more likely to occur in areas with a greater edge density (per 100 ha) than ocelots which will be more likely to occur in areas with a lower edge density.
Parameter estimates for each hypothesis were estimated using 85% confidence intervals (CIs) (Arnold, 2010). We compared each set of candidate models with Akaike's information criterion (AIC) in R 3.6.1 (R Core Team, 2019) using the difference in AIC to determine which model best explained each candidate model selection.

| RE SULTS
Over 250,000 photographs were recorded over 3,920 trap months from 2011 to 2018. Of the three species, we documented >2,000 coyote detections, 1,529 bobcat detections, and 1,076 ocelot detections (Table A2). Camera stations on the ranch occurred within 55.1% woody cover, of which 60.7% contained mixed woody canopies and 32.1% dense woody canopies (Table 1). For our first model set, we found that coyotes did not negatively influence ocelots or bobcats, rather each species mutually co-occurred in the study area.  (Table 3). Detection was best explained by a species-specific seasonal trend ( Figure 4) and not by interactions with other species. The odd of ocelots occurring in a cell was estimated to be 4-5 times higher in areas with bobcats (and vice versa) ( Figure 5). For ocelots and bobcats, the odds of occupancy were 6-7 times greater in areas with coyotes, and the likelihood of all three co-occurring was cumulative at the probability scale ( Figure 5). For our habitat models, we assumed detection was a function of a species-specific F I G U R E 4 Odds ratio with 95% CI of predicted seasonal detection for ocelot (Leopardus pardalis), bobcat (Lynx rufus), and coyote (Canis latrans) for a seasonal interaction model (a) and a seasonal distance to high-speed roadway (km) model (  Presence of coyotes was a positive indicator of bobcat and ocelot occurrence-where the likelihood for each felid was greater (6-7.5 fold) when coyotes were also present. The positive effects were likely due to an abundance of preferred cover, high availably of food resources, and olfactory cues, which would allow the three species to coexist in the same areas, despite sharing a considerable overlap in body size and trophic level. Coyote interactions with felids have been studied across their range with mixed results regarding potential negative effects such as interference competition, avoidance, predation, and aggression (Neale & Sacks, 2001;O'Donoghue, Boutin, Krebs, Murray, & Hofer, 1998;Logan & Sweanor, 2001).

| D ISCUSS I ON
Hunter (2019) suggested coyotes serve as a potential predator for ocelots across their shared geographic range from South Texas to Panama (Hidalgo-Mihart et al., 2004;Hody & Kays, 2018;Schmidly & Bradley, 2016). We did not find evidence of ocelots avoiding areas where coyotes were present. In many studies within the United States, bobcats and coyotes often shared space and bobcats did not exhibit spatial or temporal partitioning (Neale & Sacks, 2001;Thornton, Sunquist, & Main, 2004;Melville et al., 2015;Lesmeister et al., 2015). Thornton et al. (2004) suggested that reduced agonistic encounters between these species might be attributed to non-  (Massara et al., 2016), but a weak negative effect of free-ranging dogs (Massara et al., 2018 (Randa & Yunger, 2006) due to their high dietary overlap.
Although we did not examine the effect of prey abundance on the occurrence of these carnivores, prey availability may also explain species coexistence in the study area. Native wildlife was not harvested and there is low habitat manipulation, which may help facilitate increased food availability on the ranch. Witmer and deCalesta (1986) suggested little competition between coyotes and bobcats occurring in areas with moderate prey populations or greater variety in food items for coyotes (Andelt, 1985).
The presence of high-speed roadways adjacent to large private working ranches may affect the occurrence of medium-sized carnivores in South Texas. These results support our hypothesis that ocelots bobcats and coyotes occur farther from roadways. Prior to this study, multiple studies have shown the negative impact of high-speed roadways on carnivore populations in urban and wildland areas (Cain et al., 2003;Haines, Janečka, Tewes, Grassman, & Morton, 2006;Klar et al., 2008;Litvaitis et al., 2015). High-speed roads adversely impact wildlife species by fragmenting habitats and populations and causing vehicle-attributed mortalities, which often lead to decreased gene flow and population declines (Cain et al., 2003;Forman et al., 2003). High-speed roadways affect the distribution and movements of wide-ranging felids including mountain lion (Puma concolor), European wildcat (Felis silvestris), and bobcat (Dickson & Beier, 2002;Tigas et al., 2002;Cain et al., 2003;Klar et al., 2008). Further, Klar et al. (2008) reported that European wildcats generally avoid areas within 200 m of roadways.
For ocelots, paved roads are strongly associated with sources of mortality in the Lower Rio Grande Valley (LRGV) and vehicle collisions remain a major mortality factor in South Texas (Haines et al., 2005). On our study site, ocelots were detected at camera stations closest to the highway, but these dense thornscrub patches along the roadway in these areas were remnants of a larger patch of thornscrub that were cleared for brush management > 35 years ago (J. Lombardi, unpub. data). High-speed roadways are also an important source of mortality for bobcats and coyotes across the country (Tigas et al. 2002;Litvaitis et al., 2015). In wildland and urban areas, bobcats avoid areas near roads (Tigas et al., 2002;Litvaitis et al., 2015). Litvaitis et al. (2015) suggested that bobcats may avoid roads because of perceived risk or limited prey in wild and urban areas of New Hampshire. Coyotes and bobcats occurring on South Texas working ranches use ranch roads as travel corridors (Bradley & Farge, 1988), but information regarding the use of these secondary roads intersecting with high-speed roadways was not reported. Hinton et al. (2015) reported that resident coyotes in eastern North Carolina significantly avoided roads. TA B L E 4 Model selection results for candidate set 2 (habitat effects) for multiseason multispecies occupancy analyses used to estimate co-occurrence (ψ) and detection ( Note: Models with a difference in AIC < 2.00 are most plausible, with associated model weight (w) and the number of parameters (K). SpA refers to ocelots, spB refers to bobcats, and spC refers to coyotes; DspABC refers to detection of species A, B, or C; SEAS refers to the seasonal effect; DistRoad refers to linear distance (km) from each camera station to Farm-to-Market State Highway 186; WPD refers to woody patch density (number of patches/100 ha) and WPLAN refers to percent of woody cover within our 500 m buffered sampling unit; Open (<25%) and Dense (>75%) refers to a classification of canopy cover measured within each sampling unit near the sampling location (i.e., camera station); and WED refers to the total length (m) of edge in woody patches per hectare within the 500 m buffered sampling unit. Detection was not influenced by positive species associations.
The importance of olfactory marking as an intra-and interspecific communication mechanism among mammalian carnivores likely plays a role in this (Allen, Wallace, & Wilmers, 2015;King, Salom-Pérez, Shipley, Quigley, & Thornton, 2017). Olfactory cues (e.g., latrines, urine, and scat) used by carnivores are usually used to indicate reproductive status, territory marking or warn individuals of their presence (Allen et al., 2015;King et al., 2017). However, it has been suggested that community scrapes and latrines may help reduce aggression and promote tolerance of neighboring individuals from the same or different species (King et al., 2017). It is plausible that while presence of such community scrapes and latrines may allow the focal species to coexist within the same areas, it may also play a role in us being unable to discern the effects of species associations in detection.
Over 7 years, the probabilities of detection between ocelots, bobcats, and coyotes varied greatly compared to the first survey season. Coyotes had the greatest probability of detection in the study area, with odds of 1.5-4 times greater compared to the first survey season. Initially, coyotes can be wary of novel objects (i.e., camera traps) in their territories, which may explain why detections increased in subsequent years (Lombardi et al., 2017). Furthermore, the social behavior of coyotes which form packs of 2-6 individuals in South Texas (Andelt, 1985), As urbanization and road networks in the adjacent LRGV increase over the next three decades, large private working ranches like our study area will provide important habitat for ocelots and other carnivore species . The use of multiseason, multispecies models with two or more interacting species gives biologists and wildlife managers the ability to conduct long-term analyses of interspecific interactions of endangered species, potential competitors, prey species, or economically valuable species. However, as the number of interacting species increases, so does the complexity of the modeling, requiring a skilled analyst to properly model and interpret the potential effects with multiple habitat covariates. The data requirements for such complex models should also be considered before commencing fieldwork to ensure sample sizes will be adequate.
Despite the absence of a larger carnivore, and perceived larger coyote and bobcat populations, ocelots do not appear to be affected by coyote and bobcat presence, which will help guide recovery efforts in areas in which all three species co-occur. However, we acknowledge the temporal scale at which we conducted this study may have been too broad to discern more fine-scale temporal dynamics not observed in this study. Further research should examine macroand fine-scale space use using GPS data, dietary analyses, and temporal segregation among these carnivores to discern any underlying effects not observed in this study.

CO N FLI C T S O F I NTE R E S T
None declared.