Food resources affect territoriality of invasive wild pig sounders with implications for control

Interest in control methods for invasive wild pigs (Sus scrofa) has increased due to their range expansion, population growth, and an improved understanding of their destructive ecological and economic effects. Recent technological advances in traps for control of pig populations facilitate capture of entire social groups (sounders), but the efficacy of “whole-sounder” trapping strategies is heavily dependent on the degree of territoriality among sounders, a topic little research has explored. We assessed territoriality in wild pig sounders on the Savannah River Site, South Carolina, USA, and examined whether availability of food resources provided by a municipal-waste landfill affected among-sounder territoriality. We estimated utilization distribution overlap and dynamic interactions among 18 neighboring sounders around a landfill. We found that although neighboring sounders overlapped in space, intensity of use in shared areas was uniformly low, indicating territorial behavior. Neighbors tended to share slightly more space when closer to the landfill waste cells, indicating availability of a super-abundant resource somewhat weakens the degree of territoriality among sounders. Nevertheless, we conclude that sounders behaved in a generally territorial manner, and we discuss implications for whole-sounder trapping programs, particularly near concentrated resources such as landfills and crop fields.


Scientific Reports
| (2021) 11:18821 | https://doi.org/10.1038/s41598-021-97798-z www.nature.com/scientificreports/ a trap, we captured an adult sow over bait via dart rifle from a tree stand (Supplementary Methods). Because we expected the home range of the dominant mature sow in each sounder to represent that of the sounder, we attached a GPS collar to the largest sow in each sounder, assuming she was the most dominant member. All GPScollared sows were either subadult (1-3 years) or adult (> 3 years). We programmed collars to acquire GPS fixes every 2 h. Capture and handling procedures were approved by the University of Georgia's Institutional Animal Care and Use Committee (IACUC protocol numbers A2012 08-004 and A2015 05-004) and were conducted in compliance with the ARRIVE guidelines for the immobilization of animals for studies conducted in the field.
Space-use estimation. We used dynamic Brownian bridge movement models (dBBMM 21 ) to estimate season-long (3.5-month period; 27 Mar to 12 Jul during each year), monthly, and weekly utilization distributions (UD) for each collared sow. Although we tracked pigs outside of the season-long date range, we selected that date range because it maximized sample sizes while maintaining a consistent season-long date range between years. The monthly time period in 2014 included Apr, May, and Jun, and in 2016 included Feb, Mar, Apr, May, Jun, and Jul. The weekly time period in 2014 included 14 7-day periods between 27 Mar and 12 Jul, and in 2016 included 26 7-day periods between 22 Jan and 31 Jul. We specified a window size of 29 and a margin of 11 when estimating dBBMMs based on the temporal resolution of GPS locations and our a priori assumptions about the temporal scale of major behavioral shifts 22 . Based on recent research reporting average GPS collar location fix errors of 10-20 m on similar study sites 23 , we used an error estimate of 20 m when estimating all UDs. We estimated all UDs on identical 50-m resolution spatial grids that encompassed all GPS locations collected during each of 2014 and 2016. We defined home ranges and core areas using 95% and 50% UD isopleths, respectively 24 . We estimated spatial overlap of home ranges and core areas for each pair (dyad) of neighboring sows across each of our temporal scales. We considered sows to be neighbors if their home range boundaries intersected during any time period 25 . For each dyad, we calculated two-and three-dimensional overlap as the proportional area of overlap of the two UDs (2D) and the volume of intersection (VI), respectively 26 . We estimated dBBMMs and calculated 2D overlap and VI in the R statistical environment (Supplementary Methods 27 ).
Modeling UD size and overlap. We used regression methods to quantify effects of various predictors on UD size and overlap at each of our temporal scales. Given 2D overlap and VI metrics are bound between 0 and 1, we used beta regression and generalized linear mixed-effects models with a beta error distribution. For each of our three response variables (i.e., UD size, 2D overlap, and VI), we developed individual models for each time period to focus our analysis on variation within time periods. We used generalized linear regression rather than mixed-models to analyze season-long UD size and overlap because we lacked repeated season-long measures for dyads. We fit sow ID and sample period (i.e., the specific week or month during weekly or monthly time periods) as random intercepts in each mixed model to account for unbalanced sample sizes and repeated sampling of individual sows 28 . Given VI estimates relate to dyads and that sows could be included in multiple dyads during a given sample period, we used the mean of all VI estimates for dyads including a given sow during each sample period as the response variable in VI models. For example, if a sow was included in four different dyads during a  www.nature.com/scientificreports/ given week, we calculated the sow-specific VI as the mean of the four weekly dyads including the sow. Similarly, we modeled 2D overlap as the mean of sow-specific 2D overlap estimates for dyads including a given sow during each sample period. We sought to understand the influence of the landfill while accounting for other variables known to affect space use by wild pigs on our study site (e.g. 29 ). Hence, for UD size, 2D overlap, and VI, we developed candidate models based on four covariate groups representing effects related to: (1) the landfill; (2) sow characteristics; (3) vegetation cover type; and (4) weather conditions (Supplementary Table S1, Supplementary Methods). The landfill covariate group included distance to the active waste cells within the landfill (m; Dist.WC) and percent of the UD within the landfill boundary (%UD.LF). Sow covariates included body mass (kg; Mass), age (adult or subadult; Age), and sounder size (total number of pigs of all ages in the group; Sounder). Vegetation cover type covariates included the percent of the UD within bottomland hardwood/wetland cover (%UD.BLHW), and weather covariates included period-specific mean temperature (°C; Temp) and barometric pressure (mb; Press).
We included combinations of and interactions among covariates in the four covariate groups, but we limited interactions to the third order to minimize the risk of overparameterizing models and to simplify interpretation. We fit UD level (home range and core area; UD.lev) and year (2014 and 2016; Year) as categorical variables in season-long, monthly, and weekly overlap models, but dropped year from season-long UD size models due to model convergence issues. We did not fit weather covariates in season-long models due to model convergence issues. Additionally, we did not fit interactions involving year or weather covariates in any models to minimize the risk of overparameterizing models. We used second-order Akaike's Information Criterion (AIC c 30 ) for model selection and considered models with ΔAIC c < 2 as plausible models. We estimated regression models in the R statistical environment (Supplementary Methods).

Dynamic interactions.
We examined attraction and avoidance within dyads during each time period using two indices of dynamic interaction that explicitly focus on locations for a dyad that are simultaneous in both space and time 31 . We considered locations of a dyad to be simultaneous in space and time if they were < 50 m and < 10 min apart, respectively. Accordingly, we specified distance and time thresholds of 50 m and 10 min, respectively, in calculation of both dynamic interaction indices. Following Benhamou et al. 32 , we calculated dynamic interaction indices only for dyads with VI estimates > 0.1 during a given time period to avoid calculating indices for dyads with little to no probability of overlapping.
First, we used proximity analysis to estimate the frequency at which pigs of a given dyad were close in time and space 31 , which is calculated as the ratio of locations simultaneous in space and time to locations simultaneous only in time. Second, we calculated the half-weight association index (HAI 33 ) to test for attraction or avoidance between individual sows of a given dyad during periods of simultaneous use within a shared area. The HAI index is a localized metric that compares the number of dyad locations simultaneous in space and time within a shared area against the number of solitary locations within a shared area 31 . The HAI is bound between 0 and 1, with values approaching 0 and 1 indicating avoidance or attraction, respectively, within the shared area. Because our study was focused on effects of the landfill resource, we calculated HAI using shared areas defined by polygons for: (1) the landfill boundary; and (2) the waste cells within the landfill boundary ( Fig. 1). We estimated dynamic interaction indices in the R statistical environment 27 (Supplementary Methods).
Wild pig capture and telemetry. We captured most members of 18 target sounders (n 2014 = 7, n 2016 = 11, total pigs captured = 127), except members of the 2 sounders from which sows were darted over bait, and we tracked one sow from each sounder. These 18 sows provided 19,508 GPS locations (n 2014 = 4723, n 2016 = 14,785). Mean sounder size during 2014 and 2016 was 8.8 ± 2.6 (SD) and 6.4 ± 6.5 pigs, respectively, and mean mass of tracked sows during 2014 and 2016 was 81.8 ± 19.1 and 72.5 ± 19.8 kg, respectively. Of the 7 sows tracked in 2014, 4 were subadult and 3 were adult, and of the 11 sows tracked in 2016, 4 were subadult and 7 were adult. Space-use. Mean home range and core area sizes were largest during the season-long time period and declined during shorter monthly and weekly time periods (Table 1). We identified 19 and 30 dyads in 2014 and 2016, respectively. At the season-long scale, dyads exhibited considerable 2D overlap (Fig. 2), but average VI estimates were generally low across all time periods. Overall, mean 2D overlap and VI of home ranges were < 0.3 and < 0.1, respectively, across all time periods, and mean 2D overlap and VI of core areas were < 0. Modeling UD size and overlap. The top season-long UD size model included UD level, landfill, vegetation cover type, and sow covariates and an interaction between UD level and percentage of UD in the landfill (Supplementary Table S2). Overall, season-long UD size at the home range level increased as distance to the active waste cells decreased (Fig. 3d), and season-long UD size decreased for sows with home ranges that included a greater percentage of the landfill ( Fig. 3a Fig. S4a.i, a.ii) and decreased for sows with UDs that included a greater percentage of the landfill (Supplementary Fig. S5a.i, a.ii). The top season-long 2D overlap model included interactions among UD level, landfill, and sow covariates (Supplementary Table S6). Overall, season-long 2D overlap increased for sows with UDs that were closer to the waste cells (Fig. 3e) and for sows with UDs that included a greater percentage of the landfill (Fig. 3b). At the home range level, season-long 2D overlap decreased for older sows but increased for heavier sows ( Table 2). The top monthly 2D overlap model included interactions between UD level and landfill covariates (Supplementary Table S7). Monthly 2D overlap increased for sows with UDs that were closer to the waste cells ( Supplementary  Fig. S4b.i) and for sows with home-ranges that included a greater percentage of the landfill (Supplementary Fig. S5b.i; Table 2). The top weekly 2D overlap model included interactions among UD level, landfill, sow, cover type, and weather covariates (Supplementary Table S8). Overall, weekly 2D overlap increased for sows with UDs that were closer to the waste cells ( Supplementary Fig. S4b.ii), that included a greater percentage of the landfill (Supplementary Fig. S5b.ii), for heavier sows, and for sub-adult sows ( Table 2).
Top season-long, monthly, and weekly VI models included interactions among UD level, landfill, sow, and vegetation cover type covariates (Supplementary Tables S9-11). Season-long, monthly, and weekly VI increased for sows with home ranges that were closer to the waste cells and decreased for sows with home ranges that Table 1. Mean and standard deviation (SD) of wild pig (Sus scrofa) home range and core area sizes (hectare) across time periods (Season-long [n season = 2; n pigs = 14], Month [n months = 6; n pigs = 18], and Week [n weeks = 26; n pigs = 17]) and utilization distribution (UD) levels representing home ranges (95% UD contour; HR) and core areas (50% UD contour; CA) for pigs tracked on the Savannah River Site, South Carolina, USA in 2014 and 2016.  Additionally, season-long and weekly VI decreased for sows with home ranges that included a greater percentage of bottomland hardwood/wetland cover and for heavier sows (Table 3). Season-long VI increased at the home range level for sows in larger sounders, but decreased for older, heavier sows.  Table 4). In 2014, pigs from only two dyads were close in space and time across all time periods, mean landfill-HAI estimates were < 0.0001 across all time periods, and mean waste-cell-HAI estimates were all 0 (Table 4). In 2016, mean HAI estimates were all < 0.031 across time periods (Supplementary Fig. S7, Table 4).

Discussion
Under a strict definition of territoriality as involving exclusive use of an area 34 , wild pig sounders in our study cannot be considered territorial. Our dyads often exhibited extensive overlap in space at the level of the home range, and to a lesser extent, even core areas. However, some researchers have recognized the degree of territoriality is characterized by a continuum, which often has been evaluated using the extent to which individual home ranges overlap, although the extent of overlap that distinguishes territorial from non-territorial behavior is unclear (reviewed by 7 ). Aspects of space use sharing other than simple 2D spatial overlap (e.g., VI, frequency of simultaneous locations within shared areas) can also inform our understanding of the degree of territoriality in a species. By considering these other metrics, we detected clear signs of among-sounder territoriality that were not adequately captured by area-based estimates of 2D overlap. Although VI estimates for dyads were often > 0, even in core areas, mean values were nevertheless quite low, indicating very low probability that dyads shared use of the same space. Similarly low values of VI have been reported for other species known to be territorial (e.g., American redstart, Setophaga ruticilla 35 ). Additionally, negligible but non-zero VI estimates can occur even in strictly territorial species, because UDs inherently include peripheral areas beyond the extent of dyad locations 36 . Much of the 2D overlap we observed likely reflects occasional excursions into a neighbor's home range, which typically is www.nature.com/scientificreports/ not frequent among territorial species and can misrepresent the degree of territoriality, as judged by the extent of space-use sharing 26 . The VI metric on the other hand is based on the intensity of use within shared areas, which reduces the influence of occasional excursions and reflects more nuanced aspects of the extent to which neighboring individuals share space. Finally, our dynamic interaction metrics demonstrated that sounders avoided each other, very rarely occurring in the same place at the same time. Likewise, low direct contact rates for members of neighboring sounders have recently been reported from other studies in the southeastern U.S. 37,38 , although Yang et al. 38 documented some direct inter-sounder contacts. We therefore believe that despite sharing as much as 60% of the area of their home ranges with neighboring sounders, our results largely support the conclusions of Gabor et al. 11 and Sparklin et al. 12 in that wild pig sounders displayed territorial characteristics by generally avoiding each other in both space and time.
The degree of territoriality we observed among sounders apparently was influenced to some extent by the tremendous food resource provided by the landfill. Our best-supported VI model indicated the intensity of use within shared areas of home ranges increased for dyads that were closer to the waste cells. This suggests that although sounders rarely came in contact at the waste cells, they nevertheless shared space to a greater extent closer to the waste cells than away from them. In contrast, the intensity of use within shared areas declined with increases in both the percentage of the landfill footprint and of bottomland hardwoods in home ranges. Sounders frequently were located within the landfill footprint during nights they were never observed accessing the waste cells. Wild pigs have been observed foraging extensively in the open fields within the landfill footprint but outside of the waste cells (K. Cox, U.S. Forest Service, personal communication), suggesting these landfill fields also represent a valuable resource for wild pigs, if perhaps not to the extent of the concentrated resources in the waste cells themselves. Given the similar decline in VI with increases in percentage of the UD in the landfill and in bottomland hardwood/wetland forests, our results suggest some preferred natural habitats (i.e., bottomland   39 . Home range sizes were generally smaller and overlapped less when they included a greater percentage of the landfill, suggesting a mechanism by which more territories could be supported in the vicinity of the landfill by the resources provided there. Similarly, resource-mediated territory packing has been shown to be facilitated by reduced home range size in other taxa 40 . However, sizes of home ranges increased when they were closer to the waste cells, contrasting with the expectation that increased resource availability would lead to decreased home range sizes. We are uncertain as to an explanation for this apparent contradiction in the effect of landfill and waste cell resources on home range size but suspect it may have been attributable to the spatial distribution of    www.nature.com/scientificreports/ cover types within home ranges. Sounders with home ranges closer to the waste cells exploited the concentration of food resources there, which were abundant to the point defense of space and resources was rendered unnecessary. However, that these sounders did not need to defend space near the waste cell to maintain access to food may have reduced costs associated with expanding their home ranges to access other resources (cover and water), despite being closer to concentrated food resources near waste cells. In contrast, sounders used less space and were more territorial when home ranges were located where the spatial configuration of cover types enabled sounders to reduce home range sizes while still maintaining access to high quality food resources in the landfill and critical food, cover, and water resources in bottomland hardwoods. Thus, sounders with home ranges that included a greater percentage of the landfill, but not necessarily home ranges closer to the waste cells, used less space because their home ranges were positioned such that they could reduce home range size while still maintaining largely exclusive access to resources in both the landfill and bottomland hardwood/wetland cover. Although we did not explicitly assess the effects of pig density on territoriality, the difference in density between 2014 and 2016 may have influenced space use of pigs around the landfill. We estimated a 67% reduction in density between 2014 and 2016. This marked reduction apparently was due to particularly intensive control efforts that occurred in the 32.4-km 2  A potential explanation of the level of spatial overlap we observed may pertain to sounder structure. Unlike Gabor et al. 11 and Sparklin et al. 12 , Boitani et al. 9 reported complete overlap of two sounders of wild boar in Italy. Gabor et al. 11 suggested that the apparently conflicting results between their study and that of Boitani et al. 9 were attributable to differing definitions of sounders between the two studies. Gabor et al. 11 considered sounders as stable groups of breeding-aged females and their young that shared a common range and often form non-random subgroups. They proposed that the female groups monitored by Boitani et al. 9 may have corresponded more closely to what Gabor et al. 11 considered subgroups of the same sounder. Like Boitani et al. 9 , we lacked genetic information on relatedness of various sounders in our study area and may therefore have monitored subgroups as separate sounders, when in fact they belonged to the same sounder. Had that been the case, however, we would have expected to observe greater VI and dynamic interaction values, as such subgroups would not likely avoid areas of shared space and would not avoid each other.
Effects of sow characteristics on overlap metrics were equivocal. Older and heavier sows, presumably dominant in interactions between sounders, had lower mean VI and older sows had lower mean 2D spatial overlap, indicating that sounders with such sows had a competitive advantage that enabled them to exclude other sounders with younger and smaller sows. Heavier sows also had greater 2D spatial overlap with neighbors, suggesting their larger size conferred a competitive advantage that lowered perceived risks of intruding into home ranges of neighboring sounders. However, contrary to expectations, larger sounders, presumably dominant to smaller sounders, had higher VI. Our results suggest increased sounder size does not necessarily confer a competitive advantage in wild pigs, in contrast to many other social species [41][42][43] . This finding could have been driven by differences in age composition among our sounders. For instance, a sounder with a single adult sow and 8 piglets would likely be a weaker competitor than a sounder with 4 adult sows and 2 yearlings, though the former is the larger sounder. However, given the generally low VI estimates across all sounders, we suspect that these apparent effects of sow characteristics may be confounded by other aspects of sounder structure (e.g., age and weight of all pigs in a given sounder, relatedness among neighboring sounders). Further research is needed to determine what aspects of individual sows and sounders are most influential on sounder territoriality.
Collectively, our findings have implications for programs aimed at control of this invasive species based on removal of entire sounders. One tenet of whole-sounder trapping assumes that sounders are territorial and have high site fidelity, the implication of which is that when a sounder is removed, pigs (other than lone boars or small bachelor groups) are then absent from that territory for some period of time 6,12 . Monitoring response to an experimental removal, Bastille-Rousseau et al. 8 reported limited evidence of substantive shifts into the removal area by adjacent GPS-collared pigs during 14 weeks of monitoring post-removal. However, despite the generally territorial behavior we observed among sounders and the fact that sounders avoided each other, our 2D home range data indicated their home range boundaries overlapped extensively in space. Thus, removal of one sounder would not necessarily render that territory free of pigs. Furthermore, several different sounders used the landfill, suggesting a concentrated resource such as a productive crop field with high nutritional value may also be used by multiple sounders. In such cases, where multiple sounders, each from a different territory, converge on a single high-value resource, trapping at or around that location (e.g., along nearby access routes) may effectively remove sounders from a much larger area than a single territory 12 . Indeed, four of our sounders in 2016 were captured at a single trap location within 125 m of the landfill over a period of 9.5 weeks. Thus, managers should not assume that when a single sounder is removed near a concentrated resource, no additional sounders can be captured there until additional monitoring reveals no more are present. However, at the greatest distance from the waste cells, 2D overlap approached zero for our season-long period, suggesting that in the absence of such a resource, additional sounders are not likely to be captured at the same trap location.
We expected that if territoriality among sounders was weak, then evidence of territoriality may only be detected over shorter time periods (e.g., weeks); i.e., the greater the duration, the greater the likelihood occasional incursions by neighbors would obscure exclusivity of space use. Although mean VI values were uniformly low, regardless of our time period duration, the magnitude of landfill predictor effects declined as the duration of time periods shortened. Such attenuation of landfill predictor effects over shorter time periods suggests the degree of territoriality among sounders is strongest over shorter time periods. Nevertheless, our 3.5-month season period www.nature.com/scientificreports/ is applicable to the length of time during which most wild pig trapping efforts typically occur, often during winter and early spring when availability of natural foods is low and baiting is most effective. Thus, we believe that control operators should expect sounders to tend to behave in a generally territorial manner for the duration of most trapping efforts, but multiple sounders may use the same space in the presence of concentrated high-quality resources, such as those provided by the landfill on SRS or agricultural fields.