Net-wire fencing built to confine livestock is common on rangelands in the Southwestern USA, yet the impacts of livestock fencing on wildlife are largely unknown. Many wildlife species cross beneath fences at defined crossing locations because they prefer to crawl underneath rather than jump over fences. Animals occasionally become entangled jumping or climbing over fences, leading to injury or death. More commonly, repeated crossings under net-wire fencing by large animals lead to fence damage, though the damage is often tolerated by landowners until the openings affect the ability to enclose livestock. The usage, placement, characteristics, and passage rates of fence crossings beneath net-wire fencing are poorly understood. We monitored 20 randomly selected fence crossings on net-wire livestock fencing across two study sites on rangelands in South Texas, USA, from April 2018–March 2019. We assessed characteristics of fence-crossing locations (openings beneath the fence created by animals to aid in crossing) and quantified crossing rates and probability of crossing by all species of animals via trail cameras. We documented 10,889 attempted crossing events, with 58% (n = 6,271) successful. Overall, 15 species of medium- and large-size mammals and turkey (Meleagris gallopavo) contributed to crossing events. Crossing locations received 3–4 crossing attempts per day on average, but the number of attempts and probability of successful crossing varied by location and fence condition. Probability of crossing attempts was most consistently influenced by opening size of the crossing and season; as crossing size (opening) increased, the probability of successful crossing significantly increased for all species. Peaks in crossing activity corresponded with species’ daily and seasonal movements and activity. Density and size of fence-crossing locations were dependent on fence maintenance and not associated with vegetation communities or habitat variables. However, crossing locations were often re-established in the same locations after fence repairs. This is one of the few studies to monitor how all animal species present interacted with net-wire livestock fencing in rangelands. Our results will help land managers understand the impact of net-wire livestock fencing on animal movement. 

Fence description and condition

We surveyed boundary net-wire fence lines at both sites to verify the presence of intact, maintained fences, with ≤7 cm between the bottom fence wire and the ground. We randomly selected a 9,146-m boundary fence at El Sauz and a 2,174-m boundary fence at Santa Rosa; fence lengths differed because of the configuration of the property boundaries. Boundary fences were selected over interior fences because they often form long, linear features with no openings (e.g., gates). Therefore, animals must go under or over the wire to pass beyond the fence. Both fence lines were standard net-wire livestock fences 1.25 m in height. Both fences had an unpaved 2-track road on both sides, with mesquite and huisache woodlands beyond the roads, except for the exterior side of the fence at Santa Rosa which was grassland. We drove a utility vehicle along target fence lines at each study site to identify and record fence-crossing locations. At each identified crossing location, we recorded the maximum height of the bottom wire (m), and width (m) of each opening. We conducted these surveys of fence-crossing locations during Autumn (October – November) 2017, 2018, and Spring (April – early June) 2018, 2019. We then calculated the opening size of each crossing (m²) as the maximum height multiplied by width. When fence crossings become large enough for livestock to pass through, a common practice at these study sites is to patch the hole by securing a panel of net-wire livestock fence over the opening to discourage further crossings. Therefore, we also recorded fence-crossing locations in relation to previous repairs or patched locations.

Landscape features

Landscape features can influence wildlife habitat use (Van Dorp and Opdam 1987, Thogmartin 2001, Zemanova et al. 2017), and thus may influence where animals choose to cross fences. We quantified woody cover at fence-crossing locations using a spatial pattern analysis in ArcGIS ArcMap 10.5.1 (ESRI©, Redlands, CA) FRAGSTATS 4.2 (McGarigal et al. 2012) based on high-resolution (1-m) aerial multispectral images from the National Agriculture Imagery Program (NAIP) for 2016. We first classified imagery into 4 land cover types: woody cover, herbaceous, bare ground, and water using unsupervised image classification in ERDAS Imagine 2018 (Hexagon Geospatial; Norcross, GA; Xie et al. 2008). We conducted an accuracy assessment with 200 random points per image until ≥85% accuracy was achieved (Jensen 2016, Pulighe et al. 2016). We created 30-m buffers at fence-crossing locations and at an equal number of random locations on the same fence line at both sites. We focused on woody cover, as the most common cover types were woody and herbaceous; there was no permanent water near the boundary fencing, and bare ground was sporadic and ephemeral. At El Sauz, random locations were adjusted to not overlap other known or random crossing buffers. This approach was not feasible on Santa Rosa because crossings were relatively abundant. We clipped the imagery to the extent of the buffers to quantify the amount and spatial structure of woody cover within buffer areas. We characterized woody cover using 6 landscape metrics (McGarigal et al. 2012): patch density (PD, number of woody patches/100 ha), percentage of the landscape in woody cover (PLAND %), the mean area of woody patches (AREA_MN), the Euclidean nearest-neighbor distance between woody patches (ENN, m), the aggregation index (AI, frequency which like patches appear side by side, %) and edge density (ED, edge length of woody cover patches per unit area, m/ha).

Crossing-site usage

To assess the usage of each crossing location at each study site, we randomly assigned 10 camera traps to fence crossings identified through the fence surveys (Reconyx© HyperFire HC500 or XR6 UltraFire, Reconyx, Holmen, WI; Moultrie© A-5 Gen2 MCG-12688 Moultrie feeders, Alabaster, AL). We fastened cameras onto 1.5-m metal t-posts at a mean height (±SE) above ground of 0.54 ± 0.02 m (range 0.43–0.63) at El Sauz, and 0.66 ± 0.03 m (range 0.40–0.80) at Santa Rosa. The mean distance (±SE) from the t-post to crossing was 3.00 ± 0.12 m (range 2.40–3.58) at El Sauz and 1.72 ± 0.15 m (range 1.04–2.80) at Santa Rosa. The boundary fence at Santa Rosa often had an unpaved 2-track road close to the fence and we could not place cameras on the road; thus, the distance between the crossing and the site of camera placement was shorter than for El Sauz. We placed the cameras higher up to angle down at the crossings to address the reduced distances between cameras and fence crossings. The cameras were focused on crossing locations where wildlife crawled underneath fencing. Depressions on the top wire of these fences were rare, so we did not assess jumps over the fence by deer or nilgai.

We first deployed cameras in January 2018 as a pilot study to assess camera placement and photo quality. During the pilot study, on March 28, 2018, two fence crossings were patched with a panel of livestock fencing at El Sauz. In response, we kept cameras at the two patched locations and added cameras to two active, un-patched fence crossings. These two patched crossings (ID: EF24 & EF25) provided an opportunity to assess wildlife response to blocking of well-established fence crossings. Both patched crossings were monitored from April 2018–March 2019. We checked cameras every two weeks to ensure functionality as extreme heat greatly reduced battery life, and frequent rubbing of the cameras by cattle increased camera failure. We programmed cameras to take a 3-photograph burst with a 10-s delay (Moultrie) or 15-s delay between triggers (Reconyx), with high motion detector sensitivity. The minimum delay interval for the Moultrie cameras was 10-s with 1-s between photo bursts. A no delay setting would minimize missed crossing attempts, but our delay was sufficient due to the open visibility on the opposite side of the fence and limited occurrences of large groups (besides turkeys) passing through the fences. During the camera checks we also measured the height (m) and width (m) of each fence-crossing location to record any changes during the study.

Data analysis

We used a Kolmogorov–Smirnov test to compare the distributions of each landscape metric between known fence-crossing locations and random locations along the fence line, implemented via the R programming language (R Core Team 2013). Known crossings included any crossing location that was recorded during the four surveys. Multiple factors likely influence the distribution of fence crossings, and certain landscape features might promote clusters of fence crossings in areas. To understand whether fence crossings were randomly spaced or clustered across the fence lines we conducted a Wilcoxon signed rank test to compare distances between known crossings sites and distances between random sites along the fence.

            We classified the first two weeks (336 hrs) of photographs each month per camera, from April 2018–March 2019. We classified all animal events by species, time of day, date, and outcome of each attempted crossing event as successful or unsuccessful. A successful crossing event was an attempted crossing event where the 3-photo burst showed an animal passing under the fence or had at least half of the body through the fence crossing. We classified “attempted crossing events” as animals in close proximity to the crossings, either between the camera and crossing (about 3 m), or on the opposite side of the fence that approached or came into contact with the fence. Attempted crossing events included animals that successfully crossed or had no resulting photos to verify a successful crossing event. Animals that clearly disregarded the fence (i.e., browsing or walking a trail nearby with photos of it walking in and out of the camera frame) were recorded but were not used in this analysis. If we had consecutive photo bursts of the same individual animal, it was not re-counted, unless it was present for >1 minute since classification. We classified unrecognizable photographs of animals as “unknowns,” and further categorized as unknown carnivore or unknown ungulate, when possible.

            We hypothesized that wildlife crossing events could be influenced by vegetation composition in the vicinity of the site and characteristics of the fence (e.g., height and width of the opening). We first conducted diagnostics to evaluate potential for multicollinearity and nonconstant variance. Preliminary analyses revealed that some of the 6 Fragstats metrics of the fence crossings were correlated (Table 1 and Supp. Info Table 1). Therefore, we conducted a principal components analysis on the standardized variables (mean of 0 and SD of 1) to reduce the dimensionality of the woody cover data and retain most of the variation in a reduced set of uncorrelated variables. We retained the top two principal components that explained most of the variation in the 6 woody cover metrics for further analyses. For fence characteristics, we used the size (height × width in m) of the opening in the fence and the proximity (distance in m) to the nearest crossing location to represent ease of crossing and the density of crossings (alternative sites to cross if unsuccessful) present.

We conducted generalized logistic regression analyses to model the probability of successful crossing events relative to biotic and abiotic covariates. We conducted separate analyses of crossing events by animal species under the assumption that species’ body size and behavioral characteristics may influence probability of successful crossing. We only analyzed species with sufficient detections to be informative: deer, nilgai antelope, collared peccary, wild pig, and coyote (Canis latrans). The final logistic regression model for each species included the binomial response (0 = unsuccessful crossing, 1 = successful crossing) and principal components of woody cover metrics, size of opening, and distance to the nearest crossing as predictors; all predictors were standardized with a mean of 0 and a standard deviation of 1. We also included season in the model because preliminary analyses revealed that the frequency of photographs was higher during winter, which indicated that probability of crossing may vary seasonally. To quantify the potential season effect, we classified June, July, and August photographs as ‘summer’ and December, January, and February photographs as ‘winter’. These two seasons were included as a categorical variable in the final model to focus on relative hot and cool times of the year which may affect crossing rates as it relates to thermoregulation. Lastly, to account for potential spatial autocorrelation, we included fence crossing ID as a random effect in the models.

            We conducted an additional suite of analyses aimed at understanding how characteristics of a crossing location may influence the number of crossing events and species that attempt to cross. For instance, are crossing-site characteristics associated with use by more individuals or species, and so on. To understand temporal activity pattern of crossing attempts by multiple species, we categorized time into 8 parts of the day, each 3 hours in duration, starting with 0500–0759 hr, since 0500 hr best encompassed dawn or the first hour of light during this study. We excluded species with <100 crossing attempts from this analysis, due to low occurrence. We calculated frequencies of crossing attempts per species, location, and time, and season. We quantified species diversity and richness, excluding cattle and unidentified animals, to account for both abundance and species evenness among crossing locations based on the Shannon-Weiner index (Shannon 1948). We also modeled the Shannon-Weiner Index relative to the woody cover principal components, size of the opening, and nearest crossing via generalized linear models to determine if characteristics of crossing locations influenced the number and diversity of species that used the site. Principal component and regression analyses were conducted with R packages factoextra (Kassambara and Mundt 2020), and lme4 (Bates et al. 2014).