This study was conducted along a select portion of Farm-to-Market 106 (FM 106), a 97 km/h road that bisects ocelot and bobcat habitat within LANWR in South Texas ( Figure 1A ). FM 106 is a two-lane highway, running from Rio Hondo to Bayview and Laguna Vista in Cameron County, Texas. The road was surrounded by grassland, mesquite, and coastal prairie habitat. It was comprised of 18 km of mitigated road with nine wildlife crossing structures (WCS) with adjacent, non-continuous 2 m high chain-link fencing of varying lengths. Percent canopy cover was measured at camera height (approximately 50 cm from the ground) at the center of each site and at 5 m in each cardinal direction using GRS Densitometer (Geographic Resource Solutions, Arcata, CA). Photos from each point were used to document canopy cover. Vegetation at each site was classified into three categories of Open = 0-30%, Mixed = 31-69%, and Dense = 70-100%.

Ongoing monitoring of mitigation structures at FM 106 consisted of 36 motion-activated camera traps at the WCS Figures 1B-E . Each crossing structure had four Reconyx Hyperfire 2 camera traps, one passive infra-red (PIR) camera, and one active infra-red (AIR) camera, for a total of 6 cameras on each side of the structure, which were installed facing either end of the WCS opening at an angle that maximized the view of the opening. Each PIR and AIR camera was set to take three photos of the initial detection of an animal, with one-second intervals between photos. AIR cameras were connected to a solar panel and battery that powered an external infrared trip located at the entrance of the WCS that was used to detect wildlife that entered. PIR cameras were powered by internal batteries. Ten PIR right-of-way (ROW) camera traps were installed in pairs on either side of the road, approximately 5 m from the ROW, facing existing game trails. Thirty-two PIR camera traps were installed at all but two fence ends (FEs) adjoining WCS to monitor wildlife circumnavigation of fencing. The FEs that were not fitted with cameras and could not be included in monitoring were those adjacent to the Port Isabel Detention Center. The entire FM 106 project incorporated 92 cameras during this study. All cameras were protected by metal lock boxes mounted on metal poles approximately 30 cm above the ground. Cameras on FM 106 were serviced every two weeks, which included changing batteries when necessary, switching SanDisk (SD) memory cards, verifying dates and times, checking trip functionality, and removing any vegetation within a 2.5 m arc that may have caused false triggers. Each camera was programmed so that photos displayed a label of date, time, and location. Camera trap data from FM 106 were collected at WCS over 18 months from December 2019 to May 2021. ROW camera data were collected for eight months from September 2020 to May 2021, and FE data were collected for seven months from October 2020 to May 2021.

All photos captured on cameras were off-loaded to hard drives and renamed using ReNamer or Special ReNamer according to date and time (https://www.advancedrenamer.com/). Once photos were renamed, they were organized by camera and sorted by species. Photographs of bobcats were reviewed to determine the number of individuals present in each photo. A file structure was simultaneously created to assess an individual’s “interactions” at each site (Interactions were classifications of bobcat activities including: crossing, entry, approach, and parallel movement to structure, with a sub classification of scent marking, that occurred at the road mitigation structures). After all individuals were accounted for, DataOrganizer Ver. 4.5 was used to create a text file that included the location, date, and time (Sanderson and Harris, 2013). Once it was ensured that all photos were correctly labeled and filed, an Excel file with the metadata from the images was created using R (v4.5.0; R Core Team, 2025). Each line within the Excel file represented each photo in the dataset and included the camera name, location, date, time, species, and number of individuals within that photo. Photos were reviewed and sorted according to bobcat interaction classifications of: crossing, entry, approach, parallel, and a sub classification of scent marking. Four additional columns were added to the Excel file by reviewers: (1) class of interaction, (2) marking, (3) directionality, and (4) reviewer initials. Each interaction was classified as a 30-min period of independence (O’Brien et al., 2003). This was later confirmed by individual identification. The final photo in the series of interaction was given a class, marking, direction, and initials. Although felids may use a variety of scent marking behaviors, such as urine spraying, scraping, defecating, body rubbing, and claw marking (Kleiman and Eisenberg, 1973; Bailey, 1974; Mellen, 1993; Smith et al., 1989; McBride and Sensor, 2012; Allen et al., 2014a, b), only urine spraying and deposition of feces were considered for this study since these marking behaviors could be consistently and accurately captured using camera traps.

A dataset of 48 distinct interactions was randomly selected from FM 106, with three interaction events at each of the three site types (WCS, FE, and ROW) for each month when cameras were active at each site type. The photos from each interaction were reviewed, and the images with the best quality and angle to see distinguished “points of interest” were used for manual visual identification of the felid(s) in that interaction. “Points of interest” were deemed locations on the individual that were more easily identifiable due to the contrast of black-spotted or striped markings against the lighter color of the body. However, some of the most consistent points of interest were the features of the inner and outer leg markings, tail, and potential patterns on the side of the body depending on the definition of an individual’s spots ( Figure 2 ; Heilbrun et al., 2003; Sequin, 2001). Matches of individuals from different interaction events were made by comparing and confirming at least three of the same features on the body, or by one feature and a radio collar as suggested by Sequin (2001). Two reviewers viewed photos of bobcats independently. To resolve any inter-observer discrepancies, additional confirmed photos of the individual(s) were used as a reference and observers either agreed on identification, or the interaction was labeled as unidentifiable (UNID). The detection of at least one inconsistent spot pattern between the photos was considered enough to declare the bobcats as two distinct individuals. The individuals identified in this process served as the “true matches” to compare the accuracy and efficiency of the HotSpotter program (Nipko et al., 2020).

Images with individuals in the best positioning of the major points of interest were selected from the previous collection of photos of the 48 interaction events used for visual ID. These select images were then uploaded to HotSpotter. One or more image “chips” were created using rectangular selections around points of interest in each photo. Queries were run on each chip to allow the program to search for similarities in patterns. Each query ended in a results window that consisted of the top six matches with a rank and a score that ranges from zero to well above the tens of thousands depending on the strength of the match. True matches show the side-by-side comparison of the selected chip and the chip that HotSpotter matched based on pattern recognition. An array of red, orange, and yellow lines drawn between chips identified proposed matching points among patterns. A “no match” result indicated that no other chip within the library contained patterns that matched the chip being run. This was depicted by a red “X” through the potential matching chip. The results of HotSpotter matches were then compared to those of visual ID to evaluate the program’s success.

The final method tested was a spatio-temporal strategy, used particularly for FE and ROW sites with lower camera coverage. Due to the disproportionate number of camera traps at WCS, FE, and ROW sites (8:1:1), there were much greater odds of capturing multiple identifiable photographs of bobcat “points of interest” at WCS than at FE and ROW. Therefore, when an unidentifiable interaction event occurred at a FE or ROW, the identification of that bobcat was determined based on when and where other interaction events occurred and including bobcat traveling speeds (0.4 ± 0.40 km/h; Elizalde-Arellano et al., 2012; O’Brien et al., 2003) into the interaction interval.

Since each WCS had two camera traps installed on either side, facing the opening of the structure, images of individual bobcats interacting with the WCS were captured at multiple angles to create a complete, or nearly complete composite of the individual to build an identification database. Interactions at WCS were all identified to individual first, to ensure the greatest possibility of capturing the maximum number of photos for each individual observed. Interactions, where only a small portion of an individual was captured in an image, were able to be compared to these large photo databases. Interactions at sites with fewer cameras and therefore less coverage were identified to an individual last after a large identification database of individual bobcats was already established. Interactions with poor photo quality or only photographs of partial body parts that could not be matched with certainty to an existing individual with an established photo database were considered unidentifiable (UNID).

Kintsch et al. (2017) categorized wildlife interactions into three categories (crossing, repel, and parallel) and calculated three rates (Crossing Rate, Repel Rate, and Parallel Rate). We modified these into four categories and calculated crossing and refusal rates. Bobcat interactions at WCS were categorized as:

Those interactions categorized as D were not included in the numerator of the crossing or refusal rate because the individual did not directly interact with the WCS.

The study period used to determine bobcat usage at multiple site types (WCS, FE, and ROW), was defined as the seven-month period in which all site types were being monitored; November 2020 to May 2021. All FE had one camera at each end and on either side of the road. Due to the great distance apart from each other (range 119 m – 1,115 m), each FE camera was treated as a monitoring site. All ROW sites had one camera located on either side of the road within the right-of-way. Each camera at ROW sites was also treated as its monitoring site. All WCS had four cameras monitoring the entire site, with two cameras on either side of the structure that faced the opening of the crossing. Since both cameras on either side of a WCS faced the same direction and had an overlap of coverage, each side of the WCS with two cameras served as one sampling unit.

Bobcat individuals were classified as residents or transients based on three criteria: 1) the number of interactions, 2) the ratio of months detected/total months in study period (Schmidt et al., 2020b), and 3) the total period detected in months (Blankenship et al., 2006). Frequency distributions of all three criteria were formed using the data from all bobcat individuals and cut-off points were created based on those distributions. To be considered a resident bobcat, at least two of the following criteria were met: 1) 50 or more interactions, 2) a ratio of 2.66 or greater for months detected/total months in study period, or 3) a span of five or more consecutive months detected. Transients were defined by meeting one or none of these criteria. Juvenile individuals were not included in the analysis, as their movement and behavior at WCS were largely dependent on their mother until their dispersal at around 9 months of age (Hansen, 2007; Reid, 2006).

To determine if there were any differences in bobcat movements between either side of the road at WCS a generalized linear mixed model (GLMM) with a Poisson distribution and fixed factors of canopy cover and direction and a random effect of identified individual was used. The study area included a bend in the road, with traffic coming from FM 510 flowing north and south, and then curving to east and west towards Los Fresnos as all crossings were labelled “east” and “west” ( Figure 1 ). Only successful “A” crossings of WCS were used in this analysis.

To determine the effect of the number of bobcat individuals present on bobcat interactions at WCS, a generalized linear model (GLM) with a Poisson distribution with canopy cover and number of identified individuals as a factor was used.

An ANOVA with canopy cover as a factor was used to determine if resident or transient bobcats were observed at WCS more often. To determine if sex or residency status affects marking events a GLM with a negative binomial distribution was used with factors of sex and residency status (resident or transient).

Since the construction of some WCS on FM106 were completed before other WCS were started, some bobcats likely habituated to WCS before our camera monitoring began. To account for this, bobcats observed in the first 25 days of camera monitoring were excluded from the dataset. This resulted in a dataset of 31 individuals that were observed at a WCS for the first time and nine individuals that interacted with two consecutive WCS for the first time. An additional consideration was made for a difference in the total number of interactions for individuals included in this study. For each additional interaction after the initial encounter at a WCS, crossing and refusal rates were summed for all individuals, and a percentage ratio was calculated for up to 100 interactions.

To determine if the number of crossings at WCS by individual bobcats increased with time since their first interaction, a linear regression was used. Linear regression was also used to determine if bobcats interacted with WCS more often as they became habituated to the structure. A maximum of the first 100 interactions were considered for individuals (mean = 25.73 ± 4.81), and the number of days between each interaction was averaged for the individuals. A Welch’s t-test was used to compare the number of interactions required for an individual to successfully cross at the first WCS they encountered vs the second WCS they encountered.

For ANOVAs and linear regressions, assumptions were tested with Q-Q plots and histograms and transformations were applied to the data where necessary (Anderson and Burnham, 2004). For significant factors in ANOVAs, pairwise comparisons were conducted with Tukey’s post-hoc test. ANOVAs and linear regressions were conducted in IBM SPSS version 28.0. GLMs and GLMMs were conducted using package “lme4” in R (v4.5.0; R Core Team, 2025). Appropriate models were selected by comparison of residuals and AICc values. A significance level of p ≤ 0.05 was used across all statistical tests to determine significance. Data are presented as mean ± SD unless otherwise noted.

During the 18-month study period, 42,297 photos of bobcats and 2,338 interactions were captured across the entire study. At WCS, 39,722 bobcat photos and 2,000 unique interactions were captured. FE and ROW sites monitored during the last seven months of the study captured 2,552 bobcat photos and 338 interactions. This resulted in a 96% success rate for identifying the individuals conducting an interaction across all sites along FM 106. Only 4% of the total interactions (93 interactions) captured were unable to be identified as individuals due to poor photo quality, inadequate angles for capturing unique markings, an insufficient number of photos to confirm an identity, or a combination of these. Seventy-eight individual bobcats were identified across the 23 sites along FM 106 including:17 residents, 52 transients, and 9 juveniles (not included in the analysis) ( Figure 4A ). At WCS, 76.4% of the 72 bobcat individuals had a higher crossing rate than refusal rate ( Figure 4B ). More bobcat individuals (44 out of 72) had a crossing rate of 75% or greater ( Figure 4C ).

Canopy cover was a significant predictor of bobcat crossings, with more crossings occurring in dense canopy cover than in mixed or open areas (GLM: χ²(2) = 192.44, p < 0.001). However, bobcat direction of travel (east–west or west–east) was not significant (χ²(1) = 0.09, p = 0.768), nor was the interaction between canopy cover and direction of movement (χ²(2) = 2.63, p = 0.269; Figure 5 ).

The number of bobcat individuals at a WCS per month was a significant predictor of the number of bobcat interactions (GLM: β = 0.19, SE = 0.04, χ²(1) = 95.90, p < 0.001). More bobcat interactions occurred at WCS per month under dense canopy cover than at mixed or open canopy cover (GLM: χ²(2) = 190.50, p < 0.001; Figure 6 ).

A total of 40 bobcat individuals accounted for 1,071 combined interactions at WCS, FE, and ROW sites. Of these, 28 bobcats used WCS, seven used FE, and four used ROW. Site type was a significant predictor of the number of interactions per individual (GLM: χ²(2) = 46.91, p < 0.001), with higher interaction rates at WCS (20.10 ± 32.82) compared to FE (4.95 ± 7.43) and ROW (1.72 ± 4.39; Figure 7 ).

There was a significant effect of residency status on the number of interactions (ANOVA: F(1, 86) = 167.71, p < 0.001), with resident individuals (26.93 ± 15.69) showing more interactions at WCS than transients (5.61 ± 6.39) per canopy cover class and month. The interaction between canopy cover class and residency status was also significant (F(2, 86) = 47.88, p < 0.001). Tukey post hoc comparisons revealed significantly more resident interactions at dense sites than at mixed and open canopy sites ( Figure 8 ).

For marking behaviors at WCS, bobcat sex was not a significant factor (χ²(1) = 0.10, p = 0.749; Figure 9 ), with neither males (19.88 ± 35.54) nor females (13.56 ± 23.30) marking significantly more per site. However, residency status significantly affected marking behavior, with residents (30.13 ± 37.74) marking more frequently than transients (3.31 ± 4.36; χ²(1) = 18.56, p < 0.001; Figure 9 ). Additionally, the number of bobcat marking events per month was a significant predictor of the number of interactions at each WCS per month (R² = 0.368, p < 0.001; Figure 10 ).

The number of interactions before a successful crossing at the first wildlife crossing structure (WCS) an individual encountered (1.6 ± 0.24) compared to the second WCS encountered by the same individual (1.1 ± 0.11) indicated that the habituation periods were not significantly different (t(8) = 2.306, p = 0.104). The number of crossings at WCS vs. time since first interaction was also not significant (R²=0.00006, p=0.941; Figure 11 ). However, the number of days between interactions decreased as the number of interactions since initial contact with a WCS increased, indicating that bobcats interacted more frequently with the structure over time (R² = 0.346, p < 0.001; Figure 12 ).

Based on the large differences in success rates between the two methods tested for identifying bobcats to individuals via camera trap photos, the expectation that HotSpotter would outperform visual ID was not supported. Nipko et al. (2020) demonstrated that HotSpotter outperformed Wild-ID in the proportion of matches for jaguar and ocelot images in Belize. However, in our study, HotSpotter only had a 3% identification success for bobcats. There are several potential reasons why it was unsuccessful. Though jaguars, ocelots, and bobcats have individually distinct natural markings (Nipko et al., 2020), jaguars and ocelots are consistently darker blotches than bobcats and their markings have a greater contrast to the lighter color of their main coat, creating distinct spots and blotches, making it easy to distinguish between individuals. The coat patterns of bobcats in South Texas can vary from dark black spots that highly contrast with their main coat (similar to a jaguar or ocelot), to a total lack of spots on their coat. Therefore, the contrasting inner leg markings are more commonly used when identifying bobcats as individuals (Heilbrun et al., 2003). However, the inner legs have fewer total markings and surface area and, therefore, fewer opportunities to make comparisons between individuals as opposed to the entire side coat pattern of an ocelot or jaguar. Another contributing factor as to why HotSpotter was unsuccessful in creating true matches compared to that of Nipko et al.’s (2020) study could be the diversity of body parts we used in identification, which contributed to 6% of chips being mismatched. Nipko et al. (2020) used strict images capturing the entire left or ride side of the felid to identify individual jaguars. These side profile images are ideal for creating rectangular chips in HotSpotter, as they leave very little background in the selection. The side coat pattern of bobcats is not always prevalent or useful in identification. Therefore, we used chips of inner leg markings, along with tail, outer leg markings, and side coat markings when present. These irregularly shaped body parts leave a greater area of background when creating a rectangular chip selection in HotSpotter. This caused issues when queries were run to discover potential matches, as HotSpotter repeatedly distinguished matches using the background/ground within a chip, whether between an outer leg spot in one chip and the ground of another chip or between the ground in both chips. Additionally, interactions at WCS do not always photograph the bobcat at precisely the same angle each time. Images of extremities are sometimes all that were captured from an interaction.

Due to the variable quantity of light, distance to the subject, and pace of the subject moving past the camera, images captured from camera traps in most studies will vary in quality. Unlike Nipko et al. (2020) we distinguish photos by image quality (1 blurred, low-quality images, 2 medium-quality images, and 3 crisp, high-quality images) because the goal of using HotSpotter was to positively identify bobcats to individuals for as many interactions events as it could, including photos that may have ranked as a 1 on Nipko et al.’s (2020) image quality scale.

Manual visual identification was highly successful in identifying 85% of bobcat images to individuals using distinct patterns along the extremities. This percentage is comparable to similar studies using camera trap photographs to visually distinguish between bobcat individuals in Texas. Schmidt et al. (2020b) were able to visually identify 65% of 4,090 detections of bobcats along FM 106 in a 6.5-year-long monitoring study. And Heilbrun et al. (2003) were able to visually identify bobcats in 86% of the 65 photographs captured from a 12-month study in Central Texas. Because bobcat spot patterns are asymmetrical on the left and right sides, a key component in successful identification is being able to link the two profiles (Karanth, 1995). This can be an issue in camera trap studies due to mutual interference between camera flashes (Karanth, 1995). However, the angled setup of camera traps installed at FM 106 ( Figure 2 ) optimized the opportunity to capture both flanks without nighttime flash interference enabling the capture of images of both the left and right sides of individual bobcats. Our camera setup also optimized captures of the inner leg markings, which, like Heilbrun et al. (2003), we found was a useful aid when identifying individual bobcats due to the stark contrast of black markings against the white fur of their ventral side.

The final identification method tested, a spatio-temporal strategy used to infer the identity of bobcats at FE sites with lower sampling efforts, proved successful. This strategy was efficient in saving time during the identification process by suggesting an ID with 90% accuracy based on when and where interactions at the closest WCS occur, and the identification of the individual performing that interaction. Although this method was intended to be confirmed by visual identification, the accuracy was high enough that it could potentially be used to determine the identity of individuals for interactions at FE who do not have sufficient photographs to use visual ID.

Individual bobcats that directly interacted with WCS along FM 106 had a greater crossing rate (75%) than refusal rate, providing an indication that ocelots would be expected to have similar crossing rates if they were not rare and endangered. While we demonstrated that bobcats successfully used the WCS, the 25% refusal rate appears high until put into context. Bobcats have complex movement patterns and exhibit a variety of behaviors (Horne et al., 2009). Because successful use of WCS has become defined as an animal passing from one side of the road to the other (Clevenger and Waltho, 2005), uses other than completely passing through are incorrectly categorized as refusals. We classified 100 refusals by bobcat that included other behaviors: 44 instances included smelling behavior either in or around the structure and 56 instances of day-bedding in the culvert. Bobcats were observed day-bedding from 0.75 to 10 h per day between March and October, the hottest months of the year in South Texas. All but one of these 100 interactions were conducted by resident individuals who were arguably familiar with the WCS. While most of these interactions resulted in the individual crossing to the other side of the WCS, 12 of them did not.

There was a positive relationship between bobcat interactions at WCS and the number of identified bobcats present. Canopy cover had a significant interaction with number of identified individuals indicating canopy cover may influence the number of bobcats interacting at a WCS. Based on these results, WCS should be evaluated separately when evaluating the “success” of a WCS based on the number of interactions that occur at individual WCS, as there are factors such as canopy cover that may influence trends. One other factor is residency status, with resident individuals observed for longer periods of time compared to transient individuals (Janecka et al., 2008; Litvaitis et al., 1987). In a previous study of bobcats along FM 106, individual interactions at WCS ranged from 1 to 172 (Schmidt et al., 2020b), with resident individuals observed most frequently and over a greater time period. During this 18-month study, individual interactions of bobcats had a similar range of 1 to 192 at a single WCS.

Exclusionary fencing is designed to funnel wildlife towards WCS for safe passage from one side of the road to the other (Clevenger et al., 2001). However, how far from a crossing to extend guiding fencing is not well studied. The fencing associated with the WCS on FM 106 was not continuous along the entire roadway and was limited at each crossing to the end of adjacent habitat, the nearest driveway or road intersection, or limited to 1000 feet. Highway crossings by bobcats on FM 106 were more frequent at WCS (under the road) than at FE and ROW (on the road). This demonstrates that the fencing associated with WCS is effective in funneling bobcats toward WCS and likely decreases the risk of bobcats entering the roadway via direct access from the habitat. While fencing has been shown to reduce wildlife-vehicle collisions, it needs to be maintained permanently, may increase mortality at fence ends, and act as a barrier to wildlife movement (van der Ree et al., 2015).

More resident individuals were observed at WCS than transient individuals. Canopy cover also influenced the number of resident and transient interactions, with significantly more resident interactions occurring at dense sites. Bobcats are territorial and use scent marking for communication: scraping, urine‐spraying, claw‐marking for communication (Allen et al., 2014a) and scent marking is thought to be primarily used by adult bobcats (Bailey, 1974). The results of the present study indicate that most marking events were conducted by resident bobcats. Allen et al. (2014a) did not collect sex data for the bobcats they observed and stated that data on sex would be important in future studies. Surprisingly, sex was not found to be a significant factor in the number of marking events at WCS in the present study.

Bobcats increased their visits to WCS after their initial interaction. However, the rate at which individuals successfully crossed the structure did not increase. This could be due to several factors; an individual may have no intention of reaching the other side of the road, merely interacting with the WCS to inspect the longevity of scent marks, drink water that often pools at some sites, or seek refuge in the shade of the structure. This would result in a “refusal” of the WCS and could be interpreted as a failure. However, these behaviors could also be interpreted as habituation themselves, as individual bobcats incorporate the WCS into their habitat obtaining resources, sheltering, and successfully crossing the road.

Bobcats could also be habituating to WCS. These individuals may be increasingly incorporating WCS into their activity patterns (Elizalde-Arellano et al., 2012), and therefore, more frequently utilizing WCS. The variation in the data could be accounted for by the unknown home ranges of individual bobcats. Those who have WCS in the middle of their home range may interact with structures more frequently than those with WCS on the fringes of their home range.