Introduction

Humans have altered terrestrial ecosystems globally and many of these impacts result in habitat fragmentation and loss (Fahrig 2017; Cross et al. 2021). Land conversion for agricultural or energy production can be detrimental due to direct causes of habitat loss (Thomas et al. 2018; Johnson et al. 2020; Lark et al. 2020) and the resulting landscape patterns likely have strong effects on wildlife communities. Cultural landscapes are a geographic area in which the relationships between human activity and the environment have created ecological, socioeconomic, and cultural patterns and feedback mechanisms that govern the presence, distribution, and abundance of species assemblages (Farina 2000; Jones 2003). While there is a robust literature on the effects of cultural landscapes on wildlife (Young 1997; Foster 2002; Fuller et al. 2017), the scale at which habitat alteration has on wildlife behavior remains an open and important question.

The issue of scale remains a primary question in ecology (Morris 1987; Levin 1992; Denny et al. 2004; Jackson and Fahrig 2015; Elderd et al. 2022). Patterns and processes occur at multiple spatio-temporal scales (Wiens et al. 1989; Liang et al. 2022) and the scale of effect is often defined as the scale, or range of scales, that explains the most variation in a given ecological response (Moraga et al. 2019; Blackburn et al. 2021; Arroyo-Rodriguez et al. 2023). Using multiple grains and extents to identify the scale of effect is a common methodology employed to understand how a given species perceives and responds to their environment (Jackson and Fahrig 2012; Moraga et al. 2019; Blackburn et al. 2021). The scale of effect of a given ecological response can be influenced by many factors ranging from a species’ biology (Miguet et al. 2016; Martin 2018), an individual’s physiology (Jackson and Fahrig 2015; Miguet et al. 2016), and external pressures like anthropogenic disturbance (Hamer and Hill 2000; Mangiacotti et al. 2013). There is a growing body of literature on how wildlife perceive and respond to landscape patterns and the scales at which this occurs (Delaney et al. 2010; Mérő et al. 2015; Šálek et al. 2015). There are numerous examples of mammals responding to urbanization metrics at various scales (Lombardi et al. 2017; Moll et al. 2020; Fidino et al. 2021; Robb et al. 2022). Species range from synanthropic to highly sensitive to human disturbance, and their success in human-modified landscapes depends upon their traits and tolerance of human disturbance (Ferreira et al. 2018).

The rangelands of South Texas are quintessential examples of cultural landscapes where the integration of multiple management objectives has resulted in novel landscape conditions. Despite relatively low human population densities, landscape patterns in these rangelands are heavily altered for agriculture and extraction of natural resources (Dodd et al. 2013; Tunstall 2015). The Eagle Ford Shale oil and natural gas reserve of southwestern Texas, one of the largest reserves in the world, has experienced unprecedented energy development in recent decades (Gilmer et al. 2012; Tunstall 2015). Energy development has resulted in increased fragmentation through the creation of roads and clearing of native landcover for energy infrastructure. The configuration of patches of brush and herbaceous vegetation is also a product of centuries of grazing and brush management practices (Fulbright and Ortega-Santos 2013). In more recent decades, the optimization of livestock production and wildlife management has resulted in the creation of systemic brush mosaics. Systematic brush mosaics are areas where brush is cleared at precise spacing to create alternating strips of brush and herbaceous ground cover increasing the interspersion of forage and thermal refuge resources for livestock and wildlife (Fulbright and Ortega-Santos 2013; Fulbright et al. 2018). Systematic brush mosaics are a representation of the integration of multiple objectives including livestock production and wildlife conservation and have proven effective management for target species such as white-tailed deer (Odocoileus virginianus) and northern bobwhite (Colinus virginianus; Webb et al. 2006; Hernández et al. 2013; Fulbright et al. 2018). However, the effects of these novel environments on carnivore species such as bobcats (Lynx rufus) that have been demonstrated to be negatively affected by fragmentation (Riley et al. 2003) have not been examined (Bradley and Fagre 1988). A key component to understanding the effects that landscape patterns have on ecological responses of bobcats is identifying the scale at which the effects are strongest.

Carnivores are often used as means of monitoring ecosystems due to their trophic position (Marneweck et al. 2022). The bobcat is an ambush predator that relies on concealment cover to hunt and therefore is likely to be impacted by brush management through the alteration of the distribution of concealment cover, increases in patch edge density, and interspersion of cover types, which provide different prey communities in close proximity and concealment cover adjacent to herbaceous plant communities. Bobcats often respond negatively to fragmentation associated with habitat loss (Crooks 2002; Riley et al. 2003; Lewis et al. 2015; Smith et al. 2020) and some evidence suggest areas with greater fragmentation are less productive for bobcats. The Habitat Productivity Hypothesis suggests there is an inverse relationship between space use requirements and habitat productivity. Animals on less productive sites require more space to meet their dietary and life history requirements (Harestad and Bunnell 1979). Numerous studies have used home range size as a measure of habitat productivity for wildlife species (Harestad and Bunnell 1979; Riley et al. 2003; Seigle-Ferrand et al. 2021; Quinlan et al. 2022). For bobcats in agricultural areas, increased fragmentation has been associated with larger home range sizes indicating fragmentation reduces habitat productivity for bobcats in that system (Tucker et al. 2008). Bobcats also respond strongly to prey abundance and increases in prey can result in smaller home range sizes (Litvaitis et al. 1986). As jaguars (Panthera onca), pumas (Puma concolor), and red wolves (Canis rufus) have been extirpated or reduced to low population densities in Texas (Daggett and Henning 1974; Nowak 2002; Harveson et al. 2012), bobcats, have become the largest remaining obligate carnivore and de facto apex predator in many ecosystems (Bradley and Fagre 1988; Lombardi et al. 2020). Bobcats are currently classified as nongame in Texas and there are no harvest limits or season restrictions, however, they are believed to occur at high densities (Heilbrun et al. 2006; Symmank et al. 2008).

We assessed bobcat resource selection within the home range (i.e., Johnson’s 3rd order of selection, Johnson 1980) in a landscape where patch composition and configuration were driven by agricultural practices, wildlife management, and energy development. We hypothesized bobcat resource selection would occur at multiple scales but that there would be a characteristic scale where the effects of landscape patterns on resource selection were strongest. We predicted the scale of effect of landscape patterns on resource selection would occur at finer scales since our ecological response variable is selection of a specific location within the home range. Processes within the home range such as foraging success often are influenced by landscape variables at finer scales than in processes such as dispersal (Miguet et al. 2016). Bobcats often select specific vegetation attributes such as concealment cover across systems (Kolowski and Woolf 2002; McNitt et al. 2020b; Zamuda et al. 2022). We predicted bobcats would select woody cover and woody vegetation patches that were more contiguous. We also predicted bobcats would avoid energy infrastructure due to human disturbance at these sites.

Methods

Study area

We conducted the study in a 5,665-ha ranch in the South Texas Plains ecoregion in La Salle County, Texas, during 2021–2022. The 40-year average (1981–2021) annual rainfall for the area was 57.20 cm (National Ocean and Atmospheric Administration 2021). The 2021 average summer (June–August) and winter (December–February) daily temperatures were approximately 29 °C and 13.4 °C, respectively (National Oceanic and Atmospheric Administration 2021).

The ranch included areas that have experienced brush management and contain systemic brush mosaics characterized by large tracts of brush strips alternating with herbaceous vegetation (Fig. 1). The ranch was managed for cattle (Bos tauros) production and wildlife with a focus on white-tailed deer (Odocoileus virginianus), northern bobwhite (Colinus virginianus), and chestnut-bellied scaled quail (Callipepla squamata). The site included substantial energy development with a coverage of 0.35 energy pads/km2 and was associated with regular maintenance resulting in frequent human disturbance. From 2009 to 2019, the southern portion of the ranch was the focus of an effort to restore 300 ha of native grasslands primarily for quail conservation. Restoration efforts targeted the removal of two dominant invasive grasses, including old world bluestem (Bothriochloa spp.) and buffelgrass (Dicanthium spp.) using prescribed fire, herbicide, and native plant seeding (Olsen et al. 2018; Fulbright et al. 2018).

Fig. 1
figure 1

The study extent of the Hixon Ranch located in La Salle County, Texas, USA with its two partitions of Hixon North and South. The area included systematic brush mosaics where brush was removed at precise spacings intervals to establish alternating strips of brush and herbaceous vegetation to promote interspersion of forage and thermal cover resources for livestock and wildlife panel A The ranch also maintains several larger continuous patches of woody vegetation panel B

Full size image

Woody plants included honey mesquite (Neltuma glandulosa), Texas live oak (Quercus virginiana), blackbrush acacia (Acacia rigidula), cenizo (Leucophyllum frutescens), huisache (Acacia farnesiana), whitebrush (Aloysia gratissima), spiny hackberry (Celtis ehrenbergiana), and guayacan (Guaiacum angustifolium). Common cacti observed included Texas prickly pear (Opuntia engelmanii var. lindheimerii) and tasajillo (Cylindropuntia leptocaulis). Herbaceous species found in the area include creeping bundleflower (Desmanthus virgatus), bristlegrasses (Setaria spp.), wild petunia (Ruellia spp.), gramas (Bouteloua spp.), and purple threeawn (Aristida purpurea) (Olsen et al. 2018, Palmer et al. 2021).

GPS collaring

We captured and collared adult bobcats with a single-door 108 × 55 × 40 cm wire box traps (Tomahawk Trap Co., Tomahawk, WI) baited with live pigeons (Columba livia) that were maintained safely in a separate enclosure. We immobilized bobcats with a mixture of medetomidine (0.6–0.8 mg per kg of bodyweight) and ketamine (2.5–4 mg per kg of body weight; Rockhill et al. 2011; Tellaeche et al. 2020). We fit bobcats with Lotek Litetrack Iridium 150 g and 250 g GPS-satellite collars (Lotek New Market, ON, Canada), and recorded locations every 2 h. We programmed GPS collars to drop-off 52-weeks following deployment. We monitored nine bobcats (five females and four males) for 12 months which resulted in a dataset with 17,881 GPS locations (Fig. 2).

Fig. 2
figure 2

GPS location of 9 adult bobcats (4 males and 5 females) located on the Hixon Ranch, La Salle, TX. Bobcats were monitored from May 2021 to May 2022 and collected 17,881 locations

Full size image

Our capturing and handling of bobcats followed recommendations by Sikes and Animal Care and Use Committee of the American Society of Mammalogists (2016) and protocols were approved by Texas A&M University- Kingsville Institutional Care and Use Committee Guidelines (Protocols: IACUC 2012–12-20B-A2, 2019–2-28A-2-28B), and Texas Parks and Wildlife Department Scientific Research permit (no. SP0190-600).

Land cover classification

We quantified landscape configuration of vegetation cover in our study area using supervised classification with random forest models (Hayes et al. 2014). We performed supervised classification in ArcPro 2.8 (ESRI, Redlands, CA) using 2020 National Agriculture Imagery Program (NAIP; U.S. Department of Agriculture 2020) 0.6 × 0.6 m imagery and categorized our study area into woody and herbaceous vegetation. After this initial classification, we digitized the highway, paved roads, energy infrastructure, and water bodies (i.e., Nueces River, ponds, cattle tanks). Water bodies were digitized based on National Wetlands Inventory Dataset (U.S. Fish and Wildlife Service 2021). We then performed an accuracy assessment using 300 random points in a confusion matrix with 2021 Google Earth imagery to assess our classification accuracy and attained a rate of 92% accuracy (Jensen 2016; Rwanga and Ndambuki 2017).

Landscape and environmental covariates

We calculated GPS collar location error using 90 test locations with collars placed in dense brush and determined an average of 5 m ± 1.5 standard deviation (SD). Therefore, we resampled the imagery to 10 m to approximately match the resolution of the collar error (Agouridis et al. 2004; Ganskopp et al. 2007; Smith et al 2021). To quantify landscape configuration, we calculated values for three landscape metrics: woody vegetation edge density (m/ha), woody vegetation patch density (number of patches/100 ha), and contagion across cover classes (0–100%) in Fragstats 4.2 (McGarigal 1995). The contagion metric is a landscape-level metric across all class-types (i.e., woody vegetation, herbaceous vegetation highway, paved roads, energy infrastructure, and water bodies) and values increase as the landscape is more aggregated (Riitters et al. 1996). We only included metrics in our analyses if they were not correlated (|r|< 0.7; Dormann et al. 2012) and the metrics gave us insights into potential landscape fragmentation. To identify the scale of effect for resource selection, we performed moving window analyses with a window size set to 10 unique values representing scales of interest. Our scale range was selected based on the home ranges of regional bobcat prey ranging from Permoyscus genera (Morris 1992) to a white-tailed deer (Williams et al. 2012). We selected 30 m, 60 m, 100 m, 200 m, 300 m, 400 m, 500 m, 600 m, 700 m, and 800 m as our representative scales for the moving window analysis.

Distance-based approaches are a useful method for assessing the effect land cover on resource selection (Conner et al. 2003). To assess the effects of cover type on resource selection, we created distance raster layers where each pixel was characterized by the distance to the nearest patch of a specific cover type. We created distance raster layers for woody vegetation and herbaceous vegetation the two primary cover types in our system. We also created a distance raster layer for water as the distribution of water in semi-arid systems can be an important determinate of wildlife space use (Ochoa et al 2021). To assess the effects of human disturbance associated with energy extraction, we created a distance raster for energy infrastructure.

Resource selection functions

We developed resource selection function (RSF) models using Design 3 described by Manly et al. (2007) at the 3rd order (i.e., selection within home range; Johnson 1980). To quantify availability, we generated 100% minimum convex polygons (MCPs) for each bobcat and generated 10 random locations per observed animal location (Dunagan et al. 2019; Mayer et al. 2021). We selected MCPs as our home range estimator to characterize availability to be as inclusive as possible in defining what was available to each bobcat (Bosco et al. 2021; Hughey et al. 2021). We then extracted all landscape and distance class metrics to each used and random location. We used the package: lme4 (R Core Team 2022) to fit RSFs using general linear mixed effect models with animal ID treated as a random intercept.

We employed a two-stage modeling approach: first we generated 10 univariate models for each landscape metric where each model represented one of the 10 scales. We then used Akaike Information Criterion adjusted for small sample size (AICc) to identify the scale of our landscape metrics that was best supported (ΔAICc < 2) within our analytical framework, which we also compared with the null model. After identifying the characteristic scale for each landscape metric, we generated a global model that included each landscape metric and distance metrics (woody vegetation, herbaceous vegetation, water, and gas infrastructure). We then projected our final model output with the raster package (R Core team 2022) by reclassifying the map into 10% equal area bins to balance the skewed distributions on our predictions (Boyce et al. 2002).

Results

Female home range size was 4.16 km2 ± 2.1 (average ± SD) and for male home range size was 19.4 km2 ± 5.4. In our assessment of the scale of effect, our univariate models revealed the scale of effect for both patch density and edge density of woody vegetation for bobcat resource selection was 30 m, the finest scale examined (Tables 1, 2). For contagion, the scale that explained the most amount of variation for bobcat resource selection was 100 m (Table 3). The null model was not competitive with the top models.

Table 1 Akaike Information Criterion (AICc)table for resource selection models identifying the scale of effect of brush management for the landscape metric woody vegetation edge density across 10 different scales for bobcats (Lynx rufus). Data collected from 10 May 2021 to 23 May 2022 on the Hixon Ranch in La Salle County, Texas, USA
Full size table
Table 2 Akaike Information Criterion (AICc) table for resource selection models identifying the scale of effect of brush management for the landscape metric woody vegetation patch density across 10 different scales for bobcats (Lynx rufus). Data collected from 10 May 2021 to 23 May 2022 on the Hixon Ranch in La Salle County, Texas, USA
Full size table
Table 3 Akaike Information Criterion (AICc) table for resource selection models identifying the scale of effect of brush management for the landscape metric contagion index across 10 different scales for bobcats (Lynx rufus). Data collected from 10 May 2021 to 23 May 2022 on the Hixon Ranch in La Salle County, Texas, USA
Full size table

Bobcat probability of use decreased by 9% for every 50 m increase in distance from woody vegetation (β = -1.21 ± 0.02 p < 0.01; Fig. 3) and decreased by 2% for every 500 m increase in distance from water bodies (β = -0.25 ± 0.01 p < 0.01; Fig. 3). Conversely, we determined probability of use increased by 30% for every 50 m increase in distance from herbaceous vegetation (β = 0.28 ± 0.01 p < 0.01; Fig. 3) and increased 0.5% for every 500 m increase in distance from gas infrastructure (β = 0.01 ± 0.01 p < 0.01: Fig. 3).

Fig. 3
figure 3

Predicted responses to distance to woody vegetation A distance to herbaceous vegetation B distance to water body C, and distance to gas infrastructure D by bobcats (Lynx rufus) from resource selection functions fitted with data collected from 10 May 2021 to 23 May 2022 on the Hixon Ranch in La Salle County, Texas, USA

Full size image

Bobcat probability of use decreased by 2.5% for every increase of 500 patches per 100 ha in patch density of woody vegetation (β = -0.30 ± 0.01 p < 0.01; Fig. 5) and increased 2% for every 500 m/ha increase in values of edge density (β = 0.05 ± 0.01 p < 0.01; Fig. 5). However, as contagion increased by 25%, probability of use increased by 2.5% (β = 0.06 ± 0.01 p < 0.01; Fig. 4). We then evaluated models that used multiple landscape metrics and identified that the global model performed the best (Table 4). We projected the global model across the landscape to visualize areas of high and low probability of use (Fig. 5, Table 5). This heat map allowed us to visualize patterns of bobcat resource selection with various configuration of brush alteration on the landscape.

Fig. 4
figure 4

Predicted responses to the landscape metrics: contagion 100 m %; A edge density 30 m (meters [m] per ha; B and patch density 30 m (number of patches [#] per 100 ha; C by bobcats (Lynx rufus) from resource selection functions fitted with data collected from 10 May 2021 to 23 May 2022 on the Hixon Ranch in La Salle County, Texas, USA

Full size image
Table 4 Akaike Information Criterion (AICc) for resource selection function models for identifying the best multiple landscape metric model for bobcats (Lynx rufus). Data collected from 10 May 2021 to 23 May 2022 on the Hixon Ranch in La Salle County, Texas, USA
Full size table
Fig. 5
figure 5

Spatially explicit visualization of bobcat (Lynx rufus) resource selection as a function of our global model across the Hixon Ranch in La Salle County, Texas, USA. Data were collected from 10 May 2021 to 23 May 2022. Higher probability of use values (darker blue) represent areas bobcats were more likely to use on the landscape. Probability of use ranged from 0 to 100%

Full size image
Table 5 Summary table from bobcat (Lynx rufus) global resource selection model including beta estimates, odd ratios, and 95% confidence intervals. Data collected from 10 May 2021 to 23 May 2022 on the Hixon Ranch in La Salle County, Texas, USA
Full size table

Discussion

Resource selection occurs at a variety of spatial and temporal scales (McGarigal et al. 2016). In our investigation of the scale of effect for bobcat resource selection, we identified characteristic scales for each of our landscape configuration metrics. Our observations surrounding the scale of effect aligned with the predictions proposed by Miguet et al. (2016) and Martin (2018) regarding to finer scales in context to breeding and foraging success within the home range. We observed bobcats responded to metrics of woody vegetation configuration at finer scales within their home range, which may be associated with the potential perception of hunting efficiency and prey availability within woody patches (Dunagan et al. 2019; McNitt et al. 2020a). Bobcats avoided herbaceous cover and selected larger patches of woody vegetation, and generally avoided interspersed areas.

Creation of systematic brush mosaics and other land management practices created a landscape configuration that influenced resource selection of bobcats. Even though bobcats selected for higher amounts of woody vegetation edges, woody patch density had a negative effect on the probability of use by bobcats. Bobcats are known ambush hunters that often hunt along edges but increased patchiness of vegetation may not be advantageous (Fuller et al. 1985; Marrotte et al. 2020a; McNitt et al. 2020a). Fragmentation of woody patches can have mixed effects on carnivores as it can create structural heterogeneity which can increase prey abundance, but also creates a mosaic of smaller patches which are generally avoided by ambush predators.

Bobcats selected for aggregation of patches across our study area. Areas with greater aggregation of patches are less fragmented and are more homogenous. Contagion has been shown to have varying effects on carnivores, with some species benefiting from fragmented areas and others selecting more aggregated patches (Dijak and Thompson 2000; Kramer-Schadt et al. 2011). Aggregation of patches of similar vegetation structure can improve habitat connectivity for a variety of species, and bobcats are known to select aggregated patches of cover (Tucker et al. 2008; Ruell et al. 2012; Poessel et al. 2014; Janecka et al. 2016).

At broader scales bobcats may select for more aggregation of patches, but as we observed bobcat resource selection within the home range was influenced by woody vegetation. Bobcats selected locations closer to woody vegetation and attributes associated with the configuration of woody cover. We observed patterns of bobcats avoiding high patch densities of woody cover; however, they did select woody vegetation patches with higher edge density. This suggests a pattern of selecting for the edges of single large patches of woody vegetation. These results support the findings of previous studies with bobcats in fragmented landscapes ranging from Mexico, Canada, and both coasts of the United States (Poessel et al. 2014; Espinosa-Flores and López-González 2017; Farrell et al. 2018; Jones et al. 2020; Marrotte et al. 2020b).

We observed selection of open water bodies by bobcats. Water availability drives carnivore distributions in many other systems (Steiner et al. 2018; Rabaiotti and Woodroffe 2019; Perera-Romero et al. 2021), especially in arid and semi-arid landscapes. It is likely bobcats select areas near water, not only to meet their own water requirements, but also as water likely congregates prey species (Webb et al. 2006; Ochoa et al. 2021). In many semi-arid environments, open water is often ephemeral and so anthropogenically sourced water is an attribute of a cultural landscape that can strongly influence the spatial patterns of many wildlife species (Smit et al. 2007; Atwood et al. 2011; Rich et al. 2019). With more unpredictable climate regimes projected for the future of these environments, it is likely these anthropogenic sources of water will be crucial resources on for many wildlife species in these systems (Ogutu et al. 2012; Ochoa et al. 2021).

Although a relatively small effect, bobcats avoided energy infrastructure. Gas and oil extraction operations are important forms of disturbance for many species from toxicological exposure to alteration in space use (Bowen et al. 2014; Laberee et al. 2014; Garman 2018; Walker 2022). Our study area contained a pad density of 0.35 pads/km2 which was relatively lower in other studies (Kalyn Bogard and Davis 2014; Johnson et al. 2015; Hethcoat and Chalfoun 2015). Energy development contributes to the fragmentation of woody and herbaceous vegetation in this system, but the high amounts of human traffic associated servicing these sites likely serve as an important mechanism of anthropogenic disturbance as well (Bowen et al. 2014; Allred et al. 2015; Pattison et al. 2016). Future investigations into the role of energy infrastructure on wildlife should evaluate the relative importance of habitat fragmentation and the direct human disturbance associated with energy development.

Bobcats, similar to other generalist carnivores, have adapted to cultural landscapes across their distribution, despite the myriad of threats to survival that human-dominated areas can pose (Roberts and Crimmins 2010; Young et al. 2019). However, despite such success, there are likely responses that occur at different spatial and temporal grains and extent that influence survival (Miguet et al. 2016; Martin 2018). Multi-scale analyses of habitat selection should remain an integral component of how scientists understand the complex patterns of selection. From our analysis, we demonstrated the effects of a cultural landscape on the behavior of a generalist carnivore. While we did not conduct a multi-order analysis (Johnson 1980), we did assess multiple scales within a single order. The scale of effect may likely vary at different orders of selection (Martin 2018), which is fertile grounds for future research. The scale of effect remains an important aspect to study when trying to elucidate the effects of landscape structure on wildlife.

Conclusion

We found bobcat resource selection of landscape patterns was more strongly influenced by patterns at finer scales. In our system, systematic brush mosaics increases interspersion of patch types and are designed to cultivate high densities of prey which would theoretically benefit carnivores. However, bobcats avoided locations that were more interspersed which suggests bobcats may forgo prey rich patches if the patch configuration is not conducive to their hunting mode. As rural landscapes continue to be altered by agriculture and energy production, there is a growing need to understand the effects of this development on wildlife. Understanding the scale at which the resulting landscape patterns influence wildlife behavior should be a key component to future studies of investigating the effects of global change on wildlife.