Cattle exclusion increases encounters of wild herbivores in Neotropical forests

Abstract

1. Ongoing habitat loss and species extinctions require managers to implement and quantify the effectiveness of conservation actions for protecting biodiversity. Fencing, when done properly, is an important management tool for conservation in landscapes where wildlife and domestic animals co-occur, potentially enhancing habitat use through selective exclusion of domestic species. For instance, the fencing of forest patches in the Neotropics is expected to reduce the degradation of understory vegetation by cattle, releasing these resources for the native community of browsers and fruit consumers.
2. Here, we implemented an ecological experiment using a before-after control-impact design to quantify the effect of cattle exclusion on encounter probability of the native community of browsers and fruit consumers, and percent ground cover in multifunctional landscapes of the Colombian Orinoquía. We built 14 km of wildlife-permeable fences along forest edges in four forest patches (i.e. blocks) containing control and fenced (treatment) sites. We installed 33 camera traps to obtain information about wildlife and cattle encounter probabilities, before and after the fences were constructed. We used Bayesian generalised linear mixed effects models to quantify the effect of fences via the interaction between the time period (before and after the fences were built) and treatment (control or fenced sites).
3. Fencing was effective at reducing encounter probabilities of cattle in the treated sites, and it had a positive impact on relative encounter probabilities of four of seven studied wildlife species (herbivores including the black agouti [dry season only], lowland tapir [dry season only] and spotted paca [both seasons] and an omnivore, the South American coati [rainy season only]). The effect of fencing was negative for the collared peccary but only during the dry season. No statistically significant effect was detected for the white-lipped peccary or white-tailed deer.
4. Synthesis and applications: We provide experimental evidence that fences are effective at selectively excluding cattle and increasing encounter rates of wild browsers and fruit consumers in forest patches where these species co-occur with cattle. Our results highlight an important application of fencing ecology in Neotropical forests, where the implementation of wildlife-permeable fences is feasible due to smaller body sizes of wildlife compared to domestic animals such as cattle.

1 INTRODUCTION

Contemporary rates of biodiversity loss compel conservation scientists to transition from documenting conservation problems to providing solutions to these problems. Fences are ubiquitous in human-dominated landscapes and might interfere with landscape connectivity and animal movement, especially for wide-ranging (O'Neill et al., 2022) or migratory species (Hering et al., 2022). However, fences can be used as a conservation tool when designed to allow wildlife movement across the landscape (Segar & Keane, 2020). For example, fences can help reduce human–wildlife conflict by controlling the movement of domestic animals (Smith et al., 2020), protect areas from disturbance (Berry et al., 2020), and reduce wildlife-vehicle collisions (Huijser et al., 2016) and road mortality in high traffic areas (Abra et al., 2020; Frangini et al., 2022). In landscapes with co-occurring wildlife and livestock species, fencing can maintain livestock in defined areas (Aaser et al., 2022), prevent undesired effects of livestock grazing on plant communities (Boyd et al., 2022) and protect livestock from predation (Becker & Farja, 2017; Pimenta et al., 2017).

Although mathematical models can sometimes be useful for predicting the effect of management actions on animal distributions when field experimentation is cost- or time-prohibitive (Fortin et al., 2020), experimental manipulations can provide critical insights into the functioning of ecosystems and the impact of disturbance or management on ecosystem dynamics (Fayle et al., 2015). Experimental designs that use randomisation (i.e. random assignment of control and treatment groups to sampling units) can also facilitate causal inferences about the effects of an intervention on ecological communities (Larsen et al., 2019; Wauchope et al., 2021). When full randomisation is not feasible, observations may be temporally stratified into before and after intervention periods (Time) and spatially into control and impact areas (Treatment) (Green, 1979). So-called before-after control-impact (BACI) designs facilitate rigorous tests of the direction and magnitude of effects resulting from management interventions, while accounting for temporal and spatial variability present within the sample domain (Fisher et al., 2019). A positive interaction between Time (BA) and Treatment (CI) terms (i.e. the BACI contrast) indicates that the response variable increased more, or decreased less, in sites that received the treatment compared to control sites (Chevalier et al., 2019). Despite the potential of BACI designs to provide accurate inference on the impacts of disturbance or management on biodiversity, they are underrepresented in the ecological literature when compared to simpler designs (Christie et al., 2019).

Long-term experiments in African Savannas have explored the effects of livestock and large mammalian herbivore exclusions on bird diversity and abundance (Ogada et al., 2008), vegetation communities (Veblen & Young, 2010), and species interactions and trophic cascades (Goheen et al., 2018). Other experimental studies focused on the effect of cattle on wildlife have also been concentrated in African Savannas (Kimuyu et al., 2017) and rangelands (Stears & Shrader, 2020) as well as in North American grasslands (Fischer et al., 2020). However, less attention has been given to Neotropical areas, where livestock production is expanding rapidly into forested areas such as the Amazon biome (Vale et al., 2019). In addition, experiments assessing the effect of livestock grazing in Neotropical systems have focused primarily on the impacts on composition and structure of vegetation (Durigan et al., 2022; Trigo et al., 2020), and estimates of the effects of fencing and cattle exclusion on forest-dwelling wildlife species are lacking.

The interactions between livestock and wild herbivores are complex and contingent on seasonal availability of forage. Previous studies have shown increased competition between cattle and African wild ungulates during the dry season (Odadi et al., 2011), due to higher resource scarcity (Tyrrell et al., 2017). By contrast, cattle can promote high-quality forage regrowth and, thus, increase access to forage for wildlife during the rainy season (Stears & Shrader, 2020). As for any management intervention, the ecological impacts of fences to maintain cattle in defined areas and establish cattle-free zones should ideally be tested experimentally. Experimental studies will improve understanding of the response of wildlife to cattle exclusion and the effectiveness of fencing as a potential conservation strategy.

Here, we use a BACI experimental sampling design to quantify the effect of cattle exclusion on wildlife encounter probabilities in fenced forest areas. We deployed camera traps on a cattle ranch in the Orinoquía region of Colombia, where cattle and wildlife co-occur within a matrix of pasture areas, natural Savannas and forest fragments (IDEAM, 2010). A wildlife-permeable fence was built along the edges of forest fragments, and camera-trap surveys were conducted before and after fence construction in control (unfenced forest edge) and fenced sites (fenced forest edge). We predicted that the encounter probability of cattle would be lower and encounter probabilities of wildlife would be higher in fenced areas compared to un-fenced areas, mainly in the dry season when cattle increase their use of the forest for water and foraging resources. This pattern would result in a negative Time × Treatment interaction coefficient for cattle and positive coefficients for wildlife species.

2 MATERIALS AND METHODS

2.1 Study area

We conducted the cattle exclusion experiment within the private natural reserve Rey Zamuro—Matarredonda (3.542242, −73.411003), which encompasses an area of 31 km² (Figure 1a), and is located in the Meta department in the Orinoquía region of Colombia. The study area is classified as a tropical humid biome, with a rainy season occurring from April to November and a dry season from December to March (Rippstein et al., 2001). Mean annual precipitation is 4500 mm, and the mean annual temperature is 26°C (Karger et al., 2017).

FIGURE 1

(a) Study area in the private natural reserve Rey Zamuro—Matarredonda located in the Meta department (light green in inset map) in the Colombian Orinoquía. Grey areas represent forest, and black triangles indicate camera-trap locations. Fenced forest areas (treatment) are shown in purple and the control areas that remained unfenced in dark green. (b) Timeline of the cattle exclusion experiment with the approximate date of construction of the fences shown with the dotted line. Camera-trap data were collected pre- and post-fencing (below and above of the dashed line, respectively) during the dry and the rainy seasons.

The multifunctional reserve contains riparian forest with patches of the Mauritia flexuosa palm, a keystone species that grows in large aggregations in swamps and provides important food resources for wildlife (Gilmore et al., 2013; Vélez et al., 2017; Virapongse et al., 2017). Hybrid cattle (Bos spp.) use these palms swamps for water and browsing, mostly in the dry season, which leads to soil erosion and understory degradation, reducing forage for wildlife. The use of these palm swamps by cattle also represents an economic problem for landowners, as cattle can sometimes become immobilised in the mud.

Medium- and large-bodied browsers and/or fruit consumers that have high use of M. flexuosa palm swamps—and that may therefore be adversely affected by the presence of cattle in the swamps—include the black agouti (Dasyprocta fuliginosa), collared peccary (Pecari tajacu), lowland tapir (Tapirus terrestris), spotted paca (Cuniculus paca), white-lipped peccary (Tayassu pecari) and white-tailed deer (Odocoileus virginianus). The South American coati (Nasua nasua), an omnivore with high consumption of fruits, plant parts and litter invertebrates (Alves-Costa et al., 2004), is also likely to be impacted by understory degradation.

2.2 Sampling design

RapidEye imagery (5 m spatial resolution, five spectral bands) collected in August and December 2018 by Planet (Planet Team, 2017) was used to generate a binary map of forest cover. Cover types including woody vegetation, grass and unvegetated cover (e.g. building infrastructure and exposed ground) were predicted using the randomForest R package (Liaw & Wiener, 2002) with training data delineated manually in QGIS (QGIS Development Team, 2018). We trained the random forest classifier with covariates that included surface reflectance in five spectral bands and change in the enhanced vegetation index between wet and dry seasons. We used the resulting map of forest cover to randomise the deployment of 33 cameras in December 2020 and January 2021 (Figure 1a).

We used Bushnell Trophy Cam HD Aggressor camera traps and unbaited stations to detect cattle and wildlife before cattle exclusion (December 2020–October 2021), and after fences were installed (November 2021–January 2023). Pre- and post-fencing surveys were conducted during the rainy and the dry seasons (Figure 1b). Within each of four forest patches (serving as blocks), one forest fragment was selected to be fenced and another fragment served as a control (Figure 1a). The location of fencing was determined according to management needs (e.g. where cattle intrusion was more problematic for landowners) for two forest blocks and assigned randomly for the remaining two forest blocks. The fences were composed of two wire strands; the lower strand was installed 70 cm above the ground to prevent cattle from accessing the forest while still allowing wildlife to use these areas. A total of 14 km of fencing was installed. Most of the fences were built between October and early November 2021. A final forest patch was fenced in April 2022.

Camera traps were randomly located within control and treatment areas and constrained to be at least 200 m apart (Figure 1a). Cameras were located along wildlife trails when possible (Cusack et al., 2015; Jansen et al., 2014), and were kept at the same location for the duration of the experiment. Cameras were deployed on trees, 30 cm above the ground, and programmed to take three images (2-s interval) when triggered by an animal passing in front of the infrared sensor. One camera-trap station within the fencing treatment group was reassigned as a control site for a period that the fence was not operational due to damage by fallen trees, permitting free movement of cattle within the area. Mean sampling effort for all sites was 256 and 378 days for the periods before and after fence deployment, respectively.

Wildlife and cattle in the camera-trap images were classified by trained technicians and experts using data processing tools on the Wildlife Insights platform (Ahumada et al., 2020; Vélez, McShea, Shamon, et al., 2023)—https://www.wildlifeinsights.org/. Areal percent ground cover was estimated visually in 4-m² plots that were randomly located within 5 m of the camera-trap station. The plots were sampled in June 2021 and September 2022, before and after the fences were built, respectively.

Field sampling was performed under research permits issued to the Universidad de los Andes (ANLA Resolution 1177, 2014; Cormacarena Resolution PM-GA.3.20.2737). This work did not require institutional ethical approval.

2.3 Data analysis

We quantified the effect of fencing on wildlife and cattle by estimating encounter probabilities using Bayesian generalised linear mixed models. Separate models were fit for each species and season (dry and rainy) using the brms package (Bürkner, 2021) in the R statistical computing environment (R Core Team, 2022). We modelled the number of days each species was detected at site i in patch j during period t, Y_ijt, as realisations of a binomial process:

$$Y_{\mathit{ijt}}~\text{Binomial}\left(n_{\mathit{ijt}},p_{\mathit{ijt}}\right),$$(1)

where $n_{\mathit{ijt}}$ represents the number of days that the camera trap was active and $p_{\mathit{ijt}}$ represents the daily encounter probability at site i within patch j during period t. The log odds of an encounter was modelled as a linear combination of predictors and site- and patch-level random effects, γ_i and δ_j, respectively:

$$\begin{array}{c}\text{logit}\left(p_{\mathit{ijt}}\right)={\mathit{\beta }}_0+{\gamma }_i+{\delta }_j+{\beta }_1{\text{Time}}_t+{\beta }_2{\text{Treatment}}_i+{\beta }_3{\text{Time}}_t{\text{Treatment}}_i\\ +{\beta }_4{\text{Distwater}}_i+{\beta }_5{\text{Distedge}}_i\end{array},$$(2)

$${\gamma }_i~\text{Normal}\left(0,{\sigma }_{\gamma }\right),$$(3)

$${\delta }_j~\text{Normal}\left(0,{\sigma }_{\delta }\right),$$(4)

where ${\text{Time}}_t$ equals 1 for the periods after fence installation (0 for the pre-fencing period), ${\text{Treatment}}_i$ equals 1 if site i is within a fenced area (0 otherwise), and ${\text{Distwater}}_i$ and ${\text{Distedge}}_i$ are the distances from site i to fourth-order streams and the nearest forest edge, respectively. Both distance covariates were centred and scaled using the mean and standard deviation of distances for all sites. The parameters ${\sigma }_{\gamma }$ and ${\sigma }_{\delta }$ represent standard deviations of site- and patch-level random effects, respectively, and were included to capture features of the study design and account for potential spatial autocorrelation. We fit the model using N (μ = 0, σ = 5) priors for the regression coefficients, β, and Half Student-t distributions for the standard deviations of the random effects:

$${\sigma }_{\gamma }~\text{Half}-\text{student}-t\left(3,0,2.5\right),$$(5)

$${\sigma }_{\delta }~\text{Half}-\text{student}-t\left(3,0,2.5\right).$$(6)

Areal proportion of ground cover sampled at site i in patch j during period t, $Y_{\mathit{ijt}}$, was assumed to follow a beta distribution

$$Y_{\mathit{ijt}}~\text{Beta}\left({\mu }_{\mathit{ijt}},\varphi \right),$$(7)

with mean ${\mu }_{\mathit{ijt}}$ and precision $\varphi$. The mean areal ground cover proportion was modelled as a logit-linear combination of predictors using a similar structure as in Equation (2) but without the distance covariates

$$\text{logit}\left(p_{\mathit{ijt}}\right)={\mathit{\beta }}_0+{\mathit{\gamma }}_i+{\delta }_j+{\beta }_1{\text{Time}}_t+{\beta }_2{\text{Treatment}}_i+{\beta }_3{\text{Time}}_t{\text{Treatment}}_i.$$(8)

The precision parameter $\varphi$ was assumed to be constant across sites, patches and design covariates,

$$\text{log}\left(\varphi \right)={\beta }_0.$$(9)

We fit the model with N(0, 1) priors for the regression coefficients, β, and Half Student-t distributions for the standard deviations of the random effects as described in Equations (5) and (6).

Models were run with four chains, 150,000 iterations per chain, a thinning rate of 50, and with 90,000 iterations discarded as a burn-in. Model convergence was assessed by inspection of traceplots, R-hat values and effective sample sizes (Gelman & Rubin, 1992; Vehtari et al., 2021). Model fit was assessed via graphical posterior predictive checks where observed data were compared to replicated data generated from the posterior predictive distribution (i.e. data generated under the model). The effect of fencing on encounter probabilities of each animal species and areal ground cover was assessed using the posterior distribution of the coefficient for the Time × Treatment interaction term, β₃ (i.e. the BACI contrast). We also compared cell means of the p_ijt for the animal species and posterior distributions of predicted proportional ground cover across treatment groups and time periods. We refer to coefficients as statistically significant when 89% credible intervals do not overlap zero.

3 RESULTS

During the dry season, parameter estimates for the Time × Treatment interaction coefficient were positive and statistically significant for the black agouti (β₃ = 0.8; 89% CI = 0.6, 1.1), the lowland tapir (β₃ = 0.8; 89% CI = 0.2, 1.3) and the spotted paca (β₃ = 2.1; 89% CI = 1.3, 3.1), and negative for cattle (β₃ = −2.5; 89% CI = −3.2, −1.8) and the collared peccary (β₃ = −0.3; 89% CI = −0.5, 0; Figure 2). Differences in dry-season encounter probabilities between sampling periods were either negligible or similar at both control and fenced sites for the South American coati, white-lipped peccary and white-tailed deer (non-significant β₃; Table 1; Figure 2). In the rainy season, the Time × Treatment coefficient was positive and significant for the South American coati (β₃ = 0.6; 89% CI = 0.2, 1.1) and the spotted paca (β₃ = 0.6; 89% CI = 0.1, 1.1), and negative for cattle (β₃ = −1.5; 89% CI = −2.1, −1; Figure 2). Credible intervals for the interaction coefficient included 0 for all other species (β₃, Table 1; Figure 2). The posterior median β₃ Time × Treatment coefficient was positive for percent ground cover, though its credible interval included 0 (Figures 3 and 4). Models fit the data well, with posterior predictive distributions largely matching the characteristics of the observed data for the dry (Figure S1) and rainy (Figure S2) seasons.

FIGURE 2

Posterior median Time × Treatment coefficient and 50% and 89% credible intervals (bold and thin lines, respectively). Species-specific Bayesian generalised linear mixed models were fit to data obtained from camera-trap surveys conducted in the private natural reserve Rey Zamuro—Matarredonda in the Colombian Orinoquía between December 2020 and January 2023, during the dry and the rainy seasons.

TABLE 1. Regression coefficients (89% credible intervals) for the effect of Time (β₁), Treatment (β₂) and Time × Treatment interaction (β₃).

Season	Species	Conservation status	β 1	β 2	β 3
Dry	Black agouti	Least concern	−1.2 (−1.4, −1)^a	0.2 (−0.3, 0.8)	0.8 (0.6, 1.1)^a
	Cattle	–	0.3 (0.2, 0.5)^a	−0.9 (−1.5, −0.4)^a	−2.5 (−3.2, −1.8)^a
	Collared peccary	Least concern	−0.1 (−0.2, 0.1)	0.7 (0.3, 1.1)^a	−0.3 (−0.5, 0)
	Lowland tapir	Vulnerable	−0.9 (−1.3, −0.4)^a	−0.5 (−1.2, 0.2)	0.8 (0.2, 1.3)^a
	South American coati	Least concern	0.2 (−0.2, 0.7)	0.3 (−0.3, 0.9)	0.1 (−0.5, 0.7)
	Spotted paca	Least concern	−2.3 (−3.2, −1.6)^a	−0.4 (−1.8, 1)	2.1 (1.3, 3.1)^a
	White-lipped peccary	Vulnerable	0 (−0.5, 0.5)	1 (0.3, 1.7)^a	0 (−0.7, 0.7)
	White-tailed deer	Least concern	0.3 (0, 0.7)^a	−0.2 (−0.7, 0.3)	−0.3 (−0.7, 0.2)
Rainy	Black agouti	Least concern	−0.3 (−0.5, −0.2)^a	0.7 (−0.1, 1.3)	0 (−0.2, 0.2)
	Cattle	–	−0.3 (−0.5, −0.2)^a	−0.3 (−1.1, 0.5)	−1.5 (−2.1, −1)^a
	Collared peccary	Least concern	0.1 (−0.1, 0.3)	1 (0.5, 1.5)^a	−0.1 (−0.3, 0.1)
	Lowland tapir	Vulnerable	−1.1 (−1.5, −0.8)^a	0.1 (−0.5, 0.8)	0.2 (−0.3, 0.6)
	South American coati	Least concern	−0.8 (−1.1, −0.5)^a	−0.2 (−0.6, 0.3)	0.6 (0.2, 1.1)^a
	Spotted paca	Least concern	−0.2 (−0.6, 0.2)	−0.3 (−1.1, 0.5)	0.6 (0.1, 1.1)^a
	White-lipped peccary	Vulnerable	1.2 (0.2, 2.3)^a	0.8 (−0.4, 2.2)	−0.7 (−2, 0.5)
	White-tailed deer	Least concern	0 (−0.3, 0.3)	0 (−0.7, 0.8)	0.3 (−0.1, 0.6)

• Note: Species-specific Bayesian generalised linear mixed models (GLMM) were fit to data obtained in the dry and rainy season from camera-trap surveys conducted in the private natural reserve Rey Zamuro—Matarredonda in the Colombian Orinoquía between December 2020 and January 2023. Conservation status from the International Union for Conservation of Nature's Red List of Threatened Species is shown for each species.
• ^a Indicate credible intervals that do not overlap zero.

FIGURE 3

Posterior median Time × Treatment coefficient and 50% and 89% credible intervals (bold and thin black lines, respectively). A beta regression model was fit to areal ground cover proportion sampled in 4 m² plots located at each sampling station in the private natural reserve Rey Zamuro—Matarredonda in the Colombian Orinoquía. The plots were respectively sampled in June 2021 and September 2022, before and after the fences were installed to exclude cattle from forest patches.

FIGURE 4

Posterior distributions of predicted areal cover proportion sampled in 4 m² plots located at each sampling station in the private natural reserve Rey Zamuro—Matarredonda in the Colombian Orinoquía. The plots were respectively sampled in June 2021 and September 2022, before and after the fences were installed to exclude cattle from forest patches. Bold and thin black lines correspond to 50% and 89% credible intervals, respectively.

4 DISCUSSION

Our study was effective at demonstrating how fences can be used to exclude cattle from forested areas, while allowing other wildlife species to continue or increase their use of those areas. We observed a positive effect of the fencing treatment (inferred via the BACI contrast) on encounter probabilities of the black agouti, lowland tapir and spotted paca during the dry season and on the South American coati and spotted paca during the rainy season. The BACI contrast was negative only for cattle (during both seasons) and the collared peccary during the dry season. Thus, we conclude that fencing can be an effective management tool for selectively excluding cattle and protecting habitat for wildlife on cattle ranches in Neotropical areas (with native fauna smaller than cattle).

4.1 Positive effect of cattle exclusion for wildlife

Positive effects of fencing were more prevalent and pronounced during the dry season, when there was higher competition between cattle and wild ungulates (Odadi et al., 2011) and increased use of the forest for forage and water resources by cattle. As cattle often degrade ground cover when they forage in the forest (Desbiez et al., 2009; Regolin et al., 2021), the relative increase in use of fenced sites by browsers and fruit consumers in the post-fencing period may be related to an increase in forage availability following cattle exclusion. For example, this might benefit species such as the lowland tapir, which is commonly found in open-canopy areas with palatable understory vegetation (Paolucci et al., 2019), also consumed by cattle.

The posterior distribution for the interaction coefficient between time and treatment (i.e. the BACI contrast) in the ground cover model overlapped zero. Detecting an effect of fencing on vegetation recovery might require a longer time than the interval between our pre- and post-fencing sampling periods (15 months). For example, abandoned pastures in humid Neotropical areas may reach similar levels of above-ground vegetation biomass as old-growth forest after 20 years (Bechara et al., 2016). Alternatively, although cattle are expected to have a larger impact on ground cover than wildlife due to their greater biomass (Greenspoon et al., 2023), if reduced impacts from cattle are accompanied by a concurrent increase in browsing by wild ungulates, we might expect differences in ground cover between fenced and control sites to be minimal.

4.2 Interaction between cattle and collared peccary

During the post-fencing dry season, encounter rates of cattle increased at the control sites and decreased at the treatment sites, whereas during the post-fencing rainy season, cattle encounter rates decreased at both control and fenced sites but more so at fenced sites. The elevated use of control sites by cattle during the post-fencing dry season may have been driven by lost access to fenced areas. Interestingly, we observed a negative Time × Treatment coefficient for the collared peccary, but only during the dry season. Collared peccaries had higher initial encounter probabilities at fenced sites, whereas cattle had higher initial encounter probabilities at control sites (Figure S3), possibly suggesting avoidance between these two species.

Previously, we documented infrequent co-occurrence of cattle and the white-lipped peccary (Vélez, McShea, Pukazhenthi, et al., 2023; Vélez, McShea, Pukazhenthi, Stevenson, et al., 2024). Given similar habitat requirements of the two peccary species (Desbiez et al., 2009), we might expect cattle and the collared peccary to also avoid each other. Alternatively, it is possible that collared peccaries benefit from cattle grazing and have a delayed response to cattle presence, favouring the occurrence of collared peccaries in cattle ranches (Flores-Martínez et al., 2022). A positive association or facilitation between cattle and wildlife might be related to an improvement in forage quality resulting from the formation of nutrient-rich patches and redistribution of soil nutrients in areas previously grazed by cattle (Augustine et al., 2011). In addition, cattle grazing can maintain short and nutrient-rich vegetation preferred by some wildlife species (Herrik et al., 2023). For example, studies of coexistence of cattle and wild herbivores have identified that small to medium-sized herbivores in large herds tend to select areas recently grazed by cattle whereas the largest species in large herds and smallest species in small herds select cattle-free zones (Herrik et al., 2023).

4.3 Strategic management of cattle ranches

Cattle ranching drives deforestation throughout the tropics (Zu Ermgassen et al., 2022) and contributes to loss of mammal diversity in areas with high rates of habitat conversion (de Souza et al., 2018). However, integration of livestock and wildlife can offer complementary ecological and economic benefits (Keesing et al., 2018), and ranches can be conceived of as potential conservation tools if managed appropriately for sustaining wildlife (Hoogesteijn & Chapman, 1997; Medici et al., 2022). For example, non-intensive cattle ranching might offer ecological benefits (e.g. by increasing forage quality for some wildlife species and reducing tick abundance as a result of periodic grazing) and provide economic benefits from meat and dairy production and from tourism (Keesing et al., 2018). However, these co-benefits would not be achieved when cattle ranching is a main cause of forest loss, with little to no connected forest left to be used by wildlife.

Effective multifunctional approaches require investment in strategic management of livestock to reduce potential competition with wildlife and maintain landscapes with functional heterogeneity represented by spatiotemporal variation in structure and composition of vegetation (Fynn et al., 2016). As an example, rotational grazing with cattle-free zones incorporates targeted management of areas for species that avoid cattle and for species that benefit from cattle grazing via ecological facilitation (Herrik et al., 2023; Odadi et al., 2017). In addition to studies of habitat use, future research should consider the effect of fencing on reducing the risk of disease transmission between domestic and wildlife species (Fernandes-Santos et al., 2019; Medici et al., 2014).

4.4 From pattern description to conservation action

We demonstrated the potential of fencing ecology and the opportunities it presents for stakeholders to develop strategic management on cattle ranches (Jakes et al., 2018). Livestock production is an important economic activity, and complete exclusion of domesticated species from forests may not be possible or desirable (e.g. cattle will still require access to water sources). Yet, livestock use of forest can be better organised and delimited. Fences can be used as a tool to reorganise ecosystems (McInturff et al., 2020) and can mitigate cattle degradation of large forest areas while continuing to provide access to wildlife. We recognise the effects we have measured represent short-term responses to cattle exclusion from forest patches (e.g. we are not measuring changes in longer-term metrics such as abundance and occupancy), and that continued monitoring is needed to quantify long-term responses resulting from habitat improvement (e.g. via an increase in understory vegetation). However, the experimental manipulation we performed in this study—cattle exclusion to protect habitat fragments in multifunctional landscapes—resulted in benefits for different wildlife species, especially those that would otherwise co-occur with cattle. Pervasive management challenges associated with food and energy production require that conservation scientists scale up evidence-based interventions to optimise the use of remaining habitat by wildlife (Montanheiro Paolino et al., 2024), beyond simply describing ecological conditions and conservation challenges (Williams et al., 2020). This study highlights the potential for evaluating ecological interventions using robust BACI monitoring designs to improve habitat for wildlife.

AUTHOR CONTRIBUTIONS

Juliana Vélez and John Fieberg conceived the ideas and designed methodology. Juliana Vélez, Juan David Rodríguez and María Fernanda Suárez collected the data. Juan David Rodríguez, José Manuel Torres and César Barrera led the building of the fences for the exclusion experiment. Juliana Vélez and John Fieberg analysed the data and led the writing of the manuscript. William McShea and Budhan Pukazhenthi contributed critically to the drafts. All authors gave final approval for publication.

ACKNOWLEDGEMENTS

We are very grateful to all the workers of the Rey Zamuro—Matarredonda Reserve for their ongoing participation in construction and maintenance of the fences, data collection and other field activities. We appreciate the help of students and volunteers with data collection. We are grateful to Pablo Stevenson, Mireya Osorio and the Universidad de los Andes for their help with institutional requirements. Thanks to María C. Santos for diagram design. This research was sponsored by the Fulbright-Minciencias Scholarship, the WWF's Russell E. Train Education for Nature Program (EFN), the Interdisciplinary Center for the Study of Global Change Fellowship, the Department of Fisheries, Wildlife and Conservation Biology at the University of Minnesota, the Big Ten Academic Alliance Smithsonian Fellowship and the Dayton Bell Museum Fund. John Fieberg received partial salary support from the Minnesota Agricultural Experimental Station.

CONFLICT OF INTEREST STATEMENT

We have no conflicts of interest.

Open Research

DATA AVAILABILITY STATEMENT

Data and code for reproducing all of the analyses have been archived in the Data Repository for the University of Minnesota https://doi.org/10.13020/k7x0-gy26 (Vélez, McShea, Pukazhenthi, Rodríguez, et al., 2024) and a GitHub repository (https://github.com/julianavelez1/baci_fencing.git).