Rethinking livestock encroachment at a protected area boundary

Significance

Protected areas borderlands are often hotspots of human–wildlife conflicts. In Kenya’s Maasai Mara National Reserve (MMNR), cattle frequently enter the reserve to graze. This “livestock encroachment” has been believed to cause wildlife declines in MMNR, leading to strict and sometimes violent measures against livestock and herders. However, our study found that the presence of cattle, at their current intensity, had no significant impacts on the presence of nearly all wild herbivores or on vegetation and soil conditions. Rather, entering MMNR presents as the only option for local pastoralists to continue their livelihoods. These findings highlight the need to rethink the current management practices in and beyond MMNR to consider both ecological integrity and social well-being.

Abstract

The presence of livestock inside protected areas, or “livestock encroachment,” is a global conservation concern because livestock is broadly thought to negatively affect wildlife. The Maasai Mara National Reserve (MMNR), Kenya, exemplifies this tension as livestock is believed to have resulted in the declining wildlife populations, contributing to the strict and sometimes violent exclusion measures targeting Maasai pastoralists. However, research embedded in the real-world setting that draws insights from the social–ecological contexts is lacking. In this study, we conducted 19 mo of ecological monitoring covering 60 sites in MMNR and found that cattle presence inside the reserve did not significantly impact most co-occurring wild herbivores at the current intensity. Using the Hierarchical Modeling of Species Communities and Gaussian copula graphic models, we showed that cattle had no direct associations—neither negative nor positive—with nearly all wild herbivores despite frequently sharing the same space. Moreover, we did not detect resource degradation correlated with cattle presence near the MMNR boundary. Given the colonial legacy and land use history of Mara, entering MMNR becomes the only viable option for many herders. These results corroborate the emerging perspective that the ecological impacts of extensively herded livestock on wildlife might be more nuanced than previously thought. To effectively balance the needs of people, livestock, and wildlife, the current rigid livestock exclusion measures need to be reassessed to holistically consider herbivore ecology, local land use history, and modern politics of protected area management.

Protected areas are the cornerstone of biodiversity conservation. However, people can settle around protected areas because of the initial displacement from the protected areas (1), economic incentives such as employment in tourism (2, 3), and land tenure changes (4). Numerous studies have shown that human activities at or across protected area boundaries can compromise wildlife populations and ecosystem functioning inside these areas (5, 6). As a result, there is a growing consensus that protected areas should be managed as a part of the larger landscapes to maintain livelihood options for local communities while maximizing conservation effectiveness (7–9). Yet, accurate understandings of the interactions among biotic, abiotic, and human components—crucial for devising integrative and inclusive management strategies—are often lacking (10).

The periphery of Maasai Mara National Reserve (MMNR) exemplifies the intricate relationships between protected areas and local livelihoods (4, 11). Located in the northern region of the Serengeti-Mara Ecosystem, MMNR harbors some of the highest densities of wild fauna on the planet and is the destination for millions of ungulates migrating from Serengeti during the long dry season from July to October (12). Since its establishment in 1948 by the British colonial administration, the reserve and its surrounding areas have experienced drastic land tenure reforms, including the displacements of Maasai pastoralists, the appropriation of land, the formation and subdivision of group ranches, and the expansions of private conservancies [(4, 11, 13), Fig. 1 and SI Appendix, Supporting Text]. Despite the long history of pastoral use of the areas dating back before the 1900s (11, 14), livestock grazing today is subject to legal restrictions. Particularly, the grazing bans in MMNR and the neighboring conservancies have effectively “squeezed” Maasai pastoralists into small and marginalized areas (15, 16). To cope with the surging pressure on the remaining grazing lands, particularly during dry and drought seasons, pastoralists frequently cross borders to graze inside the reserve to access forage and water resources that are scarce elsewhere (11, 17).

Fig. 1.

Study area and sampling design. The left panel shows the timeline of key events about the establishment and management of Maasai Mara National Reserve (MMNR, green bounded area in the map) and neighboring private conservancies (orange bounded areas) in Kenya. For more detailed information on the land history, see SI Appendix, Supporting Text. The right panel demonstrates the 60 sampling sites (purple dots) along five transects extend from the MMNR boundary 12 km into the reserve.

Concurrently, wild herbivore populations have declined inside MMNR over the past four decades (18, 19)—a trend largely attributed to competition with “encroaching” livestock (5, 17, 20, 21). Three types of evidence were commonly used to support the wildlife–livestock competition narrative in MMNR: 1) occurrence correlation, where the absence of wildlife correlates with the presence of livestock (19–21); 2) spatial displacement, with wildlife voluntarily or involuntarily avoiding areas used by livestock (22, 23); and 3) resource compression, which links livestock presence with resource degradation at the edge of MMNR (5). These claims have led to severe measures against pastoralists entering MMNR, including harassment, heavy fines, arrestments, and violence (11).

Beyond MMNR, recent evidence has challenged the prevailing belief that wild herbivores and extensively herded livestock (as opposed to intensive livestock farming) necessarily compete for space and resources (24, 25). For example, recent molecular evidence revealed that the diets of cattle and wild herbivores in Africa were more divergent than previously believed (26, 27), pointing to a potential coexistence mechanism through fine-scale spatial and temporal partitioning (28). Additionally, studies found that while livestock reduces forage quantity, they could benefit wildlife by enhancing forage quality and diversity, especially when their grazing intensity is low (29–32). Hence, different wild species may interact with livestock differently in response to the forage quantity-quality trade-off (33). Ultimately, the interactions between wild and domestic herbivores may form a competition-facilitation spectrum with the direction and magnitude of highly species-specific (24) and dependent on biotic and abiotic contexts such as grazing intensity (31), seasonality (31, 34), and management practices (28, 35).

Given these emerging perspectives, could the effects of livestock presence inside MMNR be more nuanced than previously believed? In theory, observed species relationships are a result of direct species associations, species’ responses to environmental processes (i.e., environmental filtering), and the influence of mediator species [i.e., a third species that two species both respond to, (36, 37)]. However, studies in MMNR primarily relied on annual population surveys at a coarse spatial scale (19–22), which can be sensitive to regional-scale processes other than livestock presence such as expansion of machinery cultivation, steep growth of tourism, and spread of fences (16, 18). Second, few studies to our knowledge have controlled for potential interspecies interactions when evaluating herbivore assembly in MMNR. As the tension between conservation and local livelihoods at the MMNR borders continues to escalate (5, 15), the extreme wildlife–livestock competition narrative warrants reassessment.

Here, we addressed the previous research issues and asked whether and how livestock present inside MMNR competed with wild herbivores. We reevaluated the three common types of competition evidence—occurrence correlation, spatial displacement, and source compression. Different from the commonly used annual population survey data, this study was based on finer-scale field data spanning 19 mo (2018 to 2019) and 60 sites extending from the boundary of MMNR into the reserve (Fig. 1). We quantified spatially explicit species association networks among cattle and eleven sympatric wild herbivores to examine occurrence correlation and spatial displacement. We then tested whether cattle presence in the MMNR has led to resource compression through a three-pronged approach to i) quantify cattle occurrence as a function of the MMNR boundary, ii) examine the effects of cattle occurrence on vegetation and soil conditions, iii) evaluate vegetation and soil conditions as a function of distance to the MMNR boundary.

Results

Wildlife and Livestock Co-occurrence and Association.

Our surveys confirmed cattle presence inside MMNR (Fig. 2A). Cattle dung was present in 5.9% (n = 67) of the 1,140 sampling events, covering 9 out of the 19 mo surveyed. The average cattle dung count was 0.39 (σ = 2.04) at a single site and 23.16 (σ = 49.91) across all sites in a month (SI Appendix, Table S1). Among all 440 cattle dung, 87.3% of them were found during the long dry season (July–October). In comparison, wildebeest was the most prevalent herbivore species in the study area. We found wildebeest dung in 69.7% (n = 795) of the sampling events, spanning 17 out of the 19 mo surveyed. The average wildebeest dung was 13.29 (σ = 18.25) at a single site and 794.68 (σ = 860.81) across all sites. At the monthly scale, cattle and wild herbivores shared habitats. Among the 67 sampling events where cattle dung was present, the average wild species richness at the same site was 4.6, with a maximum richness of 7.0 (SI Appendix, Fig. S2). Additionally, all eleven wild ungulates have used sites used by cattle (SI Appendix, Table S2).

Fig. 2.

Occurrence of cattle and eleven wild herbivores using dung counts as the proxy in the Maasai Mara Nature Reserve from May 2018 to November 2019. (A) Wildlife and livestock occurrence summarized over the 60 sample sites overlapped with monthly precipitation (mm, gray dash line). Lower panels show the predicted cattle (B), zebra (C), and wildebeest (D) occurrence within the 95% credible interval inside MMNR as a function of distance to the reserve boundary. The prediction was conducted for the driest month when cattle use of the reserve is typically at its peak and when wildlife–livestock conflicts were believed to be the greatest (July, average precipitation 13.4 mm, calculated from the 20-y TerraClimate precipitation history). Each line represents a single draw from the posterior distribution of the Hierarchical Modeling of Species Communities model, and a total of 4,000 predictions were made for each species.

Despite cooccurrence, our species association network found no evidence for spatial displacement. While raw occurrence correlations based on Pearson’s r showed that cattle were negatively associated with six wild herbivores (Fig. 3A and SI Appendix, Fig. S3A), these associations became absent after controlling for environmental covariates and spatiotemporal structures (Fig. 3B and SI Appendix, Fig. S3B). Further controlling for potential effects of mediator species consistently showed the absence of cattle-wild herbivores associations except for buffalo, which exhibited a negative, albeit negligible, association (r = −0.01, Fig. 3C and SI Appendix, Fig. S3C).

Fig. 3.

Species association networks derived from three different methods showing positive (blue), negative (red), or no association (white) among the 12 herbivore species in MMNR. (A) Raw species occurrence correlations based on Pearson’s correlation; (B) Species correlation estimated using the residual association structure among species from the Hierarchical Modeling of Species Communities, which controls for effects of environmental variables; (C) Direct species association estimated as the conditional dependence among species from Gaussian copula graphic models, which controlled for effects of environmental variables as well as mediator species. The width and shade of links represent the magnitude of correlation (SI Appendix, Fig. S3).

Cattle Concentration and Effects on Resource Conditions at MMNR Boundary.

Cattle tended to concentrate near the MMNR boundary as the Hierarchical Modeling of Species Communities (HMSC) revealed a significant negative relationship between cattle occurrence and the distance to the boundary (β = −0.302, posterior probability = 97.6%, SI Appendix, Fig. S4). Specifically, cattle occurrence decreased sharply as the distance to MMNR boundary increased (Fig. 2B and SI Appendix, Fig. S4). Such concentration near the boundary was also found in Thomson’s gazelle, Grant’s gazelle, Topi, and Eland (SI Appendix, Figs. S4 and S6). The occurrence of the rest of the species, including zebra and wildebeest, the two most prevalent wild species, showed no significant response to the distance to MMNR boundary (Fig. 2 C and D and SI Appendix, Figs. S4 and S6).

While cattle significantly affected vegetation quantity, quality, and soil N (Table 1), we found no evidence of resource compression by cattle. Based on spatial mixed-effect models [spaMM, (38)], the occurrence of all wild herbivores or all herbivores, rather than cattle alone, better predicted the variations in resource conditions (SI Appendix, Table S3) and showed stronger effects (Table 1). Although the null model and linear model were competitive (i.e., within 2 marginalized ΔAIC) when examining the relationships between distance to boundary and resource conditions (SI Appendix, Table S4), the linear effects of distance to the boundary were not significant in all models (Table 1).

Table 1.

Coefficient estimates of grazing effect models and linear spatial compression model

		Response variable
Model	Effect	Veg. quantity	Veg. quality	Veg. greenness	Soil N
Grazing effect	Cattle	−0.08	0.10	0.00	−0.12
		[−0.13, −0.03]	[0.05, 0.16]	[−0.05, 0.04]	[−0.18, −0.06]
	All wild ungulates	−0.52	0.37	−0.06	0.28
		[−0.56, −0.47]	[0.31, 0.42]	[−0.11, −0.01]	[0.22, 0.34]
	All ungulates	−0.52	0.38	−0.06	0.27
		[−0.57, −0.48]	[0.32, 0.42]	[−0.11, −0.01]	[0.21, 0.34]
Linear spatial compression	Distance to boundary	−0.003	−0.002	−0.006	0.000
		[−0.052, 0.046]	[−0.056, 0.053]	[−0.050, 0.038]	[−0.057, 0.058]

Grazing effect models examined the effects of cattle occurrence, total herbivore occurrence, and total wild herbivore occurrence on four resource conditions, namely vegetation quantity (grass height), quality (crude protein content), greenness (NDVI), and soil N. The linear spatial compression model examined the linear effect of distance to the boundary of Maasai Mara National Reserve on the four resource conditions. Each coefficient was derived from a mixed-effect model with spatially correlated random effects. All models also controlled for the effects of precipitation and time of year. Significant effects are bolded.

Discussion

The nature of wildlife–livestock relationships is widely considered to be negative around the world (24, 25, 39). Hence, livestock presence in protected areas, termed “encroachment” or “incursion,” was perceived as a major conservation challenge, incentivizing broad support for strict exclusionary policies against livestock grazing in protected areas (25, 40, 41). MMNR serves as an extreme example where violent measures have been exercised (11, 41). However, we found that the current level of cattle presence in MMNR did not adversely affect wild herbivores. Using intensive field data and recently developed modeling approaches, we found little to no direct associations between cattle and all sympatric wild species in MMNR. Despite the concentration of cattle near the MMNR boundary, we found that no species avoided the boundary, and the vegetation and soil conditions were comparable at the edge and the core of the reserve. Altogether, our study questioned the ecological foundation of the current rigid livestock exclusion measures. Joined with growing evidence of complex relationships between wildlife and extensively herded livestock (26, 28, 29), the one-sided “livestock encroachment” narrative warrants more sophisticated evaluation in and beyond MMNR to consider the ecological, historical, political, and socioeconomic dimensions holistically.

Effects of Cattle Presence on Wildlife and Resource Conditions.

Consistent with previous studies (4, 5, 19), we confirmed that cattle frequently enter MMNR primarily during the long dry season between July and October (Fig. 2A). While the raw occurrence correlation between cattle and wildlife was predominantly negative (Fig. 3A and SI Appendix, Fig. S3A), their associations became largely nonsignificant after controlling for environmental variables, spatiotemporal structures, and the effects of mediator species (Fig. 3 B and C and SI Appendix, Fig. S3 B and C). In other words, cattle-wildlife co-occurrence patterns might be dominated by the effects of environmental filtering and biological interactions with other species, rather than direct attraction or repulsion. The comparison between the three community networks demonstrated that wildlife–livestock relationships drawn from numeric occurrence correlation could be misleading and that previous conclusions on cattle-wildlife competition in MMNR solely based on negative occurrence correlations could be biased.

Cattle did not spatially displace wild herbivores at the current presence intensity inside MMNR, as supported by 1) the absence of direct species associations between cattle and most wild herbivores (Fig. 3 B and C) and 2) the HMSC results that showed none of the wild herbivores avoided the boundary (SI Appendix, Fig. S6) despite the concentration of cattle near the MMNR boundary (Fig. 2A). Notably, buffalo, the only wild species that showed a negative direct association with cattle (Fig. 3C and SI Appendix, Fig. S3C), showed no significant responses to the distance to the boundary (SI Appendix, Figs. S4 and S6). It is plausible that the observed negative association may manifest as avoidance at a spatial scale finer than the 1-km resolution of our surveys (Fig. 1). While our results are at odds with Veldhuis et al. (5), which showed a contraction of wildebeest habitat use 15 km into MMNR, they are consistent with other studies that documented the extensive presence of wild herbivores at the MMNR edge habitats (42, 43). The varying results might be partly due to the differences in the focal scales [i.e., patterns based on large-scale aerial population surveys in the entire Serengeti-Mara Ecosystem, as in Veldhuis et al. (5) vs. fine-scale intensive field surveys particular to the MMNR north border, as in this study]. Temporally, cattle and wildlife may partition at different times of day, which may not be captured by our monthly surveys. Future studies should leverage advancements in animal tracking technology by concurrently monitoring both wildlife and livestock with high-resolution, long-term tracking, which allows assessment of species interactions across scales (44).

We did not detect resource compression correlated with the concentration of cattle near the MMNR boundary, as all resource conditions we explored showed no significant relationships with distance to the MMNR boundary (Table 1). Consistent with results from the Kenya Long-term Enclosure Experiment (31, 45), we found that total herbivory rather than isolating cattle effects better explained resource patterns (SI Appendix, Table S3), likely a result of the low intensity of cattle occurrence in our study. Our future work plans to expand into adjacent conservancies and unprotected areas to examine resource patterns across a full gradient of cattle usage. It is important to note that grazing intensity is among the most important factors governing the effects of herbivores on ecosystems (24). While cattle presence in MMNR currently appears benign, an increase in grazing pressure could lead to significant environmental changes (35), as has been documented in Laikipia, Kenya (40).

Herder Decisions May Influence Wildlife–Livestock Interactions.

Importantly, pastoral herding strategies are likely pivotal in mediating the patterns and consequences of cattle presence inside MMNR. Inside the reserve, pastoralists tend to avoid park rangers and tourists where wildlife tends to gather (11, 46, 47) and graze strategically in areas with sufficient, but not necessarily the highest, forage quality and quantity (46). Our results showed that only 25.5% of cattle occurrence variance was explained by the environmental variables, the lowest among all studied species (SI Appendix, Fig. S5). This finding suggests that additional factors beyond the environment, such as herding decisions, could have shaped cattle occurrence. In addition, pastoralists attempt to avoid wildlife that are potentially dangerous for humans and livestock (46, 48), such as elephants and buffalos [(49), Fig. 3C]. Altogether, herders’ attempts to balance the need for resource access (gravitating toward areas also preferred by wild herbivores) and the necessity of avoiding hazards (including park rangers, tourists, and dangerous animals) may have neutralized the positive and negative correlations between cattle and most wild herbivores (Fig. 3 and SI Appendix, Fig. S3). This may also explain why cattle exhibited a negative response to forage quality (SI Appendix, Fig. S4).

Although our study lacks information to quantify the effects of herding strategies, it uniquely positioned cattle as a component of, rather than an external factor influencing, the herbivore assembly. This species-specific community approach is an important step toward acknowledging the complex spatial ecology of livestock as well as the key role of herders, which were typically omitted in wildlife–livestock interaction studies (35, 48, 50). Participatory approaches and knowledge coproduction are essential for future studies to properly incorporate herder’s knowledge to holistically understand patterns and consequences of wildlife–livestock interactions (46, 51).

Toward a Reassessment of “Livestock Encroachment”.

While our study found few negative impacts of cattle presence under its current, relatively low intensity within MMNR during our study period, the reserve—along with the residing wildlife and the Indigenous people—is unquestionably grappling with conservation and sustainability challenges. In the broader Mara region, resident wildlife have been progressively replaced by livestock (19), primarily contributed by the increase of sheep and goats as opposed to cattle, which are central to the competition narrative (SI Appendix, Fig S7). Other activities such as land division, fencing, cultivation, and the development of tourist infrastructure are encroaching up to the reserve’s edge (13, 16, 18), creating stark landscape contrasts and threatening the connectivity and viability of wildlife within the reserve (7, 42). Yet, rather than mitigating human–wildlife conflicts, the current grazing exclusions in the reserve and the conservancies are likely only to exacerbate these issues as they reinforce the fortress conservation scheme, foster divergent land management trajectories across jurisdiction boundaries, and further alienate Indigenous communities from conservation efforts (3, 6, 8).

Moving forward, managers and policymakers must recognize that the conflicts revolving around livestock presence inside protected areas are coproduced by the ecology of resource utilization, the history of land use, and the politics of PA management (11, 35, 48). MMNR harbors key resource areas for pastoralists during dry seasons, which coincides with the arrival of millions of migratory wildlife [hence tourists, (46, 47)]. After being displaced, many Maasai feel morally justified to access resources inside PAs because of ancestral claims and customary rights to resources (52). However, as land tenure is being reorganized across the Mara (e.g., exclusionary membership of conservancies, property disputes among neighbors), the landscape has become ripe with fines, fences, and violence for herders and grazing options become increasingly limited [SI Appendix, Supporting Text (11, 15)]. Too often, entering MMNR is the only survival choice (11). Meanwhile, MMNR is a significant source of tourism revenue for the state. Studies have conclusively linked grazing restrictions and the ensuing conflicts to the rise of tourism, which is predicated on the fetishism of “pristine wilderness”—the presence of wildlife and the absence of cattle (11, 52, 53). In fact, restrictions on livestock movements are spatially and temporally patterned in ways that reflect the changing seasonality of tourist visits, and conflicts between park rangers and pastoralists are more likely to occur in areas frequented by tourists on safari (52). Fundamentally, narratives around wildlife–livestock competition and human–wildlife conflicts in MMNR, although often appearing to be apolitical, cannot be examined separately from historical and contemporary injustices.

Pastoralists around the world have long been marginalized in conservation discourses (54). MMNR likely represents one of many cases where the relationships between wildlife and livestock are not solely negative. As the global conservation scheme increasingly strives to balance ecological integrity and social well-being (7, 9), there is a pressing need to reassess the one-sided narrative of “livestock encroachment” and to update prevailing exclusionary management practices.

Materials and Methods

Study Area and Sampling Design.

MMNR is an unfenced PA and composed primarily of open grassland interspersed with riparian woodland vegetation. The precipitation is characterized by a bimodel pattern, with a long dry season between July and October and a short dry season between January and February. We conducted monthly surveys of animal, vegetation, and soil surveys at 60 sites inside MMNR for 19 mo from May 2018 to November 2019, resulting in 1,140 sampling events (Fig. 1). The 60 sampling sites were established along five transects, starting 1 km from the northern border and extending 12 km into the reserve. Each transect had 12 sites spaced 1 km apart. Each site contained one 100-m subtransect, running perpendicular to the transect and in alternate directions. A total of 15 local assistants were hired across the study period, all with extensive knowledge of local flora and fauna.

Species Occurrence Data.

We used species dung counts to approximate the occurrence of wild and domestic herbivores (29). Dung counts were obtained using a belt transect method by three observers walking in parallel and visually identifying dung piles to species five meters on either side of the subtransect. Each counted dung was destroyed in situ to avoid double-counting. For analysis, we summarized the monthly total number of dung piles per species per site. We excluded species detected less than three times, resulting in a dataset including one domestic species, cattle (Bos taurus), and eleven wild species including wildebeest (Connochaetes taurinus), zebra (Equus quagga), Thompson’s gazelle (Eudorcas thomsonii), impala (Aepyceros melampus), topi (Damaliscus lunatus), eland (Taurotragus oryx), buffalo (Syncerus caffer), Grant’s gazelle (Nanger granti), waterbuck (Kobus ellipsiprymnus), dikdik (Madoqua kirkii), and elephant (Loxodonta africana).

Vegetation and Soil Surveys.

We measured three vegetation properties: site utilization, quantity, and quality. Site utilization was measured by walking along the 100-m subtransect, visually evaluating the presence of bite marks on vegetation every 5 m and thereby calculating the percentage of bites at each site. Vegetation quantity was measured using a rising plate pasture meter every 5 m along the subtransect, which estimates standing biomass. Because our sampling sites are within the same habitat type, the pasture meter readings provided a simple way to represent relative vegetation quantity across sites. For analysis, we calculated the average pasture meter reading at each site. To measure vegetation quality, we clipped vegetation samples defined by a randomly placed a 1-ft² frame. Samples were sent to Crop Nutrition Laboratory Services Ltd. (CROPNUTS) for protein content analysis using infrared analysis. For soil, we measured the N content of soil samples at a depth of 2 to 7 cm, analyzed by CROPNUTS. We focused on soil N because it was shown as the limiting nutrient on grasses in a similar East African savanna (55).

Additional Environmental Data through Remote Sensing.

We obtained monthly precipitation (mm) for the study area from TerraClimate (56) to match the study period from May 2018 to November 2019. We also obtained the precipitation history from 2001 to 2020 and calculated the average monthly precipitation, which was later used for model prediction (see below). To measure vegetation greenness, we extracted and calculated monthly NDVI for each sampling site from the 250-m Terra Moderate Resolution Imaging Spectroradiometer (MODIS) Vegetation Indices (MOD13Q1 v006). Both precipitation and NDVI data were extracted through Google Earth Engine (57).

Occurrence Correlation and Spatial Displacement.

We first constructed a nonspatial species occurrence correlation network using pairwise Pearson’s correlation coefficients of raw dung counts, representing the commonly used correlation methods in previous studies (e.g., refs. 4, 17, and 29). To examine whether livestock spatially displaced wildlife, we then constructed spatially explicit species association networks between cattle and sympatric wild large herbivore dung counts using two methods. The first network was based on the residual covariance matrix estimated from a joint species distribution model fitted using Hierarchical Modeling of Species Communities (HMSC, 58), with R-package “Hmsc” (59). HMSC models environmental filtering by variation and covariation in the responses of individual species to their environment and captures biotic assembly rules by species-to-species association matrices through a multivariate Bayesian generalized linear mixed modeling framework (58). Specifically, we fitted a Poisson-lognormal model using dung counts as the response variable, site utilization, vegetation quantity and quality content from the previous month, distance to MMNR boundary, precipitation, and month as the fixed effects. We applied a trigonometric transformation to the “month” variable to account for its circular nature. To account for the spatial and temporal autocorrelation, we included spatial and temporal latent factors as random effects (59). We assumed the default prior distribution and fitted the models with four Markov Chain Monte Carlo chains, each of which consisted of 3,950,000 iterations, out of which we discarded the first 50,000 as the burn-in period and thinned the remainder by 1,000, yielding 4,000 samples per chain (16,000 samples in total). We evaluated the effective sample size to ensure adequate independence of samples and potential scale reduction factors to assure model convergence (SI Appendix, Fig. S1).

Most existing model-based methods for analyzing species cooccurrence, including HMSC, do not consider the potential effect of mediator species. Hence, we constructed a second network using Gaussian copula graphic models (GCGMs, 36), with R-package “ecoCopula” to quantify direct species association through conditional dependence among species. GCGMs are a flexible type of graphical model that accommodates a broad range of data types for exploring whether co-occurrence patterns among species may be explained by indirect mediator species, by environmental variables, or neither (36). Because ecoCopula does not support mixed-effect models, we used the same set of response variables and environmental fixed effects as the HMSC model with “transect ID” as an additional fixed effect to account for the spatial structure of the dung count dataset.

Resource Compression.

According to Veldhuis et al. (5), if resource compression exists, we predicted that 1) cattle occurrence was concentrated near the MMNR boundary, 2) cattle significantly affected resource conditions, and 3) resource conditions were compromised near the boundary compared to the core of MMNR. To examine the first prediction, we used the HMSC model to quantify and predict cattle occurrence as a function of the distance to the MMNR boundary. The prediction was made in a dry month when the wildlife–livestock competition was assumed to be the greatest (5, 30), and we used the average precipitation of the driest month from 2001 to 2020 (July, 13.4 mm, calculated from the TerraClimate precipitation history), for model prediction to account for the interannual precipitation variations. We also conducted model prediction for all wild species occurrence as a reference.

To examine the second prediction, we fitted spatial generalized linear mixed models to quantify the effects of cattle occurrence on vegetation quantity, vegetation quality, vegetation greenness, and soil N using the R-package “spaMM” (38). The spaMM is robust to converge and fast to fit. We log-transformed and normalized all four resource conditions to obtain standardized coefficient estimates. For each resource condition, we used cattle dung counts, precipitation, and trigonometrical-transformed month as fixed effects and a Matérn correlation structure to account for spatial autocorrelation. We also fitted the same sets of models replacing cattle dung counts with total wild herbivore dung counts and total herbivore dung counts. This set of models was referred to as the “grazing effect models” (Table 1). We compared which grazing effect models better predicted resource condition variations based on marginal AIC (SI Appendix, Table S3).

Finally, we evaluated the third prediction using spaMM models with distance to boundary, precipitation, and trigonometrical-transformed month as fixed effects and a Matérn correlation structure (Table 1). Based on marginal AIC, we compared a null model (i.e., no distance effect) with a linear model as well as a quadratic model (Table 1), given that resource condition in relation to distance to boundary might follow a curvilinear form where effects might be the strongest between 0 and 12 km (5). This set of models was referred to as “spatial compression models.” Annotated R code and data to replicate the analyses are available on Zenodo (10.5281/zenodo.10688128) (60).

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

Comma-separated values (csv) files, R data file, and R code have been deposited in Zenodo (https://zenodo.org/doi/10.5281/zenodo.10688128) (60).