Humans induce differential access to prey for large African carnivores

Wildlife adaptively respond to human presence by adjusting their temporal niche, possibly modifying encounter rates among species and trophic dynamics that structure communities. Here we show that these human-induced modifications to behaviours are prolific among species and alter apex predators’ access to prey resources. We assessed human-induced changes to wildlife diel activity and consequential changes in predator-prey overlap using 11,954 detections of three apex predators and 13 ungulates across 21,430 trap-nights in West Africa. Over two-thirds of species altered their diel use in response to human presence, and ungulate nocturnal activity increased by 6.8%. Rather than traditional pairwise predator-prey comparisons, we considered spatiotemporally explicit predator access to a suite of prey resources to evaluate community-level trophic responses to human presence. Although leopard prey access was not affected, lion and hyena access to 3 prey species significantly increased when prey increased their nocturnal temporal niche to avoid humans. Ultimately, humans considerably altered the composition of available prey, with implications for prey selection, demonstrating how humans perturb ecological processes via behavioural modifications.

in predator-prey overlap using 11,954 detections of three apex predators and 13 ungulates across 21,430 23 trap-nights in West Africa. Over two-thirds of species altered their diel use in response to human 24 presence, and ungulate nocturnal activity increased by 6.8%. Rather than traditional pairwise predator-25 prey comparisons, we considered spatiotemporally explicit predator access to a suite of prey resources to 26 evaluate community-level trophic responses to human presence. Although leopard prey access was not 27 affected, lion and hyena access to 3 prey species significantly increased when prey increased their 28 nocturnal temporal niche to avoid humans. Ultimately, humans considerably altered the composition of 29 available prey, with implications for prey selection, demonstrating how humans perturb ecological 30 processes via behavioural modifications. and resulting encounter rates are likely to be changed 11 , thus altering predator access to a suite of prey 48 resources (Fig. 1). Such perturbations to predator-prey dynamics can have cascading impacts that alter 49 population regulation, habitat structure, and various ecosystem processes, such as carbon storage, herbivory and seed dispersal [12][13][14][15][16] . 51 52 Figure 1: Conceptual framework illustrating the community-level effects of human presence on predator-prey 53 temporal interactions. a) Circles represent the temporal niche occupied by each species (3 prey and 1 predator), and 54 shaded regions indicate temporal overlap between the predator and a prey species (i.e., shared temporal niche 55 space). Dotted lines within each circle depict the species' temporal activity distribution. b) The diel activity patterns 56 of both predators and prey are expected to shift in response to human presence, generally increasing nocturnal 57 activity to avoid humans during the day. c) As wildlife diel activity changes, so does predator access to individual 58 prey. This can lead to intensified or relaxed predation pressures on an individual prey species depending on the diel 59 responses of the prey compared to the predators and other sympatric prey.
Overall, our threshold of human occupancy designating low vs. high pressure did not alter our 133 interpretation of sensitivity from resultant shifts in diel activity. The only change we observed was 134 detecting significance when reducing the threshold from 0.54 to 0.44 for 2 species: leopard and roan 135 antelope (Table S3). Our results highlight that most species respond to human occurrence by modifying 136 their behaviours and reducing their realized temporal niche to incorporate more night-time activity, 137 potentially altering predator-prey encounter rates. 138 139 Figure 3: Temporal activity kernel density curves for apex predators and ungulates in areas of low and high human 140 use (threshold human occupancy = 0.54). Nocturnal diel periods (2 hours after sunset to 2 hours before sunrise) are 141 shaded using the average times of sunrise and sunset during our study period, and lighter shading represents the diel-142 specific nocturnal activity that is different between low and high human areas. Significance levels for bootstrapped 143 randomization test of differences in diel distributions between human zones: * < 0.05, ** < 0.01, *** < 0.001. Plus 144 signs (+) represent species with p-values < 0.1 which achieved significance when the human occupancy threshold 145 was adjusted ± 0.1 (Table S3). 148 between low and high human zones for carnivore (dashed lines) and ungulate (solid lines) species and overall guilds.

149
Increases and decreases in nocturnality from low to high human use areas are indicated by plus (+) and minus (-) 150 labels, respectively, next to species' names. Stars (*) above colored lines indicate species that showed significant 151 differences in nocturnal activity between human zones based on bootstrapped 95% CIs of nocturnality.

Changes in predator-prey overlap 154
Human-induced diel activity shifts among species did not result in significant changes in individual 155 predators' overlap with prey when we aggregated their prey species (Fig. S2). However, high human use 156 reduced the mean overlap of lions with their prey by 0.08 (Δhigh = 0.718 ± 95% CI 0.08; Δlow = 0.797 ± 157 0.11). In contrast, leopards may be experiencing some benefit from human use, as their temporal overlap 158 with prey exhibited a mean increase of 0.17 where human activities were high (Δhigh = 0.699 ± 0.12; Δlow 159 ΔPAUChyena = +1.88 ± 0.87%), duiker (ΔPAUClion = +1.56 ± 1.31%, ΔPAUChyena = +0.84 ± 0.73%), and 167 kob (ΔPAUClion = -2.24 ± 1.58%, ΔPAUChyena = -1.45 ± 0.88%) (Fig. 5). All three species to which 168 predator access increased significantly also exhibited increased night-time activity as a human avoidance 169 strategy. In contrast, kob reduced nocturnality and experienced lower overlap with lions and hyenas (Fig.  170 4). Additionally, lion and hyena access to 2 prey species showed near significant changes (buffalo 171 ΔPAUClion = +1.33 ± 1.41%, ΔPAUChyena = +0.87 ± 0.97%; and waterbuck ΔPAUClion = -1.66 ± 1.73%, 172 ΔPAUChyena = -1.67 ± 1.71%). 173 All three apex predators showed strong similarities in prey access differences with comparable 174 changes in access to all prey between human use levels (Fig. 5). However, changes in leopard access to 175 prey items were not significant based on 95% CIs, with only aardvark (ΔPAUCleopard = -4.6 ± 4.7%) and 176 bushbuck (ΔPAUCleopard = 1.7 ± 1.8%) access nearing significance (Fig. 5). We suspect this is due to 177 leopards' differential response to human presence (-4.6 ± 19.1% change in nocturnality) compared to 178 lions (+11.9 ± 16.3%) and hyenas (+3.7 ± 6.5%; Fig. 4). 179 Although overlap with total available prey did not differ, the variation in species-specific prey 180 accessibility (PAUC estimates) was higher for lions as a result of human presence (Fligner-Killeen test, p-181 value = 0.03), indicating reduced diversity of available prey and therefore increased access to certain prey 182 species compared to others. Leopards and hyenas showed no significant changes in access variability as a 183 response to humans.  Wildlife responses to human activities have the potential to reshape natural ecological processes and 196 trophic dynamics 8,9,23 . When anthropogenic pressures are heterogeneous, the resultant dynamism 197 promotes a myriad of adaptive strategies to manage and mitigate threats including behavioural shifts in diel activity that redefine species' temporal niches 10,24,25 . Such shifts in diel activity may lead to increased 199 prey vulnerability to nocturnal predators, thus altering probabilities of encounter and diets in consumers 200 ( Fig. 1). We found that over two-thirds of the assessed species altered their overall diel activity patterns as 201 a response to humans in the study area. Most species increased nocturnal activity, consistent with 202 previous works and supporting our hypothesis of human avoidance 10,25 . Valeix et al. 26  Human presence appears to be limiting temporal refugia from risks for many species and driving 206 increases in ungulate activity when predators are also active, possibly decoupling anti-predator 207 behaviours from predation risks 28,29 . Patten et al. 11 also presented evidence of human avoidance driving 208 increased predation risks in North American white-tailed deer (Odocoileus virginianus). 209 Heterogeneity in species' responses to human pressure, however, indicates different sensitivities to 210 humans among the carnivores and ungulates in our study system. Some species did not exhibit shifts 211 towards nocturnality as expected (e.g., kob and aardvark). These species may be benefitting from the 212 observed human avoidance in many sympatric species that potentially reduces risks of predation and 213 competition, commonly referred to as a human shield response 24,30 . For example, Atickem et al. 31 reported 214 mountain nyala (Tragelaphus buxtoni) leveraging predator avoidance of humans during the day as a 215 temporal refuge in Ethiopia. The ability to exploit human presence as a shield from predatory or 216 competitive encounters may be due to the life history traits of a species that reduce sensitivity to humans, 217 such as body size, energetic requirements, dispersal abilities, social structure, or foraging strategies 19,32 . 218 Similarly, these species' temporal niches may be constrained by inherent characteristics that were evolved 219 for diurnal activity, making night-time activity more costly despite refuge from human pressures and 220 limiting their adaptive capacity to avoid humans 33 . In contrast, the amount of wildlife persecution (i.e., 221 trophy hunting and poaching) in the system may induce stronger human-avoidance behaviours in hunted 222 species. For instance, Vanthomme et al. 34 attributed the negative associations of 10 mammals with human positive associations. Though the mechanisms driving differential responses to humans were not 225 explicitly investigated here, our study demonstrates non-uniform responses of large mammals to human 226 presence. As such, future work can assess the drivers of species-specific responses and sensitivities to 227

humans. 228
We showed that human presence modified the availability of prey species relative to the overall pool 229 of available prey, which is an important driver of prey selection in apex predators, and thus provide new 230 insights into community-level repercussions of human sympatry with wildlife 17,18 . While we expected 231 overall predator-prey overlap and the diversity of available prey to increase due to human avoidance, the 232 combination of human avoidance and human shield strategies observed in our system resulted in little 233 change in overall overlap but substantial changes to apex predator access to individual prey species. 234 Specifically, our new community-level approach to predator-prey temporal overlap revealed that prey 235 species experienced intensified overlap with predators when they increased their nocturnal temporal niche 236 (e.g., duiker, reedbuck, bushbuck) to avoid humans, while overlap was lessened for species that did not 237 (e.g., kob). For lions, this resulted in reduced diversity of available prey, likely intensifying predation 238 pressures on a smaller subset of species which could contribute to destabilizing trophic dynamics 35 . The 239 predators in our study are largely opportunistic night-time hunters, and temporal overlap is often strongest 240 between predators and their preferred prey species 20,21,36,37 . Thus, we expect that species experiencing the 241 highest overlap with apex predators relative to other prey to be integrated into the predators' diets in 242 higher proportions, and consequently expect varied prey selection by predators between low and high 243 human use areas. Buffalo are a common prey item of lions in other systems, and our results suggest they 244 may be vulnerable to intensified selection by lions due to human presence increasing access to buffalo in 245 our study area 38 . Additionally, human disturbance can increase the predation rates and carcass 246 abandonment by large carnivores as well as alter mesopredator foraging behaviours, potentially 247 increasing mortality rates on preferred prey species and providing augmented carrion resources that may potentially lead to alterations in predators' diets with consequences for ungulate and mesopredator 250 community regulation 41 . 251 Though protected areas are the primary strategy for biodiversity conservation worldwide, human 252 exploitation of protected areas is pervasive and in many cases necessary for the sustenance of human 253 populations 42,43 . By accounting for imperfect detection to understand human intensity of use, we 254 contribute to a more comprehensive understanding of human impacts within coupled human-natural 255 ecosystems that is imperative to effectively manage for the conservation of ecological processes, 256 biodiversity, and human needs. However, human activities observed in our study system may not impact 257 species uniformly. Because we aggregated a variety of human activities to depict human use, there might 258 be activity-specific responses by wildlife that were not captured. Humans exploit resources in national 259 parks in many ways including livestock herding, resource gathering, subsistence poaching, hunting, and 260 recreation, all of which impact the system and wildlife to varying degrees 43-45 . Overall, human impacts 261 encompass a variety of disturbances that impact ecosystems, both in our study and more broadly, and thus 262 disentangling the responses of wildlife to specific human pressures may facilitate designing more 263 effective conservation interventions 42,46 . 264 Our results demonstrate prevalent disruptions to wildlife temporal activity patterns from human 265 presence, leading to overall reductions in diurnal activity and modified community dynamics. Because 266 both carnivores and ungulates serve fundamental roles in regulating African ecosystems via predation and 267 herbivory, respectively, the pervasiveness of their responses to human occurrence demonstrates the 268 capacity for humans to disrupt essential ecological processes that facilitate coexistence among wildlife, in 269 this case reshaping predator-prey interactions. As the human footprint continually expands, spatial refugia 270 from anthropogenic disturbance become more limited, stimulating an increasing need to exploit temporal 271 partitioning to avoid human pressures. We show that the community-level implications of these 272 behavioural modifications must be considered in light of complex higher-order interactions that govern 2a). The complex contains 5 national parks (54% of total area), 14 hunting concessions (40%), and 1 280 faunal reserve (6%). Our study area within WAP comprised three national parks and 11 hunting 281 concessions in Burkina Faso and Niger across ca. 13,100-km 2 (Fig. 2a). Trophy hunting of many ungulate 282 processing). We created independence of species triggers using a 30-minute quiet period between 304 detection events using the 'camtrapR' package in R 3.5.1 (http://www.r-project.org) 52 , and we assumed 305 detections to be a random sample of each species' underlying activity distribution 21 . 306

Human occupancy models 307
We constructed single-season, single-species occupancy models to discriminate WAP into areas of 308 low and high human use. We separated detection/non-detection data for humans into 2-week observation 309 periods, which were modelled as independent surveys to account for imperfect detection. Our occupancy 310 models first modelled the detection process (p) using covariates expected to influence detection while 311 holding occupancy (Ψ) constant, and then modelled human occupancy by incorporating grouping 312 variables among which Ψ may vary. occupancy covariates using the Akaike information criterion corrected for small sample sizes (AICc). We 323 selected the top-performing detection and occupancy models as those with ΔAICc < 2 compared to the 324 lowest AICc model. We assessed goodness-of-fit of the top-performing models using 1,000 parametric 325 bootstraps of a χ 2 test statistic appropriate for binary data and estimated the ĉ statistic to ensure the data AICc) occupancy model corrected for imperfect detection 54 . From those estimates, we categorized grid 331 cells as either low or high human use. We delineated the threshold for human use using the mean value of 332 human occupancy, which presented a bimodal distribution at low and high levels of human site use (Fig.  333   2b). 334

Temporal analyses 335
Using detection timestamps from our camera survey, we compared the temporal activity patterns for 336 apex predators (lions, leopards, and hyenas) and sympatric ungulates between areas of low and high we excluded topi (Damaliscus korrigum jimela) and red-fronted gazelle (Eudorcas rufifrons) from 344 analysis in our study. Duiker species were aggregated due to difficulty distinguishing the two in camera 345 trap images, resulting in 13 total ungulate species in our analyses. 346 We used kernel density estimation to produce diel activity curves representing a species' realized 347 temporal niche in both human use zones for each of the 16 species. We first tested for differences in these 348 activity distributions between low and high human use areas for all individual species and overall for each 349 guild by calculating the probability that two sets of circular observations come from the same distribution 350 with a bootstrapped randomization test 55 . Significant differences in temporal activities were evaluated as under the diel activity curves to determine the proportion of each species' activity that occurred during 355 nocturnal hours (two hours after sunset to two hours before sunrise). We used the sunrise (05:51) and 356 sunset (18:06) times from the median date of our surveys (April 4, 2018) at the survey area centroid to 357 define nocturnal hours. To test if wildlife nocturnality increased to avoid human presence, we compared 358 the bootstrapped 95% confidence intervals (CIs) of the change in nocturnality for each species and overall 359 guilds between low and high human areas where a significant change was observed when the CI did not 360 overlap 0. 361 We used the coefficient of overlap (Δ) to quantify the total temporal overlap between each apex 362 predator and their associated prey from circular activity distributions. Elephant and hippopotamus were 363 not included in the prey list for any predator and buffalo was excluded from the prey list of leopards due 364 to large body sizes. All other prey species were aggregated to produce a single diel activity curve of all 365 prey for comparison to predator activity. We chose the specific estimator based on the minimum sample 366 size of detections for both guilds to contrast human use levels (Δ1 if N < 75, Δ4 if N > 75). Values of Δ 367 range from 0-1 where 0 represents no temporal overlap and 1 represents complete overlap or identical 368 temporal niche between predators and their prey. We used 10,000 bootstrapped estimates to extract the 369 bias-corrected 95% CIs of Δ. We compared CIs of Δ between human use levels for each species to assess 370 differences in predator-prey overlap in response to human occurrence. Non-overlapping CIs between 371 human use levels indicated that the overall temporal overlap of predators with their prey was significantly 372 altered by human presence. Temporal analyses were conducted using the 'activity' and 'overlap' 373 packages in R. 374

Predator access to prey 375
After determining overlap between predators and their prey as well as shifts induced by humans, we to assess species-specific prey access for predators that is temporally explicit over the diel period, 378 enabling assessment of changes to the composition and diversity of accessible prey for predators resulting 379 from responses to humans in both guilds. We first combined (i.e., stacked) the bootstrapped temporal 380 kernel density curves for individual prey to produce a total diel activity curve for prey, but this time 381 maintaining each species' contributions to overall prey activity. We then multiplied each prey species' 382 proportional contribution to prey activity at a given point in the diel cycle by the corresponding kernel 383 density activity value of each respective apex predator. This method produced a discrete area under the 384 predator temporal activity curve for each prey species of a given apex predator (percent area under curve, 385 PAUC), where each prey species' value represents the relative temporal overlap between the apex 386 predator and that prey species throughout the day. We used these PAUC values to assess whether predator 387 access to individual prey species, relative to all available prey, was altered by human presence by 388 calculating the difference in prey access (ΔPAUC) for each predator/prey combination between areas of 389  Buffalo not included as prey for leopard, and elephant and hippopotamus not included for any predator. 11 Error bars represent bias-corrected, bootstrapped 95% confidence intervals of coefficient estimates. Photo 12 credit: Applied Wildlife Ecology Lab (AWE), University of Michigan, images from camera trap survey 13 within the study area. 14