Habitat complexity and lifetime predation risk influence mesopredator survival in a multi-predator system

Variability in habitat selection can lead to differences in fitness; however limited research exists on how habitat selection of mid-ranking predators can influence population-level processes in multi-predator systems. For mid-ranking, or mesopredators, differences in habitat use might have strong demographic effects because mesopredators need to simultaneously avoid apex predators and acquire prey. We studied spatially-explicit survival of cheetahs (Acinonyx jubatus) in the Mun-Ya-Wana Conservancy, South Africa, to test hypotheses related to spatial influences of predation risk, prey availability, and vegetation complexity, on mesopredator survival. For each monitored cheetah, we estimated lion encounter risk, prey density, and vegetation complexity within their home range, on short-term (seasonal) and long-term (lifetime) scales and estimated survival based on these covariates. Survival was lowest for adult cheetahs and cubs in areas with high vegetation complexity on both seasonal and lifetime scales. Additionally, cub survival was negatively related to the long-term risk of encountering a lion. We suggest that complex habitats are only beneficial to mesopredators when they are able to effectively find and hunt prey, and show that spatial drivers of survival for mesopredators can vary temporally. Collectively, our research illustrates that individual variation in mesopredator habitat use can scale-up and have population-level effects.


Results
Short-term spatial drivers of survival. We included 133 cheetahs in our survival analyses for a total of 110 months. Within these cheetahs, 28 individuals were only in the adult state, 78 individuals were only in the cub state, and 27 individuals were included in the models as both in the cub and adult states.
Cheetah survival was most sensitive to short-term environmental conditions within the 50% home range contour and the same top model was supported regardless of home range contour (Appendix S2). Thus, we only present the results from the 50% HR models. At the short-term time scale, cheetahs exhibited variation in environmental conditions within their home range with regards to EVI (mean = 0.29; range = 0.14-0.44), lion encounter risk (mean = 0.36; range = 0.16-0.90), and prey density (mean = 37.8 prey/km 2 , range = 24.0-80.3 prey/km 2 ).
For our short-term survival models, survival was best described by the average EVI within the core of a cheetah's home range (Table 1). In contrast to our predictions under the habitat complexity risk mediation hypothesis, for both adults and cubs, EVI had a negative influence on survival, with higher survival occurring at the lowest EVI values ( Fig. 1). At the lowest EVI values, adult monthly survival was 0.99 (85% CI = 0.98-1.00) and cub monthly survival was 0.98 (85% CI = 0.97-0.99), whereas at the highest EVI values, adult monthly survival was 0.97 (85% CI = 0.94-0.98) and cub monthly survival was 0.84 (85% CI = 0.75-0.91). Cub survival was more sensitive than adult survival to changes in EVI, with the probability of surviving decreasing 2.8% for every 1-unit increase in EVI.
When considering environmental covariates across individuals' lifetimes, survival was best described by a model including the average EVI and the average risk of encountering a lion within a cheetahs' home range during their lifetime (Table 2). Models including EVI and lion encounter risk separately were also competitive ( Table 2), however we only present results from the top model given that the covariate relationships were similar in all competitive models. In contrast to our predictions under the habitat complexity risk mediation hypothesis, lifetime EVI had a negative influence on adult and cub survival. At the lowest lifetime EVI values, adult monthly survival was 0.99 (85% CI = 0.98-1.00) and cub monthly survival was 0.96 (85% CI = 0.93-0.98), whereas at the highest lifetime EVI values, adult monthly survival was 0.96 (85% CI = 0.89-0.98) and cub monthly survival was 0.87 (85% CI = 0.75-0.93). As predicted by the spatial top down hypothesis, lifetime lion encounter risk had a negative influence on cub survival, although there was not a significant effect of lifetime lion encounter risk on adult survival (Fig. 2

Discussion
We evaluated support for effects of spatial influences of top-down predation risk, bottom-up prey availability, and habitat complexity on cheetah survival, and found the most consistent support for survival being influenced by habitat complexity across multiple temporal scales. However, our results contradict the habitat complexity risk mediation hypothesis, which predicts that mesopredators should experience increased survival in areas of high habitat complexity 20 . Instead, our results show that subordinate predators do not always benefit from structurally complex habitats, potentially because subordinate predators might only benefit from habitat complexity if they are able to avoid predation, and effectively obtain prey, in complex habitats. In predator-prey systems, using specific areas as refuges from predation can come at a cost to the prey, because although they might reduce predation risk, the resource availability of the refuge might be lower than non-refuge areas because of increased competition or sub-optimal conditions 30,31 . In systems of multiple predators, the quality of a predation refuge habitat might be related to the subordinate predator's ability to find and hunt prey, which is a function of the subordinate predator's hunting mode 32 .
There are two main explanations as to why we did not find support for the habitat complexity risk mediation hypothesis in our study system. First, vegetation complexity might increase or reduce the probability of cheetahs being predated upon. Ambush predators in a variety of systems have been found to have enhanced hunting abilities in structurally-complex areas because of decreased sight lines for prey 33,34 . Although lions use a variety of habitat types, they kill prey more frequently in areas of dense vegetation 35,36 . Closed habitat types could act as a predation refuge for cheetahs by providing cover to enhance concealment from lions, but these habitats could also increase predation risk because of the hunting preferences of lions. Conversely, open habitat types might reduce the probability of predation by improving cheetahs' ability to detect nearby lions, compared to closed habitats 37 . However, when we analyzed locations where cheetahs were killed by other predators (Appendix S3), we did not find evidence to suggest that vegetation complexity increased the risk of cheetahs being predated upon (Fig. 3a). Therefore, it seems that cheetahs experience predation independent of vegetation complexity, and higher predation in areas of higher EVI might not be the mechanism driving our observed patterns of spatial survival.
Second, vegetation complexity might influence cheetahs' hunting ability, which in turn could affect survival. Cheetahs are coursing predators, as opposed to ambush predators, and can reach high speeds when chasing prey 38 . Therefore, open areas could improve cheetah hunting success by allowing cheetahs to see prey easier and facilitating high-speed chases. Although cheetahs are able to hunt in areas of dense vegetation 39 , they are more likely to initiate hunts, and have higher hunting success, in open habitats 40 . Prey availability can be an important driver of carnivore demography 41 , so the use of areas to facilitate hunting, rather than the density of prey themselves 42 , could influence mesopredator survival. Indeed, when we analyzed cheetah kill site locations (Appendix S3) we found that cheetah kill sites were more likely to be located in areas with low EVI (Fig. 3b).  In addition to habitat complexity affecting cheetah survival, we found that cheetah survival was also influenced by duration of exposure to top-down predation risk. Our results indicate that long-term risk of encountering a lion, rather than short-term risk of encountering a lion, influenced cheetah survival, with cubs having a lower probability of survival if there was a higher probability of encountering lions in their home range during the entire time when they were a cub. We likely did not observe effects of lion encounter risk on short-term cheetah survival because cheetahs have adapted behaviors to minimize short-term predation risk 25 . For example, cheetahs use the same general areas as apex predators such as lions and use fine-scale spatial partitioning to reduce the probability of lion encounters 10,26,27 . The use of spatial or temporal partitioning by mesopredators affected by apex predators has been found in a variety of other systems, such as red foxes (Vulpes vulpes) avoiding coyotes (Canis latrans) in North America 43 , and European badgers (Meles meles) avoiding wolves (Canis lupus) in Italy 11 .
Although fine-scale partitioning or other predator avoidance behaviors might be beneficial to reduce shortterm risk for mesopredators, our results show that the long-term risk of co-occurring with an apex predator can negatively influence mesopredator survival. In the long-term, the risk of encountering lions could be associated with the direct effects of predation, with an increased probability of antagonistic encounters 44 . Long-term risk can also be associated with non-consumptive effects of predation related to reduced foraging 45,46 , or reduced parental care 47 . We could not explicitly investigate whether long-term exposure to predation risk reduced cub survival through direct or indirect effects. The majority of cub mortality in our study system is a result of predation 48 , but cubs also experience non-predation mortality such as starvation or injury 22,48 , and indirect effects of predation risk could have reduced cub body condition, which might increase susceptibility to predation. In the absence of direct predation, the long-term risk of predation has been found to cause changes in the morphology 49 , behavior 50,51 , physiology 52,53 , and demography 54,55 of a variety of species. Our results build on the growing literature on long-term risk of predation to demonstrate how long-term predation risk can affect the demography of mesopredators.
Understanding spatial variation in survival can help inform wildlife conservation actions by focusing efforts on environmental factors that improve the survival of imperiled species. Specific for cheetahs in southern Africa, bush encroachment has caused the transition from open grasslands to closed habitats dominated by woody plants 56 . Bush encroachment can be caused by a number of factors including climate, fire, and herbivore distributions, but is predicted to increase based on future climate change models 57 . Based on our results, increased bush encroachment could be detrimental to cheetah populations that do not have adequate open areas for hunting. Thus, the persistence of this species in the southern portion of their range could be improved by prescribed burning or mechanical vegetation removal in order to maintain open habitats 58,59 . One limitation of our study was that we focused on only one cheetah population. Cheetahs experience variable conditions throughout their range, including differences in habitat composition, predator and prey communities, and conservation practices 60,61 . Therefore, further research is needed to better understand range-wide variability in spatial drivers of cheetah survival.
Our research shows how individual space use of mesopredators can scale-up and influence population-level processes, and illustrates the importance of understanding spatial drivers of survival on different temporal scales 62,63 . We propose that, in systems with apex predators and mesopredators, the survival of mesopredators in the short-term is driven by vegetative complexity likely associated with prey acquisition, whereas long-term survival depends on both top-down and bottom-up influences. Additionally, our results show that complex habitats might only be beneficial for mesopredators when they allow mesopredators to avoid apex predators, and effectively find and hunt prey, at the same time. Understanding how individual space use can influence populationlevel processes of mesopredators can offer insight into how communities with multiple predators are structured and can provide recommended conservation actions to ensure the future persistence of mesopredator species.  Carnivore monitoring. We monitored the cheetah and lion populations by subdividing the reserve into seven sections. Trained monitors usually drove the roads in each section at least once a week. In addition, monitors frequently followed-up on sightings reported by game rangers conducting game drives within the reserve. Cheetahs and lions can be individually recognized using their spot patterns, whisker spots, and scars, which allowed us to monitor the populations based on sightings alone 66 . We obtained an average of 40 ± 6 locations per individual cheetah during adult states and 27 ± 2 locations during cub states. When cheetahs or lions were observed, we recorded the location, behavior, and number of individuals present.

Cheetah habitat use.
We quantified coarse-scale cheetah habitat use by estimating lifetime home ranges for individual cheetahs. Although small-scale differences in habitat use might occur seasonally, cheetahs in our study area had stable home ranges across their lifetimes (Appendix S1). However, cheetahs will often shift home ranges when they become independent from their mothers, so for individuals that were included in the study as both cubs and adults, we estimated cub and adult home ranges separately. We estimated home ranges by calculating a utilization distribution (UD) using a fixed-kernel estimator and the plug-in method of bandwidth selection 67 . We only included cheetahs in our analyses with > 10 locations. For cubs died that before reaching the minimum number of locations, we used covariates associated with their mother's home range, or the home range of their surviving littermates. For each cheetah's home range, we extracted time-varying covariates of lion encounter risk, prey spatial density, and vegetation complexity (see below). To account for temporal differences in spatial drivers of survival, we extracted covariates within home ranges corresponding to each season, and also averaged covariates within home ranges across the lifetime of individual cheetahs. For cheetahs that were included in the analyses as both cubs and adults, we calculated separate "lifetime" covariate values for cub and adult periods separately. To identify the spatial scale most influential to survival, we extracted these covariates within the 50%, 75%, and 95% UD isopleths.
Lion encounter risk. We estimated lion encounter risk by analyzing the spatial distribution of lions in each season 68,69 . We collected sightings data on the location of lion prides, rather than individual lions, from 2000-2019. For each pride of lions in a given season, we calculated a utilization distribution (UD) using a fixedkernel estimator using the plug-in method of bandwidth selection 67 . To account for differences in pride size, we multiplied the UD for each pride by the average number of lions in that pride within a specific season 70 . To obtain a reserve-level measure of lion encounter risk, we added the individual pride UDs and rescaled the resulting values such that a value of 0 indicated no risk of encounter, and 1 indicated the highest risk of encounter.
Prey spatial density. We estimated spatial variation in prey density in the reserve by collecting distance sampling data on impala, and nyala during the dry season (April-September), and the wet season (October-March) from 2010-2015 48 . We limited our prey analyses to these species because they comprised 82% of cheetah kills in the study area 65 . We estimated prey abundance using hierarchical distance sampling models with spatial covariates on both the abundance and detection processes, and used our top model to extrapolate prey abundance over our entire study period 48  www.nature.com/scientificreports/ monitored cheetah was sighted or recovered dead as adults or cubs 48 . If a cheetah was removed from the reserve for management purposes, we censored that individual animal from analyses 48 . Because lion and cheetah density can vary greatly within a season, and because cubs can be born and become independent at any time during the year, we conducted our analysis on a monthly timescale to best reflect the conditions that might be driving survival 48 .
Because survival of cubs from the same litter might not independent, we first ran a Chi-square test of independent survival 72 to test this assumption, with the null hypothesis being that survival of cubs is independent. To run this analysis, we randomly selected half (n = 19) of the monitored litters and ran a survival model (see below for information on model structure) without any individual covariates, to estimate monthly cub survival. We used the results of this model to estimate the expected number of living cubs at independence, which we defined as 16 months post-birth 48 . When then repeated this procedure 50 times and ran a Chi-square test on the observed vs. expected number of survived cubs. We found that fates of cubs within the same litter were independent (X 2 = 34.3; p = 0.12), so for our subsequent survival models we treated each individual cheetah cub as an independent sample. Male cheetahs in the same coalition might also have non-independent fates, so we used the same Chi-square test of independent survival modeling framework with 50 replicates to test independence of males in coalitions (n = 11 coalitions). We found that that fates of males within the same coalition were independent (X 2 = 6.5; p = 0.28), so for our subsequent analysis we treated each individual male cheetah as an independent sample.
Model structure. Similar to previous research on cheetah survival in this system 48 , we analyzed cheetah survival using multi-state joint live-encounter dead-recovery models 73 using the rmark R package 74 . This model made use of our frequent re-sightings and mortality data, and allowed for survival estimation based on individuals with unknown fates. Additionally, because juvenile cheetahs stay with their mothers for variable amounts of time 75 , we could not incorporate a regular age structure into our models. Thus, we used a multi-state approach to estimate survival for both cubs and adults simultaneously 48 . We specified the two model states as cub (juvenile cheetahs dependent on their mother) and adult (cheetahs that were independent from their mother) and did not incorporate immigration or emigration because our population was a closed population.
Hypothesis testing. We previously determined that cheetah survival was best described using a structural model with resighting rate varying by year, and survival varying by season 48 . Therefore, we used the same structural model for these analyses to test for the effects of spatial covariates on cheetah survival. We used a two-stage approach to evaluate our hypotheses of interest. We first ran models to determine the spatial scale most influential to survival by running models with spatial covariates corresponding to the 50%, 75%, and 95% UD isopleths (Appendix S2). We considered spatial covariates on two temporal scales: short-term (seasonal), and long-term (spatial covariates within a home range averaged over an individual's lifetime). The spatial scale associated with the best-fit model at both the short-term and long-term scales was retained and used for the hypothesis-testing portion of our analysis (Appendix S2).
To test our hypotheses of interest, we developed 11 a priori models that included covariates of average lion encounter risk, average prey density, and average EVI, as well as additive and multiplicative models with the same covariates. Similar to the first stage of our analysis, we considered spatial covariates both at the short-term (seasonal) and long-term (spatial covariates within a home range averaged over an individual's lifetime) temporal scales. Because adults and cubs are known to have different survival rates 75 , we did not consider any models in which state was not included. We compared models separately for each temporal scale using Akaike's Information Criterion corrected for sample size (AIC c ; 76 ), considered models within 2 ΔAIC c of the top model to be competitive, and evaluated if covariates were informative by calculating 85% confidence intervals 77  www.nature.com/scientificreports/ Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creat iveco mmons .org/licen ses/by/4.0/.