Human activities can have complex effects on the antipredator behaviour of wildlife, and an understanding of the intricacies can provide important information for conservation management. In some cases, ‘human shields’ that attract wild ungulates may form around human settlements due to a lower density of large predators. However, human presence may also be associated with increased exposure to a range of anthropogenic threats, such as poaching, predation by dogs, and costly interactions with livestock and their herders. Here we compare the antipredator behaviour of two savannah antelopes, topi (Damaliscus lunatus) and Thomson’s gazelle (Eudorcas thomsonii), between the relatively undisturbed areas in the interior of the Maasai Mara National Reserve (Kenya) and the peripheral areas next to human settlements by the reserve boundary. We found that both antelope species were less vigilant in the more human-impacted peripheral areas, suggesting a reduction in the overall predation risk. However, both species also responded more strongly to conspecific (but not heterospecific) alarm calls in the human-impacted areas. We suggest that alarm calls in the human-impacted areas may be elicited by a more variable and unpredictable set of threats, many of which are anthropogenic, and that these require more careful assessment by the antelopes. Human presence can thus have opposite effects on different aspects of antipredator behaviour, emphasising the need to better understand how animals perceive threats in their environment and the consequences for population performance.

Selection of disturbed and undisturbed areas

Two open plains along the boundaries of the Mara with the Talek Enclave were selected as “disturbed areas”. These areas were exposed to livestock grazing, according to maps of livestock distribution made available by the Mara Predator Conservation Programme (www.marapredatorconservation.org), published evidence (Butt 2010, 2014), and personal observations. All playback trials and vigilance observations were conducted within 1.5 km of the heavily settled Talek town. In order to represent undisturbed areas, I chose three open plains, all > 15 km away from the border with the Talek Enclave, well beyond the maximum penetration range of pastoralists and their herds (~ 6 km; Pangle & Holekamp 2010, www.marapredatorconservation.org). This distance also ensured limited movements between disturbed and undisturbed areas for both study species, based on estimated home range diameter (~ 3 km for resident populations of Thomson’s gazelles; Walther 1972; 7 km for the topi; Bro-Jørgensen 2003). Individuals in the two areas were thus assumed to have experienced markedly different levels of human activity over their lifetime.

Baseline vigilance observations

I recorded vigilance behaviour during grazing bouts (duration = 5 minutes) on a digital video camera (Sony HDR-PJ810E) for ten different topi and gazelles (5 males, 5 females) in disturbed and undisturbed areas, respectively. Observations were limited to individuals in small herds (≤ 4 individuals), and at sites with grass ≤ 20 cm, in order to exclude the confounding effects on vigilance exerted by group size (Hunter & Skinner 1998; Creel, Schuette & Christianson 2014; no significant associations were found between vigilance and group size, topi: z = 0.258, n = 10, p =0.796; gazelle: z = 0.326, n = 10, p = 0.745) and grass height (with taller grass providing cover for ambush predators; Funston, Mills & Biggs 2001; Meise et al. 2018). The rate of head-lifts (at shoulder level and above) during grazing bouts was considered as an indicator of baseline vigilance levels.

Playback stimuli

Exemplars of alarm calls from six different individual gazelles and topi (three from each sex) were recorded during the course of a previous study (Meise et al. 2018). Each of these two species has a highly stereotypic alarm call, without obvious differences in acoustic structure between calls elicited by different predator species (Estes 1991; Meise et al. 2018). Recorded vocalizations of the ring-necked dove (Streptopelia capicola), comparable in volume to the alarm calls, were used as control sounds (Meise et al. 2018; Fig. 1). All playback stimuli were standardized to natural amplitudes measured in the field at a 35 m distance, using a handheld recorder (UNI-T, model UT352).

Playback experiments

We conducted a total of 144 playback trials. Each stimulus type (topi, gazelle, and dove) was presented to 11-13 different adult individuals per species in each area (disturbed and undisturbed), balancing trials between males and females (stimulus type per sex per area range = 5-6). Adult topi and gazelles were located while driving along the existing road network of the Mara. We selected the closest grazing animal in a relaxed and stationary herd as the focal individual, and played back the stimulus after a 20s period of uninterrupted grazing (Meise et al. 2018). Stimuli were broadcasted using a digital audio recorder (Tascam H2-P2) connected to a loudspeaker (Mipro MA707) positioned at ground level and hidden by the car silhouette (ungulates in the Mara are habituated to vehicles; Bro-Jørgensen & Pangle 2010). All trials were recorded on a digital video camera (Sony HDR-PJ810E). To ensure stimulus detection, playbacks were conducted at distances of 45-80 m (measured with a laser rangefinder, Bushnell Scout DX 1000 ARC), and at a wind speed of ≤ 3 m/s (measured with an anemometer, Proster Digital Lcd Meise et al. 2018). We also estimated grass height, distance from vegetation cover, and group size at each playback site (Appendix). In order to minimize the risk of pseudo-replication, individual exemplars, which were presented in a randomized order, were not played more than three times to the same species in the same area, and we noted down visible morphological traits (shape of the horns, forehead markings, presence of albinisms, visible scars; Walther, Mungall & Grau 1983; Bro-Jørgensen & Durant 2003) to ensure that each playback trial was conducted on individuals not previously targeted. For male antelopes, we focused on territorial individuals (recognizable by distinctive behaviours; Estes 1991; Bro-Jørgensen 2003), and played back stimuli on different days at distances larger than the estimated territory diameters (Walther 1972; Bro- Jørgensen 2003). Combined with the very large populations of the two study species in the Mara (Bhola et al. 2012), these strategies contributed to reduce the risk of pseudo-replication to a minimum. To avoid priming other individuals to the playback stimuli, we travelled >500 m between playback trials on the same day, beyond the active space of alarm calls (Meise et al. 2018). We also visited the same plains at a minimum of five-day intervals, in order to prevent habituation of the study population.

Behavioural analyses

Videos from playback trials were processed in BORIS (Behavioural Observation Research Interface Software; Friard & Gamba 2016) using frame-by-frame analysis (temporal window length = 0.04 s). A response was scored as occurring if focal individuals lifted their head at shoulder level within 10 seconds from the onset of the stimulus. Response intensity was measured as: (i) response latency (time to head- lifting); and (ii) response duration (interval between head-lifting, and first head-lowering to resume grazing; Meise et al. 2018).

Responses to controls

I found that topi and gazelles lifted their heads significantly more often after playback of alarm calls, than after playback of the control dove sound (topi: χ 2 = 32.237, p < 0.001; gazelle: χ 2 = 27.036, p < 0.001). Additionally, individuals of both study species were not more likely to lift their heads after playback of the control sound than during randomly-selected 10 second intervals from grazing bouts (topi in disturbed areas: χ 2 = 0.002, p = 0.961; topi in undisturbed areas: χ 2 = 0.000, p = 1.000; gazelle in disturbed areas: χ 2 = 0.552, p = 0.458; gazelle in undisturbed areas: χ 2 = 0.000, p = 1.000). These patterns support that the playback itself does not elicit responses and that heightened responsiveness is due to the information contained in the alarm calls.

Statistical analyses

All statistical analyses were conducted in R v. 3.5.2 (R Development Core Team, 2019). I compared response latency to alarm calls (re-scaled from seconds to milliseconds) between disturbed and undisturbed areas using the log-rank statistic of time-to-event Kaplan-Meyer survival analysis, in the packages survival and survminer (Therneau & Lumley 2014; Kassambara et al. 2017). In case of no response, data were considered as right-censored, and entered in the model with a value of 10,000 milliseconds (the maximum cut-off point I allowed for a response to occur). Response duration to conspecific and heterospecific alarm calls was compared between areas using unpaired Wilcoxon rank-signed sum tests (statistical significance was set at p ≤ 0.05). Wilcoxon rank-signed sum tests were also applied to compare differences in the number of head-lifts during grazing bouts according to the level of human disturbance. The relatively small sample size (23-25 playback trials per each combination of stimulus-receiver) prevented the use of multivariate analyses to account for the simultaneous effects of multiple predictors on the intensity of the observed responses. In order to determine whether the impact of human disturbance could be confused by other factors, we therefore tested for differences in socio-ecological variables at playback sites between disturbed and undisturbed areas (distance to focal individual, group size, grass height, distance to vegetation cover, and wind speed), using Wilcoxon rank-signed sum tests. We then explored the potential correlations between these variables and the two measures of response strength (latency and duration) using Kendall’s rank correlation tests.