Data were collected from the 18th of April 2016 until the 28th of June 2017 (see Chapter 2). The mean (± SD) number of observation days per month was 14 ± 5. An area was sampled by driving randomly selected routes, based on accessibility due to weather
and previous reports of elephant sightings by guides, and by communicating with field guides about elephant presence and then searching for those elephants (see Chapter 2).
Thirty-three lodges were spread across the reserve and conducted game drives in the morning from sunrise (approximately from 5:30 am) until approximately 11 am, and in the afternoon and evening between approximately 3:30 – 8 pm.
For the behaviour analysis, 26 individuals (14 males, 12 females) were identified based on distinguishing features. Herds included in travel direction analyses were herds which were encountered throughout data collection but were unknown and not
individually identified. Upon spotting elephants, the researcher aimed to keep 30 m distance from the nearest elephant. If the animal was spotted at <30 m distance, the vehicle was slowly reversed to 30 m from the nearest elephant before the engine was switched off. When animal/s moved parallel to the road used by the researcher without displaying signs of distress (such as vigilance, body posture changes such as ‘ears out’
threats, or moving away whilst repeatedly looking back at the vehicle), the researcher followed at a distance before switching the engine off again. A bull group was defined as several bulls within a 500 m radius of each other, whilst a mixed group was defined as an adult bull within 200 m of a cow-calf group. All distances were approximated visually.
Data were collected on a Lenovo TAB 2 A8-50F tablet using the Prim8 app
(McDonald & Johnson, 2014). The researcher classed elephants as juvenile or adult based on size (Moss, 1996; Poole & Granli, 2009). Adult females had mammary glands and an angled forehead, whilst adult males had a rounded forehead, wider skulls, and could be twice the size of adult females. Juveniles were smaller than adult females, moving and foraging independently, and had tusks of approximately ten centimetres in length. Once a sighting was made from the road, the researcher randomly selected an elephant of any age, which was in full view and close to the research vehicle, as the focal elephant to
observe, using continuous focal animal sampling (Altmann, 1974) for five minutes. The researcher noted identity, if known, along with additional factors such as sex, age, and season (Table 4.1). If identity was not known at the time the observation was carried out, photographs of the individual were taken and later compared against a database for identification.
Table 4.1. Factors recorded for five-minute continuous behavioural observations of African elephants, Loxodonta africana, carried out in Madikwe Game Reserve, South Africa.
Factor Levels Description
Sex Female, male Sex of focal individual
Age Adult, juvenile Age of focal individual Herd type Lone male, bull group,
cow-calf group, mixed group
Type of herd in which focal individual was observed
Type of habitat the focal individual was observed in
Season Dry, wet Season during which observation took
place
Vehicle 0 – 3 Number of GD vehicles present during
the focal observation
ashrub= various bushes and trees in observed area but not obscuring observation noticeably; dense shrub= shrub and trees in observed area, growing so densely that observation only possible at close distance and dense enough to cover view of large areas of the body of the focal animal; open grassland= observation area vastly open with only occasional bushes or trees; waterhole= water accumulated either naturally or pumped artificially with enough water for one or more elephants to drink
The researcher noted the direction of travel of the whole herd by visually comparing herd location at the start and end of the focal observation and inferring direction of travel. If the centre of the herd (assessed visually through assessment of all visible elephants within the surroundings) increased its’ distance from the observer or, if present, the closest GD ≥10 m (without simultaneously approaching another GD), we classed it as ‘retreat’, otherwise, when herds did not increase their distance by more than 10 m or approached one GD whilst retreating from another we classed it as ‘stay’. Five-minute focal animal observations (Altmann, 1974) were conducted for as long as a herd
stayed within the vicinity of the observer, or until all individuals in a herd had been
observed once. We only recorded one herd movement observation per encounter, during the first five minutes after a herd was encountered or after a GD(s) arrived, as a measure of immediate reaction of herd movement to the potential stressor. If a herd had already been observed before a GD arrived, the herd movement observation used for analysis was the one during which the GD arrived.
According to previously published ethograms as well as following communication with an elephant expert, Dr Y. Pretorius, behaviours were categorised as stress-related (running, trunk touching own body or face, trunk twirling), vigilance (smelling, touching trunk to mouth, observing surroundings) or aggressive (charging, displacing another, redirecting aggression, having ears out or flapping, slapping, tusking, head shaking, pushing an object, standing tall, trunk swinging or aggressive sparring; see Chapter 1 and 2 for more detail). Because several aggressive behaviours could be directed at either humans or conspecifics, we made note of the direction of the recipient of the threat, and excluded all aggression explicitly directed at the researcher (n=5 occasions, where n=3 were ‘ears out’ and n=2 were ‘head shake’) in the analysis of conspecific-directed aggression.
Season was defined as wet or dry based on mean monthly rainfall measured at four stations within Madikwe by the South African Weather Service. Mean total rainfall during the study period was 189.69 mm. Wet season was defined as the period in which 95% of precipitation for the study year fell (Loarie et al., 2009a, b) and therefore wet season lasted from October 2016 – February 2017 and dry season lasted from April 2016 – September 2016 and March 2017 – June 2017. North West Parks Board, South Africa, provided the total number of tourists visiting Madikwe each month. Tourist number across seasons can be seen in Figure 2.3 in Chapter 2. As each GD carries a maximum of
ten tourists, higher tourist number overall generally leads to more GDs being used to accommodate all tourists during GD times and therefore directly relates to a higher number of vehicles on the road (unless tourists chose to stay at the lodges instead of conducting game drives, which is very unusual and would only represent a small number of guests (P. Hatting, C. Cotton, K. Potgieter, pers. comm.)).
4.2.3 Data analysis
Only focal observations where the animal was visible for >4 minutes 30 seconds (s) were retained for analysis. In order to assess the effect of monthly tourist numbers on elephant behaviour in the reserve, we analysed only those observations of individuals in herds with no vehicles present besides the research vehicle. This was to avoid potentially confounding effects on behaviour related to immediate presence of tourists and because the sample size of behavioural observations on individually identified elephants with and without game drive vehicles present was insufficient for analysis. We included individuals that had a minimum number of n=2 observations over the complete time period of the study.
For herd movement direction analysis, we included observations of unidentified herds with and without GDs present. Where a single GD arrived or left within the five-minute observation but was present for less than 60s, the observation was excluded from analysis. If GDs were present for more than 60s, the herd movement was considered to be in response to the number of GDs present for that time. This means that if one GD was present from the beginning, but a second GD arrived and stayed for over 60s, herd
movement was assumed to be in response to two GDs present. If a second GD arrived but left in under 60s, the whole observation was considered to be in response to one GD.
Although continuous focal animal observations were carried out, data were zero inflated and therefore we decided to score specific behaviours as occurring or not within each five-minute observation. This presents a conservative estimate, as even
observations during which several instances of one specific behaviour occurred, were still only recorded as occurring, rather than adding weight to it. We analysed data using R v.3.4.1 (R Core Team, 2000). We scored each behaviour as occurring or not, and elephant herd travel as away or stay, forming binary response variables. First, we assessed factors to rule out collinearity using variance of inflation factor analysis (see Chapter 2; Fox &
Monette, 1992), using a cut-off value of four, where all values were below two. We specified General Linear Mixed Effects Models (package lme4, Bates et al., 2014; Bolker, 2009) including all possible two-way interactions to analyse the effect of tourist pressure on stress-related and vigilance behaviour:
𝑔𝑙𝑚𝑒𝑟 (𝑓𝑜𝑟𝑚𝑢𝑙𝑎
= 𝐵𝑒ℎ𝑎𝑣𝑖𝑜𝑢𝑟𝑎𝑙 𝑐𝑎𝑡𝑒𝑔𝑜𝑟𝑦 ~ 𝑇𝑜𝑢𝑟𝑖𝑠𝑡 ∗ 𝐻𝑒𝑟𝑑 𝑡𝑦𝑝𝑒 + 𝑇𝑜𝑢𝑟𝑖𝑠𝑡 ∗ 𝑆𝑒𝑥 + 𝑇𝑜𝑢𝑟𝑖𝑠𝑡 ∗ 𝐴𝑔𝑒 + 𝑇𝑜𝑢𝑟𝑖𝑠𝑡 ∗ 𝐻𝑎𝑏𝑖𝑡𝑎𝑡 𝑡𝑦𝑝𝑒 + 𝑇𝑜𝑢𝑟𝑖𝑠𝑡 ∗ 𝑆𝑒𝑎𝑠𝑜𝑛 + 𝐻𝑒𝑟𝑑 𝑡𝑦𝑝𝑒 ∗ 𝐻𝑎𝑏𝑖𝑡𝑎𝑡 𝑡𝑦𝑝𝑒 + 𝑆𝑒𝑥 ∗ 𝐴𝑔𝑒 + 𝑆𝑒𝑥 ∗ 𝐻𝑎𝑏𝑖𝑡𝑎𝑡 𝑡𝑦𝑝𝑒 + 𝑆𝑒𝑥 ∗ 𝐻𝑒𝑟𝑑 𝑠𝑖𝑧𝑒 + 𝑆𝑒𝑥 ∗ 𝑆𝑒𝑎𝑠𝑜𝑛 + 𝐴𝑔𝑒 ∗ 𝑆𝑒𝑎𝑠𝑜𝑛 + 𝐻𝑎𝑏𝑖𝑡𝑎𝑡 𝑡𝑦𝑝𝑒
∗ 𝐻𝑒𝑟𝑑 𝑠𝑖𝑧𝑒 + 𝐻𝑎𝑏𝑖𝑡𝑎𝑡 𝑡𝑦𝑝𝑒 ∗ 𝑆𝑒𝑎𝑠𝑜𝑛 + 𝐻𝑒𝑟𝑑 𝑠𝑖𝑧𝑒 ∗ 𝑆𝑒𝑎𝑠𝑜𝑛 + (1|𝐼𝐷), 𝑓𝑎𝑚𝑖𝑙𝑦 = 𝑏𝑖𝑛𝑜𝑚𝑖𝑎𝑙, 𝑑𝑎𝑡𝑎 = 𝐷𝑎𝑡𝑎)
Due to insufficient data to support a more complex model, we ran the following global model to assess all possible models including two-way interactions for conspecific-directed aggression:
𝑔𝑙𝑚𝑒𝑟 (𝑓𝑜𝑟𝑚𝑢𝑙𝑎
= 𝐶𝑜𝑛𝑠𝑝𝑒𝑐𝑖𝑓𝑖𝑐 𝑑𝑖𝑟𝑒𝑐𝑡𝑒𝑑 𝑎𝑔𝑔𝑟𝑒𝑠𝑠𝑖𝑜𝑛 ~ 𝑇𝑜𝑢𝑟𝑖𝑠𝑡 ∗ 𝑆𝑒𝑥 + 𝑇𝑜𝑢𝑟𝑖𝑠𝑡 ∗ 𝐴𝑔𝑒 + 𝑇𝑜𝑢𝑟𝑖𝑠𝑡 ∗ 𝑆𝑒𝑎𝑠𝑜𝑛 + 𝑆𝑒𝑥 ∗ 𝐴𝑔𝑒 + 𝑆𝑒𝑥
∗ 𝐻𝑒𝑟𝑑 𝑠𝑖𝑧𝑒 + 𝑆𝑒𝑥 ∗ 𝑆𝑒𝑎𝑠𝑜𝑛 + 𝐻𝑒𝑟𝑑 𝑠𝑖𝑧𝑒 ∗ 𝑆𝑒𝑎𝑠𝑜𝑛 + 𝐻𝑒𝑟𝑑 𝑡𝑦𝑝𝑒 + 𝐻𝑎𝑏𝑖𝑡𝑎𝑡 𝑡𝑦𝑝𝑒 + (1|𝐼𝐷), 𝑓𝑎𝑚𝑖𝑙𝑦
= 𝑏𝑖𝑛𝑜𝑚𝑖𝑎𝑙, 𝑑𝑎𝑡𝑎 = 𝐷𝑎𝑡𝑎)
We scaled and centred the tourist pressure and herd size variables (see Chapter 2). Using the ‘dredge’ command in the MuMIn package (Barton, 2018), we compared the conditional Akaike Information Criterion (AICc) values (Burnham & Anderson, 2002; see Chapter 2). This function produced a table with all possible candidate models and their associated AICc values from which we identified the top model. Where several models were within 2 AICc of each other, we further followed the criteria detailed in Leroux (2019; see Chapter 2). However, as the top model gives no indication on how weak or strong each individual effect in this model affects the response variable (Mundry, 2011;
Symonds & Moussalli, 2011; Harrison et al., 2018), we further analysed significance of each fixed effect in the top model with a type II ANOVA (Langsrud, 2003). Where categorical fixed effects were significant, we assessed differences between the levels using a Tukey post-hoc test in the multcomp package (Hothorn et al., 2008), checking that 95% confidence intervals did not cross zero.
For the direction of travel dataset, we excluded the open grassland habitat type from analysis as only n=5 observations had one GD present, resulting in poor model fit.
The following a priori Generalised Linear Model was used:
𝑔𝑙𝑚 (𝑓𝑜𝑟𝑚𝑢𝑙𝑎
= 𝑇𝑟𝑎𝑣𝑒𝑙 ~ 𝐻𝑒𝑟𝑑 𝑡𝑦𝑝𝑒 ∗ 𝐺𝐷 𝑛𝑢𝑚𝑏𝑒𝑟 + 𝐻𝑎𝑏𝑖𝑡𝑎𝑡 𝑡𝑦𝑝𝑒
∗ 𝐺𝐷 𝑛𝑢𝑚𝑏𝑒𝑟 + 𝑆𝑒𝑎𝑠𝑜𝑛 ∗ 𝐺𝐷 𝑛𝑢𝑚𝑏𝑒𝑟 + 𝐻𝑒𝑟𝑑 𝑠𝑖𝑧𝑒, 𝑓𝑎𝑚𝑖𝑙𝑦
= 𝑏𝑖𝑛𝑜𝑚𝑖𝑎𝑙, 𝑑𝑎𝑡𝑎 = 𝐷𝑎𝑡𝑎)
We scaled and centred GD number and herd size. To account for non-independence in the data due to potential pseudoreplication, we performed 1000 iterations of bootstrapping, using the package boot (Canty & Ripley, 2018) to obtain bootstrapped 95% confidence intervals. We considered fixed effects significant if confidence intervals did not cross zero. We plotted all graphs using the effects- (Fox, 2003) and ggplot2 (Wickham, 2016) packages (see Chapter 2).
4.3 Results
A total of 156 observations of known individuals were collected (mean ± SD = 6 ± 6 per individual, Table 4.2). These observations were from ten adult males (18 observations as lone males, eight in bull groups, three in mixed groups), nine adult females (56
observations in cow-calf groups, 37 in mixed groups), three juvenile females (eight observations in cow-calf groups, two in mixed groups), and four juvenile males (16
observations in cow-calf groups, eight in mixed groups). We did not observe the following behaviours during our behavioural observations of individuals: slap, pushing object, standing tall, aggressive sparring (see Chapter 2). Removal of individuals with a small sample size or nesting ID in social group did not change the effect of tourist pressure reported below. We recorded travel direction of herds during 479 observations (81 bull groups, 141 cow-calf groups, 100 mixed groups, and 157 lone males).
Table 4.2. Information on number of observations collected of individually identified African elephants, Loxodonta africana, in Madikwe Game Reserve, South Africa. Sex, age, ID of the individual, as well as which herd it belonged to is presented alongside number of observations collected during the wet and dry season and the range of herd sizes the individual was observed in.