As detailed in Section 2.2ii, three 62-day camera trap surveys were undertaken in different parts of the farm block (see Figure 2.10) which generated 22,675 photographs in total. All pictures of domestic livestock (cattle, donkeys, horses and goats), people, vehicles and camera misfires were discarded so that only images of wildlife remained. This amounted to a total of 8,789 photographs. All images were entered into Camera Base v. 1.3 (Tobler 2007) and the date, time, camera station and species pertaining to each picture recorded. For analysis of prey species abundance and occupancy the data were filtered to exclude carnivores and small mammals (weight < 5kg) such as hares and small rodents. This left five naturally occurring ungulate species – greater kudu; common duiker; steenbok; warthog and one large rodent – the porcupine (Hystrix africaeaustralis). The data were further filtered to exclude photographs of the same species at the same station within a period of 60 minutes (following Bowkett et al. 2008; Tobler et al. 2009) in order to ensure that photographic events were independent. A period of one hour was selected as the criterion for independence in this instance as browsing and grazing species, such as kudu, duiker, steenbok and warthog, sometimes remained in front of the camera for long periods. The capture frequency for a species was calculated as the number of trap days/independent photograph (RAI1:Carbone et al. 2001), and the number of independent photos/100 trap days (RAI2: O'Brien et al. 2003). A camera trap day was defined as a period of 24-h when at least one camera at a station was operational. If the batteries had failed or the camera had malfunctioned then it was classed as a „missed day‟. Capture
frequencies and detection histories were calculated by Camera Base. Statistical analysis was undertaken using Minitab v. 15.0 (2007) and where samples were not normally distributed nonparametric statistics were used (Siegel & Castellan Jr. 1988).
There is some concern that the abundance of group living species may be underestimated when doing line transect surveys (O'Brien et al. 2003), and Treves et al. (2010) found that there was a strong correlation between detectability and both abundance and distribution for gregarious species in their camera trap data. Following their lead therefore, photographs were studied to obtain both mean and maximum group sizes for species that had been captured in all surveys. It should be noted however that kudu, being large bodied animals, are likely to have had their group sizes underestimated by this method due to the limited number of animals that will fit into the frame of a photograph. Spearman‟s rank correlation coefficient was used to test for association between indices of abundance and spatial distribution and for the effect of group size on detectability.
47 For analysis in PRESENCE the three datasets of 62 days were each divided into 10 six-day sampling periods in order to maximise detection probability. As a result this meant that two days worth of data had to be discarded for each survey. If a camera station had malfunctioned for more than one day during any of the sampling periods, and there had been no detection of the species on other days in that sampling period, the data point for that period was entered as a missing observation. Visual analysis of the data indicated that there was heterogeneity of detection between camera stations in all surveys for all species; therefore it was deemed appropriate to employ the RN model. The primary goal of the RN model is to estimate two parameters r, the inherent detection probability of the species and lambda
, the mean abundance of animals at all sites. Occupancy
is not estimated directly, but is derived from
as
1e
. In order to be able to use the figure given for lambda to estimate density, the size of the area surveyed in each case had to be established. The criteria used to estimate survey area can differ depending on the analysis being carried out. In this instance it was calculated using the mean distance between camera stations in each survey. A buffer of that distance was created around the camera station grid in ArcGIS v. 9.3 (ESRI Inc. 2008) and the size of the whole area within the buffer then calculated.Differences in
were tested for using three covariates: camera trap survey, vegetation type (hardveld or sandveld) and land use (cattle or game farms). Covariates used to test for differences in r were those of season (wet or dry) and of site-specific (camera station) habitat. These were classed as: thick bush, medium bush, mixed woodland, water point and pan. Kudu were modelled twice, first using all three surveys and second using only data from the cattle farms in which they are free-ranging. Models were ranked in PRESENCE using AkaikeInformation Criteria (AIC) (Burnham & Anderson 2002), which measures the weight of evidence for model choice amongst a set of models, the lowest value of AIC indicating the most
parsimonious model.
Ungulates vary in size across their ranges and field guides are equally variable in their weight estimations. For example Kingdon (1997) gives a weight range for a female kudu of 120-215 kg while Mills & Hes (1997) merely give a weight of up to 210 kg. Hayward et al. (2007) used weights from Stuart and Stuart (2000) and followed George Schaller‟s (1972) method of using ¾ of mean female weight to calculate biomass in order to allow for predation on juveniles and sub-adults. In order to facilitate the use of their equations the same protocols were followed here except that weights were taken from the third edition of Stuart and Stuart (2006) (Table 4.1).
48
Table 4.1 Mass of prey species used to calculate prey abundance
Species ¾ mean mass of adult female (kg)
Kudu (Tragelaphus strepsiceros) 135 Duiker (Sylvicapra grimmia) 16 Steenbok (Raphicerus campestris) 8 Warthog (Phacochoerus africanus) 45 Porcupine (Hystrix africaeaustralis)* 13
* Porcupines are not sexually dimorphic so ¾ mean weight of the species as described by Stuart and Stuart (2006) was used.