The influence of scale and count method on prey selection of a large carnivore

Get Complete Project Material File(s) Now! »

Study area

We conducted our study in a 1,000-km2 area in the central region of the KNP, South Africa, centred on Satara rest camp (31.77° E, 24.39° S). The study area was mainly open-tree savanna with a moderate to sparse shrub layer and a dense grass layer. Dominant tree species were marula (Sclerocarya birrea) and knobthorn (Acacia nigrescens) with red grass (Themeda triandra) and stinking grass (Bothriocloa radicans) dominating the grass layer (Gertenbach 1983, Venter et al. 2003). The area comprised the northern component of wildebeest (Connochaetes taurinus) and zebra (Equus quagga) migrations, resulting in high densities of these species in the wet months (Gertenbach 1983). Buffalo (Syncerus caffer), kudu (Tragelaphus strepsiceros), giraffe (Giraffa camelopardalis), and waterbuck (Kobus ellipsiprymnus) occurred in large numbers, providing a high prey density and consequently a high lion density (Gertenbach 1983, Mills & Funston 2003).

We collared 5 female lions with GPS/global system for mobile communications (GSM) units (i.e., GPS/GSM, GPS units with mobile phone capabilities; Hawk105 units, African Wildlife Tracking cc, Pretoria, South Africa) between May 2005 and April 2007. Lions were captured using standard techniques by South African National Parks (SANParks) veterinarians (Smuts et al. 1977). Collars recorded GPS locations on 2 schedules: a) once per hour every night between 1800−0600 hours and during the day at 0900 hours, 1200 hours, and 1500 hours (16 fixes attempted) and b) once per hour over the full 24-hour period (24 fixes attempted). Collar schedules were therefore identical at night (1800−0600 hr) but differed during the day. Due to the high rate of GPS location recordings, collars had a reduced battery life and we replaced some during the study. We deployed 9 collars during the study, 4 of which attempted fixes 24 hours/day and 5 of which attempted fixes 16 hours/day. Lions in the KNP are active predominantly at night so most kills are nocturnal, with daylight hours spent resting (Mills & Biggs 1993). The combination of datasets using both collar schedules incorporated balanced sets of hourly readings at night (when lions are most active) across the entire dataset, but missed some data points during the day. The unbalanced daytime schedules should not have created any significant error due to inactivity of lions during daytime. We nevertheless performed all analyses with both schedules separately as well as with the entire pooled dataset. We downloaded data remotely via the GSM network when each collared individual entered an area of GSM coverage. We calculated distances between successive locations using the Animal Movements Extension (Hooge et al. 1999) in ArcView® 3.2 and treated missed GPS locations as stationary locations. We defined a GPS aggregation cluster (hereafter a cluster) as ≥2 consecutive recorded fixes with each consecutive pair of fixes <100 m apart (Including both day and night data; See Appendix 1). If a GPS location was not recorded and distance to the following location was >100 m, a GPS cluster was not created (even though measurement of the intervening missing points may have resulted in definition of a cluster being met). We uploaded cluster coordinates onto a hand held GPS unit and investigated clusters on foot. We investigated an area of approximately 25-m radius around GPS points (See Appendix 1) that encompassed all GPS points for that cluster. Therefore, we included any GPS point outside the 25-m radius of the first GPS point as the centre of a new search radius so that we investigated all GPS points at the cluster (See Appendix 1). We attempted to investigate as many clusters as possible, however to maximize number of clusters investigated, we investigated clusters occurring near each other first. By conducting investigations based on proximity of many clusters, we may have biased cluster investigation to areas readily accessible by roads. We assessed this possible bias by comparing distance that clusters with or without a kill occurred from a road, as well as distance that checked and unchecked clusters occurred from a road using Wilcoxon rank sum tests. We identified predation events at clusters from presence of prey stomach contents, teeth, bones, horns, or hair and determined prey species, age, and sex when possible (age – horns and teeth development; sex – horns and canine presence in zebra). The GSM coverage in the study area was not uniform and areas existed without coverage; therefore, collared females re-entered areas of coverage sporadically, resulting in an irregular pattern of cluster investigation over time (i.e., time between cluster occurrence and investigation depended on when the female moved into GSM coverage).
We separated field observations into 2 investigative periods; initially (Jun 2005 – Feb 2006) we only noted number of days between cluster occurrence and cluster investigation for kills, but subsequently (Mar 2006 – Apr 2007) we recorded number of days between cluster occurrence and cluster investigation for clusters with and without kills. During this second period, we investigated clusters between 0−671 days ( x = 54.2 days, median = 6 days) after cluster occurrence. Using Wilcoxon rank sum tests, we investigated our ability to identify species, age, and sex of the kill as time between cluster occurrence and investigation increased.

Statistical Methods

We measured 8 predictor variables for each GPS cluster. 1) Hours: length of time (hr) lions spent at a cluster from the first point of the cluster until they left the cluster for the last time. Hours was an indication of total time spent at on a carcass. 2) Return: a categorical variable describing a return visit to a cluster within 12 movement steps of leaving that cluster indicating the possibility that lions returned to carcass sites following initial movements away from the carcass site. 3) Ratio_24: ratio of distance moved during the 24 hours prior to the cluster beginning against distance moved during the 24 hours following termination of the cluster, where cluster duration included all return events. We based the variable ratio_24 on the premise that lions predominantly search for prey when hungry and rest following a kill, therefore resulting in higher ratios for successful hunts. 4) Dist_1: distance moved by lions during the first GPS interval of a cluster (i.e., between the first and second recorded GPS coordinate); a short dist_1 indicates that the female remained close to the carcass at the beginning of a cluster and did not move around the cluster site. 5) Dist_2: distance moved by lions during the second GPS interval of a cluster (i.e., between the second and third recorded GPS coordinate); a short dist_2 indicates that the female remained at the carcass, as apposed to moving away from the carcass while remaining at the cluster. 6) Drain: distance from the cluster to the nearest drainage line (classes 1-4, S.MacFadyen, KNP Scientific Services GIS department), which we calculated using ArcMap 9.0 and treated as a continuous variable, giving a measure of distance to available cover for hunting lions. 7) Road: distance from the cluster to the nearest road (S. MacFadyen, KNP Scientific Services GIS department), which we calculated as for drain, giving a measure of bias of investigating clusters near roads. 8) Dark: a 5-valued categorical variable that accounted for the combined effect of the sun and moon at the start of the cluster. The 5 categories were: 1 = daytime clusters, 2 = twilight clusters, 3 = night-time clusters with the moon up and full, 4 = night-time clusters with the moon up and in the first or third quarter, and 5 = night-time clusters with the moon up in the new moon phase or night-time clusters when the moon was down, as lions have been shown to have a greater kill success on dark nights We used generalized linear models (GLMs) to investigate variables related to probability of a binary response (kill = 1, no kill = 0) occurring at a GPS cluster (Hosmer & Lemeshow 2000). We identified 2 possible time delays between cluster occurrence and cluster investigation that suggested declines in our ability to identify kills at cluster sights (see results). Therefore, we developed models based on data from clusters investigated during the first 4 weeks and first 16 weeks following a cluster occurrence. We used a forward stepwise α-to-enter approach (Quinn & Keough 2002) with a cut-off of α = 0.05 as the criteria for entering parameters into the model. We tested parameters for collinearity and found that hours and return were correlated (c = 0.52), and subsequent investigation revealed that return was not an important variable if hours was already included in the model. We assessed model discrimination using the area under the curve (AUC) based on the receiver operator characteristic (ROC) curve, which is a plot of the proportion of true positives (i.e., sensitivity of discrimination) as a function of the proportion of false positives (i.e., one minus the specificity of discrimination). With this definition the AUC score represents the percentage of time that a true positive will have a higher probability of being a kill than a true negative when both are drawn at random (Zweig & Campbell 1993, Fielding & Bell 1997). We assessed the relative individual predictors’ influence on the response variable using hierarchal partitioning analysis (MacNally 2000).

READ  AN EVALUATION OF LIME EFFECTS ON TEMPORAL CHANGES IN SOIL ACIDITY PROPERTIES AND MAIZE GRAIN YIELDS

Chapter 1: Summary  
Chapter 2: Methods for Locating African Lion Kills using Global Positioning System Movement Data 
-Abstract
-Introduction
-Methods and Materials
-Study area
-GPS collars and cluster investigation
-Statistical Methods
-Results
-Discussion
-Management implications
-Tables
-Figures
Chapter 3: Feasibility of using proximity tags to locate female lion (Panthera leo) kills  
-Abstract
-Introduction
-Materials and methods
-Results
-Discussion
-Figures
Chapter 4: Combining GPS locations and scat analysis to estimate the diet of a large carnivore  
-Abstract
-Introduction
-Materials and Methods
-Study area
-Data acquisition and datasets
-Scat analysis procedure
-Data analysis
-Results
-Discussion
-The additive approach
-Diet estimates from scats
-Conclusions
-Tables
-Figures
Chapter 5: The influence of scale and count method on prey selection of a large carnivore
-Abstract
-Introduction
-Methods
-Study area
-Estimating prey availability
-Estimating lion diets and range
-Prey selection
-Results
-Prey availability
-Diet estimates
-Prey selection
-Discussion
-Conclusion
-Tables
-Figures
Chapter 6: Fine scale determinants of lion predation on buffalo  
-Abstract
-Introduction
-Methods and materials
-Study area
-Datasets
-Predation risk
-Spatial analysis
-Temporal analysis
-Buffalo demographic parameters
-Results
-Predation risk
-Spatial distribution of kills
-Temporal impact
-Demographic selection
-Discussion
-Fine scale habitat mediated spatial vulnerability to predation
-Climatic influence of predation and the impact of water availability
-Conclusion
-Tables
-Figures
Chapter 7: Quantification of predator diets for conservation  
-The GPS cluster approach to quantifying predator diets
-The use of robust large carnivore datasets in a shrinking world
-Use of the GPS cluster technique
References

GET THE COMPLETE PROJECT

Related Posts