Sampling procedure

Each of the 6 study sites (see Chapter 2, same abbreviations) was visited once per season between July ’07 and July ’08 (winter = July/Aug 2007; spring = Oct/Nov 2007; summer = Jan/Feb 2008; autumn = April/May 2008), except for Shamwari, which was not visited during spring. For each site and season, visual body condition scores (as described in Chapter 2) were recorded (see Appendix 1 for sample sizes), and 5 fresh (<2 days old) faecal samples collected and placed in labelled brown paper bags. Additionally, fresh plant clippings were obtained from the AMC and NCA seasonally, and from all other sites for summer and winter. Plant clippings were taken from at least three plants per species per season at different locations for each reserve to account for interplant variation. The plants collected formed part of the known elephant diet for each site, obtained from previous dietary studies for Asante Sana (Minnie 2006), Addo (Landman et al. 2007), Nyathi (Davis 2005), Shamwari (Roux 2006) and Kariega (Wolmarans 2006), and from direct observations by reserve managers for Blaauwbosch. All the principal dietary items (PDI’s; those species contributing 3% or more to the diet) identified for AMC and NCA were collected, but only overlapping dietary items were collected from the other reserves. These were plant species that formed part of the known elephant diet for at least three of the sampled sites, and included Azima tetracantha, Schotia afra, Cynodon dactylon and Acacia karroo.

All samples collected were placed in labelled brown paper bags and oven dried at 60°C for a minimum of 3 days. All plants of each species per season were added together to create one sample per species per season. These samples were then analysed at the Grootfontein Agricultural Development Institute, Middleburg, Eastern Cape. Samples were ground to 2 mm mesh size, homogenised and subsets of each sample were then analyses for phosphorus (P), nitrogen (N) and neutral detergent fibre (NDF). Nitrogen was analysed using the Kjeldahl method as set out by the Association of Official Agricultural Chemists (AOAC 1995), whilst NDF was determined by the Ankom Technology

method filter bag technique (www.ankom.com) and phosphorus by the dry ashing method (ultraviolet-visible spectrophotometry, AOAC 1995). Nitrogen values were subsequently converted to crude protein (%N x 6.25) by standard practice (Robbins 1983). All values are expressed as percentage dry weight. In order to test the precision of the sample analysis, 5 blind replica samples were submitted without the knowledge of those doing the analysis.

In addition to the body condition and dietary quality data collected, rainfall (both long term and for the study period and the preceeding six month time), reserve size, history of elephant occupation and elephant density data were collected at each site. Where rainfall data were not recorded on site, the nearest town/weather station rainfall data were used. Study period rainfall was recorded as rainfall over the six month prior the commencement of the study to six months before the end of the study to allow a more accurate representation of vegetative responses in productivity. Unless stated otherwise, where rainfall during the study period is mentioned, it refers to this recorded rainfall.

4.3.4 Data analysis

4.3.4.1 Body Condition

As body conditions in the initial AMC analysis (Chapter 3) were found to differ significantly between life stages, a heterogeneity chi-squared test (Zar 1999) was performed for each season to determine whether energy stressed life stages (i.e. old age, lactating and weaning) could be pooled for subsequent body condition comparisons with other sites. As body conditions in the initial AMC analysis also were found to differ significantly between seasons, these data, and those from other sites, were kept separate in subsequent analyses.

A four-way log linear analysis (Statistica 8.0; StatSoft Inc. 2007) was used to determine whether there was a significant difference in body condition between the various sites and seasons. Factors were site, season, energy class (non- energy stressed and energy stressed) and body condition category. Data were

then explored graphically using Box-and-Whiskers plots (Statistica 8.0; Statsoft Inc. 2007), with whiskers representing the 90% confidence interval (CI).

A proportional odds ordinal logistic regression model was used to investigate which site specific factors (i.e. rainfall, density, reserve size and occupational history) would best explain the body conditions observed. Sites were weighted by the relative frequency of occurrence of each body condition class observed. Backward stepwise selection was used to select the most parsimonious model. This process involved first fitting the model with all variables, removing non- significant variables, and re-fitting the model using only the significant variables (elephant density and rainfall over the course of the study period). Model

goodness-of-fit was assessed using Nagelkerke’s R2 (Nagelkerke 1991). The

proportional odds logistic regression model was fitted using the package “Design” (Harrell 2008) on the statistical platform R (R.D.C.T. 2008).

4.3.4.2 Dietary quality

A paired t-test (Statistica 8.0; StatSoft Inc. 2007) of the blind replicate samples was used to determine the precision of the dietary quality analyses obtained from the Grootfontein Agricultural Institute.

The three components (protein, phosphorus and NDF) of the faecal dietary quality data were then analysed using factorial ANOVAs (Statistica 8.0; StatSoft Inc. 2007) to investigate whether any differences existed between the seasons and sites. Significant results were further analysed using Tukey’s post hoc tests in order to explore the significant interactions. Subsequently, a proportional odds ordinal logistic regression model was used to investigate which of the dietary quality measures best explained the ordering and observed body conditions. The best model was again selected using the backward stepwise method (as

explained in 4.3.4.1) and goodness of fit was assessed using Nagelkerke’s R2

using the package “Design” (Harrell 2008) on the statistical platform R (R.D.C.T. 2008).

The various diet quality components of overlapping dietary plant species were analysed using paired t-tests (Statistica 8.0; StatSoft Inc. 2007) to investigate any differences between seasons. The dietary quality values obtained were also explored graphically to investigate any differences between sites. Two-way ANOVAs (Statistica 8.0; StatSoft Inc. 2007) were used to determine whether any differences in quality between the PDI’s of the AMC and NCA could be detected.

Where data were not normally distributed, log10-transformations were done to

allow conformation with the requirements of parametric statistics. Where data were log transformed for statistical analyses, the data were back transformed to report results in the original units. Except where stated otherwise, tests were assessed at the 95% probability level (Zar 1999).

4.4 Results