Chapter 6. Diet selection by elephants: the influence of short-term rate of protein intake
6.2. Method
6.2.5. Model fitting and hypothesis testing
A maximum likelihood estimate of the parameter for the short-term rate of protein intake was obtained by fitting the multinominal logit form of the discrete choice model, with choice of primary food as the dependent variable and short-term rate of protein intake as the independent variable, to the data using Systat 9 (SPSS, 1998). The significance of the parameter estimate was tested using a t-test, and the change in the odds (odds ratio) of selecting the primary food type, given a 1 gs-1 increase in the short- term rate of protein intake, was calculated using the following equation:
(
)
(
x)
x x x p p p p − − = + + 1 1 1 1 ωwhere ω is the odds ratio, and px is the probability of being chosen as the primary food type when the
short-term rate of protein intake is x gs-1.
6.3.
ResultsA total of 72 observations (62 from bulls and 10 from cows) were made across 24 land units (Figure 6.1). The temporal distribution of the observations were as follows: Apr 2002 = 2, May 2002 = 7, Jun 2002 = 3, Jul 2002 = 6, Aug 2002 = 7, Sep 2002 = 1, Oct 2002 = 8, Nov 2002 = 11, Dec 2002 = 8, Jan 2003 = 8, and Feb 2003 = 11. Fewer observations were obtained from cows than bulls because family units were more difficult to approach on foot. The mean (n = 72) number of feeding stations and trunkloads per observation was 15.9 and 81.0 respectively (Figure 6.2).
Figure 6.1 Map showing the spatial positions of the feeding observations used in the discrete choice analysis. The
positions were recorded by saving a track on a GPS.
(a) (b) Figure 6.2 Histograms showing the frequency of (a) number of feeding stations and (b) number of trunkloads per
observation (n = 72).
Elephants were recorded feeding from 101 plant species. However, this figure is an underestimate because for many trunkloads the plant species consumed could not be identified. Within each food type, certain plant species were eaten noticeably more often than others. These included:
Pupalea lappacea (forb); Panicum maximum (green and mixed grass); Urochloa mosambicensis (green
grass); Grewia bicolor (bark from canopy braches, green leaves, leaves and stems, and roots);
Colophospermum mopane (green leaves, leaves and stems, bark from canopy branches, bark from main
stem, and stems); Acacia nigrescens (green leaves, leaves and stems, and bark from roots); Albizia
petersiana (green leaves); Ehretia amoena (green leaves); Acacia tortilis (leaves and stems, roots, and
bark from main stem); Dalbergia melanoxylon (leaves and stems); Dichrostachys cinerea (leaves and stems); Hippocratea indica (leaves and stems); and Gymnosporia pubescens (leaves and stems) (Figure 6.3).
Green grass was chosen as the primary food most often (32 times), followed by green forbs (21 times) and green leaves (17 times). Both mixed grass and bark from canopy branches were only chosen as the primary food once.
(a) (b) (c)
(d) (e) (f) (g)
(h) (i) (j)
Figure 6.3 Charts showing the number of trunkloads consumed of each plant species for (a) forbs, (b) green grass, (c) canopy bark, (d) green leaves, (e) roots, (f) bark from roots, (g) mixed green and dry grass, (h) leaves and stems, (i)
The parameter estimate for the short-term rate of protein intake (β = 2.27, S.E. = 0.55) was highly significant (t-ratio = 4.16, P = 0.00003) indicating that the rate of protein intake had a strong positive influence on choice of the primary food type (Figure 6.4). The odds ratio was 9.68, with upper and lower 95 % confidence intervals of 28.19 and 3.33 respectively. This meant that an increase in the rate of protein intake of 1 gs-1 increased the odds of a food type being chosen as the primary food source by a
multiplicative factor of 9.68. The effect was considered genuine because the lower bound of the confidence interval for the odds ratio was greater than 1.
Figure 6.4 Chart showing the effect of varying levels of short-term rate of protein intake on the relative
probability of a food type being chosen as the primary source of food. The relative probability was calculated using the estimated value of β (2.27) in the equation for the multinominal logit model given in section 6.2.1.
The highest rates of protein intake were achieved when feeding on green forbs and green grass, with green leaves, mixed grass and canopy bark supplying progressively lower rates (Figure 6.4). Because of the relationship between short-term rate of protein intake and relative probability of being chosen as the primary food type, the above ordering of food types was also reflected along the probability of choice gradient.
There was substantial variation in the probability of choosing canopy bark despite very little variation in its rate of protein intake. Consequently, choice of canopy bark may be influenced by factors other than rate of protein intake.
6.4.
Discussion6.4.1.
Ecological insightsThe results show the rate of protein intake to be a potentially important factor influencing diet choice by elephants and, in so doing, provide support for the hypothesis that elephants take advantage of a high passage rate of ingesta and process a large amount of food with cell solubles rich in energy and nutrients per unit time to meet their nutritional demands.
The seasonal change in the diet of elephants, which has been consistently reported by studies of elephant feeding behaviour (Buss, 1961; Bax & Sheldrick, 1963; Laws, 1970a; Field, 1971; Wyatt & Eltringham, 1974; Williamson, 1975; Field & Ross, 1976; Guy, 1976; Barnes, 1982; Owen-Smith, 1988; Kabigumila, 1993; de Boer et al., 2000; Osborn, 2004), is well explained by the influence of protein intake rate on diet choice. During the wet season, grass and forbs provide the highest rates of protein intake and, therefore, make up the bulk of the diet at this time. As the dry season progresses, the rate of protein intake from herbaceous food types declines below the level supplied by woody browse, which results in a switch from a diet predominantly composed of grass and forbs to a diet predominantly composed of leaves and stems from woody plants. Once woody plants have shed their leaves, elephants have no alternative other than to eat bark and roots, despite the very low rates of protein intake achieved when feeding on these food types.
It could be argued that preference for grass as opposed to browse during the rainy season is explained by the higher levels of chemical defence present in the latter (Shipley, 1999). However, this hypothesis is not supported by the fact that elephants consume large amounts of forbs during the rainy season, many of which are known to be chemically defended (e.g. Acanthospermum (Ali & Adam, 1978), Tribulus (Botha & Penrith, 2008), Heliotropium (Mohanraj et al., 1981; Asibal et al., 1989; Agarwal et al., 1995; Guntern et al., 2001), Crotalaria (Botha & Penrith, 2008), Indigofera (Garcez et al., 1989), Sida (Driemeier et al., 2000) and Blumea (Ahmad & Alam, 1996)). The consumption of forbs with high levels of secondary metabolites may adversely affect the health of elephants. For example, Flaccid Trunk Paralysis of elephants at Matusadona National Park in Zimbabwe (Kock et al., 1994) is thought to be caused by consumption of large quantities of Heliotropium ovalifolium and Blumea
gariepina (Guntern, 2003).
Although the rate of protein intake proxy was able to explain choice of green forbs, green leaves, and mixed grass at both the coarse and fine scale, it was only able to explain choice of canopy bark at the coarse scale. This is possibly explained by a failure of the rate of protein intake to adequately represent
the rate of intake of digestible energy by elephants when feeding on bark (i.e. the assumption that crude protein content and digestible energy are positively correlated is unlikely to hold for bark). However, it should be noted that in this study bark was never high on the preference hierarchy (relative probability of bark being chosen as the primary food choice did not exceed 0.2). This is possibly because the long handling time of bark makes it an unprofitable food type regardless of the potentially high levels of digestible energy it contains.
Dependency on an adequate rate of protein intake may provide an explanation for reports of coprophagy by elephants (Guy, 1977; Leggett, 2004) because when elephants eat their faeces they may be making use of undigested microbial protein contained therein. However coprophagy has been rarely recorded for elephants and therefore is unlikely to play a significant role in their nutrition.
Most published studies of resource selection by animals treat habitat and diet selection as separate processes because habitat selection occurs at the coarse scale whereas choice of diet occurs at the fine scale (Schaefer & Messier, 1995; Ginnett & Demment, 1997; Rolstad et al., 2000; Rettie & Messier, 2000; McLoughlin et al., 2002; Fortin et al., 2003; Martinez et al., 2003; Morin et al., 2005; Friar et al., 2005; Whittingham et al., 2005). In this thesis, the opposite stance was taken. It was postulated that the nutritional choices made by elephants at different spatial scales are directed towards achieving a common nutritional goal, and that the integrated effect of the nutritional choices made during habitat and diet selection can be represented in a statistical model by a single, functionally derived variable, namely the short-term rate of protein intake (see chapter 1). The results of this chapter, and those of chapter 5, show that the short-term rate of intake of digestible energy and nutrients (as indexed by the rate of protein intake) potentially has a strong influence on both habitat and diet selection by elephants and, in so doing, provide support for the idea that habitat and diet selection by animals should not be treated as separate processes but rather as interconnected processes that are driven by a common nutritional goal.
6.4.2.
Limitations and future improvementsIn this study, unlike most other studies of resource selection, the focal animals were not presented with each food type simultaneously because the choice set was delimited by the boundary of the relevant land unit as opposed to an area defined by a short radius around each feeding station. This approach rests on the assumption that the animal under study has prior knowledge of the potential availability of different food types in the land unit. This is because an animal may pass up an opportunity to feed on a particular
food item in its path if it knows that there is a good chance it will encounter something better further on. In other words, each food type is still part of the choice set regardless of whether it is physically present when a decision is made about what or what not to eat. Prior knowledge of availability could be gained during previous bouts of foraging, and may improve with the number of days or years spent feeding in the land unit. Given that adult (> 15 years old) elephants were chosen as the focal animals, it is likely that the elephants under study had considerable knowledge of the availability of food types in each land unit and therefore the use of a discrete choice model without simultaneous presentation of each food type should not be inappropriate. Furthermore, patterns of diet selection that emerge from an analysis where each data point represents an accumulated set of feeding observations from within a land unit are likely to be more robust than those from a data set where each data point is a single observation from a feeding station. However, the effect of prior knowledge of food availability on diet choice requires further investigation.
In this study, the choice set was limited to five food alternatives whereas in reality elephants utilised in excess of 10 different forage types. The existence of a strong relationship between short-term rate of protein intake and diet choice justifies further study with a more comprehensive choice set.
The paucity of data from adult cows prevented an analysis of sex related differences in diet selection. An understanding of the differences in feeding behaviour between the sexes is important for the management of elephant-woodland systems because (1) adult bulls are more destructive feeders than cows (Barnes, 1982), and (2) density dependent effects that may act as natural controls on elephant populations are likely to be manifested when nutritional constraints are imposed on family units (O'Connor et al., 2007). For these reasons, future studies should focus on investigating whether the effect of the rate of intake of digestible energy and nutrients on diet choice differs between the sexes.
Use of the rate of protein intake to predict diet choice by elephants over large areas is limited by the fact that calculation of intake rate is highly data intensive. In this study, spatio-temporal change in the density of food patches was quantified using a ground-based approach. The development of remote sensing techniques that reduce the effort required to quantify patch density of different food types should be the focus of future research.