Mass harvested per trunkload as a constraint to forage consumption by the African savanna elephant (Loxodonta africana)

Estimation of trunkload mass for grass

When feeding on grass elephants generally uproot and consume a whole tuft with each trunkload. However, for robust perennials only the upper parts of a tuft are eaten, the roots and bases of the tillers remaining unharvested or being discarded. Consequently, for robust perennial grasses the mass of a trunkload was represented by the average mass of the upper portions of the tuft and for other species by the average mass of the whole plant (Clegg, 2010).

Elephants avoid eating senescent forage, so a grass tuft is selected if it offers a sufficient ratio of green to dry material. It takes twice as long for elephants to harvest, clean by shaking, chew and ingest a trunkload of mixed green and dry grass compared to one of green grass only (Clegg & O’Connor, 2016). Consequently, separate estimates were derived for trunkloads of green grass, and mixed green and dry grass.

The mass of a trunkload when feeding on green grass, S_grass green, varies both spatially and temporally because of differences between grass species, soil types and stages of growth. To capture this variation, S_grass green was estimated daily for each landscape unit as: ${\displaystyle S_{grassgreen}=a{H}_{green}B,}$ where H_green is the average height (cm) of the tallest green grass leaf, B is the average basal area (cm²) of a tuft, and a is a constant. Trunkload mass when feeding from mixed grass was estimated in the same way except values for H_green were substituted by values for mixed green and dry grass patches.

An estimate of a was derived for each vegetation type using the following approach. A list of grass species was compiled by recording the dominant perennial grass species in each vegetation type (Clegg & O’Connor, 2012). Twenty-five tufts, covering a range of leaf heights, basal areas, and vegetation types, were selected for each species on the list. For each tuft, the height above the ground of the tallest leaf and the circumference of the base was measured to the nearest centimetre. The tuft was then uprooted, dried to constant mass after removing the soil from the roots, and weighed to the nearest gram. The data for individual grass species were then used to construct data sets for each vegetation type by pooling the data for each of the dominant grass species found growing in a vegetation type. Estimates of a were derived for each vegetation type by fitting the above equation to the pooled data sets using the linear regression module of Systat 9 (SPSS, 1998).

To derive daily estimates of H_green and H_mixed, each landscape unit was sampled between November 2001 and July 2003 at approximately 3-month intervals. On each occasion, five sample points were positioned in each landscape unit using a stratified random sampling strategy and located in the field using a GPS. At each sampling point, the height of the tallest green and dry grass leaf was measured to the nearest centimetre in 25 × 1 m² quadrats positioned along a 50 m tape. A time series for each landscape unit was constructed by plotting the average height data (H_green and H_dry) against time and interpolating between sample points using the smoothing spline regression module of Kyplot 4.0 (KyensLab Inc., 2002). Data for H_mixed were calculated as the average of H_green and H_dry for each sample point.

The average basal area of tufts in each landscape unit was estimated by randomly selecting between 50 and 100 tufts of each of the dominant perennial grass species and measuring the circumference of each tuft (nearest centimetre). The point at which the coefficient of variation stabilised determined the number of tufts sampled for each species. There is a limit to the size of tufts elephants will uproot in a single trunkload. To estimate the preferred maximum circumference of tufts for adult bulls and members of breeding herds, 100 tufts of Setaria incrassata that had been uprooted, fed on and discarded were collected from beds of Setaria incrassata where adult bulls only and members of breeding herds only had been feeding. Setaria incrassata was chosen as the target species because, of all the grass species at Malilangwe, it has individuals with the largest tuft circumference and therefore provides an opportunity to estimate the upper limit of tuft size selection. The mean tuft circumferences selected by adult bulls, members of breeding herds, and within the plant community as a whole were calculated, and differences between means were tested using a Kruskal–Wallis test and a Dunn post hoc test (chosen because data were non-normal and heteroscedastic). The analysis was conducted using R version 4.1.2 (R Core Team, 2021), with the ‘FSA: Simple Fisheries Stock Assessment Methods’ package (version 0.9.5; Ogle et al., 2023) being used for the Dunn test. The mean values were used to adjust the tuft circumference data sets for each grass species by setting a ceiling on the size of tufts selected by adult bulls and members of breeding herds. The adjusted tuft circumference data were then used to calculate the average basal area for each species. The average basal area of a tuft selected by adult bulls, B_bull, was estimated for each landscape unit as: ${\displaystyle B_{bull}=\sum _{i=1}^n{B}_{i,bull}{C}_i,}$ where B_i,bull is the average basal area of a tuft of the ith species selected by adult bulls and C_i is the proportional aerial cover of the ith species (refer to Clegg & O’Connor, 2012 for how the aerial cover of grass species was estimated). The average basal area of a tuft selected by members of breeding herds, B_cow, was calculated in a similar way except that B_i,bull was replaced by B_i,cow.

The herbaceous layer of some vegetation types was entirely composed of annual grasses and because elephants also utilise annual grasses the most abundant of these were selected for measurement. However, the seasonal variation in mass of annual grass plants was ignored on account of their rapid natural attrition following cessation of growth. Consequently, only the dry mass of 25 randomly selected plants was measured, from which the average dry mass of a plant was calculated. For these vegetation types, S_grass green was estimated as: ${\displaystyle S_{grassgreen}=\sum _{i=1}^n{W}_i{C}_i,}$ where W_i is the average dry mass of the ith annual grass species, and C_i is the relative per cent aerial cover of the ith annual grass species.

Estimation of trunkload mass for forbs

When feeding on forbs (defined in this study as herbaceous dicotyledonous vascular plants), elephants uproot and consume a whole plant with each trunkload. It was assumed that the mass of individual forb plants was greatest at the end of the growing season and declined as the dry season progressed due to browsing, trampling and senescence. It was assumed that the decline in mass was proportional to the change in average forb height and, consequently, trunkload mass when feeding on forbs, S_forb, was calculated, for each landscape unit, as: ${\displaystyle S_{forb}=\sum _{i=1}^n{W}_i{H}_{forb}{C}_i,}$ where W_i is the average dry mass of the ith forb species at the end of the growing season, H_forb is the average height of green forbs relative to the maximum average height measured for the landscape unit, and C_i is the proportional aerial cover of the ith forb species (refer to Clegg & O’Connor, 2012 for how the aerial cover of forb species was estimated). H_forb was estimated daily for each landscape unit by measuring the height of the tallest forb in the quadrats used to measure grass height and interpolating between average height estimates using the method employed for grass.

In contrast to the pattern of abundance of grasses, the forb component of the herbaceous layer was seldom dominated by a few species. Instead, many forb species occurred in low abundance. Consequently, to reduce data collection to manageable proportions, the average dry mass of each forb species, W_i, was estimated, at the end of the growing season, from the dry mass of five randomly selected individual plants.

Estimation of trunkload mass for leaves from woody plants

The mass of an individual leaf or leaflet influences the mass of leaves gathered with each trunkload from woody plants. When feeding from plants with small leaves, elephants harvest small trunkloads because the mass of individual leaves is low and because small leaves are difficult to handle and many fall from the trunk’s grasp. The converse is true for woody plants with large leaves. To account for variation between species of woody plants, the mass of a trunkload of leaves, S_leaf, was estimated for each woody species using a linear regression model with the mass of a leaf unit as the predictor variable: ${\displaystyle S_{leaf,i}=a{M}_{leaf,i}}$ where S_leaf,i is the dry mass of a trunkload of the ith woody species, M_leaf,i is the dry mass of a leaf unit of the ith woody species and a is a constant. Separate regressions were created for adult bulls and members of breeding herds, and the difference between the mass of leaves harvested by each group was tested across a range of leaf sizes using analysis of covariance (ANCOVA), with leaf size specified as the covariate in the model. The analysis was conducted using R version 4.1.2 (R Core Team, 2021), and the ‘tidyverse’ (Wickham et al., 2019), ‘ggpubr’ (version 0.6.0; Kassambara, 2023a), and ‘rstatix’ (version 0.7.2; Kassambara, 2023b) packages. A leaf unit was defined as the smallest unit held by the trunk and was most often represented by a leaflet. The data used to create the regression models were collected by observing adult bulls (n = 17) and members of breeding herds (n = 18) stripping or plucking leaves from a range of woody plant species. When an elephant moved off, five samples were gathered from the plant it had been feeding on by simulating the amount of leaf it had extracted at each trunkload. The average dry weight of the five samples provided an estimate of S_leaf. In this way, estimates of S_leaf were obtained for a range of plant species, across a number of different adult bulls and members of breeding herds. The average dry mass of a leaf unit was calculated for each woody plant species by randomly selecting five individual plants of each species and collecting between 10 and 100 leaf units, depending on the size of a unit, from each individual, and drying them to constant mass. The average mass of a trunkload of green leaves for adult bulls was then estimated for each landscape unit as: ${\displaystyle S_{leaf,bull}=\sum _{i=1}^n{S}_{leaf,i,bull}{D}_{i,bull},}$ where S_leaf,bull is the dry mass of a trunkload of green leaf gathered by an adult bull from the ith woody species and D_i,bull is the density (no. individuals ha⁻¹) of the ith woody species with canopy volume below six m and >25% canopy volume with green leaf. S_leaf,cow was calculated in the same way except S_leaf,bull was substituted by the value for members of breeding herds and D_i,bull was replaced by D_i,cow which was the density (no. individuals/ha) of the ith woody species with canopy volume below four m and >25% canopy volume with green leaf (see Clegg & O’Connor, 2017 for how density was calculated). Heights of six m and four m were used as cut offs because these represent the maximum feeding heights for adult bulls and adult cows respectively (Clegg, 2010). Only woody plants with >25% green leaf were considered available for foraging because elephants avoid eating senescent leaves (Clegg, 2010).

Estimation of trunkload mass for bark from canopy branches

Elephants remove bark from the canopy branches of shrubs and trees by breaking off a branch, placing it in the mouth and then chewing off the bark along the length of the branch. The average mass of bark, W, extracted from a branch was estimated for the most commonly used species in the following way. For each species, 10 to 13 branches, across a range of diameters and lengths, were collected. Each branch was gathered from a separate, randomly chosen plant. For each branch, the bark was removed, dried to constant mass, and weighed to the nearest gram. Once the bark had been removed, the length (cm) and diameter (mm) at 10 cm intervals was measured for each branch, from which an average diameter was calculated for each branch. Equations predicting the mass of bark from a branch were developed using Systat 9 (SPSS, 1998), for each species, by estimating the parameters of the following model using non-linear regression: ${\displaystyle W=aLD+b{\left(LD\right)}^2,}$ where W is the dry mass of bark, L is the length of the branch, D is the average diameter of the branch and a and b are constants. One hundred branches that had been fed on by adult bulls and 100 branches that had been fed on by members of breeding herds were collected for each plant species. For each branch, the length and average diameter was measured for the part from which bark had been removed. When a forked branch was debarked, each fork was measured separately. The mass of bark removed from each branch was then estimated using the regression equations. The average mass of bark removed from branches of each species was calculated for adult bulls and members of breeding herds by calculating the average of the respective samples. Differences between means were tested using an aligned ranks transformation ANOVA with post hoc multiple comparisons with Tukey’s adjustment (chosen because data were non-normal and heteroscedastic). The analysis was conducted using R version 4.1.2 (R Core Team, 2021) and the ‘ARTool’ package (version 0.11.1; Kay et al., 2021).

The average trunkload mass of bark for bulls, S_bark,bull, was then calculated for each vegetation type as follows: ${\displaystyle S_{bark,bull}=\sum _{i=1}^n{W}_{i,bull}{D}_{i,bull},}$ where W_i,bull is the average dry mass of bark removed by bulls for the ith species, and D_i,bull is the relative density of the ith species with canopy below six m. S_bark,cow was calculated in a similar way except W_i,bull was replaced by W_i,cow and D_i,bull by D_i,cow which was the relative density of individuals of the ith species with canopy below four m.

Running the models

The models for each forage type were run in Idrisi GIS software (Eastman, 2001) at a daily time step for one annual cycle (12th of March 2002 to the 10th of March 2003). Predictor variables, which were constructed by adding attribute data to a raster map of the 65 land units, were included in the regression equations as raster layers. Once the models had been run, the mean, minimum and maximum mass potentially harvested per trunkload over the study area were calculated from the daily output raster layers, and timeseries plotted for each forage type. Daily differences in the mean mass harvested across all paired combinations of forage types were tested for significance using a Wilcoxon signed rank test (assumptions for paired t-tests were not met). The analyses were conducted using R version 4.1.2 (R Core Team, 2021). Comparisons were conducted separately for forage harvested by adult bulls and members of breeding herds, and for the differences between adult bulls and members of breeding herds.

Model parameterization

Estimation of trunkload mass for grass and forbs

The height of green grass, mixed green and dry grass, and forbs varied seasonally and across landscape units (Fig. 1). The height of grasses and forbs decreased during the dry season due to herbivory, trampling and senescence. Within the same land unit, mixed grass was usually taller than green grass, and grasses were generally taller than forbs.

When feeding on Setaria incrassata, a robust, perennial grass, the mean tuft circumference utilised by adult bulls and members of breeding herds was 35.5 cm and 22.3 cm, respectively, compared to 52.7 cm for Setaria tufts in the plant community (Fig. 2). A Kruskal–Wallis test, followed by a Dunn post-hoc test showed significant differences between the three groups (χ² = 122.34; df = 2; P = 2.2e−16; adult bull vs member breeding herd, Z = −8.898, P = 1.14e−18; adult bull vs plant community, Z = −2.398, P = 1.65e−02; member breeding herd vs plant community, Z = −9.614, P = 2.1e−21), and consequently the mean values for adult bulls and members of breeding herds were used to set upper size limits for tuft utilisation.

In each landscape unit, there was a strong linear relationship between the mass of a grass plant and the product of leaf height and tuft basal area (average R²_adj (± SD) = 0.74 ± 0.12) (Fig. 3).

Sixty-eight species of forb were included in the analysis with the average dry mass of a fully developed plant ranging from two g to 418 g. The weighted average dry mass of a fully developed forb plant (assumed to represent the maximum mass of a trunkload) growing in a landscape unit ranged from 14 g to 160 g, with a mean of 62.7 g (Fig. 4). Sixty per cent of the landscape units had an average forb dry mass between 40 and 80 g.

Estimation of trunkload mass for leaves from woody plants

There was a positive linear relationship between the mass of a trunkload of leaves harvested from woody plants and the mass of an individual leaf unit (Fig. 5). Adult bulls harvested heavier trunkloads of leaves than members of breeding herds (ANCOVA, with dry mass of a leaf unit as the covariate: F(1,32) = 8.262; P = 0.007).

Estimation of trunkload mass for bark from canopy branches

The mass of bark removed from canopy branches was successfully modelled for Colophospermum mopane, Grewia bicolor and Grewia monticola (the most commonly utilised species for canopy bark) using quadratic regression equations with the product of the length and diameter of the debarked branch as the predictor variable (Fig. 6). The aligned ranks transformation ANOVA to test for differences in the mass of bark harvested by adult bulls and members of breeding herds across the three plant species showed a significant effect for elephant group (F(1,630) = 81.69; P < 2.22e−16), plant species (F(2,630) = 3.84; P = 0.022), and for the interaction between the elephant group and plant species (F(2,630) = 6.56; P = 0.0015). Adult bulls harvested heavier trunkloads (25.6–41.1 g) than members of breeding herds (19.8–22.1 g) for all comparisons between the plant species (P < 0.001), but there were no significant differences between the mass harvested across the plant species for adult bulls or members of breeding herds (Fig. 7).

Model outputs

The average mass potentially harvested per trunkload changed seasonally for green grass, mixed grass, green forbs, and green leaves but was assumed constant over the annual cycle for canopy bark (Fig. 8). The mass of a trunkload of green grass and green forbs was three times heavier during the late-wet season than at the height of the dry season for both adult bulls and members of breeding herds (Table 1). Trunkload mass was also heavier during the wet season for mixed grass and green leaves from woody plants but only by a factor of one and a half. The variation in trunkload mass across the study landscape was greater for the herbaceous forage types than those from woody plants. For example, for adult bulls during the late wet season there was a difference of 218 g between the minimum and maximum estimates of a trunkload of green grass but only a 33 g difference for a trunkload of green leaves from woody plants. The variation between minimum and maximum estimates was greater for adult bulls than for members of breeding herds.

Grass and forbs yielded heavier trunkloads than leaves and bark from woody plants during the wet season for both adult bulls and members of breeding herds (Wilcoxon signed rank test, P < 0.01), but this was reversed in the dry season (Figs. 9 and 10). Of the herbaceous forage types, green grass produced the heaviest trunkloads during the late-wet season but trunkloads of mixed grass and green forbs were similar or heavier than green grass in the early- and mid-wet season.

Differences between adult bulls and members of breeding herds were apparent. Firstly, trunkloads were heavier for adult bulls than members of breeding herds across all forage types (Wilcoxon signed rank test, P < 0.01), except for forbs, which were assumed to yield a similar mass for both groups (Fig. 11). Secondly, trunkloads of grass were noticeably heavier than trunkloads of forbs for adult bulls, while grass and forbs yielded a similar mass for breeding herds. Thirdly, trunkloads of canopy bark were significantly heavier than trunkloads of forbs for a large portion of the dry season for adult bulls but not for members of breeding herds. Lastly, during the wet and cool-dry seasons, green leaves yielded heavier trunkloads than canopy bark for members of breeding herds but the two forage types yielded a similar mass for adult bulls over the same periods.