Behavioral Ecology and Sociobiology

, Volume 57, Issue 4, pp 327–338

Spatial relationships and matrilineal kinship in African savanna elephant (Loxodonta africana) clans


    • Bioacoustics Research Program, Laboratory of OrnithologyCornell University
  • Rob Roy RameyII
    • Dept. of ZoologyDenver Museum of Nature & Science
  • William R. LangbauerJr.
    • Pittsburgh Zoo and PPG Aquarium
  • Katharine B. Payne
    • Bioacoustics Research Program, Laboratory of OrnithologyCornell University
  • Rowan B. Martin
    • Greendale
  • Laura M. Brown
    • Dept. of ZoologyDenver Museum of Nature & Science
Original Article

DOI: 10.1007/s00265-004-0867-5

Cite this article as:
Charif, R.A., Ramey, R.R., Langbauer, W.R. et al. Behav Ecol Sociobiol (2005) 57: 327. doi:10.1007/s00265-004-0867-5


African savanna elephants, Loxodonta africana, live in stable family groups consisting of adult females and their dependent offspring. During the dry season, “clans” consisting of several family groups typically share a common home range. We compared spatial relationships and mitochondrial DNA (mtDNA) haplotypes among 14 adult female elephants within 3 clans during the dry season in northern Zimbabwe. Spatial relationships were studied by radio-tracking. Home-range similarity was quantified by correlating the estimated utilization distributions of all pairs of elephants. Clans were identified by cluster analysis of the home-range similarity values. All three clans contained at least two of the five mtDNA haplotypes that were found, indicating that clan members are not necessarily matrilineally related. Within clans, home ranges of elephants with the same haplotype were not significantly more similar to each other than those of elephants with different haplotypes. Most elephants within each clan used their shared home ranges independently of each other: the distribution of distances between their positions at any given time did not differ from the distribution expected by chance. However, 8 out of the 26 within-clan pairs exhibited long-term coordination of space use by remaining within known hearing distance of each other’s low-frequency calls significantly more often than expected by chance. At least four of these coordinated pairs consisted of animals in different family groups. Elephants in three of the four different-family pairs whose movements were coordinated had different haplotypes. Further research is needed to determine the relationship between these coordinated movements and conventionally defined bond-group behavior.


African elephantLoxodonta africanaHome rangeUtilization distributionMitochondrial DNA


African savanna elephants (Loxodonta africana, Roca et al. 2001) live in complex societies characterized by several levels of social and spatial relationships among individuals and groups (Douglas-Hamilton 1972; Moss and Poole 1983; Moss 1988). The basic social unit is the “family group,” consisting of one or more (typically two or more) adult females and their dependent offspring. Members of a family group spend most of their time together engaging in the same behaviors (e.g., feeding, drinking, mud-bathing), with brief periods of separation for foraging. A female typically spends her entire life in her natal family group (Moss 1988). The individuals in a family group are therefore presumed to be closely related. On rare occasions, apparently unrelated family groups fuse (Moss 1988), raising the possibility that certain family groups may contain some unrelated elephants.

Male elephants leave their natal groups upon reaching sexual maturity in their mid-teens. Thereafter, males spend some of their time alone, some in association with other males in impermanent groups, and some in short-term associations with various family groups. As they mature, males engage in contests through which each one establishes and repeatedly tests his rank in the local population’s dominance hierarchy. Starting in their late twenties, males spend a portion of each year in a hormonal and behavioral condition called musth, which increases their aggressiveness toward other males and their success in mating (Poole and Moss 1981; Poole 1987).

Female elephants and their associated family groups maintain stable relationships with each other and stable home ranges over many years, whereas relationships of males are more transitory. Family groups that share a common home range during the dry season constitute a “clan” (Moss and Poole 1983; Moss 1988). Clans have been defined on a purely geographical basis, without regard to behavioral interactions or possible genetic relationships. Prior to the work reported here, there have been no quantitative descriptions of the patterns of home-range use within a clan, or of whether family groups within a clan use their shared space independently or in some temporally coordinated way. There have also been no molecular data available on the genetic similarities of families within clans.

Some family groups within a clan spend much of their time in close association with each other, and engage in synchronized behaviors when together. A group of two or more such closely associated family groups constitutes a “bond group” (Moss and Poole 1983; Moss 1988). Bond-group members often engage in excited “greeting” behaviors when they meet each other after a period of separation. During such an episode, members of the groups run together, tapping tusks and entwining trunks, rumbling, trumpeting and bellowing in chorus, twirling around in circles, bumping parts of their bodies together, streaming from their temporal glands, urinating and defecating vigorously and synchronously, and smelling and stroking each other with intense interest. These same “greeting ceremonies” occur within families, when members reunite after separations (Moss 1988). The occurrence of greeting ceremonies and synchronous behaviors within bond groups, as well as family groups, suggests that the family groups within a bond group are genetically related to each other (Douglas-Hamilton 1972; Moss 1988). Douglas-Hamilton (1972) regarded bond groups as extended family, and his term for this level of association was “kinship groups.” Moss (1988) described one case where a single large family group gradually split into two family groups that then behaved as a bond group. Further observations from the long-term Amboseli elephant research project suggest that such fission is the usual, but not the only, way that bond groups form (C. Moss and J. Poole, personal communication regarding Amboseli Elephant Research Project long-term records).

In this paper, we provide the first quantitative description of spatial relationships among adult female elephants from different family groups. We consider 2 types of spatial relationship for each possible pair of 14 elephants that we radio-tracked during the 1990 dry season in northern Zimbabwe. First, we quantify home-range similarity by use of a spatial correlation coefficient (“static interaction,” Doncaster 1990). The values that are correlated are the probabilities of occurrence for each animal at each point in space (i.e., the animals’ utilization distributions), which we estimate from radio-tracking data. We use cluster analysis of the pairwise home-range correlations to develop a quantitative foundation for identifying clans. Second, we assess “dynamic interaction,” the tendency of two animals to be closer together or farther apart than expected under the null hypothesis that each animal’s position in its home range at any given time is independent of the other animal’s location. Positive interaction occurs when one animal is more likely to be at a location if the other is nearby, suggesting either that the animals are attracted to each other, or that both are attracted to some resource or other factor that varies over space and time. Negative interaction occurs when an individual’s probability of being at a given location decreases with the proximity of the other animal to that point, suggesting that the animals are actively avoiding each other (Doncaster 1990). We refer to dynamic interaction assessed over the entire duration of our field study as “long-term coordination,” to distinguish it from short-term coordination limited to periods of a few hours to a few days. Tests for short-term coordination call for different statistical methods using position data that are autocorrelated over time, and will be considered in a separate publication.

We also investigate whether possible matrilineal kinship (as indicated by mitochondrial DNA haplotype sharing) is correlated with clan membership, home-range similarity, and long-term dynamic interaction within clans.


Field and laboratory methods

Field site and study population

We conducted the fieldwork from early August through October 1990 at the Sengwa Wildlife Research Area (SWRA) in northwest Zimbabwe (Fig. 1). The SWRA encompasses 373 km2 at approximately 900 m above sea level, situated between 28°03′ and 28°20′E and 18°01′ and 18°13′S. It is bordered on three sides by sparsely settled communal lands where marginal farming is practiced, and on the north by the Chirisa Safari Area in which, as in Sengwa, management by the Department of National Parks and Wildlife Management affords partial protection to wild game species. A fence runs along the western, southern and eastern boundaries, but on the north, elephants freely move between Sengwa and Chirisa. Three rivers—the Sengwa, the Manyoni and the Lutope—converge near the center of the research area, which is a broad forested basin bordered on the east and west by hills and escarpments.
Fig. 1

Location of the Sengwa Wildlife Research Area. Shaded areas are sparsely settled farmland. The black dots indicate the locations of the two radio-tracking stations. The contour lines show where the mean location error is equal to 0.5 and 1.0 km.

The climate of Sengwa is hot and dry, with an average rainfall of 680 mm (data from 1965–1981) and temperatures ranging from 5° (July) to 35°C (October). During the dry season, which lasts from May to October, the rivers stop flowing in most places. However, surface water remains available in pools formed by old oxbow meanders, in springs below the major escarpments, and in thousands of small sand- and clay-walled wells that elephants dig in the dry riverbeds with their trunks and feet (K. Payne 1998 and unpublished data).

The dominant vegetation types are Miombo (mixed Brachystegia-Julbernardia) woodland on sandy soils in higher areas and mopane (Colophospermum mopane) woodland on clay soils in lower areas (Guy 1974). The sandy riverbeds are bordered by strips of open grassland up to several hundred meters wide.

The elephant population in and around the SWRA has been the subject of considerable research and management work. Radio-tracking studies of elephants in the SWRA were first undertaken in 1971 (Martin 1978), and continued through 1985. Concerns about elephant damage to vegetation led to seven culling operations between 1978 and 1986, which removed a total of 1,415 elephants (Osborn 1998). The culls were intended to remove entire family groups. The last cull before the present study was in 1986, when 94 elephants were killed. At the start of the present work, there were an estimated 500–750 elephants in Sengwa.


We placed radio collars on 16 mature female elephants in 14 herds. We use the term “herd” to refer to a group of elephants found in close proximity to each other at a particular time, without reference to any social bonds that may exist among those individuals. In most cases, a herd corresponds to one or more family groups within a bond group. Individual elephants were assigned names for convenience; animals are identified in this paper by name and collar number (cN). Two of the collared elephants (Chisamiso, c12, and Manengi, c2) were still wearing collars from Martin’s (1978) previous studies when they were immobilized for this project. Three others (Makweru II, c3; Chipo II, c9; and Lutya II, c8) resembled elephants collared in previous studies and may have been the same. Each of these five elephants was found in the same home range as that observed in previous studies (R.B. Martin, unpublished data). Three elephants (Chisamiso, c12; Makweru II, c3; and Shena, c15) were collared in one herd and two (Crooked Tusk, c11, and Babe, c16) in another. In each of the 11 remaining herds, we collared 1 adult female.

We fitted each elephant with a collar carrying a dual radio transmitter package. Each package contained both a location beacon and an audio transmitter, for use in a study of long-range acoustic communication. The transmitters operated on different frequencies so that the identity and location of each calling elephant could be determined. The completed transmitter package was a sealed fiberglass box that measured 30×15×15 cm, and weighed 6 kg. The transmitter package was bolted to the middle of a length of nylon-reinforced rubber machine belting 2.5 m long, 15 cm wide, and approximately 0.6 cm in thickness.

Elephants were immobilized using M99 (etorphine hydrochloride) tranquilizer darts fired from a helicopter or on foot. Once an elephant was immobilized, blood was drawn for genetic analysis, the animal was measured (shoulder height, tusk and tooth measurements), and the collar attached. The tranquilizer antidote (M5050, diprenorphine) was then administered, and the elephant regained her feet within several minutes.

Collars were attached by fastening the machine belting around the neck of an immobilized elephant so that the transmitter package rode on the top of the animal’s neck. A 7-kg counterweight that hung below the elephant’s neck prevented the transmitter package from slipping down the side of the neck. To allow visual recognition of collared elephants, each transmitter package was painted with a unique color combination, the collar was painted with an identifying number, and a unique pattern of notches was cut into the edges of the collar belting that hung below the animal’s neck.

DNA extraction, mtDNA amplification, and sequencing

We collected blood samples from elephants that had been immobilized for radio-collaring. Blood was drawn from a leg vein into EDTA-treated vacutainer tubes, centrifuged to separate cells from serum, and the leukocytes drawn off with a pasteur pipette. These were frozen at −20°C until DNA was extracted using a salt/chloroform extraction procedure (Mullenbach et al. 1989).

We enhanced the mtDNA fraction of DNA samples prior to PCR amplification. A 75-µl aliquot of DNA in TE buffer was passed through a 30-gauge needle 30 times to shear genomic DNA. Then 250 µl of 95% ethanol was added, mixed by vortexing, and spun at 1,000 g for 1 min to precipitate genomic DNA. The supernatant was removed and chilled to –20°C for 30 min and then spun at 10,000 g for 30 min to pellet the enhanced mtDNA fraction. The pellet was dried completely at 60°C, and then 100 µl of 10 mm Tris (pH 8.5) heated to 65°C was added and mixed by vortexing.

We identified mitochondrial haplotypes by sequencing part of the mitochondrial control region. We amplified a 626-bp fragment of the 5′ end of the control region using primers Laf Cr1 and Laf Cr2, as described in Nyakaana and Arctander (1999) with the following modifications to PCR conditions: 94°C denature for 2 min, followed by 37 cycles of 95°C for 30 s, 47°C for 30 s, and 72°C for 45 s. Cycle sequencing was performed on both strands at an annealing temperature of 48°C with primers Laf Cr1, Laf Cr2, and an internal primer BETH (ATGGCCCTGAAGAAAGAACC) that we designed for the first conserved sequence block of the control region. Unincorporated dye terminators were removed with DyeEx spin columns (Qiagen), and the reactions run on an Applied Biosystems 373XL automated DNA sequencer. Chromatograms were compiled and edited using the program Sequencher 4.0 (GeneCodes). The final edited sequences were 399 bp in length and homologous to those analyzed by Nyakaana and Arctander (1999).

Since mtDNA is maternally inherited, elephants that share the same mtDNA haplotype may or may not be members of the same matriline; individuals with different haplotypes must be members of different matrilines. Fernando et al. (2000) estimated the sequence divergence rate of the mitochondrial control region between African savanna elephants and Asian elephants (Elephas maximus) to be ≈1%/Myr. At this rate, and assuming a generation time of 20 years, we would expect a single nucleotide substitution to occur in a 399-bp sequence on average only once every 12,500 generations. We therefore assume that any difference in haplotype between family groups reflects a divergence in matrilines that occurred in the remote past relative to the time scale of elephant social behavior.

Radio-tracking and visual observations

Bearings to each collared elephant were taken every 3 h during the day and night from two receiving stations located 7 km apart (Fig. 1). One station (RS1) was at the SWRA headquarters; the other (RS3) was on top of a mesa that rises steeply 125 m from the valley floor near the confluences of the Sengwa, Manyoni, and Lutope rivers. Each station used two parallel directional Yagi antennas atop a rotatable mast (8 or 12 m tall) connected to a null-peak switchbox. When used in “peak” mode, the switchbox combined the signals from the two directional antennae in phase. In “null” mode, the phase of one of the signals was reversed before the two signals were combined; an operator monitoring the signal on headphones as the mast was rotated perceived an abrupt drop in signal intensity when the antennae were pointed directly at the transmitter being tracked.

Data from the RS3 tracking station were transmitted by radio back to the SWRA headquarters as soon as bearings were obtained for all animals. Bearings from both stations were immediately entered into a spreadsheet that calculated coordinates for the intersecting bearings. This immediate processing allowed us to identify occasional tracking errors that resulted in non-intersecting bearing lines, and to re-track the animal in question within a few minutes. Determination of bearings to all collared animals from both stations typically required no more than 20–30 min; locations determined for all animals during one such tracking session are treated here as simultaneous. A custom data visualization program was used to plot the locations and recent tracks of animals on a computer monitor.

We gathered information on the composition of groups containing collared animals by opportunistic visual observations from the top of the radio-tracking mesa, from five aerial census flights at roughly 1-month intervals, and from occasional ground encounters with elephants.

Analysis of radio-tracking data

Radio-tracking location errors

Earlier tests of the accuracy of the radio-tracking system demonstrated that bearing errors from each tracking station were roughly normally distributed around 0°, and that 80% of the errors were within ±2°, corresponding to a standard deviation of 1.6° (R.B. Martin, unpublished data). Based on this bearing-error distribution, we estimated the mean location error for each radio-tracking location, as detailed in the Appendix. The accuracy with which an animal could be located varied systematically throughout the study area, depending on the angle at which the bearing lines from the two tracking stations intersected. Location error was minimal where the bearings were perpendicular to each other, and became very large as the difference between the two bearings approached zero.

Home-range analysis

For each elephant, we estimated the utilization distribution (UD) during the time of the study using the kernel method described by Worton (1989), with a fixed smoothing parameter chosen by least-squares cross-validation (Worton 1995; Seaman and Powell 1996). This method estimates the probability of finding an animal at any given point on the plane. Probability estimates were calculated at points on a 0.25-km grid. All of the data for each animal were used for estimating the individual’s home range. Strictly speaking, this violates the kernel method’s requirement of independent observations. However, because the observations are uniformly distributed over this period, the data should provide an unbiased representation of the home range during the time of the study (White and Garrott 1990; De Solla et al. 1999).

Home-range similarity (static interaction)

We calculated a product-moment correlation coefficient between the home ranges of every possible pair of radio-collared elephants. The correlation coefficient provides an index of the similarity or “static interaction” between the patterns of space use of two animals (Doncaster 1990). This approach provides a better measure of home-range similarity than simple home-range overlap values because it takes into account the intensity of use of each part of an individual’s range. The values that were correlated for each pair were the probability estimates of each animal’s occurrence at every point on a 0.25-km grid covering the entire research area, as generated by the kernel method described above. The correlation coefficient can vary between 0 and 1. Values close to 1 indicate extensive home-range overlap and similar use of space within the overlap area. Values close to 0 indicate minimal range overlap. The home-range correlation ignores the temporal dimension of patterns of space use. Two animals may have a high home-range correlation because they are always closely associated with each other (e.g., members of the same family group), or they may be members of different groups that use the same areas with similar intensity, but not necessarily at the same times. We used STATISTICA (StatSoft 1994) to perform a cluster analysis (using Ward’s linkage method) on the home-range correlations among all possible pairs of elephants.

Long-term dynamic interaction

To assess the degree to which elephants were associated in time as well as space, we used the nonparametric randomization technique described by Doncaster (1990). This method compares the distribution of observed distances between simultaneous tracking fixes on two animals to a distribution of distances expected under the null hypothesis of independent movements within the animals’ actual home ranges (no dynamic interaction). The expected distribution is generated empirically by calculating the distances between all possible pairs of non-simultaneous fixes on the two individuals. This method accounts for the actual pattern of space use of each animal, without making any assumptions about the form of the utilization distributions.

The Doncaster method requires that the successive distances between animals be independent. In the present study, successive fixes in the complete tracking data set for an animal are usually not independent, because the elephants moved only a short distance (relative to their home-range sizes) between successive fixes. We used the method described by Swihart and Slade (1985) to determine the time-to-independence (TTI) for each animal. The TTI is the minimum time k, such that the animal’s position within its home range at time t+k is not a function of its position at time t. Successive distances for a pair of animals are independent as long as successive fixes on at least one of the animals are independent. The lesser of the two TTIs of a pair is thus called the “pairwise TTI”.

For each pair within a clan (see Results for clan identification), we generated a set of independent distances by extracting every Nth fix from each animal’s complete set of fixes, where N is the pairwise TTI expressed in number of 3-h tracking intervals. We then calculated the distances between all possible pairs of independent fixes. Distances between simultaneous fixes comprise the observed distance distribution; distances between non-simultaneous fixes comprise the distance distribution expected under the null hypothesis of independent movement (no interaction).

We tested whether pairs of animals were within 4 km of each other more frequently than expected under the null hypothesis of independent movements within their respective home ranges. We chose 4 km as an appropriate distance threshold because this is the greatest distance at which elephants have been demonstrated to detect each other’s loud low-frequency calls (Langbauer et al. 1991). We first calculated the probability that the distance between the two animals would be ≤4 km at any given time, based on the expected distribution of distances. We then used the binomial distribution to calculate the probability that at least the observed number of distances would be ≤4 km (Doncaster 1990). We consider the movements of a pair of elephants to be coordinated if the individuals are within 4 km of each other significantly more often than expected by chance (P<0.05).

Comparison of genetic and spatial relationships

We used the Mantel test (Dietz 1983; Schnell et al. 1985) to determine whether there was a positive correlation between mtDNA haplotype sharing and clan membership. The Mantel test is a non-parametric permutation test that compares two square similarity (or difference) matrices in a way that accounts for the interdependencies among pairs of individuals. To test whether mtDNA haplotype sharing was positively associated with home-range similarity and long-term coordination of movements within clans, we used a modified version of the Mantel test, in which permutations are performed only within specific square submatrices of the entire matrix, with each submatrix corresponding to one clan. The cell products of the matrices being compared are then summed only for the same-clan pairs. In all tests, haplotype sharing, clan membership, and long-term coordination were represented by matrices of ones and zeros. Mantel tests were implemented in MATLAB (MathWorks 1992). All Mantel test results reported here are based on 10,000 iterations.


Identification of mtDNA haplotypes

Mitochondrial DNA sequence analysis identified five haplotypes that differed from each other by one to ten base pairs, which we refer to as haplotypes A, B, C, D, and E (Fig. 2). The sequences for these haplotypes have been deposited in the GenBank database under accession numbers AY577804 through AY577808.
Fig. 2

Minimum spanning network representing relationships among mtDNA haplotypes A–E.  Each circle represents the given haplotype, with the area of the circle proportional to the number of individuals with that haplotype. Hatch marks on connecting branches indicate the number of nucleotide substitutions separating haplotypes. Identifying collar numbers (cN) of individuals with each haplotype are shown within each circle. Background shading indicates the geographic clan membership of individuals: white Sengwa clan; light gray Manyoni clan; dark gray Lutope clan.

Home ranges and clan composition

We obtained adequate data to estimate home ranges for 14 of the 16 elephants collared. One transmitter evidently failed; signals from a second were detectable only intermittently. Figure 3 shows the locations of the 14 home ranges on a map of the Sengwa Wildlife Research Area. Home-range correlations among all 91 possible pairs of collared elephants ranged from 0 to 0.968. Cluster analysis of the home range correlations identified three distinct clusters of home ranges (Fig. 4). Geographically, these three clusters are centered on the Lutope River, the Manyoni River, and on the confluence of the Sengwa, Manyoni, and Lutope. We refer to the elephants represented in each of these areas as the Lutope, Manyoni and Sengwa clans. Within each clan, all pairs (with one exception) have home range correlations greater than 0.31.
Fig. 3

Over-smoothed 50% home ranges of 14 radio-collared elephants in 3 clans. In order to show range overlaps clearly, these contours were generated using a fixed smoothing parameter larger than any of the optimal smoothing parameters used to generate the utilization distributions for home-range correlation analysis. The actual utilization distributions used in the home-range correlation analysis are shown in Fig. S1. Collar numbers (cN) identify individual elephants; uppercase letters AE identify mtDNA haplotypes. Hatched backgrounds of labels identify pairs of elephants that are likely to be members of the same family group.

Fig. 4

Dendrogram generated by clustering elephants according to home-range correlation. Elephants are identified by collar number (cN). Boxes around pairs of collar numbers indicate pairs of elephants that may be in the same family group. Arrows at left indicate pairs of elephants in different families that coordinated their use of shared space by remaining within 4 km of each other significantly more than expected. The home-range correlations used in the clustering are based on the entire home range, not just the 50% core illustrated in Fig. 3.

The most intensively used portions of each home range center on water sources. The area at the center of the home ranges of the Sengwa clan contained a unique water source—a series of large, closely spaced permanent pools in an old oxbow of the Sengwa River. No other water source in the study area was comparable in terms of size and permanence.

Table 1 shows numbers of locations, distribution of estimated location errors, and home-range sizes for each elephant. The 50% and 95% home-range sizes are the areas included within the 50% and 95% cumulative probability contours. Fifty percent home-range size varied between 3.9 and 31.4 km2 (median=13.5 km2). Ninety-five percent home-range size varied between 22.1 and 136.3 km2 (median=65.6 km2). Fifty percent and 95% home-range sizes in the Sengwa clan were significantly smaller than the corresponding home-range sizes for the pooled Lutope and Manyoni clans (50% ranges: P=0.0278; 95% ranges: P=0.0136; Mann-Whitney U-test), which did not differ significantly from each other (50% ranges: P=0.1417; 95% ranges: P=0.3272; Mann-Whitney U-test).
Table 1

Numbers of location fixes, location errors, and home- range sizes for 14 radio-tracked elephants in 3 clans. Within each clan, elephants are listed in order of decreasing 95% home-range size


Elephant ID

Location fixes

Median location error (km)

% of location errors >1 km

50% home-range size (km2)

95% home-range size (km2)

































































































Dynamic interaction and family membership

Table 2 (columns 5–8) summarizes, for each of the 26 same-clan pairs, the comparisons of the distributions of independent distances actually observed to those expected under the null hypothesis of no coordination of movements. The number of independent distances available for each pair varied between 10 and 54. This variation is a result of differences among animals in time to independence (which varied between 10 and 51 tracking sessions) and variations in the completeness of the tracking record for each animal.
Table 2

Home-range correlations, mtDNA haplotypes, comparisons of observed and expected distance distributions, and proportion of distances ≤4 km for all pairs of elephants in each clan. Pairs are identified by the numbers on their radio collars. Boldface type indicates pairs for which the proportion of independent distances ≤4 km is significantly greater than expected by chance. Pairs in which one animal is considered a redundant representative of its family group are listed last within each clan and are shown in italics. Otherwise, pairs are listed within each clan in order of decreasing home-range correlation



HR correlation

mtDNA haplotypes

Independent distance distribution

Proportion of distances ≤4 km



Median (km)


Median (km)




P (binomial)








































































































































































































































































*P<0.05; **P<0.01; ***P<0.001

Figure 5 shows two representative cumulative-frequency plots from the analysis of independent distances used to assess long-term coordination. Part (a) shows an example of a pair with no significant difference between the observed and expected distributions; part (b) shows a pair with a significant difference.
Fig. 5

Examples of cumulative distribution for observed and expected independent distances for a two members of the Sengwa clan (c6, Computer, and c11, Crooked Tusk) who use their home ranges independently, even though their ranges have a correlation of 0.798, and b two member of the Manyoni clan (c4, Miss Piggy, and c8, Lutya) whose movements often appeared to be coordinated. The continuous curve shows the expected distribution; the points show the observed distribution

Based on their independent distance distributions, none of the same-clan pairs showed any sign of avoiding each other within their shared home ranges.

Eight pairs of collared elephants were within 4 km of each other significantly more often than expected by chance (binomial test: P<0.05; Table 2, column 11), and are thus considered “coordinated” in their movements. Overall, there was a significant correlation between the occurrence of such coordination and home-range similarity (Mantel test, P=0.0356).

In order to investigate spatial relationships among different family groups, we must consider data from only one collared elephant per family group. Elephants at Sengwa are intolerant of close approach by human observers and inhabit mopane/acacia woodland where visibility is limited, and thus we were not able to collect the type of extensive visual observations on which determinations of family and bond group membership are usually based (Moss and Poole 1983; Moss 1988). Although we cannot unambiguously determine that two individuals are in the same family group, our radio-tracking data can be used to distinguish between pairs of elephants that might be members of a single family group, and pairs that must be in different family groups. We consider that animals may be in the same family group if and only if: (a) their movements are coordinated, and (b) their median distance is less than or equal to 2 km. Any pair that fails to meet either of these conditions must represent two distinct family groups. However, a pair that is not in the same family group could also satisfy both conditions if the groups coordinate their movements and have very similar home ranges (e.g., elephants that are in different family groups within the same bond group).

Three of the eight coordinated pairs meet this criterion for possible membership in the same family group (c3/c12, Makweru and Chisamiso in the Lutope clan; c11/c16, Crooked Tusk and Babe in the Sengwa clan; c9/c10, Chipo and Sijamba in the Manyoni clan). The first two pairs were also together when they were collared. We consider one elephant from each of these pairs (c3, c9, and c16, selected by coin flip from the three pairs) as potentially redundant representatives of their families. If we omit these 3 potentially redundant elephants, only 4 of the remaining 15 within-clan pairs of elephants exhibited significant long-term coordination of movements (Fig. 4), too few to allow statistical testing of whether such coordination is correlated with home-range similarity.

In analyses of mtDNA haplotype sharing and spatial relationships, we present results both for the complete set of all 26 same-clan pairs of elephants, and for the restricted set of 15 pairs in which data from the 3 potentially redundant animals have been omitted. The restricted data set thus constitutes all pairs whose members undoubtedly belong to different family groups.

Spatial relationships and haplotype sharing

All three geographically defined clans contained animals of two or three haplotypes (Table 2, Figs. 3, 4). Pairs of elephants within clans were no more likely to have the same haplotype than expected by chance (Mantel test, P=0.8115).

Within clans, there was no significant tendency for elephants with the same haplotype to have more similar home ranges than elephants with different haplotypes, in either the complete or the restricted data set (Mantel test, complete: P=0.2081; restricted: P=1.000). When data from all 14 elephants are included, the mean±SD home-range correlation for all 9 same-haplotype pairs within clans was 0.619±0.312; for all 17 different-haplotype pairs, the mean±SD was 0.471±0.129. When data from the 3 potentially redundant elephants were omitted, the 4 same-haplotype pairs had mean±SD home-range correlation of 0.417±0.307; the 11 different-haplotype pairs had mean±SD of 0.484±0.148.

When data from all 14 elephants are included, 4 of the 8 coordinated pairs of elephants had the same haplotype. In this complete data set, there was no significant association within clans between a pair having the same haplotype and coordinating their long-term movements (Mantel test: P=0.1081). When the three potentially redundant elephants are omitted, there are only four coordinated pairs, of which one has the same haplotype. This sample size is too small to permit statistical testing of any association between long-term coordination and haplotype sharing among elephants in different families.


Clan “membership” and matrilineal kinship

The dry-season home ranges of the 14 elephants that we radio-tracked fell into 3 distinct geographic areas. Within each area, core portions of home ranges overlapped extensively, as indicated by high home-range correlation values. We consider the elephant family groups in these areas to constitute three distinct clans, similar to those observed at Amboseli National Park in Kenya (Moss and Poole 1983). Identification of clans was quantitatively supported by cluster analysis of the pairwise home-range correlations (Figs. 3, 4). Clans have also been observed in Asian elephants in southern India (Sukumar 2003). However, some populations of African savanna elephants and Asian elephants appear to lack geographically distinct clans (Thouless 1996; Fernando and Lande 2000).

It has been unclear from the Amboseli research whether clan “membership” implies any particular kinship relationships (Moss 1988, pp 132, 134). Douglas-Hamilton (1972) raised the question of whether shared home ranges were the result of matrilineal kinship, but concluded—lacking access to genetic data—that “only a lifetime of monitoring could produce a definite answer.” Our data on mitochondrial haplotypes show that clans at Sengwa are not composed of matrilineally related family groups. All three clans included animals of different haplotypes, and there was no significant association between two elephants having the same haplotype and being in the same clan. The genetic structure of clans in this population may not be typical of natural conditions, however, because of the history of repeated culling at Sengwa. Also, because we collared only a subset of all of the elephant families at Sengwa, we cannot draw conclusions about the overall degree of genetic variation within clans there. Further research will be required to determine whether African elephant clans in less disturbed populations tend to consist of more genetically similar individuals than expected by chance.

Spatial relationships among families within clans

Previous research has not addressed questions of how similar the ranges of elephants within a clan are, or whether there is any coordination between family groups in how their shared space is used over time. We found extensive variation among pairs within clans in the degree of home-range similarity. Among pairs of elephants certain to be in different family groups (the “restricted data set”), home-range similarities within clans varied between 0.312 and 0.798. Among these elephants, there was no significant relationship between haplotype sharing and home-range similarity. Indeed, four of the five highest home-range correlations were for pairs of elephants that had different haplotypes (Table 2).

Within clans, most pairs of elephants used their shared space independently, even in cases where their home ranges were extremely similar. For example, Computer (c6) and Crooked Tusk (c11) had the highest home-range correlation of all different-family pairs (HR correlation=0.798), with a median distance of only 1.7 km. Nevertheless, their independent distance distribution did not differ from that expected by chance (Fig. 5a). Furthermore, in 14 and 17 sightings of these 2 animals respectively, they were not seen together, supporting the inference that the frequent close proximity revealed by the radio-tracking data was simply the result of their similar compact home ranges centered on the same small water resource, and did not reflect any close social bond. In making this inference, we assume that close social bonds would be reflected in a non-random pattern of spatial association. We found no evidence of families actively avoiding each other. These observations are consistent with behavioral observations indicating that most family groups within a clan interact amicably when they meet, but exhibit no close social bonds (Moss 1988).

In contrast, 4 out of 15 pairs of elephants in different family groups seemed to coordinate their use of space by spending more time than one would expect by chance within hearing distance of each other’s low-frequency calls. Three of these four coordinated pairs had only moderately similar home ranges. For example, Piggy (c4) and Lutya (c8) (HR correlation=0.431; see Table 1, Fig. 5b) frequently occupied unshared portions of their ranges, but when they were both in shared areas, their movements often were coordinated for periods of hours or days at a time. Coordination of movement took several forms. In some cases, the animals would travel together (i.e., within 0.5 km of each other) or in parallel (often up to several kilometers apart), or one animal would follow the other at distances of up to several kilometers.

The most likely mechanism to account for such coordinated use of shared home ranges is low-frequency acoustic signaling. African elephants produce a variety of vocalizations with fundamental frequencies between 14 and 35 Hz, and source sound pressure levels of up to 117 dB re 20 µPa at 1 m (Poole et al. 1988). Playback experiments have demonstrated that elephants can detect some conspecific calls over distances of 2.5 to 4 km (Langbauer et al. 1991; McComb et al. 2003), and that females at Amboseli are typically familiar with the contact calls of around 14 other family groups, or around 100 individual elephants (McComb et al. 2000). Computer models of sound propagation suggest that the maximum detection range for elephant calls varies widely depending on atmospheric conditions, but can exceed 10 km under conditions that are not uncommon in African savanna environments (Larom et al. 1997). This range is large enough to cover all or most of each of the home ranges that we observed (Fig. 2).

The coordinated movements described here constitute a previously undocumented aspect of elephant social behavior. Given the close social relationships among families within bond groups, we might expect that bond groups would often use low-frequency acoustic signals to coordinate their movements when separated by several kilometers. Behavioral observations at Manyara and Amboseli suggest that the families within a bond group are typically members of the same matriline (Douglas-Hamilton 1972; Moss and Poole 1983; Moss 1988; C. Moss and J. Poole, personal communication re Amboseli Elephant Research Project long-term records). In the present study, however, three out of four coordinated pairs of elephants had different haplotypes, and were thus from different matrilines. Three explanations seem plausible. (1) Coordination of movements is not restricted to members of the same bond group. (2) Coordination of movements is restricted to bond groups, and non-matrilineal bond groups are generally more common than has been presumed. (3) Coordination of movements is restricted to bond groups, and non-matrilineal bond groups are more common at Sengwa than at other sites. Non-matrilineal bond groups might be more common at Sengwa because of the history of culling there. Although the policy at Sengwa has been to cull entire family groups, it is not known how often some individuals or sub-groups survive from culled families. Such cull survivors would be likely to establish social bonds with each other or with intact family groups, leading to the formation of apparent family groups and bond groups that do not represent single matrilines (Moss 1988; Nyakaana et al. 2001). Resolving which of these explanations is correct will require further study of genetic and spatial relationships among family groups at multiple research sites.

Another factor that undoubtedly influences elephants’ uses of space—and perhaps secondarily, their social relationships—is resource distribution. The lack of coordination between most pairs of elephants within a clan, coupled with the placement of core areas for all elephants at water sources, suggests that the locations of dry-season home ranges, and their tendency to form geographically distinct clusters or clans, reflect the patchy distribution of resources, rather than any centripetal social influence. Our results thus support the suggestion that clan “membership” may be an artifact of grouping families that fortuitously share dry-season home ranges (Moss 1988, pp 132, 134). Patchy resource distribution may also contribute to the variation in home-range size and shape that we observed, the largest home range being more than 6 times the size of the smallest. The home ranges in the Sengwa clan were significantly smaller than those in the other two clans, and all had a single activity peak centered on the largest and most reliable source of surface water in the study area. With one exception, families in the Manyoni and Lutope clans made relatively long treks between smaller water sources, resulting in large, irregular-shaped home ranges with multiple or very broad activity peaks. The exception is Jabula (c7) in the Lutope clan, who had a very small home range centered on a small (≈9 m2) permanent elephant-dug waterhole. These observations suggest that the sizes and shapes of dry-season home ranges at Sengwa are determined largely by the locations and quality of water sources. Additionally, the presence of a very young calf or calves may restrict the family’s movements to areas closer to a permanent water source. Long-term studies at Amboseli show that dry-season home ranges typically remain stable from year to year, suggesting that our data probably reflect longer-term patterns of dry-season space use, even though they were collected only during the latter (driest) portion of a single dry season. During the wet season, elephants at Sengwa are typically observed in larger groups, and their space use is less concentrated around a few distinct locations (Guy 1974; R.B. Martin, unpublished data). Similar differences between dry and wet seasons in patterns of space use and group size have been observed at several sites with highly seasonal rainfall (Moss and Poole 1983, and references therein).

In conclusion, our data show that dry-season home ranges of elephant family groups at Sengwa are clustered into geographically distinct areas occupied by clans that do not represent single matrilines. The occurrence of clans at this site is probably a result of patchy distribution of resources rather than social alliances, for within clans, most pairs of family groups use their shared ranges independently of each other. Nevertheless, clans provide the demographic framework within which most social interactions occur during the dry season. Specifically, clans may be important in defining the set of families with which any given family group is most familiar. Such familiarity is likely to affect the type of interactions that occur when families encounter each other (McComb et al. 2000). A few pairs of family groups within clans coordinate their use of space by staying within 4 km of each other more than expected by chance. Matrilineal relationship is not a prerequisite for these coordinated movements. Further research will be required to determine the relationship between coordinated movements and conventionally defined bond-group behavior.


This research was supported by grants from the National Science Foundation (grant no. BNS-8910482), World Wildlife Fund-US, the National Geographic Society (grant no. 3610-87), a Guggenheim Fellowship to Katharine Payne, and private contributions from the late John S. McIlhenny and Caroline Getty. The Zimbabwe Department of Parks and Wild Life Management made all of the facilities and personnel of the Sengwa Wildlife Research Institute (SWRI) available in support of the field work. This work would have been impossible without the able assistance in the field of the late Andrew Masarirevhu and twelve other game scouts of the SWRI. The late Ian Coulson, then director of the SWRI, and the entire staff of the SWRI supported this work logistically and scientifically. Loki Osborn and Lillie Wilson assisted with all aspects of the field work. Mike Kock supervised the immobilization of the elephants. Deborah Gibson flew aerial surveys. Helicopter time and fuel during collaring operations were donated by Mobil Oil Corporation. We thank Charles Walcott and Christopher Clark for their support of this work. Naomi Altman provided statistical advice. Kathy Dunsmore, Kurt Fristrup, Leila Hatch, Irby Lovette, Patricia Parker, Joyce Poole, Sandra Vehrencamp, and an anonymous reviewer made helpful comments on earlier versions of the manuscript. The research described here complies with the current laws of Zimbabwe.

Supplementary material

Fig. S1 Utilization distributions (UDs) for 14 radio-collared elephants in three clans. Grayscale darkness of the plot indicates probability of occurrence at a given point. The contour lines on each map encompass the innermost 50% and 95% of each animal’s UD. The values plotted here are those used to calculate home range correlation values; Figure 3 in the main text shows a simplified representation of these data. Labels (cN) indicate each animal’s collar number. Letters (A – E) indicate mtDNA haplotypes

supp.pdf (528 kb)
(PDF 529 KB)

Copyright information

© Springer-Verlag 2004