Wild elephant (Loxodonta africana) breeding herds respond to artificially transmitted seismic stimuli
- First Online:
- Cite this article as:
- O’Connell-Rodwell, C.E., Wood, J.D., Rodwell, T.C. et al. Behav Ecol Sociobiol (2006) 59: 842. doi:10.1007/s00265-005-0136-2
- 273 Views
Seismic communication is known to be utilized in insects, amphibians, reptiles, and small mammals, but its use has not yet been documented in large mammals. Elephants produce low-frequency vocalizations, and these vocalizations have seismic components that propagate in the ground, but it has not yet been demonstrated that elephants can detect or interpret these seismic signals. In this study, we played back seismic replicates of elephant alarm vocalizations to herds of wild African elephants in their natural environment and observed significant behavioral changes indicating that they had detected these signals. Seismic communication may provide an important complement to existing communication modes used by elephants. Seismic sensitivity may also provide elephants with an additional modality for sensing important environmental cues such as changes in weather patterns or seismic disturbances.
KeywordsElephant seismic communicationLoxodonta africana
Both African and Asian (Elephas maximus) elephants produce high-amplitude [103 and 90 dB sound pressure level (SPL) at 5 m, respectively], low-frequency vocalizations in the range of 20 Hz (Poole et al. 1988; Payne et al. 1986), and Asian elephants are capable of detecting acoustic signals at 20 Hz when received above 60 dB SPL (Heffner and Heffner 1982). Although the threshold of hearing sensitivity for the African elephant is not yet known, it is reasonable to expect that it is similar to that of the Asian elephant, given the similar ear anatomy and vocalization frequency range. African elephants have also been shown to respond to acoustic playbacks of their own low-frequency vocalizations at a distance of up to 2 km (Langbauer et al. 1991). When both African and Asian elephants vocalize with these low-frequency rumbles, correlates of these vocalizations propagate in the ground (O’Connell-Rodwell et al. 2000a, 2001; Gunther et al. 2004).
The use of seismic cues for assessing the environment in the context of prey detection, predator avoidance, and communication has been documented over short distances in a diverse range of arthropods, amphibians, reptiles, and small rodent species (see O’Connell-Rodwell et al. 2000a for review). If elephants can detect and interpret the seismic replicates of their vocalizations as signals, they may be communicating and sensing the environment in a modality not previously considered.
Working under the hypothesis that the natural behavior of wild elephants would change significantly if they detected meaningful seismic stimuli, we played back seismic replicates of low-frequency elephant vocalizations to wild herds of African elephants in their natural environment in Etosha National Park and monitored their behavior for significant changes.
To demonstrate that elephants respond to seismic stimuli, we first chose to play back an elephant vocalization that had the greatest potential for a behavioral response. We then generated this vocalization seismically with enough amplitude to be detected by an elephant. Finally, the seismic signal had to be monitored to ensure that there was no coupling of the signal with the air.
Material and methods
Trials took place between 4 p.m. and 12 a.m. with an average temperature of 16.4°C and average wind speed of 4.6 km/h. Each trial began after breeding herds had sufficient time to drink (2–5 min), consisting of 5 min of control period, and 3 min of playback, where a 15-s string of alarm calls was played each minute, followed by 45 s of silence. Elephants were then monitored for 4 min after the trial or until they left the water hole.
All data were collected in real time in notebooks as well as on videotape for confirmation as needed and for later analysis. Night time trials were conducted through low light and generation 1 and 2 night vision binoculars (Zeiss 8×56, Bushnell 7×50 binoculars; Rigel Optics 3200 Generation 1 Night Vision goggles with a 3.6× lens conversion, Voyager 2nd Generation Night Vision Monocular with 1.5× magnification). A night vision lens was attached to the video recorder to document these trials. Herd size and composition (adult, three fourths, one half, one fourth, and babies; sizes are relative to adult females) were noted for each trial. This information was used to identify herds being tested to confirm that they were indeed independent groups. The direction in which herds arrived and departed from the water hole was also noted.
During the trials, we monitored four classes of behavioral data: herd spacing, vigilant behaviors (smelling, visual orientation toward source of stimulus, freezing, rumbling, foot lifting, and scanning), body orientation relative to the seismic signal source, and duration of time spent at the water hole. For a control on the duration of time spent at the water hole, we also measured the duration of time that 16 breeding herds spent at the water hole when no seismic signals were played back. The observer noting real-time behaviors did not know the onset and timing of the trials.
A Neumann KM131 microphone with a Sound Devices MM1 preamplifier was placed directly above each of the two 4.5-Hz Mark Products geophones to test for evidence of any acoustic correlate of our seismic playback signals (Fig. 1b). Recordings were made on a TASCAM 2 channel digital recorder. Background noise activity was measured during trials prior to turning on the shakers.
Behavioral analyses were based on real-time data collection in the field, and body orientation analysis was based on video analysis. Individual occurrences of each type of behavior were noted for each trial period and summed. This number was divided by the number of animals present in the herd and then divided by the minutes in that period to give a measure of occurrences per animal per minute. Another observer recorded the inter-individual spacing of the herd every 15 s during the trial by noting the most common inter-individual spacing between the pairs of animals. This was noted in three categories: less than 1 body length apart (0), 1 body length apart (1), and greater than 1 body length apart (2). Average body lengths apart were then calculated for each period (control and post-playback) by summing the recorded spacing (either 0, 1, or 2) and dividing by the number of recorded spacings for that period. The duration of time spent at the water hole was measured from the time of arrival of the whole herd at the water hole until the time that the first individual left the water hole. To standardize the duration of time spent at the water hole (because the start of playback trial varied from 2 to 5 min after the herds’ arrival), we subtracted the time from the onset of playback for each trial. For the controls, since there was no playback, we subtracted the average onset from the playback (183 s).
We predicted that the inter-individual spacing between pre-playback and post-playback periods would decrease and that vigilance behaviors would increase due to an elevation in defensive behaviors in reaction to the nature of the seismic stimulus. We therefore used a directional Wilcoxon signed rank test to evaluate these variables (as the direction of the outcome was predictable a priori).
An analysis of covariance (ANCOVA) was used to determine if the occurrence of a seismic playback had an effect on the time an elephant herd spent at the water hole. We also included potential confounding variables such as herd size in the model.
Video analyses were conducted on the seismic playback trials to compare the body orientation of elephants before and during stimulus presentation. Videotapes were digitized and analyzed (Noldus Observer software) for body orientation (north, south, east, or west). The video analysis team was ignorant of the purpose, details, and timing of the experiment.
Video data were collected only on adult elephants (defined as any individual whose size was at least 75% of the size of the largest elephant in the group). Because the entire group was filmed on one camera, elephants were distant in the camera’s view and often obscured by one another; orientation was determined by a suite of cues, including the direction of trunk and ears, leg locations, and the overall trajectory of the spine from the head to the base of the tail. Each trial was coded using instantaneous scan samples every 10 s. Data were subsequently split into the control period (5 min) before playbacks began and the experimental period (3 min). Means were calculated for the average proportion of elephants facing each direction per interval for the control and experimental periods. A chi-square test was performed to determine whether elephants were randomly oriented before and during the presentation of the seismic stimulus.
To calculate the presentation levels of the seismic signals and to confirm that there was no signal in the air during seismic presentations, the field-recorded geophone and microphone signals for each trial were transferred from the digital audiotape (DAT) recorder and calibrated as shown below.
Geophone and microphone calibrations were performed a few degrees higher than the average field trial temperature (20 vs 16°C), and the equipment was set up in an identical manner to that in the field, albeit with shorter cables. The microphone was calibrated using a Bruel & Kjaer (type 4320) calibrator. The calibrator tone frequency of 1 kHz, at 94 dB SPL (re 20 μPa), was recorded onto a TASCAM DAT (digital) and converted to computer .wav files via a Digigram Vxpocket interface using Cool Edit Pro. Spectrum analyses were performed using customized software. A scaling factor of 6.83 Pa/full-scale resulted in 94 dB SPL for the calibration tone. Because the microphone sensitivity is flat to within 2 dB in the 20-Hz to 20-kHz region, the calibration at 1 kHz is assumed valid in the 20-Hz to 20-kHz region. Below 20 Hz, the microphone sensitivity decreases.
The geophone was calibrated by comparing its response to an accelerometer. The geophone and a Bruel & Kjaer accelerometer (type 4381) were mounted within 9 cm on a 91×54×0.5-cm solid aluminum table, mounted with a Buttkicker 5- to 200-Hz shaker below, and powered by a Buttkicker 400-W preamplifier. The geophone signal was simultaneously fed into a TASCAM DAT recorder, while the accelerometer signal was fed into a Bruel & Kjaer charge amplifier (type 2635), selected to convert accelerations to velocities. The output of the charge amplifier was the input to the SYSid (Version 7.0) measurement and analysis system (Puria et al. 1993). A tone at 100 Hz was synchronously generated by SYSid to vibrate the table. The DAT signal was transferred to a computer, and spectrum analyses were performed. A scaling factor of 423 μm/s RMS/full-scale resulted in the same geophone level as that from the accelerometer. The geophone is expected to be flat to within 0.6 dB in the 10- to 100-Hz frequency region.
The silent periods between signals within the 15-s playback were deleted (i.e., 4–7 s in Fig 1b), and the 10-s segment from each signal was combined across the three playbacks, resulting in a 30-s continuous signal that was spectrally averaged. The analysis consisted of windowing 16,384 points, in window lengths of 372 ms, and taking its Fast Fourier transform (FFT), with a corresponding frequency resolution of 2.7 Hz, and squared to convert to energy. In a 30-s period, about 80 windowed energy buffers were averaged at each frequency point. The averaged energy components were integrated across approximately one-sixth-octave band frequencies following Heffner and Heffner (1982). Additionally, all energy components were summed to obtain the decibel SPL as a function of time.
As an additional confirmation that there was no airborne coupling in any of the trials, we ran a series of matched filters on the microphone recordings during our seismic playback trials, using the playback signal recorded in the NG to create the matched filter. The matched filtering was done in Matlab 7.0 after the recordings were down-sampled from 44,100 to 1,000 Hz. The matched filter output was displayed as a spectrogram for visual inspection. All spectrograms were calculated using an FFT of 512 and a Hanning window with 98% overlap. There was no evidence of coupling in any of the trials (data not shown).
The seismic signals were successfully played back to 11 different elephant breeding herds over a 3-week period. A total of three trials were excluded from the herd spacing, vigilance behavior, and body orientation analyses due to either the arrival of another herd or the herd leaving before the start of the postplayback period.
The duration of time that herds spent at the water hole while exposed to the seismic alarm calls (n=11, mean herd size 19.18, range 3–35; Fig. 3) was significantly less than the duration of time spent at the water hole by herds that were not exposed to the seismic stimulus (n=16, mean herd size 18.5, range 6–32; Fig. 3; F1,24=8.11, p=0.009). The effect of herd size on the duration of time at the water hole was not significant (F1,24=3.74, p=0.065). There was no interaction between herd size and treatment (F1,23=0.02, p=0.889), and herd size was therefore excluded from the final ANCOVA model.
SNR calculations for all eight trials at 10 m, 90 m and at the elephant distance of 25 m from the seismic source
Signal level (dB re 1μm/s)
Noise alpha (/m)
Noise at source (10−6m3/2/s)
Noise level (dB re 1μm/s)
Noise level (dB re 1μm/s)
Signal level (dB re 1μm/s)
Noise level (dB re 1μm/s)
At the location of the elephants, the SNR is expected to be significantly lower due to a decrease in seismic signal level and an increase in the background activity level. When averaged across all eight trials, the mean seismic signal level is −13.6 dB re 1μm/s, and the mean background activity level is −12 dB re 1 μm/s for an overall SNR at the location of the elephants of −1.6 dB. There was no detectable acoustic signal from the shakers in the microphones in the frequency range we played back (Fig. 4b). In some trials, there was a slight increase in acoustic activity in the 100- to 500-Hz frequency range during presentation, most likely due to increased animal activity in response to the seismic stimuli, as this noise persisted following the presentation of the 15-s seismic signal (Fig. 4c). In the 20-Hz to 1.5-kHz range, the acoustic signal at both locations was typically near the threshold of hearing for Asian elephants (Fig. 4b,c).
In this study, we played back seismic replicates of elephant alarm vocalizations to wild elephant herds in their natural environment in Etosha National Park, Namibia, and observed significant behavioral changes in all eight of the herds tested. Elephants responded to the seismic stimuli presented in this study by clumping into tighter groups, changing their orientation from a random orientation to perpendicular to the playback source, and by leaving the water hole sooner than the control herds that were not presented with seismic stimuli.
Although the seismic alarm call playbacks did not elicit the same dramatic response as when they were transmitted acoustically, they nevertheless elicited significant defensive behaviors (clumping, orientating in a nonrandom direction, and leaving early from the water hole). Since breeding herds actively protect their young (Dublin 1983), alarm calls from breeding herds would be of particular relevance. In this case, receiving a seismic signal, but not its acoustic counterpart, may have been interpreted by the elephants as a signal originating from a distance source, which could explain why their responses were not as dramatic as when they detected the acoustic counterpart of the signal. Another possible explanation for the responses being less dramatic is that the elephants may have detected the signal but may not have been able to interpret it correctly. While elephants in this study appeared to respond appropriately to the meaning of the antipredator signals by taking defensive postures, orienting and leaving, the study was not designed to assess whether elephants can discriminate meaning from seismic stimuli, and thus, this is yet to be determined.
Humans are able to detect a signal with an SNR of −2 dB, below 200 Hz (Zwicker and Feldtkeller 1999), and it is reasonable to assume that elephants have at least this level of signal detection capability and probably better at the lower frequencies (see Gunther et al. 2004 for review). With regard to the strength of the seismic stimuli, the average SNR was low at the location of the elephants (mean −1.6 dB), but, since elephants were positioned in a range of anywhere from 10 to 25 m away from the source, at least some elephants would have been in a detectable range.
While a mechanism for seismic detection in elephants has not yet been elucidated, one possible pathway of detection would be through bone conduction to the elephants’ ears via the toes, as the hypertrophied malleus of the elephant middle ear is well suited to facilitate bone conducted seismic cues (Reuter et al. 1998). It has also been suggested that seismic signals could be detected through as-yet-undescribed somatosensory structures in the feet (O’Connell et al. 1999) or by using an organ, such as the trunk, that has an extraordinary number of pacinian and Meissner’s corpuscles sensitive to low-frequency vibrations (Rasmussen and Munger 1996) and could be placed on the ground at appropriate times (unpublished data).
This study provides the first demonstration that a large mammal can detect seismically propagated playbacks of conspecific vocalizations. Limitations in the technology needed to recreate seismic signals artificially in the field meant that we had to keep the distance between our seismic source and the elephants within the range of about 30 m. The range over which a natural seismic signal created by an elephant might be detected, however, has yet to be determined empirically. Our most recent field studies, supplemented by modeling, indicated that elephants might be able to detect seismic vocalizations as far as 2 km away from their source (Gunther et al. 2004), while previous models predicted a maximum possible range of 16 km, depending on soil transmission characteristics (O’Connell-Rodwell et al. 2000a). Seismic cues could supplement acoustic information, replace acoustic information under poor atmospheric conditions, or under ideal seismic conditions, and may even extend the elephant’s range of communication.
If elephants can monitor both the airborne and seismic components of other elephants’ vocalizations simultaneously, they could also receive information regarding the distance and bearing of the vocalizing animal. Seismic signals most often travel at a different velocity than their acoustic counterparts, which produces a differential time-of-arrival cue that increases with distance from the source and maximizes the potential for determination of range and bearing of the vocalizing individual. At two of our study sites, we measured a 17-m acoustic wavelength and a 12-m seismic wavelength at 20 Hz, which indicate ideal conditions for signal localization, 12-m wave having a more discrete phase angle, providing the feet or the ears better information on locating the source of the signal.
Intraspecific eavesdropping on acoustic, visual, or chemical cues to facilitate the detection of relevant signals has been described as an advantageous strategy in a variety of contexts. Examples include object assessment by dolphins (Xitco and Roitblat 1996), food location by whales (Madsen et al. 2002), territoriality in birds (Peake et al. 2002; Dabelsteen et al. 1997), combat between fighting fish (Oliveira et al. 1998), predation by bats (Heller 1995), and predation and parasitization in arthropods (Stowe et al. 1995). Seismic eavesdropping is also thought to be a successful strategy for predation on termites by the golden mole (Narins et al. 1997), on kangaroo rats by snakes (Randall and Matocq 1997), and in territoriality, predator avoidance and mate finding (Cokl and Virant-Doberlet 2003), predation (Sturzl et al. 2000), or parasitization (Fischer et al. 2001) in many arthropod species. Eavesdropping by keeping an “ear” to the ground could pay off for the elephant as well in the context of localizing a caller and estimating distance from the caller, coordinating herd movements, mate finding, predator avoidance, or detecting substrate changes to facilitate finding water or possibly even changes in weather patterns and seismic disturbances.
We would like to acknowledge the National Geographic Society, US Fish and Wildlife Service, the Seaver Institute, Stanford University’s Morrison Institute and Bio-X Interdisciplinary Research award, as well as University of California at Davis internal grants. We would also like to acknowledge the Namibian Ministry of Environment and Tourism, particularly Pauline Lindeque for supporting the project, Conrad Brain for his field assistance, and Ginger Mauney for her logistical support. We thank the Namibia Nature Foundation for ground and administrative support in Namibia. We thank Heinrich Steuber of SolTec, c.c., in Windhoek, Namibia for his generous support. We also acknowledge the videography of Lianna Jarecki, the statistical consultation of Neil Willits, software advice of Joe Olson, and technical support of John McCormack of Rainbow Electronics and Marvin Clamme of The Guitammer Company. We thank Denise Nicholas for her input on video analysis. We would also like to thank Ben Hart, Karen McComb, and Don Owings for their helpful comments on earlier manuscripts.