Module 5: Infrasound & Seismic Communication

In 1984, Katy Payne — who had previously discovered humpback-whale song harmonic structure — noticed a faint throbbing in the air while visiting the Portland, Oregon zoo. Spectral analysis confirmed that the resident Asian elephants were producing acoustic energy below 20 Hz, inaudible to humans but radiating up to ~105 dB at 1 m from the source. Payne, Langbauer, and Thomas (1986, Behavioral Ecology and Sociobiology) published the first definitive account of elephant infrasound — overturning a century of assumption that megaherbivore vocal repertoires were limited to the audible trumpeting and bellowing familiar from zoos.

Subsequent field recordings in Amboseli (Poole, Langbauer, Payne, and Moss 1988) and Etosha (Langbauer 1991) confirmed that the core elephant vocal repertoire is dominated by “rumbles” with fundamental frequencies of 14–24 Hz and significant harmonic content extending into the audible band up to ~120 Hz. Total bandwidth per call is ~1–2 octaves; durations range from 0.5 to 10 s.

The vocal folds of an adult elephant measure ~7 cm in length — roughly three times the human vocal fold, and the body-vibration mode of the entire larynx (not just the folds) contributes to the lowest frequencies. Herbst et al. (2012) demonstrated using an excised elephant larynx preparation that the phonation is aerodynamically driven, powered by the same myoelastic-aerodynamic theory that governs human phonation.

Atmospheric absorption of acoustic energy is dominated by two mechanisms: classical viscothermal losses(molecular diffusion of momentum and heat away from compressional wavefronts) and molecular vibrational relaxation (energy exchange with the rotational-vibrational modes of O₂ and N₂, modulated by humidity). Both produce attenuation coefficients that scale as f², so low-frequency sound propagates much further than high-frequency sound.

At 20 Hz, typical daytime attenuation is on the order of \(10^{-7}\) dB/m — negligible — meaning the range is set almost entirely by geometric spreading (20 log₁₀ r) and atmospheric refraction. At 2 kHz the same air attenuates at \(10^{-3}\) dB/m, roughly 10 000× stronger, so a 2 kHz whistle cannot compete over long range with a 20 Hz rumble.

Refractive ducting at dawn and dusk

A second atmospheric effect is even more important over kilometres: temperature-inversion ducting. Larom, Garstang, and colleagues (1997, Journal of Experimental Biology) observed that African elephant communication range expands dramatically in the cool dawn and dusk hours because the atmospheric temperature profile inverts (cool air near the ground, warmer air aloft). This bends sound rays downward, trapping acoustic energy in a surface duct that converts spherical spreading (−20 log₁₀ r) to cylindrical spreading (−10 log₁₀ r). The result is an effective detection range of 10–20 km for a 110 dB source, compared with ~3 km during midday upward-refracting conditions.

Correspondingly, elephants concentrate their long-distance vocal communication into the dawn and dusk windows — behavioural calibration to the atmospheric physics. Poole (1987) and Langbauer (1991) documented dramatic bimodal vocal-call diel patterns that correlate almost perfectly with the periods of favourable atmospheric ducting.

The same vocal-cord pulsations that radiate into the atmosphere also couple into the ground through the body-wall contact and through the feet, producing Rayleigh surface waves that propagate radially away from the animal. Foot-stomps and musth-rumble vibrations couple even more strongly.

Rayleigh waves are a fundamental solution of the elastic-half-space wave equations: particle motion is retrograde-elliptical in the vertical plane, wave energy decays exponentially with depth (concentrated within one wavelength of the surface), and the horizontal phase velocity is close to but slightly slower than the shear-wave speed:

For savannah hardpan with shear-wave speed v_S ≈ 250 m/s, the Rayleigh speed is ~230 m/s. A 20 Hz vibration therefore has a seismic wavelength of ~12 m. Attenuation along the surface goes as cylindrical spreading (\(-10\log_{10} r\)) plus anelastic decay \(\alpha = \pi f / (v_R Q)\) where the quality factor Q is ~30 for dry sandy soils.

O’Connell-Rodwell et al. (2000, Animal Behaviour; 2006, Journal of the Acoustical Society of America) measured the ground velocity 2 km from a trumpeting elephant at ~3×10⁻⁷ m/s — comfortably above the elephant Pacinian detection threshold — and inferred a maximum seismic detection range of approximately 16 km for alarm-intensity calls, roughly comparable to the atmospheric-inversion range.

The elephant’s footpad is a gel-like cushion of fibrous adipose tissue permeated by a dense array of Pacinian corpuscles — lamellar mechanoreceptors that transduce skin deformation into afferent nerve firings. Bouley et al. (2007) counted roughly 2 000 Pacinian corpuscles per footpad in Loxodonta, an order of magnitude more than in human fingertips or feline paws, and distributed so as to optimally sample vertical ground motion.

Each corpuscle is a 1–4 mm ellipsoid of concentric lamellae enclosing a single mechanosensitive nerve ending. The lamellae act as a mechanical high-pass filter, damping slow static loads and transmitting rapid transients through to the central axon. In humans the peak sensitivity is at ~250 Hz; in elephants O’Connell-Rodwell (2007) and Reinwald (2021) show it has shifted down to 20–40 Hz, matching the elephant call band almost exactly.

The afferent signals travel up the trigeminal/spinal somatosensory tracts to a specialised enlarged somatosensory cortical representation of the foot in the posterior parietal lobe (Shoshani et al. 2006). Functional MRI on captive Asian elephants (Dahle et al. 2019) confirmed hemodynamic activation of foot-dedicated cortex during playback of seismic-delivered species-specific calls.

Behaviourally, elephants exposed to playback alarm rumbles deliveredonly through the substrate (no airborne component) respond with appropriate freezing, bunching, and vigilance postures — demonstrating that the seismic channel alone is informative enough to drive defensive responses (O’Connell-Rodwell 2006, 2007). Discrimination between alarm and contact rumbles has been demonstrated with seismic-only playback, and response varies with transmitter identity (matriarch vs unfamiliar).

Poole (1987, 1999) and the Amboseli Elephant Research Project have catalogued 31+ distinct call types in the African elephant vocal repertoire. Major functional categories include:

• Contact rumble — the most common call, used to maintain cohesion of a matrilineal family group during movement. Fundamental ~14 Hz, 2–4 s.
• Let’s-go rumble — initiated by the matriarch to signal departure. Often triggers group co-ordinated movement within 30 seconds.
• Greeting ceremony rumble — produced when related family groups reunite after separation. Accompanied by trumpeting, touching, and secretions from the temporal gland.
• Alarm rumble / trumpet — loud, wideband, often harmonically rich above 100 Hz. Triggers freezing or mobbing response.
• Mobbing rumble — low-frequency broadband with characteristic rolling modulation, produced when a predator (lion, human) is identified.
• Estrus rumble — produced by females in oestrus; attracts bulls from many kilometres.
• Musth rumble — a deep-frequency pulsed call produced by bulls in musth (Module 7); advertises reproductive readiness and intimidates non-musth bulls.
• Discipline rumble — matriarch signal to younger group members; usually accompanied by a physical shove or trunk slap.

Each call category has a distinctive spectrotemporal signature. Stoeger and colleagues (2012) applied machine-learning classifiers to field recordings and achieved >95% call-type discrimination accuracy, comparable to human experts.

McComb et al. (2000, 2001) performed a landmark set of playback experiments in Amboseli National Park. Using recordings of individual elephant rumbles played back from concealed speakers, they established that family groups reliably distinguish the calls of hundreds of individual conspecificsby voice alone. The response — approach, defensive bunching, or indifference — depends on the social relationship between the caller and the receiver.

Most remarkably, McComb et al. (2001, Science) showed that recognition of a matriarch’s calls persists in her family for approximately two years after her death. Surviving family members still respond with approach-and-greet behaviour to playbacks of her recorded voice well after she has passed — documenting long-term auditory memory in a non-human mammal, and providing one of the most moving experimental demonstrations of grief-related cognition in wildlife (foreshadowing Module 6).

Population-specific vocal dialects are now known to exist (Poole 2011; Soltis 2010). Calves raised in a different population from their biological origin acquire the dialect of the foster population — strong evidence of cultural transmissionof vocal form, a rare phenomenon outside cetaceans and songbirds.

Shoshani, Kupsky, and Marchant (2006) documented the gross and cytoarchitectural anatomy of the African elephant brain. Noteworthy features include:

• Massive overall brain mass (~5 kg), second only to sperm whales among mammals.
• Large temporal lobe with expanded primary and secondary auditory representations.
• Enlarged representation of the footpad somatosensation, with cortex directly opposite the contralateral pedal region showing disproportionate volume.
• Dense interhemispheric integration, possibly supporting binaural and bipedal seismic-wave localisation.

The elephant brain therefore appears anatomically pre-adapted to integrate atmospheric and seismic streams of the same acoustic signal. Exactly how a vocalisation arriving at the two modalities with different time delays (air 343 m/s, seismic 230 m/s) is fused into a unified percept is an active area of research.

Classical acoustic source localisation relies on interaural time and level differences, which depend on wavelength being comparable to or smaller than head width. For a 20 Hz airborne signal, the wavelength is ~17 m — many times larger than the elephant’s head — and atmospheric binaural cues are therefore extremely weak.

Seismic Rayleigh waves, however, travel at ~230 m/s, giving a wavelength of ~12 m at 20 Hz. The elephant’s foot-to-foot span is 2–3 m, so inter-foot time differences of 4–13 ms are available to the central nervous system. O’Connell-Rodwell et al. (2007) argued that elephants triangulate seismic sources using precisely such a four-foot array, analogous to a seismic geophone network. Behavioural orientation toward seismic sources has been observed in playback experiments (Gunther 2004).

Localisation accuracy is improved by deliberate postural adjustments: the animal often leans, lifts one foot, or subtly shifts stance to sample the wave at different body points. The trunk tip, also rich in Pacinian corpuscles, may act as an additional mobile seismic probe pressed against the ground when localising weak signals.

The figure below summarises the full source-to-percept signal flow for an elephant alarm rumble. The vocal fold generates a pulsatile pressure wave that simultaneously couples into the atmosphere (through the trunk and mouth) and into the substrate (through the chest wall and forelimbs). The two streams propagate with different speeds and different attenuation laws, and are re-integrated in cortex.

A critical feature of this dual-channel transmission is the time offset between the two arrival modes. At 10 km range, atmospheric arrival is ~29 s after the source event, and seismic arrival is ~43 s — a 14 s lag. This difference could in principle encode range directly if the elephant’s neural circuitry performs cross-modal correlation; there is preliminary behavioural evidence supporting this hypothesis (O’Connell-Rodwell 2006) but no definitive neurophysiological demonstration yet.

Understanding elephant infrasound has direct conservation applications. Bioacoustic monitoring arrays deployed in Central African forest (Wrege et al. 2017) use passive infrasonic loggers to survey elephant populations in dense canopy where visual methods fail. Machine-learning classifiers distinguish elephant rumbles from gunshots (relevant for poaching early warning), forest-elephant vs savannah calls, and call types.

Seismic-network analogues are under active development. Reinwald et al. (2021, Current Biology) demonstrated that a small geophone array can detect and classify elephant activity at ranges exceeding 100 m even in the presence of moderate anthropogenic noise. Extended to a park-wide network, such a system could detect and localise poaching activity by monitoring unusual seismic signatures (gunshots, fleeing elephants).

Finally, the 2004 Indian Ocean tsunami anecdote — elephants in Thailand moving to high ground approximately one hour before the tsunami struck — is consistent with long-range infrasonic/seismic detection of the initial underwater earthquake. Whether this reflects true directional sensitivity or simply reaction to an unusual low-frequency ambient field remains debated, but it is at least plausible given the sensitivity limits we have derived.