Classifying elephant behavior with seismic detection and modeling

Geological, environmental, and biological sources continuously excite the Earth's surface at frequencies from free oscillations (mHz) to microcracks (kHz). Seismology is well equipped to disentangle these complex vibrations owing to the efficient long-range propagation of elastic waves, sensitive instrumentation and advanced modeling and inversion techniques. Here, we present an application of modern seismic techniques such as full-waveform modeling, data processing, stacking, and deconvolution to detect and identify signals from elephants in Kenya. We argue that such signals can in principle be detectable at a kilometer range for realistic geological settings and noise environments. This allows us to remotely classify different elephant behaviors such as walk, run, and rumble. This opens doors to non-intrusive seismic monitoring of wildlife in the savannah, for instance to address the poaching problem. Future directions include deployments of denser arrays and economic accelerometers, machine-learning techniques for classification, and advances on the physiological and ecological side to understand possible two-way communication based on elastic wave propagation.