Towards seismic monitoring of terrestial ecosystems: an exploratory data analysis of the SeisSavanna dataset

Our planet experiences ongoing unrest across various scales, from human footsteps to the powerful forces of volcanic eruptions and megathrust earthquakes. Seismic sensors, typically employed for geophysical studies, record diverse phenomena, including ground vibrations caused by the movement of terrestrial animals, known as footfall signals. The recently released SeisSavanna dataset comprises approximately 70,637 footfall signals from 11 different species in the African savanna. Consequently, ground-based vibrations might represent an underexplored sensory mode for continuously monitoring habitat usage and undisturbed animal behavior. To gain a deeper understanding of footfall signals, we conduct exploratory data analysis on the SeisSavanna dataset. Utilizing a scattering transform, we capture the distinctive features of footfall signals, creating a high-level and interpretable data representation for subsequent analyses. Seismogram atlases and clustering enable us to group similar types of footfall signals and investigate the signal-altering path and site effects, providing a comprehensive overview of the entire dataset. Moreover, this data-driven approach serves as a quality check for the species labels retrieved from co-located camera traps with a limited angle of view.