Rumble in the Jungle: Convolutional Neural Networks demonstrate accurate footfall identification of terrestrial mammals

Rumble in the Jungle: Convolutional Neural Networks demonstrate accurate footfall identification of terrestrial mammals This dataset comprises seismic footfall recordings collected from captive populations of four North American mammal species—Black Bear (Ursus americanus), Cougar (Puma concolor), Grey Wolf (Canis lupus), and White-tailed Deer (Odocoileus virginianus)—over a nine-day period at the Greater Vancouver Zoo. Recordings were captured using a custom-built seismic node featuring a 4.5 Hz geophone and processed through an automated pipeline to isolate individual footfall events. These data were used to train convolutional neural network models for species identification based on seismic signals. The raw seismograms were processed to detect events, extracting event clusters with at least 2s of event silence in-between, excluding clusters that contain only one event, and then centre-padded to achieve standardised 6s audio clips. The dataset is organised into training, testing and validation datasets for multiple allocations (21) in order to carry out Monte Carlo cross-validation. The filenames contain a timestamp in PST and are annotated with the following species codes: BB - Black Bear C - Cougar GW - Grey Wolf WTD - White-tailed Deer