Decoding the footsteps of the African savanna: Classifying wildlife using seismic signals and machine learning

Methods in Ecology and Evolution, Год журнала: 2025, Номер unknown

Опубликована: Март 17, 2025

Abstract In recent years, seismic sensors, traditionally used in geophysical studies, have been utilized to record waves generated by wildlife locomotion, providing new ways monitor non‐invasively and continuously. Given the novelty of this approach, numerous research questions, unexplored potentials, methodological challenges remain. study, we investigate signal properties African savanna species during locomotion employ machine learning techniques classify based on these footfall signals. We utilize SeisSavanna dataset, which contains over 70,000 three‐component seismograms paired with labelled images from co‐located camera traps. To create a graphical overview entire combine scattering transform uniform manifold approximation projection (UMAP). While different categories display patterns, local geological conditions known as site effects significantly alter frequency content those address issue effect, trained models data recorded various sites. For multi‐class classification task involving signals elephants, giraffes, hyenas, zebras, achieved balanced accuracy 87% at maximum animal‐sensor distance 50 m. The decreases 77% when is extended 150 m due decreasing label quality. demonstrate that can generalize stations if similar are present training data. Our results indicate potential for using monitoring conservation, complementing other existing passive sensors such traps or acoustic loggers observables about silent species. However, further advancements larger datasets essential approach become reliable tool monitoring.

Язык: Английский

Probing the evolution of fault properties during the seismic cycle with deep learning

Nature Communications, Год журнала: 2024, Номер 15(1)

Опубликована: Ноя. 20, 2024

We use seismic waves that pass through the hypocentral region of 2016 M6.5 Norcia earthquake together with Deep Learning (DL) to distinguish between foreshocks, aftershocks and time-to-failure (TTF). Binary N-class models defined by TTF correctly identify seismograms in test > 90% accuracy. raw records as input a 7 layer CNN model perform classification. Here we show DL successfully pre/post mainshock accord lab theoretical expectations progressive changes crack density prior abrupt change at failure gradual postseismic recovery. Performance is lower for band-pass filtered (below 10 Hz) suggesting learn from evolution subtle elastic wave attenuation. Tests verify our results indeed provide proxy fault properties included trained wrong time those using far mainshock; both degraded performance. Our demonstrate have potential track zone during cycle. If this result generalizable it could improve early warning hazard analysis. Artificial Intelligence technique based on used differentiate before after earthquake. The classifies aftershocks, time-to-failure, providing insights into how earthquakes.

Язык: Английский