Seismological data can provide timely information for slope failure hazard assessments, among which rockfall waveform identification is challenging for its high waveform variations across different events and stations. A rockfall waveform does not have typical body waves as earthquakes do, so researchers have made enormous efforts to explore characteristic function parameters for automatic rockfall waveform detection. With recent advances in deep learning, algorithms can learn to automatically map the input data to target functions. We develop RockNet via multitask and transfer learning; the network consists of a single-station detection model and an association model. The former discriminates rockfall and earthquake waveforms. The latter determines the local occurrences of rockfall and earthquake events by assembling the single-station detection model representations with multiple station recordings. RockNet achieves macro F1 scores of 0.990 and 0.981 in terms of discriminating earthquakes and rockfalls from other events with the single-station detection and association models, respectively.

The seismic waves emitted during granular flows are generated by different sources: high frequencies by inter-particle collisions and low frequencies by global motion and large scale deformation. To unravel these different mechanisms, an experimental study has been performed on the seismic waves emitted by dry, dense, quasi-steady granular flows. The emitted seismic waves were recorded using shock accelerometers and the flow dynamics were captured with a fast camera. The mechanical characteristics of the particle collisions were analyzed, along with the intervals between collisions and the correlations in particles’ motion. The high-frequency seismic waves (1-50 kHz) were found to originate from particle collisions and waves trapped in the flowing layer. The low-frequency waves (20-60 Hz) were generated by particles’ oscillations along their trajectories, i.e. from cycles of dilation/compression during coherent shear. The profiles of granular temperature (i.e. the mean squared value of particle velocity fluctuations) and average velocity were measured and related to each other, then used in a simple steady granular flow model, in which the seismic signal consists of the variously attenuated contributions of shear-induced Hertzian collisions throughout the flow, to predict the rate at which seismic energy was emitted. Agreement with the measured seismic power was reasonable, and scaling laws relating the seismic power, the shear strain rate and the inertial number were derived. In particular, the emitted seismic power was observed to be approximately proportional to the root mean square velocity fluctuation to the power $3.1 \pm 0.9$, with the latter related to the mean flow velocity.

Seismic waves generated by rockfalls contain valuable information on the properties of these events. However, as rockfalls mainly occur in mountainous regions, the generated seismic waves can be affected by strong surface topography variations. We present a methodology for investigating the influence of topography using a Spectral-Element-based simulation of 3D wave propagation in various geological media. This methodology is applied here to Dolomieu crater on the Piton de la Fournaise volcano, Reunion Island, but it can be used for other sites, taking into account local topography and medium properties. The complexity of wave fields generated by single-point forces is analyzed for different velocity models and topographies. Ground-motion amplification is studied relative to flat reference models, showing that Peak Ground Velocity (PGV) and total kinetic energy can be amplified by factors of up to 10 and 20, respectively. Simulations with Dolomieu-like crater shapes suggest that curvature variations are more influential than depth variations. Topographic effects on seismic signals from rockfalls at Dolomieu crater are revealed by inter-station spectral ratios. Results suggest that propagation along the topography rather than source direction dominates the spectral ratios and that resulting radiation patterns can be neglected. The seismic signature of single rockfall impacts is studied. Using Hertz contact theory, impact force and duration are estimated and then used to scale simulations, achieving order-of-magnitude agreement with observed signal amplitudes and frequency thresholds. Our study shows that combining Hertz theory with high-frequency seismic wave simulations on real topography improves the quantitative analysis of rockfall seismic signals.

Automatic identification of debris flow signals in continuous seismic records remains a challenge. To tackle this problem we use a machine learning approach, which can be applied to continuous real-time data streams. We show that a machine learning model based on the random forest algorithm recognizes different stages of debris flow formation and propagation at the Illgraben torrent, Switzerland, with an accuracy exceeding 90%. In contrast to typical debris flow detection requiring instrumentation installed directly in the torrent, our approach provides a significant gain in warning times of tens of minutes to hours. For real-time data streams from 2020, our detector raises alarms for all 8 independently confirmed Illgraben events and gives no false alarms. We suggest that our seismic machine-learning detector is a critical step towards the next generation of debris-flow warning, which increases warning times using both simpler and cheaper instrumentation compared to existing operational systems.