Deep learning forecasting of large induced earthquakes via precursory signals.

Precursory phenomena to earthquakes have always attracted researchers' attention. Among the most investigated precursors, foreshocks play a key role. However, their prompt identification with respect to background seismicity still remains an issue. The task is worsened when dealing with low-magnitude earthquakes. Despite that, seismology and, in particular real-time seismology, can nowadays benefit from the use of Artificial Intelligence (AI) to face the challenge of effective precursory signals discrimination. Here, we propose a deep learning method named PreD-Net (precursor detection network) to address precursory signal identification of induced earthquakes. PreD-Net has been trained on data related to three different induced seismicity areas, namely The Geysers, located in California, USA, Cooper Basin, Australia, Hengill in Iceland. The network shows a suitable model generalization, providing considerable results on samples that were not used during the network training phase of all the sites. Tests on related samples of induced large events, with the addition of data collected from the Basel catalogue, Switzerland, assess the possibility of building a real-time warning strategy to be used to avoid adverse consequences during field operations.

Ground motion prediction equations as a proxy for medium properties variation due to geothermal resources exploitation.

Sub surface operations for energy production such as gas storage, fluid injection or hydraulic fracking modify the physical properties of the crust, in particular seismic velocity and anelastic attenuation. Continuously measuring these properties may be crucial to monitor the status of the reservoir. Here we propose a not usual use of the empirical ground-motion prediction equations (GMPEs) to monitor large-scale medium properties variations in a reservoir during fluid injection experiments. In practice, peak-ground velocities recorded during field operations are used to update the coefficients of a reference GMPE whose variation can be physically interpreted in terms of anelastic attenuation and seismic velocity. We apply the technique to earthquakes recorded at The Geysers geothermal field in Southern California and events occurred in the St. Gallen (Switzerland) geothermal field. Our results suggest that the GMPEs can be effectively used as a proxy for some reservoir properties variation by using induced earthquakes recorded at relatively dense networks.

Subduction age and stress state control on seismicity in the NW Pacific subducting plate.

Intermediate depth (70-300 km) and deep (> 300 km) earthquakes have always been puzzling Earth scientists: their occurrence is a paradox, since the ductile behavior of rocks and the high confining pressure with increasing depths would theoretically preclude brittle failure and frictional sliding. The mechanisms proposed to explain deep earthquakes, mainly depending on the subducting plate age and stress state, are generally expressed by single parameters, unsuitable to comprehensively account for differences among distinct subduction zones or within the same slab. We analyze the Kurile and Izu-Bonin intraslab seismicity and detail the Gutenberg-Richter b-value along the subducted planes, interpreting its variation in terms of stress state, analogously to what usually done for shallow earthquakes. We demonstrate that, despite the slabs different properties (e.g., lithospheric age, stress state, dehydration rate), in both cases deep earthquakes are restricted to depths characterized by equal age from subduction initiation and are driven by stress regimes affected by the persistence of the metastable olivine wedge.

Clock advance and magnitude limitation through fault interaction: the case of the 2016 central Italy earthquake sequence.

Faults communicate with each other. Strong earthquakes perturb stress over large volumes modifying the load on nearby faults and their resistance to slip. The causative fault induces permanent or transient perturbations that can change the time to the next seismic rupture with respect to that expected for a steadily accumulating stress. For a given fault, an increase of stress or a strength decrease would drive it closer to - or maybe even trigger - an earthquake. This is usually perceived as an undesired circumstance. However, with respect to the potential damage, a time advance might not necessarily be a bad thing. Here we show that the central Italy seismic sequence starting with the Amatrice earthquake on 24 August 2016 advanced the 30 October Norcia earthquake (MW = 6.5), but limited its magnitude by inhibiting the rupture on large portions of the fault plane. The preceding events hastened the mainshock and determined its features by shaping a patch of concentrated stress. During the Norcia earthquake, the coseismic slip remained substantially confined to this patch. Our results demonstrate that monitoring the seismicity with very dense networks and timely analyses can make it feasible to map rupture prone areas.

Seismic signature of active intrusions in mountain chains.

Intrusions are a ubiquitous component of mountain chains and testify to the emplacement of magma at depth. Understanding the emplacement and growth mechanisms of intrusions, such as diapiric or dike-like ascent, is critical to constrain the evolution and structure of the crust. Petrological and geological data allow us to reconstruct magma pathways and long-term magma differentiation and assembly processes. However, our ability to detect and reconstruct the short-term dynamics related to active intrusive episodes in mountain chains is embryonic, lacking recognized geophysical signals. We analyze an anomalously deep seismic sequence (maximum magnitude 5) characterized by low-frequency bursts of earthquakes that occurred in 2013 in the Apennine chain in Italy. We provide seismic evidences of fluid involvement in the earthquake nucleation process and identify a thermal anomaly in aquifers where CO2 of magmatic origin dissolves. We show that the intrusion of dike-like bodies in mountain chains may trigger earthquakes with magnitudes that may be relevant to seismic hazard assessment. These findings provide a new perspective on the emplacement mechanisms of intrusive bodies and the interpretation of the seismicity in mountain chains.

Combining stress transfer and source directivity: the case of the 2012 Emilia seismic sequence.

The Emilia seismic sequence (Northern Italy) started on May 2012 and caused 17 casualties, severe damage to dwellings and forced the closure of several factories. The total number of events recorded in one month was about 2100, with local magnitude ranging between 1.0 and 5.9. We investigate potential mechanisms (static and dynamic triggering) that may describe the evolution of the sequence. We consider rupture directivity in the dynamic strain field and observe that, for each main earthquake, its aftershocks and the subsequent large event occurred in an area characterized by higher dynamic strains and corresponding to the dominant rupture direction. We find that static stress redistribution alone is not capable of explaining the locations of subsequent events. We conclude that dynamic triggering played a significant role in driving the sequence. This triggering was also associated with a variation in permeability and a pore pressure increase in an area characterized by a massive presence of fluids.