Pasquale De Gori

The junction between the central and southern Apennines represents a high-seismic-hazard region in the Mediterranean. Its seismotectonic setting is characterized by a complex, poorly understood interplay between SW-NE regional extension along the range axis and E-W mid-to-lower crustal shear zones in the Adria plate to the east. Although the range axis hosted several M6-7 historical earthquakes, their causative faults remain mostly debated. Monitoring by the Italian National Seismic Network (Rete Sismica Nazionale, RSN), with a station spacing of 10-30 km and a detection threshold of about ML1.2 in the region, has proved insufficient to pinpoint and fully characterize source faults for recent low-to-moderate magnitude (M < 4) sequences.To address these limitations, we conducted the first comprehensive study of background seismicity as part of the MOSAICMO project, an inter-disciplinary initiative investigating tectonic evolution and seismogenesis of this region. This study integrates a 2-year passive seismic experiment (2023-2025) with a re-analysis of the 2016-2022 RSN seismicity. Our objectives were to improve knowledge of the active faults and relationship between seismogenesis and physical properties of the crustal rocks. The seismic experiment integrated 13 temporary stations with 20 permanent stations of the RSN over an area of 60x60 km2, reducing station spacing to 4-12 km. Initial analysis of the first nine months of the new dataset using a standard STA/LTA algorithm identified 470 events (0.2 < ML < 2.8), representing a 220% increase over the RSN catalog. For these earthquakes, P-and S-phases were manually picked. For the 2016-2022 seismicity, we revised and augmented the phase picks for 1,400 selected events and applied cross-correlation template matching to a prolonged swarm-like sequence (2016-2017; Mw 4.3) to produce a high-resolution catalog.We utilized these phase picks to construct catalogs, through: i) absolute locations using the probabilistic location software NonLinLoc and a new optimized 1D velocity model, based on a non-linear approach ii) high-precision relative locations using the double-difference technique HypoDD; iii) absolute 3D re-locations alongside with Vp and Vp/Vs crustal models derived from Local Earthquake Tomography on a 3 × 3 × 2 km grid.Our results show that seismicity deepens eastward, from 3–12 km beneath the inner range to 15–22 km under the outer range. While the upper crust exhibits mixed extensional and strike-slip focal mechanisms, deeper eastern events are almost exclusively strike-slip. Most seismicity occurs in small, short-lived clusters. Along the inner range, seismicity concentrates at 5-10 km depth within high-Vp (6.0-6.7 km/s), low-Vp/Vs (1.70-1.85) zones. Here, high-precision relocations reveal NW-striking, NE-dipping alignments consistent with known Quaternary normal faults. Integrating these results with a subsurface geological model based on seismic commercial profiles and exploration wells, and a 2D magnetotelluric tomography, we find that: (i) axial seismicity is mainly hosted within the high-velocity, high-resistivity Mesozoic carbonates of the Apulia Platform, (ii) the 2016–2017 swarm-like seismicity also clusters within the Apulian Platform but correlates with a low-resistivity anomaly, suggesting a fluid-driven seismogenic mechanism.

The transition from manual to deep-learning automated seismic phase picking has revolutionized seismology applications such as seismic catalog building and fault structures analysis. However, the reliability of these AI-driven catalogs is often hindered by a "black-box" approach to model selection and decision thresholds. While deep learning models like PhaseNet (Zhu and Beroza, 2019) offer unprecedented efficiency, their performance is sensitive to the data they were trained on and to the probability thresholds used to define a "phase pick". In this work, we present a comparative study focused on the Amatrice–Visso–Norcia 2016-2017 seismic sequence in Central Italy. We investigate the influence of the training models and the threshold variation on the phase picking detections. In particular, we compare the performance of the default PhaseNet model (STEAD) against i) a model trained on the AQ2009 dataset (specific for the Central Apennines, from Bagagli et al., 2023) and ii) a model obtained through Transfer Learning on STEAD fine-tuned with the INSTANCE model (Michelini et al., 2021) via the SeisBench platform (Woollam et al., 2022). We also analyze how varying the confidence threshold (from 0.1 to 0.9) affects the final catalog's completeness and precision.Preliminary results show that regional training significantly outperforms default models in specific noise conditions and that the optimal threshold is influenced by station geometry and signal-to-noise ratios. By providing a statistical framework for automated threshold calibration, this study offers a roadmap for more objective and reproducible signal detection, applicable not only to seismology but to any domain dealing with continuous time-series classification.

The geological and structural heterogeneity of the crust exerts a primary control on the development of large earthquakes and seismic sequences. The Central Apennines is a clear example of how earthquake occurrence is entangled with the complexity of the upper crustal structure. In this study, we make new inferences on this interaction for the 2009 L’Aquila seismic sequence, combining geological and structural data with 3D high‐resolution V _p and V _p /V _s velocity models. The pervasive segmentation of normal faults inherited from the Pliocene compression is a primary factor that controls the evolution of large ruptures. We propose that geometrical heterogeneity and misorientation of fault segments pervading thrust sheets conditioned the dynamic interaction during the rupture evolution. Our results emphasize how the structural legacy of previous tectonics influences fault activation. Pre‐existing heterogeneous structures compartmentalize fluid‐filled volumes, which control the development of coseismic ruptures and the spreading of aftershocks. These findings raise questions about the challenges of directly correlating surface observations and aftershock data with fault length and call for a reassessment of the seismic hazard of the Central Apennines.

The site response is critical for accurately estimating source parameters, as it directly influences the spectral characteristics of seismic signals. Achieving reliable estimates of these parameters requires clearly distinguishing between source effects, path attenuation, and site response, which are often interdependent and subject to significant trade-offs. As highlighted in recent studies, site response reflects the amplification or damping effects of the shallow subsurface layers on seismic waves, and its accurate characterization is essential to correct the observed spectra and to improve the source parameter estimations. By accurate analysis of the site response, we aim to mitigate these uncertainties and achieve a more robust parameterization of seismic events, particularly for low-magnitude earthquakes, where these effects are more pronounced.We analyzed the site response of seismic events during the 2016 Amatrice-Visso-Norcia, Central Italy, seismic sequence focusing on earthquakes with local magnitude Ml

Since magma and gas uprise and accumulation into the crust differently affect seismic velocities, seismic tomography is revealing a potential tool to detect how melt versus exsolved magmatic volatile phase reservoirs are distributed at depth into the volcanic plumbing systems. 4D tomography (in space and time) can detect temporal changes in seismic velocities, enabling to follow gas and magma accumulation, a crucial aspect in monitoring open vent volcanoes and unresting calderas. To achieve this goal we combined fully non-linear tomography with improved and fast seismic phases detection allowed by machine-learning. The use of a fully non-linear approach allows overcoming some limits of standard linearized methods, which can obscure details because of damping, smearing and blurring of seismic anomalies, due to the need of global regularization of the inverse problem. We are developing a Bayesian approach to local earthquake tomography that erases the dependence on arbitrary starting velocity models, providing more reliable absolute velocity values and model uncertainties. This is an important aspect when observed changes in seismic velocity have to be compared with theoretical predictions from petrophysical modeling. Furthermore, a data-driven self-adapting parameterizations of the earth structure strongly enhances the resolution capability in regions where high gradients in velocity are expected and in regions poorly illuminated by seismic rays, permitting to reveal seismic anomalies not detectable by standard approaches because of pre-determined model parametrization. Here, we present our novel approach and the application to the Campi Flegrei caldera, where the discrimination between gas and magma injection at shallow depth is crucial for unraveling the causal process of the unrest and in multirisk assessment and forecasting.

We perform the attenuation local earthquake tomography (P,S, waves) of the complex fault system that ruptured a wide portion of the central Apennines, during the long lasting seismic sequence that started in 2016. Three mainshocks (the 24 August Mw 6.1, the 26 October Mw 5.9, and the 30 October Mw 6.5) hit the towns of Amatrice, Norcia and Visso causing several casualties and diffuse damages. The ruptured faults spread over an 80 km north-northwest-elongated section of a normal-faulting system. A huge amount of seismic data has been collected by permanent and temporary seismic stations since the onset of the sequence.About 230,000 seismic events have been analyzed to retrieve P and S seismic waves arrival times, which allowed us to compute 3-D velocity structure and precise earthquake locations.For all of the earthquakes that meet strict selection criteria based on the signal to noise ratio, we computed the low frequency spectral level and the decay of the amplitude spectra (t*) of both P- and S-waves at about 150 recording sites. In order to avoid source complexity, we selected only events with M

Machine-learning algorithms are widely applied to facilitate human tasks. For instance, in seismology, they can help deliver high-resolution seismic catalogs including very small magnitude events that usually remain undetected by human analysts and by standard monitoring procedures. The new frontier of modern seismology is to exploit deep neural networks (DNN) to automatically detect P- and S-wave arrival times to obtain good-quality seismic event locations in terms of hypocentral errors. Increasing the number of events and seismic phases is essential to build complete earthquake catalogs to be used in seismological analyses (such as seismic hazard estimation, seismic tomography, fault zone structure determination, rupture mechanism study, etc.). Machine Learning methods are being integrated into large Research Infrastructures (RIs), like the European Plate Observing System (EPOS ERIC), which brings together 10 different scientific domains in Solid Earth Sciences. In this contribution, we present results from a specific Sponsored Research Activity promoted by the RI EPOS and dedicated to ML-driven methods for phase picking in seismic time series.To ensure the correct recognition of seismic waves, neural networks trained on large and representative training datasets are essential. Here, we investigate the influence of the training dataset on the DNN PhaseNet performance, applying the method to three case studies. In the first case, the DNN trained with the Italian seismicity dataset (INSTANCE) is used to build a catalogue on the Fucino basin (Central Italy) study area; in the second case, we use the AQ2009 training dataset (based on the L’Aquila 2009 aftershocks) to analyse the 2016-2017 Amatrice-Visso-Norcia seismic sequence; and finally, the CREW training dataset (that contains P- and S-waves reflected on the mantle Earth and recorded at large epicentral distance) to detect P- and S- waves of teleseismic (regional) data acquired by the Adria Array project network. The use of different training datasets greatly improves the performance of the neural network in recognizing P and S phases, reducing the number of false positives and providing more accurate and precise P and S-phase arrival times.

In the Central Apennines (Italy), the project RETRACE-3D provided a reliable 3D model of the crust in the area affected by the 2016-17 Amatrice-Visso-Norcia seismic sequence, highlighting that the coseismic rupture at the surface can involve old inherited normal faults while the seismogenic sources lay at depth, possibly reactivating and inverting previous thrust faults, as in the case of the Mw 6.5 Norcia earthquake (30 october 2016). Here we present a 2D gravity model across the Central Apennines, spanning from the Tyrrhenian coast to the Adriatic Sea, aimed at completing and verifying the crustal geometries resulting from the 3D model itself. The cross-section was built integrating different types of data, such as surface geology, hydrocarbon wells, seismic lines, and results from receiver function analysis. It was then checked against gravity anomalies and the velocity distribution from Local Earthquake Tomography (LET), adding further details, and, finally, against seismicity recorded during the 2016-2017 sequence. The results substantiate the reliability of the geometries proposed in the RETRACE 3D model, as they fit well, except for some local misfits, with the other independent data, such as the Bouguer anomalies and the velocity distribution from LET. Furthermore, the integration of different types of data allowed us to describe in detail the structural setting of the Apennine chain also in the surroundings of the RETRACE study area, where the cross-section length exceeds the 3D model, and to add some new elements at seismogenic depths, that exceed those typical of hydrocarbon exploration. In particular, we were able to investigate the nature of the basement top and its relationship with seismotectonics.

The progressive increase of ground deformation, seismicity, and gas emission is marking a remarkable unrest at Campi Flegrei caldera. The direct involvement of magma has been invoked to explain the deformation and space/time changes of velocity anomalies at shallow crustal depths. A challenging aspect is to forecast possible scenarios for the upward migration of magmatic fluids from the source at depth. Here, we show that the most recent seismicity (period 2023–2024), derived by a machine-learning-based earthquake detection procedure, aligns on a continuous set of caldera rim faults and on top of an inflating magma source. Direct channeling of magma through such ring faults can be a way to feed future eruptions, as observed in other calderas and inferred for the Mt. Nuovo historical eruption.