Seismic Network Research Articles

Towards real-time earthquake forecasting in Chile: Integrating intelligent technologies and machine learning

This paper proposes an innovative approach towards real-time earthquake forecasting in Chile by integrating intelligent technologies and machine learning methods. Earthquakes pose significant risks to communities and infrastructure in Chile, making accurate and timely forecasting crucial for disaster preparedness and mitigation. Traditional forecasting methods have limitations in providing real-time insights into seismic activity. In contrast, intelligent technologies such as artificial intelligence (AI) and machine learning offer promising avenues for enhancing prediction accuracy and speed. A new earthquake forecasting method using a modified clustering approach LMSCAN(Local Maxima-based Spatio-Cluster Analysis Network) and enhanced neural network time series analysis LSTM-IC (Long Short Term Memory Inverse Correlation) is presented in this paper. This neural network-based technology has been utilized to forecast earthquakes in the Chile region, one of the countries with the largest seismic activity. This study explores the integration of intelligent systems with machine learning algorithms to analyze seismic data and predict earthquake occurrences in Chile. By leveraging historical seismic data, sensor networks, and advanced predictive models, our approach aims to provide timely warnings and insights into seismic events, thereby improving disaster response and resilience. The proposed framework holds the potential to revolutionize earthquake forecasting by enabling real-time monitoring and proactive measures to safeguard communities and infrastructure in Chile and beyond. The remarkable 95 % accuracy achieved by this model is a testament to its exceptional learning process, which sets it apart from other models. Its ability to learn and adapt to new data is unparalleled, allowing it to forecast incredibly precisely and produce highly reliable results.

Introducing ViDA3, An Earthquake Early Warning Algorithm for Offshore Hypocenter Determination Using Onshore Seismic Networks

ABSTRACT The objective functions adopted by earthquake early warning (EEW) location algorithms are inadequate for out-of-network earthquakes. As a result, the real-time locations of these earthquakes are often erroneous. The consequences of mislocating out-of-network earthquakes are that their magnitudes are miscalculated, and the loci of their shaking predictions map are shifted. Given that the largest earthquakes occur in subduction settings, improving real-time out-of-network earthquake location is of great importance. In this study, the Virtual Dynamically Assembled Array Algorithm (ViDA3) is introduced, which addresses the location issue of offshore and off-network earthquakes. The guiding principle underlying the new EEW location algorithm is that standard seismic networks may be viewed as a collection of medium-sized seismic arrays, with each array consisting of three or more network stations. The potential of array seismology for EEW against out-of-network earthquakes stems primarily from the slowness vector, which points at the direction of the epicentral region. Thus, this region may be constrained merely by intersecting two or more such vectors. In addition, the length of the slowness vector depends on the hypocentral distance and depth and is thus vital for addressing an acute problem in a subduction setting—discriminating between upper crust and deep slab earthquakes. Furthermore, when the slowness of the P phase is known, the slowness of the S phase is deduced, and the S-phase arrival is searched for using the shift-and-sum practice. What makes ViDA3 so attractive is that, in locations where a real-time network is already in place, these added values may be achieved without extra hardware or substantial budget requirements. We present the result of ViDA3 real-time operation on a shallow earthquake offshore Vancouver Island and the result of its replay on a deep slab earthquake in northern Chile. ViDA3 performance is further assessed using a dataset of seismograms from the Mendocino Triple Junction area. It is concluded that ViDA3 location scheme outperforms currently available EEW location algorithms for out-of-network earthquakes.

Study on Seismic Source Parameter Characteristics of Baihetan Reservoir Area in the Lower Reaches of the Jinsha River

The source parameters of earthquakes (stress drop, corner frequency, seismic moment, source size, radiant energy, etc.) provide important information about the source features, the state of stress, and the mechanism of earthquake rupture dynamics. Using the digital observation data obtained from a high-density seismic monitoring network deployed in the Baihetan reservoir area of the lower Jinsha River, we obtained Brune source parameters of the 459 earthquakes ranging in magnitude ML 1.50~4.70. The results revealed seismic moments M0 within the range of 2.03 × 1012~1.45 × 1016 N·m, corner frequencies between 2.00 and 10.00 Hz, and source dimensions varying from 130.00 to 480.00 m, with stress drops spanning from 0.12 to 61.24 MPa. It is noteworthy that the majority of the earthquakes had stress drops less than 10.00 MPa, with as much as 73.30% of these events having stress drops within the range of 0.10 to 2.00 MPa. We found that stress drop, corner frequency, and source size in the study area exhibited positive correlations with earthquake magnitude. Earthquakes occurring at shallower depths for the same magnitude tended to have smaller stress drops and corner frequencies, but larger rupture scales. During the first 2 years of impoundment with significant water level fluctuation, earthquakes beneath or near the reservoir released higher stress drops relative to pre-reservoir conditions, with average stress drops significantly elevated from 5.52 to 13.562 Mpa for events above ML3 since the impoundment. The radiated energy released by earthquakes with magnitudes below ML3.0 are significantly more than before impoundment, indicating that earthquakes of similar magnitudes in the reservoir area may produce greater intensity and perceptibility following the impoundment. According to our result, the triggered seismicity will continue to be active under annual regulation changes in the water level of the Baihetan Dam at high elevations in future years.

Realtime Selection of Optimal Source Parameters Using Ground Motion Envelopes

It is increasingly common for seismic networks to operate multiple independent automatic algorithms to characterise earthquakes in real-time, such as in earthquake early warning (EEW) or even standard network practice. Commonly used methods to select the best solution at a given time are simple and use ad hoc rules. An absolute measure of how well a solution (event origin and magnitude) matches the observations by the goodness-of-fit between the observed and predicted envelopes is a robust and independent metric to select optimal solutions. We propose such a measure that is calculated as a combination of amplitude and cross-correlation fit. This metric can be used to determine when a preferred solution reaches an appropriate confidence level for alerting, or indeed to compare two (or more) different event characterisations directly. We demonstrate that our approach can also be used to suppress false alarms commonly seen at seismic networks. Tests using the 10 largest earthquakes in Switzerland between 2013 and 2020, and events that caused false alarms demonstrate that our approach can effectively prefer solutions with small errors in location and magnitude, and can clearly identify and discard false origins or incorrect magnitudes, at all time scales, starting with the first event characterisation.

GRAPES: Earthquake Early Warning by Passing Seismic Vectors Through the Grapevine

AbstractEstimating an earthquake's magnitude and location may not be necessary to predict shaking in real time; instead, wavefield‐based approaches predict shaking with few assumptions about the seismic source. Here, we introduce GRAph Prediction of Earthquake Shaking (GRAPES), a deep learning model trained to characterize and propagate earthquake shaking across a seismic network. We show that GRAPES’ internal activations, which we call “seismic vectors”, correspond to the arrival of distinct seismic phases. GRAPES builds upon recent deep learning models applied to earthquake early warning by allowing for continuous ground motion prediction with seismic networks of all sizes. While trained on earthquakes recorded in Japan, we show that GRAPES, without modification, outperforms the ShakeAlert earthquake early warning system on the 2019 M7.1 Ridgecrest, CA earthquake.

Video camera and seismic monitoring of water bulge explosion at Strokkur Geyser, Iceland

Eruptions of volcanoes and geysers share many fundamental similarities: for example, they are manifestations of Earth’s geothermal energy, involving the pressure-driven expulsion of fluids from the Earth’s interior. However, while volcanoes can produce spectacular lava bubbles that burst, water bubbles are rarely observed on the surface of geysers. It is still unclear why some of these low-viscosity geyser systems produce none, while others produce them regularly. There is no quantification of the size, speed, and height of these bubbles at geysers, which is the gap we fill here. Strokkur creates a water bulge in its surface pool (bulge stage). When the bulge bursts, water is ejected into the air (jet stage). The steam then continues to rise buoyantly and drift away (drift stage). Here we study the evolution of the three stages using records from video camera campaigns and a local seismic network. We find that larger bulges are associated with larger ascent velocities and cause larger jet heights. As energy is channeled into a high jet, small seismic ground motions are recorded. The bulge formation itself is barely visible seismically. Our work suggests that the 0.74±0.27 s-long bulge stage can be used as a first-order proxy for predicting eruption height. This study might also be relevant for understanding fluid dynamics in volcanic systems.

Estimation of Geometrical Spreading Factor and Coda-Wave Attenuation Characteristics for the Saurashtra Horst in Western Deccan Volcanic Province, Gujarat, India

ABSTRACT In the present study, we have estimated the coda Q (QC) and the geometrical spreading factor (gamma) for the Saurashtra region that is located in the western part of the Deccan Volcanic Province, Gujarat, India. The waveform data from the seismic network operated by the Institute of Seismological Research, Gandhinagar, are used for the analysis. We have used records of a total of 221 local earthquakes with epicentral distances less than 70 km from the seismic stations. The moment magnitude range of the earthquakes is 2.5–5.3, and the depth range is 1–22 km. The QC has been estimated at five central frequencies (1.5, 3, 6, 12, and 20 Hz) using 20, 30, and 40 s window lengths for vertical, north–south, and east–west components, respectively. The obtained QC values show that the attenuation in the region has a strong dependence on frequency and points toward a heterogeneous medium. The geometrical spreading factors are also estimated for two cases, that is, with and without Q at the same central frequencies and window lengths. The average geometrical spreading factors with Q are 1.37, 1.48, 1.40, 1.08, and 1.04, at central frequencies 1.5, 3, 6, 12, and 20 Hz, respectively, whereas it is 1.59, 1.53, 1.47, 1.19, and 1.14, respectively, in the absence of Q. The estimated values also indicate that the medium is heterogeneous in the Saurashtra region. The estimated region-specific coda Q and the geometrical spreading factor may help in the modeling of strong ground motions and assessment of the seismic hazard for the Saurashtra region.

Inferring the Focal Depths of Small Earthquakes in Southern California Using Physics-Based Waveform Features

ABSTRACT Determining the depths of small crustal earthquakes is challenging in many regions of the world, because most seismic networks are too sparse to resolve trade-offs between depth and origin time with conventional arrival-time methods. Precise and accurate depth estimation is important, because it can help seismologists discriminate between earthquakes and explosions, which is relevant to monitoring nuclear test ban treaties and producing earthquake catalogs that are uncontaminated by mining blasts. Here, we examine the depth sensitivity of several physics-based waveform features for ∼8000 earthquakes in southern California that have well-resolved depths from arrival-time inversion. We focus on small earthquakes (2<ML<4) recorded at local distances (<150 km), for which depth estimation is especially challenging. We find that differential magnitudes (Mw/ML–Mc) are positively correlated with focal depth, implying that coda wave excitation decreases with focal depth. We analyze a simple proxy for relative frequency content, Φ≡log10(M0)+3log10(fc), and find that source spectra are preferentially enriched in high frequencies, or “blue-shifted,” as focal depth increases. We also find that two spectral amplitude ratios Rg 0.5–2 Hz/Sg 0.5–8 Hz and Pg/Sg at 3–8 Hz decrease as focal depth increases. Using multilinear regression with these features as predictor variables, we develop models that can explain 11%–59% of the variance in depths within 10 subregions and 25% of the depth variance across southern California as a whole. We suggest that incorporating these features into a machine learning workflow could help resolve focal depths in regions that are poorly instrumented and lack large databases of well-located events. Some of the waveform features we evaluate in this study have previously been used as source discriminants, and our results imply that their effectiveness in discrimination is partially because explosions generally occur at shallower depths than earthquakes.

Machine learning prediction of the mass and the velocity of controlled single-block rockfalls from the seismic waves they generate

Abstract. Understanding the dynamics of slope instabilities is critical to mitigate the associated hazards, but their direct observation is often difficult due to their remote locations and their spontaneous nature. Seismology allows us to get unique information on these events, including on their dynamics. However, the link between the properties of these events (mass and kinematics) and the seismic signals generated is still poorly understood. We conducted a controlled rockfall experiment in the Riou Bourdoux torrent (southern French Alps) to try to better decipher those links. We deployed a dense seismic network and inferred the dynamics of the block from the reconstruction of the 3D trajectory from terrestrial and airborne high-resolution stereophotogrammetry. We propose a new approach based on machine learning to predict the mass and the velocity of each block. Our results show that we can predict those quantities with average errors of approximately 10 % for the velocity and 25 % for the mass. These accuracies are as good as or better than those obtained by other approaches, but our approach has the advantage in that it does not require the source to be localised, nor does it require a high-resolution velocity model or a strong assumption on the seismic wave attenuation model. Finally, the machine learning approach allows us to explore more widely the correlations between the features of the seismic signal generated by the rockfalls and their physical properties, and it might eventually lead to better constraints on the physical models in the future.

Intra-plate Jal Az-Zor Strike-Slip faulting deciphering the enigma of local earthquakes in Kuwait

Abstract Kuwait National Seismic Network (KNSN) data shows the clustering of earthquakes in northern and southern clusters. Spatial correlation between these clusters and oil fields led previous studies to declare that oil production/injection triggered earthquakes, but some suggested the possibility of tectonic causes. This study addresses the genuine objective of uncovering the origin of Kuwait earthquakes by analyzing relationships between earthquake spatial and temporal patterns, oil production/injection, structural and tectonic setting, and subsurface fluid pressures. A kinematic model of the Jal Az-Zor dextral-slip fault was presented as a decipherer to the earthquake's origin. The northern and southern clusters represent leading quadrants, where increased mean stress causes earthquakes. Dibdibah Trough and Kuwait Bay represent trailing quadrants with decreased mean stress and a lack of earthquakes. The small percentage of earthquakes falling outside clusters are caused by the concentration of regional compressive stresses related to Arabian Plate motion on pre-existing faults. Triggering earthquakes by oil field operations requires 10% pressure increase above the original pressures, which never occurs in Kuwait oil fields. The results of this study emphasize the significance of understanding fault kinematics to assess earthquake hazards and the need to focus on engineering requirements for developments in the leading quadrants areas.

A new moment tensor catalogue of light to moderate earthquakes in Palu-Koro and Matano faults, Sulawesi, based on the regional network in Indonesia: Fault analysis and insight

Starting from 2009, substantial improvements in the regional seismic network operating in Sulawesi, as part of the Meteorological, Climatological, and Geophysical Agency (BMKG) station deployment, allowed us to determine earthquake mechanism for 140 shallow (5–60 km) light to moderate size (3.8 ≤ Mw ≤ 6.1) earthquakes between 2009 and 2018 along the Palu-Koro and Matano faults, through moment tensor inversion. Ten (10) out of the 140 solutions obtained in this study were found in the International Seismological Centre Bulletin and used as validation. The double-couple mechanism of the best solutions (variance reduction ≥0.4 and condition number ≤ 7.5) is further used to analyze the Palu-Koro and Matano segments. We observed that the Donggala segment of the Palu-Koro fault shows variable earthquake mechanism types. Whereas the two subsequent segments, the Palu and Saluki segments, show predominate oblique strike-slip with normal and reverse mechanisms. The Matano fault is characterized by strike-slip mechanisms, with normal mechanisms also observed. The normal mechanisms depict the pull-apart system in the Matano segment, manifested as Matano Lake. The results of this work improve our understanding of regional tectonics.

Reactivation of Ramadi-Musaiyib Fault-Najd Faults System, Iraq

Nine earthquakes inside Mesopotamia Plain from 2016 to 2018 were analyzed. Three seismic networks have been relied on to relocate these events and know their responsible fault. These networks are; Mesopotamian Seismological Center, Al-Refaei Seismic Array and Iranian Seismological Center. The results showed that the root mean square (RMS) values for all relocated events ranged between 0.002-0.997. The results of the error ellipse ranged from 0.24 - 4.17 km with longitude and 0.76 - 8.24 km with latitude. The new locations of these events are on or near the Ramadi-Musaiyib Fault. This means that the Ramadi-Musaiyib Fault, which belongs to the Najd fault system, is currently in reactivation.

The First Network of Ocean Bottom Seismometers in the Red Sea to Investigate the Zabargad Fracture Zone

In the last decades, the slow-spreading Red Sea rift has been the objective of several geophysical investigations to study the extension of the oceanic crust, the thickness of the sedimentary cover, and the formation of transform faults. However, local seismology datasets are still lacking despite their potential to contribute to the understanding of the tectonic evolution of the Red Sea. The Zabargad Fracture Zone is located in the Northern Red Sea and significantly offsets the rift axis to the East. Thus, it is considered a key tectonic element to understand better the formation of the Red Sea rift. To fill the gap in the dataset availability, we deployed the first passive seismic network in the Red Sea, within the Zabargad Fracture Zone. This network included 12 Lobster OBSs from the DEPAS pool, 2 OBS developed and deployed by Fugro, and 4 portable seismic land stations deployed on islands and onshore on the Saudi Arabian coast. Our data-quality analysis confirms that the head-buoy cable free to strum, as well as other additional elements of the DEPAS OBSs, generate seismic noise at frequencies $>$ 10 Hz. However, the Fugro OBSs show high-frequency disturbances even if they lack vibrating elements. Comparison between land and OBS stations reveals that noise between 1 and 10 Hz is due to ocean-generated seismic noise, and not due to resonance of the OBS elements. We also found that waveforms of teleseismic earthquakes recorded by the Fugro OBSs, islands, and onshore stations have comparable signal-to-noise ratios. Instead, differences in signal-to-noise ratio for local earthquakes are affected more by site and path effects than instrument settings.

Development of improved short-period geophone: Implementation of low-frequency compensation

Short-period geophones are widely used instruments in dense seismic array investigations. However, inherent limitations in capturing low-frequency signals restrict their versatility. This study implements a low-frequency compensation algorithm via a state-variable filtering circuit to enhance the frequency response of a traditional short-period geophone. Through simulation experiments and vibration testing, we validated the performance of the proposed bandwidth-extension (BE) system. Additionally, time–frequency earthquake signal analysis compared outcomes between our BE seismometers, benchmark short-period seismometers, and high-fidelity broadband seismometers across three seismic events with epicentral distances up to 5,186 km. Results demonstrate the BE system successfully extends the natural frequency cutoff from 4.5 Hz to below 0.2 Hz while maintaining strong consistency with broadband counterparts. Moreover, negligible power demands (10 mW) facilitate embedding the BE circuit directly within acquisition hardware. Given these advantages, the BE method shows promise for upgrading future sensors of seismic networks to record low-frequency content with minimal costs.

Time-varying damping ratios and velocities in a high-rise during earthquakes and ambient vibrations from coda wave interferometry

Coda wave interferometry is applied to data from Community Seismic Network MEMS accelerometers permanently installed on nearly every floor of a 52-story steel moment-and-brace frame building in downtown Los Angeles. Wavefield data from the 2019 M7.1 Ridgecrest, California earthquake sequence are used to obtain impulse response functions, and time-varying damping ratios and shear-wave velocities are computed from them. The coda waves are used because of their increased sensitivity to changes in the building’s properties, and the approach is generalized to show that a building’s nonlinear response can be monitored through time-varying measurements of representative pseudo-linear systems in the time domain. The building was not damaged, but temporary nonlinear behavior observed during the strong motions provides a unique opportunity to test this method’s ability to map time-varying properties. Reference damping parameters and velocities are obtained from a month-long period during which no significant seismic activity had occurred. Damping ratios measured over narrow frequency bands increase by up to a factor of 4 over short time durations spanning the main shock, as well as M > 4.5 aftershocks and a foreshock. The largest damping ratio increases occur for the highest frequencies, and the increase is attributed to friction associated with structural and non-structural surface discontinuities which experience relative motion and impact during shaking, resulting in energy loss. Shear-wave velocities in the building’s east–west and north–south directions are found by applying a waveform stretching method to the direct and coda waves. The broadband velocities are reduced by as much as 10% during building shaking, and their restoration to pre-earthquake levels is found to be a function of shaking amplitudes. Until recently, these techniques had been limited by temporal and spatial sparsity of measurements, but in this study, variations of the impulse response functions are resolved over time scales of tens of seconds and on a floor-by-floor spatial scale.

Magnitude mb: Reducing Processing-Related Variability

Abstract It is well known that magnitude mb shows a large variation between different stations and agencies for the same event. In this study we investigate the various causes of this variation and we have found that apart from the regional variation caused by the Earth’s 3D structure, a major contribution to the variation is the way the data are processed: automatic or manual determination of amplitude and period is different for the same signal at different agencies or processing systems and/or there are errors in the amplitude measurements when correcting the signal from the simulated World Wide Standard Seismic Network Short Period (WWSSN-SP) response. Part of the problem to objectively investigate the causes for the variation is that few agencies use the International Association of Seismology and Physics of the Earth’s Interior standard for reporting the observations. To get less variability in mb, a standardized way of reading amplitude and period must be developed and used.

Estimating Kappa within a Low-Seismicity Region in Northern New Mexico Using Data Recorded by the Los Alamos Seismic Network

ABSTRACT We quantify the total attenuation, κ, and the attenuation component due to near-surface site effects, κ0, in a region in northern New Mexico using data recorded by the Los Alamos Seismic Network. The area is characterized by low seismicity, where most of the well-recorded earthquakes have magnitudes between 1 and 2. This magnitude range poses a challenge for commonly used kappa methods because the high-frequency attenuation cannot be confidently isolated from the bandwidth in which the corner frequency roll-off occurs. We determine through synthetic experiments that estimates of κ within this range have quantifiable biases that depend on source (corner frequency), site (κ magnitude), and data quality characteristics (fitting bandwidth), which can be used to correct estimated κ from three commonly used kappa methods. Using 412 recorded earthquakes, we show that a bias correction results in κ distributions and κ0 estimates that are more consistent between the three methods, suggesting that the bias correction results in κ values with higher fidelity. Using the bias-corrected κ, we find κ0 between 0.038 and 0.049 s within the Valles Caldera and between 0.026 and 0.066 s on Los Alamos National Laboratory property, values near those commonly used in the western United States. We find that a main limitation in the quality of κ0 is the small number of usable waveforms at some stations, which will to improve as more earthquakes are recorded. This contrasts with other aspects, such as fitting bandwidth and source and path variability, which are unlikely to change in the future and will ultimately be the limiting factor in κ0 resolution. Overall, our results suggest that the bias-correction scheme presented here could potentially be used in other regions where small-magnitude earthquakes are prevalent. However, future work should look to verify that bias-corrected κ estimates show consistency with those retrieved from higher magnitude earthquakes.

Statistically significant difference between earthquake size distributions of independent and triggered seismicity

The Alto Tiberina Fault system, located in Central Italy, is an active structure about 60 km long composed of a principal low-angle normal fault and several minor synthetic and antithetic splays. The system is monitored by a dense seismic network, giving us the opportunity to construct high-definition seismic catalogs with a low completeness magnitude. We analyze the clustering properties of the 2010-2015 seismicity by using a 3D stochastic declustering algorithm that also includes the earthquakes’ depth. We demonstrate that the earthquake size distribution is strongly correlated with the clustering of seismic events and their depth; in particular, the principal fault and secondary faults show an opposite behavior both in terms of earthquake size distribution and clustering properties.

A dense micro-electromechanical system (MEMS)-based seismic network in populated areas: rapid estimation of exposure maps in Trentino (NE Italy)

Abstract. The micro-electromechanical system (MEMS)-based seismic network of Trentino (NE Italy) consists of 73 low-cost accelerometers installed close to inhabited areas. These sensors have a suitable sensitivity to detect moderate-to-strong earthquakes but are able to record even weaker seismicity. The densely distributed peak ground acceleration values recorded by MEMS and other types of stations are integrated within the existing seismic monitoring procedure in order to automatically obtain a complete set of strong motion parameters a few minutes after the origin time. The exposure of the resident population and critical buildings is estimated by quantifying the different levels of shaking, which is expressed according to the Mercalli–Cancani–Sieberg intensity scale. These types of results, summarized in synthetic portable document format (PDF), can be useful for civil protection purposes to evaluate the state of emergency after a strong earthquake in a timely manner and to choose how and where to activate first aid measures and targeted structural monitoring.

Seismic Signal Discrimination of Earthquakes and Quarry Blasts in North-East Italy Using Deep Neural Networks

AbstractSeparation of seismic sources of seismic events such as earthquakes and quarry blasts is a complex task and, in most cases, require manual inspection. In this study, artificial neural network models are developed to automatically identify the events that occurred in North-East Italy, where earthquakes and quarry blasts may share the same area. Due to the proximity of the locations of the active fault lines and mining sites, many blasts are registered as earthquakes that can contaminate earthquake catalogues. To be able to differentiate various sources of seismic events 11,821 seismic records from 1463 earthquakes detected by various seismic networks and 9822 seismic records of 727 blasts manually labelled by the Slovenian Environment Agency are used. Three-component seismic records with 90 s length and their frequency contents are used as an input. Ten different models are created by changing various features of the neural networks. Regardless of the features of the created models, results show that accuracy rates are always around 99 %. The performance of our models is compared with a previous study that also used artificial neural networks. It is found that our models show significantly better performance with respect to the models developed by the previous study which performs badly due to differences in the data. Our models perform slightly better than the new model created by using our dataset, but with the previous study’s architecture. Developed model can be useful for the discrimination of the earthquakes from quarry blasts in North-East Italy, which may help us to monitor seismic events in the region.