Seismic Catalog Research Articles

Are the magnitudes of earthquakes in Southern California, with incompleteness removed, correlated?

SUMMARY No. The correlation between the magnitudes of earthquakes is a scientific question usually investigated by statistical seismologists. In the last decades, two opposite answers have been given to the problem of correlation of magnitudes: there is no correlation, that is the magnitude of earthquakes in a seismic catalogue can be considered as random sampling from the magnitude distribution; or, there is a correlation, that is the magnitude of a seismic event can influence the magnitude of the successive event. Here we used the Southern California seismic catalogue, properly treated to remove incompleteness, to answer the question, finding no significant correlation between magnitudes.

Анализ локальной сейсмичности западной части полуострова Таймыр по данным сейсмической станции «Колба»

In recent years, active development of coal deposits and works related to the creation of appropriate infrastructure (ports, roads, etc.) have been carried out in the western part of the Taimyr peninsula. The main objects are the port Buhta Sever (oil loading terminal), the Syradasay coal deposit and the port Yenisei. Work is also underway in the Lembersk coal mine and the associated port Dixon. The single seismic station “KOLBA” of the Arkhangelsk seismic network installed in October 2020 near Dixson regularly registers local seismic events, the natural nature of which is questioned. From October 2020 to February 2023, 150 seismic events with a magnitude range of ML from 0.4 to 2.4 were registered at distances up to 200 km from the seismic station. The purpose of our research is to give a reasonable answer on the basis of selected criteria about the nature of registered local events: earthquakes or man-made events. Cleaning seismic catalogs from man-made events has always been an urgent task, especially after the installation of each new seismic station in a little-studied area of the Arctic. The difficulty in determining the event nature lies in the absence of a regional hodograph and any reference data such as samples of waveforms, reliable information about the place, time and methods of industrial explosions, the absence of these events on the records of other seismic stations. In this regard, we have defined a set of criteria for determining the nature of events, which are reduced to the consideration of indirect signs. These are the location of an event near an industrial facility, the analysis of spectral-time diagrams and waveforms of seismic events, the analysis of the amplitudes ratio of volume and / waves, the daily distribution of seismic events and the magnitude range of recorded events. Despite the fact that individually each criteria is not a direct proof of the man-made events nature, their totality can be taken as a decision to exclude the event from the seismic catalog. In the process of determining the epicenters of events, we identified several groups related to various industrial facilities. The daily distribution diagram shows that local seismic events were recorded in the daytime with peak values at 13–14 hours and at 19–20 hours, which indicates the technogenic nature of events. The / ratio is < 3.0 for all events, which most likely also indicates a technogenic nature, as with the presence of “bands” in the high frequency range on the spectral-time diagrams. The developed set of criteria for the identification of technogenic events allowed us to clean up the seismic catalog and attribute all 150 events to the manifestation of technogenic seismicity. The most effective criteria are the analysis of spectral-time diagrams, the / ratio and the daily distribution of events. In some cases, by the waveform types, it is also possible to determine whether an event belongs to a certain industrial facility (in particular, to the port of Buhta Sever).

Panel Review of the USGS 2023 Conterminous U.S. Time-Independent Earthquake Rupture Forecast

ABSTRACT This report documents the assessment by the U.S. Geological Survey (USGS) Earthquake Rupture Forecast (ERF) Review Panel of the draft ERF for the conterminous United States (CONUS-ERF23) proposed for the 2023 update of the National Seismic Hazard Model (NSHM23). Panel members participated with the ERF Development Team in several verification and validation exercises, including spot checks of the hazard estimates at key localities. The ERF23 forecast is substantially different from its predecessor, yielding relative differences in hazard that exceed ±50% in some low-hazard areas. These stem primarily from the new model ingredients—new faults, revised deformation rates, and updated seismicity catalogs—rather than from changes in the modeling methodology. The panel found that the main hazard changes are scientifically justified at the long return periods (≥475 yr) for which NSHM23 is applicable. Based on its evaluation of the model, the panel offered six actionable recommendations for improvements to the draft ERF23 for the western United States and two for the Cascadia subduction zone. All eight recommendations were adopted by the USGS for the revised ERF, as documented by Field et al. (2023). The panel concluded that CONUS-ERF23 represents a significant scientific advance over ERF18 and should be incorporated, after suitable revision, into NSHM23. The panel also considered changes to the CONUS-ERF that cannot be feasibly implemented in NSHM23 but could lead to future improvements. Among these aspirational recommendations, the panel prioritized the development of time-dependent extensions of ERF23 that include models of seismic renewal and clustering. The panel endorsed USGS efforts to extend the NSHM to a national earthquake forecasting enterprise capable of continually updating and disseminating authoritative information about future earthquake occurrence through a well-designed hazard-risk interface. Operational earthquake forecasting will place new and heavy demands on USGS cyberinfrastructure, requiring a more integrated approach to software development and workflow management.

Scaling of earthquake waiting time distributions in northern Chile

SUMMARY In this study, we examine the scaling of earthquake waiting time distributions in Northern Chile. For this purpose, we analysed 7-yr high spatial resolution and low completeness magnitude IPOC seismic catalogue and 43-yr USGS catalogue. A unified earthquake moment–space–time scaling relation is empirically evaluated by calculating waiting times for different ranges of magnitude and epicentral area linear dimension and estimating the associated scaling coefficients, β analogue to b-value and γ, the correlation fractal dimension. We find a scaling function that can be characterized with three distinct regions, regions whose behaviour depends on whether seismicity is in the coastal area or at intermediate depth. Moreover, high resolution localizations from the IPOC catalogue allows us to further observe differences: the lower plane seismicity of the double seismic zone behaves just like that at intermediate depth, while the crustal upper plate and the upper plane from the double seismic zone seismicity behaves more like interface seismicity. Thus, the earthquake waiting time distribution primarily depends on whether seismicity is located in a crust-on-crust or in a crust–mantle system: having, respectively, high/low correlated behaviour in the short scale region, non-exponential/exponential decay in the transition middle region and, in all cases, long-term clustering with a slower than exponential decay in the long scale.

Research Catalog of Inland Seismicity in the Southern Korean Peninsula from 2012 to 2021 Using Deep Learning Techniques

Abstract A seismicity catalog spanning 2012–2021 is proposed for the inland and near-coastal areas of the southern Korean Peninsula (SKP). Using deep learning (DL) techniques combined with conventional methods, we developed an integrated framework for compiling a comprehensive seismicity catalog. The proposed DL-based framework allowed us to process, within a week, a large volume of data (spanning 10 yr) collected from more than 300 seismic stations. To improve the framework’s performance, a DL picker (i.e., EQTransformer) was retrained using the local datasets from the SKP combined with globally obtained data. A total of 66,858 events were detected by phase association using a machine learning algorithm, and a DL-based event discrimination model classified 29,371 events as natural earthquakes. We estimate source information more precisely using newly updated parameters for locations (a 1D velocity model and station corrections related to the location process) and magnitudes (a local magnitude equation) based on data derived from the application of the DL picker. Compared with a previous catalog, the proposed catalog exhibited improved statistical completeness, detecting 21,475 additional earthquakes. With the newly detected and located earthquakes, we observed the relative low seismicity in the northern SKP, and the linear trends of earthquakes striking northeast–southwest (NE–SW) and northwest–southeast (NW–SE) with a near-right angle between them. In particular, the NE–SW trend corresponds to boundaries of major tectonic regions in the SKP that potentially indicates the development of fault structures along the boundaries. The two predominant trends slightly differ to the suggested optimal fault orientations, implying more complex processes of preexisting geological structures. This study demonstrates the effectiveness of the DL-based framework in analyzing large datasets and detecting many microearthquakes in seismically inactive regions, which will advance our understanding of seismotectonics and seismic hazards in stable continental regions.

Reappraisal of volcanic seismicity at the Kirishima volcano using machine learning

Volcanic earthquakes provide essential information for evaluating volcanic activity. Because volcanic earthquakes are often characterized by swarm-like features, conventional methods using manual picking require considerable time to construct seismic catalogs. In this study, using a machine learning framework and a trained model from a volcanic earthquake catalog, we obtained a detailed picture of volcanic earthquakes during the past 12 years at the Kirishima volcano, southwestern Japan. We detected ~ 6.2 times as many earthquakes as a conventional seismic catalog and obtained a high-resolution hypocenter distribution through waveform correlation analysis. Earthquake clusters were estimated below the craters, where magmatic or phreatic eruptions occurred in recent years. Increases in seismic activities, b values, and the number low-frequency earthquakes were detected before the eruptions. The process can be conducted in real time, and monitoring volcanic earthquakes through machine learning methods contributes to understanding the changes in volcanic activity and improving eruption predictions.Graphical

Nearly Simultaneous Pairs and Triplets of Historical Destructive Earthquakes with Distant Epicenters in the Italian Apennines

Abstract Historical seismic catalogs of Italy record several instances of pairs or triplets of large earthquakes (Mw>6.7) along the Apennine chain, occurring on the same date or within a short time frame (days or weeks). Some of these events have mesoseismic areas tens of kilometers apart and/or seismogenic structures located more than 1–3 times the fault length away. Although in the case of aligned and/or contiguous faults, their cascading activation can be explained by variations in static Coulomb stress, in the case of distant faults, this mechanism could sometimes be replaced by what is known as dynamic triggering, which is caused by the passage of seismic waves generated by a remote source. In this study, I analyze three significant ancient seismic sequences that occurred in the south-central Apennines, suggesting that the extent of the destructive effects of these earthquakes can be attributed to remote dynamic triggering, causing the activation of different and unrelated seismogenic structures within a specific time frame.

Using GNSS Phase Observation Residuals and Wavelet Analysis to Detect Earthquakes

Abstract Global Navigation Satellite Systems (GNSSs) are one of the techniques that can be used for the deformation monitoring caused by earthquakes. GNSSs enable the monitoring of specific areas affected by earthquakes in real-time and post-processing modes. This paper provides a novel method for the short-term displacement detection based on the phase residuals of GNSS observations. The study examines a natural earthquake from October 26, 2016, in Italy with Mw 6.1. In the proposed methodology, the residuals of the GNSS phase observations are derived using Precise Point Positioning. Then, residuals are transformed to the frequency domain using a wavelet transform, and the earthquake moment is identified using the anomalies detected in the scalogram, which do not appear in stable conditions. The described methodology allows for detecting the starting and ending moments of the earthquake. The methods described detected that the moment of the earthquake is visible in the residues 8 seconds after the seismic catalog time. The conducted experiments show that the use of signal analysis tools allows it to properly detect the ground vibrations in the residuals of GNSS phase observations and thus confirms the registration of ground vibrations in satellite observation data recorded by a GNSS receiver.

A novel algorithm for seismic events multiplets search

A common practice of seismology is to analyze earthquake occurrence in terms of events catalogues, with the aim to either find useful correlations between internal mechanisms under study and their outcome in the spatial/temporal series of the events or, more directly, to assess some statistical rules from observations. With this approach, catalogues are often searched for some recognizable patterns or behaviors: in this work we present a software tool created to reveal a particular kind of events sequences.The idea follows from the concept of multiplets, a well known events pattern often found in seismic series. A multiplet is defined as a sequence of events, all near in space and time and exhibiting similar magnitudes. The amount of multiplets in seismic series is related, as it is for other clustering mechanisms, to underlying correlations in the physics of the events.The software, built from scratch, scans seismic catalogues in search of events clustered as “multiplets”: this is done through the thorough application of comparison tests whose parameters thresholds are both user defined and semi-automated. The tool is however more “general” in the sense that by varying values of the filtering parameters it can reveal other kind of patterns too.While we think that this tool can be thought as a general purpose space–time series analyzer, we have found it particularly useful when applied to the results of a seismic simulator with the purpose of assessing their adherence with the observed seismicity. It can be used as a sort of metric to quantify the simulation predictions effectiveness in terms of presence of similar multiplets distributions in simulated vs. real catalogues.The software has been entirely developed in the Wolfram Language (Mathematica), a commercial powerful environment for scientific calculus and results report, but the main computational routine has been also ported to python for open-source, copyleft usage.

Deterministic seismic hazard analysis from physics-based earthquake simulations in the Eastern Betics (SE Iberia)

The low frequency of large destructive earthquakes in the Eastern Betics (SE Iberian Peninsula), not recorded in the instrumental seismic catalog, complicate the estimation of the maximum ground motion that could be reached. Knowing the characteristics and behavior of these large seismic ruptures is crucial for the study of seismic hazard. In this work, we have used a physics-based earthquake simulator to generate a 100 kyr catalog of seismic ruptures associated with seismogenic faults bordering the quaternary Guadalentin and Bajo Segura basins. From these ruptures, we have selected maximum earthquake scenarios, and those associated with certain probabilities of exceeding a severe magnitude from the synthetic catalog. The estimation of maximum peak accelerations (PGA) by means of ground motion prediction equations include a possible site amplification that we have calculated from a detailed geotechnical classification of soils. According to the simulations, the modeled faults have the capacity to generate maximum earthquakes with magnitudes ranging MW 6.7–7.4. Densely populated urban areas located along the Guadalentin valley, such as Lorca, could be subject to site-dependent PGA values close to 0.6 g on the Alhama de Murcia fault trace. Similar ground motion would affect the coastline of SE Alicante province when evaluating the maximum simulated event in the Bajo Segura offshore fault. But already from magnitudes around MW ∼ 6, PGA estimations exceed critical values of 0.5 g at the south side of the Bajo Segura basin and the city of Murcia, highly dependent of ground-motion amplification induced by thick deposits of very soft and saturated soils. Our scenarios reflect that other factors such as fault geometry, kinematics and the capability observed in the simulations to generate multi-fault ruptures also affect the spatial distribution of maximum ground motion in this region. We expect that this evaluation could contribute as an improvement guideline for future strong motion prediction studies prior to the settlement of new urban infrastructures in the Eastern Betics.

Evaluation of Station Performance of the Indonesian Seismic Network Using the Primary Location Parameter

Abstract Indonesia’s tectonic setting has resulted in a high level of seismic activity, which makes seismic networks and early warning systems vital for this country. Indonesia’s seismic stations, a total of 411 stations in 2021, are maintained by the Meteorological, Climatological, and Geophysical Agency (BMKG). This study attempts to evaluate station performance, because these are crucial in monitoring seismic activity. We utilized information from the BMKG 2020 and 2019 seismic catalogs to analyze the performance of the stations. A total of 13,798 events were reported in 2019 with magnitudes (MLv) of up to 8, whereas 10,163 events with similar magnitudes were reported in 2020. Stations in the eastern part of Indonesia (Sulawesi, Lesser Sunda Islands, the Moluccas, and Papua) detected more events compared to stations in the western part of Indonesia (Borneo, Sumatra, and Java). Up to 1651 events were detected by a single station in eastern Indonesia, compared to 847 events in western Indonesia. BMKG stations in eastern Indonesia have recorded ample amounts of events, which is consistent with the high seismic activity found there. In terms of station coverage, events in the eastern part have better station coverage due to smaller azimuthal gaps, considering the geographic location. We also analyzed station performance and the number of recorded stations in terms of a function of time for each event in the Sumatra, Sulawesi, Lesser Sunda Islands, and Papua regions. Although the number of recordings obtained by stations was high after 15 s since the origin time, the azimuthal gap progression in Sumatra and Java remained large. Therefore, an additional ocean-bottom seismometer needs to be installed in the Sumatra–Java trench.

Physically motivated moment-tensor decomposition for mining-induced seismicity

SUMMARY Compared to existing schemes, the decomposition of moment tensors for mining-induced seismic events into closing-crack and double-couple components has the advantage that each can be interpreted in terms of a physical source process (excavation convergence and slip/shear, respectively). Obviously, not every moment tensor permits such a decomposition, and we translate existing bounds on those that do to the Hudson source-type plot. For moment tensors falling within these bounds, it has previously been noted that there will be a infinite set of possible decompositions in general. We derive an implicit equation defining this set and suggest physically motivated criteria that can be used to select a single decomposition from it. Furthermore, for moment tensors falling outside the source-type bounds, we present a simple geometric scheme for determining the closest moment tensor within them (which can then be decomposed). To demonstrate the methods developed in this paper, we apply them to two catalogues of mining-induced seismicity.

Unveiling the pre-eruptive seismic series of the La Palma 2021 eruption: Insights through a fully automated analysis

We have analyzed the seismic series that preceded the La Palma 2021 eruption using a fully automated workflow based on Deep Learning. A new seismic catalog has been obtained that contains twice the number of events as the original manual catalog. The new catalog consists of 5059 absolute locations obtained using a non-linear location method and a 3D velocity model of the island, as well as relative locations by means of double differences method. Our enhanced catalog contains detailed features of seismic migration through the crust. We differentiate the sequence into three stages leading up to the eruption onset using independent seismic parameters: inter-event time variation coefficient, b-value spatial distribution, hypocentral migration and magnitude distribution. Our analysis identifies two magmatic intrusions: The first migrates from the center of the Cumbre Vieja area to the west coast of the island (stage 1), where it stagnates. After two days of seismic quiescence (stage 2), a second intrusion begins near the center of the island and rises to the surface leading to the eruption (stage 3). These results were further validated by GNSS observations and previous geodetic studies. Our findings highlight the usefulness of this type of automated methodology, offering a real-time and accurate response during intense seismic episodes.

Induced Seismicity Hazard Assessment for a Potential CO2 Storage Site in the Southern San Joaquin Basin, CA

California’s Central Valley offers vast opportunities for CO2 storage in deep saline aquifers. We conducted an induced seismicity hazard assessment for a potential injection site in the southern San Joaquin Basin for 18 years of injection at 0.68 MtCO2/yr and 100 years of monitoring. We mapped stress, faults, and seismicity in a 30 km radius around the site to build a geomechanical model and resolve the stresses on major faults. From a 3D hydromechanical simulation of the CO2 plume, we calculated the change in pressure over time on these faults and determined the conditions for safe injection. Lacking any subsurface imaging, we also conducted a probabilistic fault slip analysis using numerous random distributions of faults and a range of geomechanical parameters. Our results show that the change in probability of fault slip can be minimized by controlling the size, migration, and magnitude of the pressure plume. We also constructed a seismic catalog for the last 20 years around the site and characterized the natural patterns of seismicity. We use these results to establish criteria for evaluating potential-induced events during the storage period and to develop a traffic light response system. This study represents a first-order procedure to evaluate the seismic hazards presented by CO2 storage and incorporate uncertainties in hydrological and geomechanical parameters.

Influence of Mid-Oceanic Ridges on Seismicity of Novaya Zemlya Archipelago

The influence of the mid-oceanic ridges (MOR), including the Gakkel Ridge and the geological system Knipovich Ridge–Trough Lena, on the seismicity of the Novaya Zemlya archipelago area for 1980‒2022 is considered. For each geological element under consideration, we compiled seismic catalogs with a single unified magnitude mbISC for an equivalent comparison of information, and the annual seismic energy was calculated, and graphs of its distribution by year were constructed. We performed analytical modeling based on the Elsasser model describing the process of local stress transfer in a rigid elastic lithosphere underlain by a viscous asthenosphere, and made quantitative calculations of the disturbance propagations from MOR. The time intervals through which disturbances from MOR reach the Novaya Zemlya archipelago are 1‒2 years for the geological system Knipovich Ridge–Trough Lena and 3‒5 years for the Gakkel Ridge. The maximum joint contribution to the level of seismic activity of various geological and tectonic structures of the MOR can reach 40‒60% of the applied disturbances values, which is a sufficient condition for the influence on seismicity of the Novaya Zemlya orogen. The most geodynamical active structures and zones of concentration of tectonic stresses are identified.

All the People of Israel Are Friends: An Unknown Mid-Fifteenth Century Earthquake in the Marche Region (Central Italian Apennines) Recorded in a Coeval Hebrew Manuscript

Abstract Although the Italian seismic catalogs are considered among those with the best historical completeness, for periods prior to the Modern era they have gaps, often concentrated in limited regions. For example, Marche, a central Apennine region with high seismicity level, lacks any macroseismic observations throughout the entire fifteenth century. The chance discovery of a destructive earthquake summarized in a note written on a Hebrew prayer manuscript from 1446, not only helps us partially fill a gap in the seismic history of Italy but also prompts us to reflect on how much we still do not know about seismogenesis even in times covered by written sources.

Fractal and Spectral Analysis of Seismicity in the Lai Chau Area (Vietnam)

The time dynamics of the instrumental seismicity recorded in the area of the Lai Chau reservoir (Vietnam) between 2015 and 2021 were analyzed in this study. The Gutenberg–Richter analysis of the frequency–magnitude distribution has revealed that the seismic catalog is complete for events with magnitudes larger or equal to 0.6. The fractal method of the Allan Factor applied to the series of the occurrence times suggests that the seismic series is characterized by time-clustering behavior with rather large degrees of clustering, as indicated by the value of the fractal exponent α≈0.55. The time-clustering of the time distribution of the earthquakes is also confirmed by a global coefficient of variation value of 1.9 for the interevent times. The application of the correlogram-based periodogram, which is a robust method used to estimate the power spectrum of short series, has revealed three main cycles with a significance level of p<0.01 (of 10 months, 1 year, and 2 years) in the monthly variation of the mean water level of the reservoir, and two main periodicities with a significance level of p<0.01 (at 6 months and 2 years) in the monthly number of earthquakes. By decomposing the monthly earthquake counts into intrinsic mode functions (IMFs) using the empirical decomposition method (EMD), we identified two IMFs characterized by cycles of 10 months and 2 years, significant at the 1% level, and one cycle of 1 year, significant at the 5% level. The cycles identified in these two IMFs are consistent with those detected in the water level, showing that, in a rigorously statistical manner, the seismic process occurring in the Lai Chau area might be triggered by the loading–unloading operational cycles of the reservoir.

The 2022 MW 6.0 Gölyaka–Düzce earthquake: an example of a medium-sized earthquake in a fault zone early in its seismic cycle

Abstract. On 23 November 2022, a MW 6.0 earthquake occurred in the direct vicinity of the MW 7.1 Düzce earthquake that ruptured a portion of the North Anatolian Fault in 1999. The Mw 6.0 event was attributed to a small portion of the Karadere fault off the main North Anatolian Fault that did not rupture during the 1999 sequence. We analyze the spatiotemporal evolution of the MW 6.0 Gölyaka–Düzce seismic sequence at various scales and resolve the source properties of the mainshock. Modeling the decade-long evolution of the background seismicity of the Karadere fault employing an Epistemic Type Aftershock Sequence model shows that this fault was almost seismically inactive before 1999, while a progressive increase in seismic activity is observed from 2000 onwards. A newly generated high-resolution seismicity catalog from 1 month before the mainshock until 6 d after was created using artificial-intelligence-aided techniques and shows only a few events occurring within the rupture area within the previous month, no spatiotemporal localization process and a lack of immediate foreshocks preceding the rupture. The aftershock hypocenter distribution suggests the activation of both the Karadere fault, which ruptured in this earthquake, and the Düzce fault that ruptured in 1999. First results on the source parameters and the duration of the first P-wave pulse from the mainshock suggest that the mainshock propagated eastwards, which is in agreement with predictions from a bimaterial interface model. The MW 6.0 Gölyaka–Düzce event represents a good example of an earthquake rupture with damaging potential within a fault zone that is in a relatively early stage of the seismic cycle.

The preparatory process of the 2023 Mw 7.8 Türkiye earthquake

To verify the existence of a preparatory process for the 6 February 2023, Mw 7.8 Kahramanmaraş earthquake, southern Türkiye, we analyze the temporal evolution of seismic catalog information for ~ 7500 earthquakes with magnitudes ML ≥ 1.5, which occurred along the main segments of the East Anatolian Fault (EAF) since 2014. We find the EAF fault segments showing different temporal patterns in the proportion of nonclustered seismicity, which we interpret as temporal variation of coupling. We also study the evolution of the b-value, fractal dimension and energy rate. These seismic features show for the Amanos and Pazarcık fault segments a long-term trend during the period 2020–2022 that might correspond to a quiescence phase. The latter is followed by a change in earthquakes clustering and characteristics that starts about eight months before the Mw 7.8 Kahramanmaraş event. Our observations confirm the existence of a long-lasting preparatory phase for the 2023, Mw 7.8 Kahramanmaraş earthquake and can stimulate new investigations on the East Anatolian Fault mechanic. Intercepting when a fault starts deviating from its steady behavior, might be the key for identifying the preparatory phase of large earthquakes and mitigate seismic risk.

An attempt to use machine learning algorithms to predict strong tremors during longwall mining of a coal seam

Strong tremors during longwall mining are usually associated with the fracturing of thick sandstone layers. The prediction of such tremors is important and can be treated as a form of rockburst prevention. The article presents the possibility of predicting the occurrence of strong tremors based on the random forest algorithm in the 48-h forecast horizon. The study concerns a longwall panel in coal seam No. 506 in one of the hard coal mines in the Upper Silesian Coal Basin, Poland. The extraction of this longwall panel was carried out at a great depth, surrounded by thick layers of sandstone and with the occurrence of edges in the previously mined seams. Therefore, seismic activity was high and strong tremors occurred. Based on the seismic catalogue and data related to the longwall face advance, 12 parameters were selected that were most significantly positively or negatively correlated with the occurrence of strong tremors. Based on 10 randomly selected datasets, models of the random forest algorithm were trained and then tested. Due to the much smaller number of cases of strong tremors compared to days when there were no strong tremors, the set of training data was balanced using the ADASYN algorithm. The obtained models had an average recall value of 0.79, and average accuracy equalled 0.912. The trained models showed high efficiency in predicting minority cases of days with strong tremors for up to 48 h and a relatively low percentage of misclassifications.