MMMarija Mustać

Marija Mustać

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Marija Mustać has a total of 38 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Enhanced View of the Mw 6.4 Petrinja (Croatia) Earthquake Sequence (2020–2022) Using Deep Learning

    by Dinko Šindija, Marija Mustać Brčić, György Hetényi, Josip Stipčević

    Machine learning approaches have shown potential to automate the creation of seismic catalogues, but their abilities and performance are still questioned and being tested. We here present the Petrinja Earthquake Sequence Machine Learning Catalogue (PESMLC), generated using the EQTransformer seismic phase picker trained on the INSTANCE dataset, PyOcto for phase association, and NonLinLoc for event relocation. Our study focuses on the 2020 MW 6.4 Petrinja earthquake sequence in Croatia, covering the period from 28 December 2020 to 30 November 2022, during which we detected 41,989 seismic events. PESMLC achieves high matching rates — 96% for the first six months, and 85% over the complete study duration — when compared with detailed manual catalogues. The seismic network densification in January 2021 significantly enhanced event detection reliability, highlighting the importance of network configuration in efficient seismic monitoring. While our machine learning approach successfully identified previously undetected microseismicity and dramatically reduced processing time, we observed systematic limitations, including magnitude estimation biases and difficulties in accurately distinguishing closely spaced events in time. Optimisation of associator parameters remains essential, and should ideally be performed specifically for each network configuration, particularly for fully automated seismic monitoring systems. We further demonstrate that raw machine learning catalogues require quality filtering prior to seismological analysis: stricter magnitude of completeness estimators result in artificially high thresholds for unfiltered machine learning catalogues, and Omori p exponents are biased downward by low-quality detections. Both effects are resolved when analysis is restricted to a quality-filtered subset, which recovers parameters consistent with high-quality manual catalogues. Nevertheless, PESMLC overall provides a detailed, comprehensive seismic dataset aligned closely with manual catalogues, demonstrating the considerable potential of machine learning to support routine seismic monitoring, particularly in resource-limited observatories and/or for long seismic sequences and large datasets.

  • Open Access

    Enhanced view of the M w 6.4 Petrinja earthquake sequence (2020-2022) using deep learning

    by Dinko Šindija, Marija Mustać, György Hetényi, Josip Stipčević

    The abilities and performance of machine learning for creating seismic catalogues are still questioned and being tested. We here present the Petrinja Earthquake Sequence Machine Learning Catalogue (PESMLC), generated using the EQTransformer seismic phase picker trained on the INSTANCE dataset, PyOcto for phase association, and NonLinLoc for event relocation. Our study focuses on the 2020 M w 6.4 Petrinja earthquake sequence in Croatia, covering the period from 28 December 2020 to 30 November 2022, during which we detected 49,730 seismic events. PESMLC achieves high matching rates - 95% for the first six months, and 85% over the complete study duration - when compared with detailed manual catalogues. The seismic network densification in January 2021 significantly enhanced event detection reliability, highlighting the importance of network configuration in efficient seismic monitoring. While our machine learning approach successfully identified previously undetected microseismicity and dramatically reduced processing time, we observed systematic limitations, including magnitude estimation biases and difficulties in accurately distinguishing closely spaced events. Optimisation of associator parameters remains essential, particularly for fully automated seismic monitoring systems. Overall, PESMLC provides a detailed, comprehensive seismic dataset aligned closely with manual catalogues, demonstrating the considerable potential of machine learning to support routine seismic monitoring, particularly in resource-limited observatories and/or for long seismic sequences and large datasets.

  • Open Access

    The Role of Salt Tectonics in the 2021 Central Adriatic Seismic Sequence

    by F. Di Luccio, M. Palano, L. Scognamiglio, A. Marchetti, I. Dasović, M. Mustać, F. Magnoni, P. Artale Harris, E. Casarotti, A. Polonia, et al.

    On 27 March 2021 a 3‐months lasting seismic sequence struck the Central Adriatic Basin that is part of the Adria plate, a relatively undeformed plate since recent times. Analyzing the waveform data acquired by the Italian and Croatian seismic networks, we computed the location parameters of 160 earthquakes and the focal mechanisms of the Mw 5.2 mainshock and the larger aftershocks. Most events align along a WNW‐ESE trending, 30 km long, narrow belt. The depth distribution of events indicates that the mainshock and a few aftershocks occurred within the upper 4 km, while most aftershocks were located below 5 km within the carbonate platform. We propose that the evolution of the 2021 earthquake distribution is primarily ruled by the top ductile salt layer. Moreover, the presence of a salt layer explains the relatively high VP/VS ratio of 1.83 in the sediment rocks surrounding the salt bodies, as also observed in similar tectonic settings. We suggest that the seismogenic fault likely responsible for the 2021 events is an inherited SW‐dipping normal fault, reactivated by reverse kinematics in response to the regional compressive stress. These results, and the understanding that salt deposits play a key role in focusing deformation and seismogenesis, represent a novel contribution to the long‐standing challenge of seismic hazard assessment of the Central Adriatic Basin, where moderate to large events could have devastating impacts along the densely populated coasts.


  • Open Access

    The role of salt tectonics in the occurrence of the 2021 Central Adriatic seismic sequence

    by Laura Scognamiglio, Francesca Di Luccio, Mimmo Palano, Alessandro Marchetti, Iva Dasović, Marija Mustać, Federica Magnoni, Pietro Artale Harris, Emanuele Casarotti, Alina Polonia, et al.

    On 27 March 2021 a three-months lasting seismic sequence struck the Central Adriatic Basin, part of Adria that is considered a relatively undeformed plate since recent times. Analyzing the waveform data acquired by the Italian and Croatian seismic networks, we computed the location parameters of 160 earthquakes and the focal mechanisms of the Mw5.2 mainshock and larger aftershocks. Most events align along a WNW-ESE, 30 km long, narrow belt. They form two clusters between 0-3 km and 4-14 km of depth, separated by a 1-2 km thick aseismic zone. Based on literature data, we suggest that such a seismic gap corresponds to a ductile salt layer, which constitutes the primary control factor for the evolution of the 2021 earthquake distribution. Moreover, the presence of a salt layer explains well the relatively high Vp/Vs ratio of 1.83 in the sediment rocks surrounding the salt bodies, as also observed in similar tectonic settings. We suggest that the seismogenic fault likely responsible for the 2021 events is an inherited SW-dipping normal fault, reactivated with prevalent reverse kinematics in response to the regional compressive stress. These results, and the recognition of a specific role of salt deposits in focusing deformation and seismogenesis represent a novel contribution to the long-standing problem of seismic hazard assessment of the Central Adriatic Basin, where moderate to large events could have devastating impacts along the highly populated coasts.

  • Open Access

    Surface displacement and source parameters of the 2020 Zagreb, Croatia earthquakes

    by Marija Mustać Brčić, Jinyin Hu, Marijan Herak, Hrvoje Tkalčić, Thanh-Son Pham

    On 22 March 2020 at 5:24 UTC, a MW 5.4 earthquake occurred on the outskirts of Zagreb and was followed by a MW 5.0 aftershock at 6:01 UTC on the same day. These moderate-magnitude events resulted in ground motion at the edge of detectability with satellite data but caused considerable damage in Zagreb’s historic centre. To illuminate the seismic sources, we compute the ground motion and invert seismic data using a state-of-the-art Bayesian inversion method that accounts for uncertainties in the data and in the Earth structure model. We compare the results to a well-established technique that uses a large number of hand-picked first-motion polarities.Sentinel-1 Interferometric Wide data both from the ascending and descending orbit is obtained and processed using the European Space Agency SNAP toolbox. The resulting images show a phase difference of about 2π corresponding to ground displacement of 3 cm. The hypocentre and centroid locations of the events suggest that both were initiated at a depth of around 10 km, the rupture propagated upward, and most of the energy was released at a depth of 4 to 5 km. The shallow depth of the earthquakes possibly led to measurable surface displacement and resulted in considerable damage in the epicentral area and the centre of Zagreb.

  • Open Access

    The Berkovići (BA) 22 April 2022 earthquake sequence – seismological and seismotectonic analysis

    by Iva Dasović, Davorka Herak, Marijan Herak, Helena Latečki, Marin Sečanj, Bruno Tomljenović, Snježana Cvijić-Amulić, Marija Mustać Brčić, Tena Belinić Topić, Josip Stipčević

    A strong earthquake, ML = 6.0 (MW = 5.7), occurred on 22 April 2022 at 21:07 UTC with an epicentre near Berkovići in Bosnia and Herzegovina, with focal depth of about 20 km. The earthquake was felt throughout Bosnia and Herzegovina, Montenegro, Croatia (especially Dalmatia), but also in Slovenia, Italy (especially the western coast of the Adriatic), Serbia, Albania and North Macedonia. The maximum intensity of the earthquake was rated as VII–VIII EMS in Berkovići and Ljubinje. A young woman in Stolac lost her life from a rock slide caused by the earthquake. In the wider epicentral area the earthquake caused a number of large or small rock falls, many chimneys were damaged, tiles fell from the roofs, plaster fell off, and there were also large cracks in the walls.By 31 October 2022, the DuFAULT project researchers located 6220 aftershocks (39 with ML≥ 3.0), with as many as 900 located in the first 12 h of the series. The strongest aftershock, ML = 4.9, occurred on 24 April 2022 at 4:27 UTC with focus at a depth of about 25 km and the epicentre also close to Berkovići. The vast majority of earthquakes have their foci relatively deep for this area, at depths between 15 and 28 km. Most of the epicentres form a compact group slightly elongated parallel to the NW-SE Dinaric strike, however two smaller groups northwest and southeast of the main group stand out with extension perpendicular to the Dinaric strike with somewhat shallower foci. The analysis of the focal mechanism and the hypocentral spatial distribution suggest that the mainshock resulted from the NE-SW directed compression and occurred on a reverse fault, on a moderately NE-dipping plane. Interestingly though, this series is also characterized by earthquakes released by a tension along the NE-SW striking and approximately 45° dipping normal faults, documented in the smaller north-western group.We will present spatio-temporal analysis of seismicity, resulting focal plane solutions and seismotectonic interpretation.

  • Open Access

    Seismic moment tensors from synthetic rotational and translational ground motion: Green’s functions in 1-D versus 3-D

    by S Donner, M Mustać, B Hejrani, H Tkalčić, H Igel

    SUMMARY: Seismic moment tensors are an important tool and input variable for many studies in the geosciences. The theory behind the determination of moment tensors is well established. They are routinely and (semi-) automatically calculated on a global scale. However, on regional and local scales, there are still several difficulties hampering the reliable retrieval of the full seismic moment tensor. In an earlier study, we showed that the waveform inversion for seismic moment tensors can benefit significantly when incorporating rotational ground motion in addition to the commonly used translational ground motion. In this study, we test, what is the best processing strategy with respect to the resolvability of the seismic moment tensor components: inverting three-component data with Green’s functions (GFs) based on a 3-D structural model, six-component data with GFs based on a 1-D model, or unleashing the full force of six-component data and GFs based on a 3-D model? As a reference case, we use the inversion based on three-component data and 1-D structure, which has been the most common practice in waveform inversion for moment tensors so far. Building on the same Bayesian approach as in our previous study, we invert synthetic waveforms for two test cases from the Korean Peninsula: one is the 2013 nuclear test of the Democratic People’s Republic of Korea and the other is an Mw 5.4 tectonic event of 2016 in the Republic of Korea using waveform data recorded on stations in Korea, China and Japan. For the Korean Peninsula, a very detailed 3-D velocity model is available. We show that for the tectonic event both, the 3-D structural model and the rotational ground motion, contribute strongly to the improved resolution of the seismic moment tensor. The higher the frequencies used for inversion, the higher is the influence of rotational ground motions. This is an important effect to consider when inverting waveforms from smaller magnitude events. The explosive source benefits more from the 3-D structural model than from the rotational ground motion. Nevertheless, the rotational ground motion can help to better constraint the isotropic part of the source in the higher frequency range.


  • Large Isotropic Component in the Source Mechanism of the 2013 Democratic People’s Republic of Korea Nuclear Test Revealed via a Hierarchical Bayesian Inversion

    by Marija Mustać, Babak Hejrani, Hrvoje Tkalčić, Seongryong Kim, Sang-Jun Lee, Chang-Soo Cho

    The 12 February 2013 nuclear test conducted by the Democratic People’s Republic of Korea stands out among other nuclear tests because it produced unusually large transversal motions. Previous studies found various percentages of isotropic components of the seismic moment tensor (MT), which opens up an important question about the reliability of the methods and assumptions we routinely use to recover the seismic MT in the point source approximation. Of particular interest is the data noise model that can be utilized to represent the uncertainty associated with the recorded data. If the noise is not accounted for, this may result in a range of unwanted effects such as overfitting waveform data, and, in turn, it may lead to erroneous conclusions. We thus scrutinize the analyses of the seismic MT of this explosion by performing a thorough analysis of the source depth and time utilizing newly developed Earth structure models to invert seismograms at regional distances at different frequency bands. In addition, we estimate the solution uncertainty within a hierarchical Bayesian framework that allows accounting for noise in the data. Our results show that the resulting MT of this event contains an expectedly large isotropic component (about 70%) and a dip-slip faulting.


  • The Variability and Interpretation of Earthquake Source Mechanisms in The Geysers Geothermal Field From a Bayesian Standpoint Based on the Choice of a Noise Model

    by Marija Mustać, Hrvoje Tkalčić, Alexander L. Burky

    Moment tensor (MT) inversion studies of events in The Geysers geothermal field mostly focused on microseismicity and found a large number of earthquakes with significant non‐double‐couple (non‐DC) seismic radiation. Here we concentrate on the largest events in the area in recent years using a hierarchical Bayesian MT inversion. Initially, we show that the non‐DC components of the MT can be reliably retrieved using regional waveform data from a small number of stations. Subsequently, we present results for a number of events and show that accounting for noise correlations can lead to retrieval of a lower isotropic (ISO) component and significantly different focal mechanisms. We compute the Bayesian evidence to compare solutions obtained with different assumptions of the noise covariance matrix. Although a diagonal covariance matrix produces a better waveform fit, inversions that account for noise correlations via an empirically estimated noise covariance matrix account for interdependences of data errors and are preferred from a Bayesian point of view. This implies that improper treatment of data noise in waveform inversions can result in fitting the noise and misinterpreting the non‐DC components. Finally, one of the analyzed events is characterized as predominantly DC, while the others still have significant non‐DC components, probably as a result of crack opening, which is a reasonable hypothesis for The Geysers geothermal field geological setting.