Search

On 10 September, 2008, an Mw 5.7 earthquake occurred under the Central Valley of northern Chile near the town of Pica at a depth of ∼33 km within the continental crust of the South America plate. We find this earthquake to be a high stress-drop, reverse-oblique event that generated unusually high ground accelerations of up to 0.67g. Overall, its observed ground motion intensities are considerably larger than those predicted by ground motion models, particularly at short periods. The source properties inferred through waveform modeling indicate reverse-oblique fault motion on a ∼75 km2 plane dipping to the northeast, which is corroborated by the located aftershock distribution. Stress-drop values of the mainshock and larger aftershocks were estimated through S-wave spectrum modeling, with values up to ∼250 MPa for the mainshock. The event occurred in a cold section of the continental crust under the Central Valley, and its fault kinematics and orientation are consistent with the dominant style of faulting and stress field under the neighboring Coastal Cordillera. Although our recurrence analysis shows that crustal events in the region occur at a lower rate than interplate and inslab events, crustal events of similar or higher magnitude than the Pica earthquake have occurred, on average, approximately once every three years in northern Chile, which could pose an important hazard to nearby populations or critical infrastructure.

Centroid moment tensor (CMT) parameters of earthquakes are routinely estimated to gain information on structures and regional tectonics. However, for small earthquakes (M

On 7 January 2020, an Mw 6.4 earthquake occurred in the northeastern Caribbean, a few kilometers offshore of the island of Puerto Rico. It was the mainshock of a complex seismic sequence, characterized by a large number of energetic earthquakes illuminating an east–west elongated area along the southwestern coast of Puerto Rico. Deformation fields constrained by Interferometric Synthetic Aperture Radar and Global Navigation Satellite System data indicate that the coseismic movements affected only the western part of the island. To assess the mainshock’s source fault parameters, we combined the geodetically derived coseismic deformation with teleseismic waveforms using Bayesian inference. The results indicate a roughly east–west oriented fault, dipping northward and accommodating ∼1.4 m of transtensional motion. Besides, the determined location and orientation parameters suggest an offshore continuation of the recently mapped North Boquerón Bay–Punta Montalva fault in southwest Puerto Rico. This highlights the existence of unmapped faults with moderate-to-large earthquake potential within the Puerto Rico region.

SUMMARYCentroid moment tensor (CMT) parameters can be estimated from seismic waveforms. Since these data indirectly observe the deformation process, CMTs are inferred as solutions to inverse problems which are generally underdetermined and require significant assumptions, including assumptions about data noise. Broadly speaking, we consider noise to include both theory and measurement errors, where theory errors are due to assumptions in the inverse problem and measurement errors are caused by the measurement process. While data errors are routinely included in parameter estimation for full CMTs, less attention has been paid to theory errors related to velocity-model uncertainties and how these affect the resulting moment-tensor (MT) uncertainties. Therefore, rigorous uncertainty quantification for CMTs may require theory-error estimation which becomes a problem of specifying noise models. Various noise models have been proposed, and these rely on several assumptions. All approaches quantify theory errors by estimating the covariance matrix of data residuals. However, this estimation can be based on explicit modelling, empirical estimation and/or ignore or include covariances. We quantitatively compare several approaches by presenting parameter and uncertainty estimates in nonlinear full CMT estimation for several simulated data sets and regional field data of the Ml 4.4, 2015 June 13 Fox Creek, Canada, event. While our main focus is at regional distances, the tested approaches are general and implemented for arbitrary source model choice. These include known or unknown centroid locations, full MTs, deviatoric MTs and double-couple MTs. We demonstrate that velocity-model uncertainties can profoundly affect parameter estimation and that their inclusion leads to more realistic parameter uncertainty quantification. However, not all approaches perform equally well. Including theory errors by estimating non-stationary (non-Toeplitz) error covariance matrices via iterative schemes during Monte Carlo sampling performs best and is computationally most efficient. In general, including velocity-model uncertainties is most important in cases where velocity structure is poorly known.

The Askja volcanic system, located in the North Volcanic Zone of Iceland, consists of a central volcano with three nested calderas (Kollur, Askja, and Öskjuvatn) and a 20 km wide and ~190 km long fissure swarm with a NNE-SSW trend. Kollur caldera is ~5 km wide and formed in the Pleistocene while the younger 8-km wide Askja caldera, the largest among the three, formed in the Holocene. The smaller (~4 km) and lake-filled Öskjuvatn caldera is located within the Askja caldera and formed following the 1875 Plinian eruption. This event was followed by several localized eruptions along the Öskjuvatn ring fault system (1921, 1922, and 1929) and the last eruption occurred in 1961 in correspondence with the Askja northern caldera border. After this eruption, the Askja caldera first underwent inflation for several years followed by slow (< 1 cm/yr) subsidence over decades. In early August 2021, the volcano entered a period of unrest with new earthquake activity located below the central volcano, and the GNSS station OLAC, located near the center of Askja caldera, started to uplift at a high rate (~3 cm/week). The uplift continued until the end of November 2021. Here we use SAR images acquired from four different orbits (two ascending and two descending) by the Sentinel-1 satellites to study the ground deformation during this unrest period. Only data from the first half of the unrest period could be used (until the end of September). Later, heavy snow resulted in the loss of interferometric coherence within the caldera, preventing retrieval of the deformation signal. The maximum ground displacement of ~10 cm (from the end of July to the end of September) was found at the center of the Askja caldera, near the western shore of Öskjuvatn Lake. Interestingly, the interferograms show an asymmetric deformation pattern that follows the ring faults in the northwestern part of Askja caldera. Analytical models suggest that a roughly 7 x 3 km2 NW-SE elongated sill inflated at a shallow depth of ~2 km below the Askja caldera. However, simple sill models cannot explain the asymmetrical deformation pattern observed in the InSAR data. Therefore, using boundary element modeling, we find that while the magmatic intrusion accounts for the broad uplift, possible ring-fault activity would localize the deformation close to the caldera rim. Furthermore, an elongated sill, like the one obtained from the first source estimation, would probably activate only a part of the ring-fault system, leading to an asymmetric deformation pattern.

Summary Data recorded by distributed acoustic sensing (DAS) along an optical fibre sample the spatial and temporal properties of seismic wavefields at high spatial density. Often leading to massive amount of data when collected for seismic monitoring along many kilometre long cables. The spatially coherent signals from weak seismic arrivals within the data are often obscured by incoherent noise. We present a flexible and computationally efficient filtering technique, which makes use of the dense spatial and temporal sampling of the data and that can handle the large amount of data. The presented adaptive frequency–wavenumber filter suppresses the incoherent seismic noise while amplifying the coherent wavefield. We analyse the response of the filter in time and spectral domain, and we demonstrate its performance on a noisy data set that was recorded in a vertical borehole observatory showing active and passive seismic phase arrivals. Lastly, we present a performant open-source software implementation enabling real-time filtering of large DAS data sets.

The Bayesian earthquake analysis tool (BEAT) is an open-source Python software to conduct source-parameter estimation studies for crustal deformation events, such as earthquakes and magma intrusions, by employing a Bayesian framework with a flexible problem definition. The software features functionality to calculate Green’s functions for a homogeneous or a layered elastic half-space. Furthermore, algorithm(s) that explore the solution space may be selected from a suite of implemented samplers. If desired, BEAT’s modular architecture allows for easy implementation of additional features, for example, alternative sampling algorithms. We demonstrate the functionality and performance of the package using five earthquake source estimation examples: a full moment-tensor estimation; a double-couple moment-tensor estimation; an estimation for a rectangular finite source; a static finite-fault estimation with variable slip; and a full kinematic finite-fault estimation with variable hypocenter location, rupture velocity, and rupture duration. This software integrates many aspects of source studies and provides an extensive framework for joint use of geodetic and seismic data for nonlinear source- and noise-covariance estimation within layered elastic half-spaces. Furthermore, the software also provides an open platform for further methodological development and for reproducible source studies in the geophysical community.

Large earthquakes are usually modeled with simple planar fault surfaces or a combination of several planar fault segments. However, in general, earthquakes occur on faults that are non-planar and e...

We present an approach for rapidly estimating full moment tensors of earthquakes and their parameter uncertainties based on short time windows of recorded seismic waveform data by considering deep learning of Bayesian Neural Networks (BNNs). The individual neural networks are trained on synthetic seismic waveform data and corresponding known earthquake moment-tensor parameters. A monitoring volume has been predefined to form a three-dimensional grid of locations and to train a BNN for each grid point. Variational inference on several of these networks allows us to consider several sources of error and how they affect the estimated full moment-tensor parameters and their uncertainties. In particular, we demonstrate how estimated parameter distributions are affected by uncertainties in the earthquake centroid location in space and time as well as in the assumed Earth structure model. We apply our approach as a proof of concept on seismic waveform recordings of aftershocks of the Ridgecrest 2019 earthquake with moment magnitudes ranging from Mw 2.7 to Mw 5.5. Overall, good agreement has been achieved between inferred parameter ensembles and independently estimated parameters using classical methods. Our developed approach is fast and robust, and therefore, suitable for down-stream analyses that need rapid estimates of the source mechanism for a large number of earthquakes.

We used teleseismic waveforms and ground deformation data from GNSS and InSAR to estimate source fault parameters of the Mw6.4 earthquake that occurred just offshore southwestern Puerto Rico on 7 January 2020. The mainshock was a part of an energetic seismic sequence that started on 28 December 2019 and led to a Mw5.8 earthquake on 6 January 2020, a day before the Mw6.4 mainshock. The ground-shaking due to the largest earthquakes of the sequence caused significant damage to buildings and infrastructures in Puerto Rico and one casualty was reported by the local media. The mainshock was followed by a strong aftershock sequence that included four Mw ≥ 5 events within the first 3 hours. In the first 40 days of the seismic sequence, data from the Puerto Rico Seismic Network were used to locate ~3800 earthquakes of magnitude > 2, illuminating an east-west elongated 30x50 km2 area, just offshore the southwestern coast of Puerto Rico. The region affected by this activity was before characterized by relatively low seismicity rates, even if a system of active faults, both onshore and offshore, had been mapped. The sequence is peculiar due to its complex development and many large aftershocks (magnitude > 4.5), with the mainshock releasing only ~60% of the total seismic moment.We estimated the key source parameters of the mainshock using teleseismic data, GNSS data from the Puerto Rico Geodetic Network, and InSAR data from the Sentinel-1 and ALOS-2 satellites. The modeled source is consistent with a ~15 km long and ~11 km wide blind fault, oriented roughly east-west and dipping 46o towards north, and with up to 1.1 m of oblique normal and left-lateral strike-slip.The optimal fault plane source indicates that it is an offshore continuation of the mapped North Boquerón Bay - Punta Montalva fault zone, supported by the large number of the aftershocks that trend along the same direction. However, most of the aftershocks, even those of magnitude > 5, occurred on other nearby faults, highlighting the complexity of this fault zone area.

On 2020 October 30, an Mw6.9 earthquake struck offshore Samos Island. Severe structural damages were observed in Greek Islands and city of Izmir (Turkey). 114 people lost their lives and more than a thousand people were injured in Turkey. The earthquake triggered local tsunami. Significant seismic activity occurred in this region following the earthquake and ~1800 aftershocks (M>1) were recorded by KOERI within the first three days. In this study, we analyze the slip distribution and aftershocks of the 2020 earthquake.For the aftershock relocations, the continuous waveforms were collected from NOA, Disaster and Emergency Management Authority of Turkey (AFAD) and KOERI networks. The database was created based on merged catalogs from AFAD and KOERI. For estimating optimized aftershock location distribution, the P and S phases of the aftershocks are picked manually and relocated with double difference algorithm. In addition, source mechanisms of aftershocks M>4 are obtained from regional body and surface waveforms.The surface deformation of the earthquake was obtained from both descending and ascending orbits of the Sentinel-1 A/B and ALOS2 satellites. Since the rupture zone is beneath the Gulf of Kusadası, earthquake related deformation in the interferograms can only be observed on the northern part of the Samos Island. We processed all possible pairs chose the image pairs with the lowest noise level.In this study, we used 25 continuous GPS stations which are compiled from TUSAGA-Aktif in Turkey and NOANET in Greece. In addition to continuous GPS data, on 2020 November 1, GPS survey was initiated and the earthquake deformation was measured on 10 GNSS campaign sites (TUTGA), along onshore of Turkey.The aim of this study is to estimate the spatial and temporal rupture evolution of the earthquake from geodetic data jointly with near field displacement waveforms. To do so, we use the Bayesian Earthquake Analysis Tool (BEAT).As a first step of the study, rectangular source parameters were estimated by using GPS data. In order to estimate the slip distribution, we used both ascending and descending tracks of Sentinel-1 data, ALOS2 and GPS displacements. In our preliminary geodetic data based finite fault model, we used the results of focal mechanism and GPS data inversion solutions for the initial fault plane parameters. The slip distribution results indicate that earthquake rupture is ~35 km long and the maximum slip is ~2 m normal slip along a north dipping fault plane. This EW trending, ~45° north dipping normal faulting system consistent with this tectonic regime in the region. This seismically active area is part of a N-S extensional regime and controlled primarily by normal fault systems.AcknowledgementsThis work is supported by the Turkish Directorate of Strategy and Budget under the TAM Project number 2007K12-873.

On 20 October 2020, Reykjavík was rocked by the largest earthquake in southwest Iceland in over a decade when a magnitude 5.6 event occurred only 25 km from the city. The earthquake caused movement on multiple surface fractures, distributed over an 8-km-long north-south oriented area, indicating the location of the underlaying right-lateral strike-slip fault rupture. We mapped the coseismic surface fractures and deformation using Sentinel-1 and TerraSAR-X InSAR data, selecting with a new method the best pre- and post-earthquake SAR scenes from analyzing the tropospheric signals on each SAR image. This method does not require masking out deformed areas when determining the InSAR covariance structure and thus yields better earthquake source estimations. As the InSAR data are primarily sensitive to east-west and vertical displacements, we additionally used split-beam interferometry to obtain more information about north-south displacements. For this, we used burst-overlap interferometry (BOI), in the case of Sentinel-1 data, and multiple-aperture interferometry (MAI) on the TerraSAR-X data. Together with the standard InSAR data, we estimated the full 3D coseismic surface displacement field of the earthquake. The results show that most of the fractures had limited surface offsets, apart from a 2-3 km long north-south trending segment just north of the epicenter that was right-laterally offset by about 15 cm. Source modeling of the earthquake shows that the deformation is consistent with a near vertical north-south striking fault with up to ~30 cm of slip located at roughly 3 km depth below the surface. The estimated geodetic moment of the model amounts to a magnitude 5.6 earthquake, consistent with seismological estimates. Most of the modeled fault slip and mapped surface fractures are located north of the earthquake epicenter, indicating that the earthquake ruptured unilaterally from south to north, which agrees with the more severe surface effects and shaking reported from near the northern end of the earthquake rupture.

Estimating fast earthquakes’ source mechanism is essential for near real time hazard assessments, which are based on shakemaps and further downstream analysis such as physics based aftershock probability calculations. The model and data uncertainties associated to the estimated source mechanism are also crucial. We propose a Baysian Machine Learning algorithm trained on normalized synthetic waveforms for estimating the full moment tensor of earthquakes almost instantaneously with associated source parameter uncertainties. A prior assumption is an appropriate location of the earthquakes along with its associated uncertainties. Here, this is obtained by already established Machine learning based algorithms, where the training data set is computed by forward calculations of synthetic waveforms based on Green’s functions calculated for a specified 1-D velocity model using the Pyrocko software package. The learned labels, which are the information learned by the Machine Learning algorithm associated to the data, are the moment tensor components, described with only five unique parameters. For predefined locations in an area of interest we train a full independent Bayesian Convolutional Neural Network (BNN). With variational inference the weights of the network are not scalar but represent a distribution of weights for the activation of neurons. Each evaluation of input data into our BNN yields therefore to a set of predictions with associated probabilities. This allows us to evaluate an ensemble of possible source mechanisms for each evaluation of input waveform data. As a test set, we trained our models for an area south of the Coso geothermal field in California for a fixed set of broadband stations at maximum 150 km distance. We validate our approach with a subset of earthquakes from the Ridgecrest 2019-2020 sequence. For this data set we compare the results of the estimates of our Machine Learning based approach with independently determined focal mechanism and moment tensors. Overall, we benchmark our approach with data unseen during the training process of the Machine Learning models and show its capabilities for generating similar source mechanism estimations as independent studies within only a few seconds processing time per earthquake. We finally apply the method to seismic data of a research network monitoring the area around two south-german geothermal power plants. Our approach demonstrates the potential of Machine Learning for being implemented in operational frameworks for fast earthquake source mechanism estimation with associated uncertainties.

Earthquake fault ruptures are typically complex and can consist of en echelon segments, have bends, large step-overs, and be curved or warped at different spatial scales. Although surface fault ruptures can be mapped using a variety of geological and geophysical techniques, the subsurface topology of faults is challenging to estimate. One of the main options is to use geodetic data (InSAR and GPS) of coseismic surface deformation to estimate the subsurface earthquake fault geometry along with the distributed slip. The general practice is to assume a planar fault surface and estimate the strike and dip of a simple rectangular fault prior to the spatially-variable slip estimation. Using such simplistic fault geometry during source fault estimations of large earthquakes rarely captures all the crustal deformation details seen in the data and can cause biased estimation results of the fault slip. Here, we show how complex non-planar fault geometry can be estimated simultaneously with spatially-variable slip from geodetic data in a Bayesian framework, where our non-planar fault geometry parametrization approach allows for various undulations of the fault surface in both the along-strike and down-dip directions.We exemplify this approach through synthetic tests considering a checkerboard-like slip pattern on a listric non-planar fault. The results show that fault slip can be over-estimated by about 50-100% when using pre-assumed planar fault geometry. In contrast, both the non-planar fault geometry and spatially-variable slip are better retrieved when using our estimation approach. We then apply this modeling approach to the 2011 MW9.1 megathrust Tohoku-Oki (Japan) earthquake. Here we use prior information like the location of the trench and earthquake hypocenters during the Bayesian estimation to reduce the extent of the model space. The resulting fault geometry shows variations in fault dip in both the along-strike and down-dip directions that compare well with Hayes’ slab1.0 model of the subduction interface. The estimated fault slip is also comparable to the results that pre-defined the fault geometry using the slab1.0 model. In the future, the proposed method could lead to more realistic source models of major earthquakes, aided by improving computational resources and spatial resolution of geodetic data.

Based on >20 years of GPS observations and seismological works, direct constraints on the strain accumulation along the Main Marmara Fault (MMF) show different characteristics from Princess Islands to Ganos Fault. Ganos and Princess Island segments identified as locked based on GPS and seismological observations while the part of the Central Marmara segment show partially creeping behaviour. Moreover, around the Kumburgaz Basin, GPS data could be explained by creep models in contrast to fully locked seismic models. Clearly, there are many puzzling questions on the nature of strain accumulation on the MMF, under the constrain of various data sets. In order to contribute to a better understanding of this fault the observation capacity of geodesic network has been increased along the northern coast of the Marmara Sea and supported by seismological stations as well as Marine Geodesy.In September, 2019, an intense earthquake activity started between Central Marmara and Kumburgaz Basin. The mainshock occurred on 26 September 2019 (Mw5.8) as a largest earthquake, since 1963 Mw6.3 Cinarcik earthquake, in the Marmara Sea. A foreshock activity started 4 days before the mainshock and the largest one (Mw4.7) observed on September 24, close to the mainshock. The mechanisms of the mainshock and the large aftershocks as well as foreshocks are dominantly strike-slip with a significant reverse component. The aftershocks are located on the north of the MMF trace.Here we investigate the geodetic data related to this event, with the aim to shed some light on the complexly segmented MMF. We observed co-seismic offsets at the nearest 6 GPS stations (~12-20 km far to the epicenter) along the northern coastline of the Sea of Marmara. The estimated offsets are not big and change between 1-3 mm on horizontal and 1-10 mm on vertical components. All of the stations are located on the northern part of the hypocenter and exhibit predominant NS-direction movement, which is inconsistent with a primarily E-W right lateral transform system. Instead, the co-seismic pattern can be explained with a complex earthquake mechanism which has a dominant reverse component while the strike-slip component is relatively insignificant, based on Okada-type elastic models and geodetic moment magnitude obtained as ~6.2. The total cumulative moment using geodesy is much higher than the total seismic cumulative moment of earthquake activity (~M5.9), and the thrust component is also more dominant in comparison the focal mechanisms from regional data. Obviously, geodetic co-seismic offsets estimated from daily-based data and they include pre-and post-earthquake deformations. In addition, the tide-gauge data (station distance is 25 km far to epicenter) was analyzed and it shows the strong variations after Mw 4.7 and they faded out after Mw5.8. This sea level change, which temporally correlates with the seismic activity, gives important evidence about the possibilities of pre-earthquake activity. Using GPS time series, we intend to explore the pre-earthquake anomalies and, to reduce the discrepancy between seismological and geodetic models. (This study is supported by TUBITAK 1001 Project no: 117Y278).

We present a modular open-source software framework - Kite (http://pyrocko.org) - for rapid post-processing of spaceborne InSAR-derived surface displacement maps. The software enables swift parametrisation, post-processing and sub-sampling of the displacement measurements that are compatible with common InSAR processors (e.g. SNAP, GAMMA, ISCE, etc.) and online processing centers delivering unrwapped InSAR data products, such as NASA ARIA or LiCSAR. The post-processing capabilities include removal of first-order atmospheric phase delays through elevation correlation estimations and regional atmospheric phase screen (APS) estimations based on atmospheric models (GACOS), masking of displacement data, adaptive data sub-sampling using quadtree decomposition and data error covariance estimation.Kite datasets integrate into forward modelling and optimisation frameworks Grond (Heiman et al., 2019) and BEAT (Vasyura-Bathke et al., 2019), both software packages aim to ease and streamline the joint optimisation of earthquake parameters from InSAR and GPS data together with seismological waveforms. These data combinations will improve the estimation of earthquake rupture parameters. Establishing this data processing software framework we want to bridge the gap between InSAR processing software and seismological modelling frameworks, to contribute to a timely and better understanding of earthquake kinematics. This approach paves the way to automated inversion of earthquake models incorporating space-borne InSAR data.Under development is the processing of InSAR displacement time series data to link simultaneous modelling of co- and post-seismic transient deformation processes from InSAR observations to physical earthquake cycle models.We demonstrate the framework’s capabilities with an analysis of the 2019 Ridgecrest earthquakes from InSAR surface displacements (provided by NASA ARIA) combined with GNSS displacements using the Bayesian bootstrapping strategy from the Grond inverse modelling tool.

Southwestern Anatolia is part of a N-S extensional regime mainly driven by rollback along the Hellenic subduction zone beneath the Aegean Sea. This seismically active area is controlled primarily by normal fault systems. The fault structures in the region are segmented and in many cases seismic interaction between these segments can be observed.2017 seismic activity along the Eastern and Western edges of Gökova Bay. Within the same year, three separate moderate sized (Mw~5) events took place near the town of Ula (Muğla) on the eastern edge of Gökova Bay. One of these earthquakes occurred in April before the Bodrum-Kos earthquake while the other pair occurred in November within two days.We relocated all the events that occurred in Ula region in 2017 and remodeled the source mechanisms from regional seismic waveforms by using Bayesian Earthquake Analysis Tool (BEAT). The surface deformations can also be clearly observed from InSAR tracks of both ascending and descending orbits. Because of the large noise margins of the interferograms, atmospheric noise corrections and high resolution DEM data were used.Due to temporal and spatial proximity of the two Mw~5 events during the November sequence, InSAR yields only cumulative deformation of the earthquakes. Therefore, we determined the contribution of each event to the cumulative static displacements observed by InSAR data, using the source models from seismic waveforms. The locations and the source mechanisms of the two Mw~5 earthquake are consistent and explain the observed surface deformation.Our results imply that these earthquakes occurred on a previously unknown normal fault rather than the southeastern branches of the nearby Muğla Fault as proposed earlier. The results are consistent with the recently mapped fault structure by Akyüz et al. (2018). The November activity implies EW trending, south dipping normal faulting system and the change in the strike direction of the fault on the eastern edge can be clearly seen both InSAR and waveform modelling results of April activity.Co-seismic and post-seismic InSAR analysis shows that the seismic activity following the 2017 Mw6.6 Bodrum-Kos propagated from western Gökova Bay where rupture occurred toward east including the Ula region. A long term comparison of seismicity beneath Gökova Bay and Ula region shows that the seismicity in these two regions are temporally correlated. Hence, while the aforementioned moderate sized earthquakes are not directly triggered by the Bodrum-Kos earthquake, increased seismic activity following Bodrum-Kos earthquake shows that the stress changes in these two regions affect each other. The location errors of the events especially the depth errors in the catalogs and the active fault structure in the area cannot be realized without any geodetic or seismic data analysis. This study claims that the interpretations of the moderate size earthquakes should be studied by using multidisciplinary data sets.ACKNOWLEDGEMENTSThis work is supported by the Turkish Directorate of Strategy and Budget under the TAM Project number DPT2007K120610.

The Mw 7.3 Sarpol-Zahab earthquake occurred on 12 November 2017 in the Lurestan arc of the Zagros Simply Folded Belt (ZSFB). It is estimated that 600 people were killed and 8000 were injured in this earthquake. This earthquake has been the largest instrumentally recorded earthquake in the ZSFB and its moment, as well as its mechanism, were unexpected. We present an earthquake source study on the Mw 7.3 Sarpol-Zahab earthquake, two large following earthquakes in the region in 2018 and their corresponding aftershock sequences to gain insight of seismotectonic of the Lurestan arc fold-thrust belt.In this study, we complement previous studies on this earthquake, by non-linear probabilistic optimization of joined geodetic and seismic data using a new, efficient Bayesian bootstrap-based optimization scheme to infer the finite fault geometry and fault slip together with meaningful uncertainty estimates of the model parameters. Our optimization is based on the modeling of ascending and descending Sentinel-1 satellite data, seismological waveform from global seismic networks and the strong motion network of Iran. The posterior mean model of the Sarpol-Zahab earthquake shows that the causative fault plane is centered at is 14±2 km depth and has a low dip angle of 17°±2° and a strike of 350°±10°. The rake angle of 144°±4° points to an oblique thrust mechanism. The rupture area of the uniform-slip, rectangular model is 40±2 km long and 16±2 km width and shows 4.0±0.5 m fault slip, which results in a magnitude estimate of Mw 7.3±0.1.Later, in August and November 2018, two large earthquakes with Mw 6.0 and Mw 6.4 occurred about 40 km east and 60 km south of the Sarpol-Zahab epicenter, respectively. These earthquakes could have been triggered by the 2017 Sarpol-Zahab earthquake. We apply the same joint inversion modeling to derive the corresponding fault plane solutions. We found strike-slip mechanisms for both events but centroid depths at 10±2 km and 16±2 km for Mw 6.0 and Mw 6.4, respectively.The 2017 Sarpol-Zahab earthquake and the following studied 2018 earthquakes were followed by a sustained aftershock sequence, with more than 133 aftershocks exceeding Ml 4.0 until December 30, 2019. We rely on the local and regional seismic broad-band stations of Iran and Iraq permanent networks to estimate full-waveform moment tensor solutions of 70 aftershocks down to Ml 4. Most of these aftershocks have shallow centroid depths between 5 and 12 km, so that they occurred in the uppermost part of the basement and/or in the lower sedimentary cover, which is ~8 km thick in this area.Our results suggest that the Sarpol-Zahab earthquakes activated low-angle thrust faults and shallower strike-slip structures, highlighting that both thin- and thick-skin deformation take place in the fold-thrust belts in the Lurestan arc of the Zagros. Such information on the deformation characteristics is important for the hazard and risk assessment of future large earthquakes in this region.Additionally, we demonstrate how the joint inversion of different geophysical data can help to better resolve the fault geometry and the earthquake source parameters.

Earthquakes have been observed to initiate and terminate near geometrical irregularities (bends, step-overs, branching of secondary faults). Rupture segmentationinfluences the seismic radiation and therefore, the related seismic hazard. Good imaging of rupture segmentation helps to characterize fault geometries at depth for follow-up tectonic, stress-field or other analyses. From reported earthquake source models it appears that large earthquakes with magnitudes above 7 are most often segmented, while earthquakes with magnitudes below 6.5 most often are not. If this observationreflects nature or if it is rather an artifact of our abilities to well observe and infer earthquake sources can not be answered without an objective strategy to constrain rupture complexity. However, data-driven analyses of rupture segmentation are not often conducted in source modeling as it is mostly pre-defined through a given and fixed number of sources. We, here, propose a segmentation-sensitive source analysis by combining a model-independent teleseismic back-projection and image segmentation methods with a kinematic fault inversion. Our approach is twofold. We first develop a time-domain multi-array back-projection of teleseismic data with robust estimations of uncertainties based on bootstrapping of the travel-time models and array weights (Palantiri software, https://braunfuss.github.io/Palantiri/). Backprojection has proven to be a powerful tool to infer rupture propagation from teleseismic data and identify irregularities of the rupture process over time.We then model the earthquake sources with the results obtained from the backprojection and additional information obtained from the application of image segmentation methods to the InSAR displacement maps. For this second step, we use a combination of different observations (teleseismic waveforms and surface displacement maps based on InSAR) to increase the resolution on the spatio-temporal evolution of fault slip. We develop a novel Informational criterion based transdimensional optimization scheme to model an adequate representation of the source complexity. We present our method on two cases study: the 2016 Muji Mw 6.7 earthquake (Pamir) and the 2008-2009 Qaidam (Tibet) sequence of earthquakes. We find that the 2008 Qaidam earthquake ruptures one segment, the 2016 Muji earthquake on two segments and the Qaidam 2009 earthquake on two or three segments.This work is based on the open-source, python-based Pyrocko toolbox and is conducted within the project “Bridging Geodesy and Seismology” (www.bridges.uni-kiel.de) funded by the DFG through an Emmy-Noether grant.

We thank Shin’ichi Miyazaki (Kyoto Univ.) for discussions and the Earthquake Research Institute (Univ. Tokyo) for hosting SJ during the early part of this work. Comments from Yukitoshi Fukahata (Kyoto Univ.) and an anonymous reviewer improved the manuscript. We thank Chiheb Hammouda (KAUST) for fruitful discussions. The ENVISAT data were provided by the European Space Agency through category-1 project 3639. Generic Mapping Tools (GMT) were used to make Figure 1. This research was supported by King Abdullah University of Science and Technology (KAUST).

Pyrocko is an open source seismology toolbox and library, written in the Python programming language. It can be utilized flexibly for a variety of geophysical tasks, like seismological data processing and analysis, modelling of waveforms, InSAR or GPS displacement data, or for seismic source characterization. At its core, Pyrocko is a library and framework providing building blocks for researchers and students wishing to develop their own applications. Pyrocko contains a few standalone applications for everyday seismological practice. These include the Snuffler program, an extensible seismogram browser and workbench, the Cake tool, providing travel-time and ray-path computations for 1D layered earthmodels, Fomosto, a tool to manage pre-calculated Green’s function stores, Jackseis, a command-line tool for common waveform archive data manipulations, Colosseo, a tool to create synthetic earthquake scenarios, serving waveforms and static displacements, and new, Sparrow, a 3D geophysical data visualization tool. This poster gives a glimpse of Pyrocko’s features, for more examples and tutorials visit https://pyrocko.org/.

Seismic source and moment tensor waveform inversion is often ill-posed or non-unique if station coverage is poor or signals are weak. Three key ingredients can help in these situations: (1) probabilistic inference and global search of the full model space, (2) joint optimisation with datasets yielding complementary information, and (3) robust source parameterisation or additional source constraints. These demands lead to vast technical challenges, on the performance of forward modelling, on the optimisation algorithms, as well as on visualisation, optimisation configuration, and management of the datasets. Implementing a high amount of automation is inevitable.To tackle all these challenges, we are developing a sophisticated new seismic source optimisation framework, Grond. With its innovative Bayesian bootstrap optimiser, it is able to efficiently explore large model spaces, the trade-offs and the uncertainties of source parameters. The program is highly flexible with respect to the adaption to specific source problems, the design of objective functions, and the diversity of empirical datasets.It uses an integrated, robust waveform data processing, and allows for interactive visual inspection of many aspects of the optimisation problem, including visualisation of the result uncertainties. Grond has been applied to CMT moment tensor and finite-fault optimisations at all scales, to nuclear explosions, to a meteorite atmospheric explosion, and to volcano-tectonic processes during caldera collapse and magma ascent. Hundreds of seismic events can be handled in parallel given a single optimisation setup.Grond can be used to optimise simultaneously seismic waveforms, amplitude spectra, waveform features, phase picks, static displacements from InSAR and GNSS, and gravitational signals.Grond is developed as an open-source package and community effort. It builds on and integrates with other established open-source packages, like Kite (for InSAR) and Pyrocko (for seismology).

Caldera unrest can lead to major volcanic eruptions. Analysis of subtle subsidence or inflation at calderas helps understanding of their subsurface volcanic processes and related hazards. Several subsiding calderas have shown similar patterns of ground deformation composed of broad subsidence affecting the entire volcanic edifice and stronger localized subsidence focused inside the caldera. Physical models of internal deformation sources used to explain these observations typically consist of two magma reservoirs at different depths in an elastic half-space. However, such models ignore important subsurface structures, such as ring faults, that may influence the deformation pattern. Here we use both analog subsidence experiments and boundary element modeling to study the three-dimensional geometry and kinematics of caldera subsidence processes, evolving from an initial downsag to a later collapse stage. We propose that broad subsidence is mainly caused by volume decrease within a single magma reservoir, whereas buried ring-fault activity localizes the deformation within the caldera. Omitting ring faulting in physical models of subsiding calderas and using multiple point/sill-like sources instead can result in erroneous estimates of magma reservoir depths and volume changes. (C) 2019 Elsevier B.V. All rights reserved.

Earthquake localization is both a necessity within the field of seismology, and a prerequisite for further analysis such as source studies and hazard assessment. Traditional localization methods often rely on manually picked phases. We present an alternative approach using deep learning that once trained can predict hypocenter locations efficiently. In seismology, neural networks have typically been trained with either single-station records or based on features that have been extracted previously from the waveforms. We use three-component full-waveform records of multiple stations directly. This means no information is lost during preprocessing and preparation of the data does not require expert knowledge. The first convolutional layer of our deep convolutional neural network (CNN) becomes sensitive to features that characterize the waveforms it is trained on. We show that this layer can therefore additionally be used as an event detector. As a test case, we trained our CNN using more than 2000 earthquake swarm events from West Bohemia, recorded by nine local three-component stations. The CNN successfully located 908 validation events with standard deviations of 56.4 m in east-west, 123.8 m in north-south, and 136.3 m in vertical direction compared to a double-difference relocated reference catalog. The detector is sensitive to events with magnitudes down to M-L = -0.8 with 3.5% false positive detections.