Im Umfeld der Enguri-Talsperre sowie im Nenskra-Tal wurde ein leistungsfähiges seismologisches Messnetz installiert, um lokale natürliche und durch Stauseen bedingte Erdbeben zu erfassen. Der Beitrag beschreibt Aufbau und Betrieb des Monitoring-Systems, die Qualität der aufgezeichneten Daten und erste Ergebnisse zur Seismizität. Ziel ist es, die Grundlage für eine verbesserte Risikoabschätzung und die Sicherheit bestehender und geplanter Talsperren zu schaffen.

In Ecuador, in the vicinity of the seismic rupture of the 2016 Mw 7.8 Pedernales earthquake, the megathrust fault is also affected by aseismic slip at shallow depths, including slow earthquakes and post-seismic slow slip. The Ecuadorian margin is an exceptional natural laboratory. Its relatively narrow marine forearc and shallow megathrust make it an ideal location for studying the relationship between ongoing subduction of topographic highs, such as ridge and seamount, and seismic and aseismic slip behaviour on the plate interface.The HIPER Project (2020–2022), which aims to better characterize the 3D structure of the Ecuadorian forearc domain, is based on an international collaboration funded by the French Oceanographic Fleet, the French ANR, Karlsruhe Institute of Technology (KIT, Germany), American NSF and IG-EPN (Ecuador). We successfully deployed a large number of OBSs (47), land stations (~200) and nodes (~500) to record both R/V L’Atalante shots and seismic activity.Here, we present preliminary results from a 3D inversion of P-wave refraction and reflection data, which was performed with TOMO3D. Thanks to our newly developed semi-automatic picking tool, DeepFB, we were able to efficiently compile the catalogue of ~230,000 picks.. DeepFB is a U-Net based neural network designed for robust automatic first-break picking in active-source seismic data. It was extensively applied to our dataset and accounts for approximately 50% of the picked first arrivals. The resulting 3D P-wave velocity model provides new insights into the lateral velocity variations of the Ecuadorian margin and the subducting plate along the trench between latitudes 1°25′N and 0°10′S.

The next generation gravitational wave detector Einstein Telescope (ET) is planned to be built at a depth of about 200 m to 300 m to significantly reduce the influence of ambient seismic noise with respect to current surface detectors. Three candidate sites for ET are currently under investigation: Sardinia (Italy), Lausitz (Germany), and the Euregio Meuse-Rhine (EMR, Netherlands, Belgium, Germany). Broadband downhole and surface seismometers have been installed at all three sites over the last couple of years which now allows the comparison of seismic noise levels and reduction with depth. Furthermore, we include the Sos Enattos mine in Sardinia, as an additional reference location. We see a significant reduction in seismic noise with depth over a broad frequency range above 1 Hz and below 0.1 Hz. The most significant noise reduction is observed in the frequency band between 3 to 30 Hz for which all sites reach a noise level below 10-7 m2 s-4 Hz-1 at depth. For the Lausitz and EMR sites we measure a reduction of seismic noise with depth of up to 40 dB while Sardinia shows an exceptionally low seismic noise above 2 Hz even below the NLNM but shows the smallest improvement with depth because noise levels are remarkably low at the surface. The noise can be attributed to various sources such as anthropogenic and ocean generated microseisms. The EMR and Lausitz sites show a clear reduction of seismic noise during nights and weekends. The day/night and week/weekend dynamic of cultural noise is not very pronounced for the Sardinia site. Our favoured explanation for the extremely low noise level in Sardinia is therefore the low level of anthropogenic noise. However, the ocean generated microseism is strongest at the Sardinia site due to the Mediterranean Sea that is located only a few tens of kilometers from the candidate site location. The exceptionally low ambient noise level in Sardinia at above 2Hz exposes the self-noise of the Trilium Slimline borehole seismometer and the noise floor of the CENTAUR digitizer as the limiting factor at frequencies above 4 Hz.

Training deep-learning picking models with several published data sets can be easily done through the Python toolbox SeisBench. Most of the data sets contain earthquakes recorded at local, regional and teleseismic distances, with only limited data in the low magnitude, close distance region. Applying current published PhaseNet models to induced seismicity data leads to only a few events being detected and trained PhaseNet models are not able to outperform well-established workflows in seismology. Here we present a new seismological data set and trained PhaseNet models for picking induced seismicity with deep-learning (piSDL). PhaseNet was trained with 171,182 three component waveforms from 40,576 events. Noise samples were added in the training data set to reduce the number of false picks. In this study, we noticed that a good earthquake training data set and noise samples from the analysed area are both important to detect more seismic events with a newly trained PhaseNet model. We validated our new PhaseNet models at a geothermal site in Rittershoffen (France). The models trained with the new data set and noise samples clearly outperform PhaseNet’s original published model and traditional methods in seismology by detecting up to 62% more events compared to a seismicity catalogue published by an agency.

The simplified view of the subduction interface is that of a single plane along which seismic and aseismic deformation occurs. In reality, however, exhumed subduction zones and geophysical imaging have shown that the seismogenic plate interface is a deformed, 100m-1km thick tabular region. Within this region, we currently do not know if seismic slip is localized on a single fault or distributed over several active faults, and how this impacts seismogenesis and the timing of deformation. Here, we use high-resolution earthquake locations to shed light on these questions. We focus on the aftershock sequence of the March 27th 2022, Mw 5.8 Esmeraldas earthquake which occurred at 19 km depth at the plate interface in Ecuador, and which was recorded by the dense temporary seismic network deployed during the HIPER2 marine campaign. We use machine learning to detect and pick over 1700 earthquakes (Mw 0-3), which we then locate using a double difference algorithm with cross-correlation times and a 3D velocity model. This allows us to obtain an exceptionally detailed image of the seismicity at the plate interface, which falls into a 200-400 m thick zone, comparable to plate interface thicknesses observed in exhumed subduction zones. Using a cross-correlation threshold of 0.75, we extract families of similar earthquakes, whose geometry we investigate using the 3-point method. These families generally occur on subparallel, sometimes superposed planes with a thickness of 0-40 m that is comparable to the thickness of individual fault zones observed within fossil subduction shear zones. These individual fault zones appear to form a network whose geometry impacts the aftershock expansion, itself controlled by afterslip rather than diffusive processes, thus demonstrating the importance of considering the 3D structure of the plate interface when modeling slip.

Located within the active continental collision between Eurasia and the Adriatic microplate, Albania is an earthquake prone country with one of the highest seismic hazard in Europe. This came into evidence when the MW=6.4 Durrës earthquake hit the country in 2019, causing 51 fatalities and widespread damage to infrastructure. Despite this stark reminder, the seismotectonics of Albania remains poorly researched, holding many unknowns regarding active seismogenic faults and 3D velocity structure. In an attempt to fill-in this knowledge gap, we conceived the project ANTICS (AlbaniaN TectonIcs of Continental Subduction) to install a temporary network of 382 seismic stations and densely monitor the abundant seismic activity in central Albania. In this paper we introduce the project goals and seismic deployment, assessing data quality and extracting valuable lessons from such a complex large-N deployment. Finally, we present some preliminary results on the detected seismicity and a receiver function profile and expand on an outlook of the project and possible next steps in the area.

The eastern Adriatic margin with the Dinarides and Hellenides orogens is one of the most hazardous areas in Europe from an earthquake hazard perspective in spite of only moderate shortening rates (e.g. less than 0.5 cm/yr across the Dinarides), as exemplified by the recent highly damaging earthquakes in Durrës, Albania (2019, M6.4) and Petrinja, Croatia (2020, M6.4). Deformation is both fairly localised on shallowly NE dipping thrust faults near the coast, and distributed, with a transition to spatially extended extensional deformation in the southern Dinarides and Northern Hellenides. and a complex regime involving strike-slip and obblique mechanisms in the eastern part of the Northern Dinarides. Making sense of this distributed deformation requires highly accurate locations both horizontally and in depth, which can only be achieved with dense seismic observations. This region is thus being explored with multi-scale dense seismic deployments. On one hand, the multi-national AdriaArray initiative has combined temporary and permanent broadband stations to achieve a nearly-uniform coverage with typical inter-station distances of 30-40 km. In Albania, where the transition of the Dinarides to the Hellenides is occurring, this regional coverage is complemented by a ultra-dense deployment of nearly 400 stations (mostly geophones with a few broadband sensors) with a nominal spacing of 5 km, which was later rearranged into three orogen-perpendicular profiles, and one along-strike profile with 1 km station spacing. Such large numbers of stations require automated processing approaches leveraging recently developed machine learning-based techniques. The presentation will review the tectonic context for these surveys and share some preliminary results from these deployments as well as an earlier more localised deployment in the area of the Durrës earthquake.

Dense seismic networks are ideally suited to detect daily or even hourly variations of the global secondary microseism via ambient noise cross-correlation beamforming (CCBF) and backprojection (BP) in the slowness-backazimuth domain. We combine the seismic recordings from Hi-net in Kyushu (HINET) network, Southern California Seismic Network (SCSN), and the Large-N AlbaNian TectonIcs of Continental Subduction (ANTICS) network to capture 3-hourly and daily northern hemisphere secondary microseism variations during 2022-2023. We calculate stable ambient noise CC with 300 s lag and 24 substacks per day. In the secondary microseism period band, 1-10s, we detect clear and vigorous high apparent velocity P phase (> 8 km/s) arrivals in 3-hourly and daily stacks for these networks. Both the 3-hourly and daily stacks show clear temporal amplitude and delay time changes in station-pair-distance and symmetry changes of causal and acausal branches, indicating the active evolution of ambient noise source location and strength. For ANTICS, the strongest energy patch emerges with back-azimuth (BAZ) 280°-330° and slowness around 8-10 s/deg. Further two energy patches appear with BAZ 90°-135° and slowness of 4-6 s/deg as well as 0° BAZ and slowness of 5-7.5 s/deg . We back-project the energy from the beamforming to the source location based on IASP91 velocity model assuming the propagation of teleseismic energy as direct P wave (including Pdiff, PKiKP and PKIKP). The back-projection results reveal that the strongest energy comes from the North Atlantic covering a broad arc-shape area (from the northeast coast of the US to the west coast of the UK, and from the south of Greenland and Iceland down to 45°N). The two other energy patches with much higher apparent velocities originate from the south Indian Ocean and the north Pacific near the Aleutian Islands. The 3-hourly and daily changes are tracked and recovered by the CCBF and BP approach for all three networks. The secondary microseism variations in the north Pacific could be improved by the SCSN and HINET whereas the north Atlantic is constrained by ANTICS and SCSN. Some small-scale autumn storms near the Japanese trench are also detected and tracked. Our results are consistent with existing wave height maps and provide a new and cheap observation for hindcasting of the state of the coupling of oceans and the solid earth.

Subduction zones generate the largest earthquakes on Earth, yet their detailed structure, and its influence on seismic and aseismic slip, remains poorly understood. Geological studies of fossil subduction zones characterize the seismogenic interface as a 100 m-1 km thick zone1-3 in which deformation occurs mostly on metres-thick faults1,3-6. Conversely, seismological studies, with their larger spatial coverage and temporal resolution but lower spatial resolution, often image the seismogenic interface as a kilometres-wide band of seismicity7. Thus, how and when these metre-scale structures are active at the seismic-cycle timescale, and what influence they have on deformation is not known. Here we detect these metres-thick faults with seismicity and show their influence on afterslip propagation. Using a local three-dimensional velocity model and dense observations of more than 1,500 double-difference relocated earthquakes in Ecuador, we obtain an exceptionally detailed image of seismicity, showing that earthquakes occur sometimes on a single plane and sometimes on several metres-thick simultaneously active subparallel planes within the plate interface zone. This geometrical complexity affects afterslip propagation, demonstrating the influence of fault continuity and structure on slip at the seismogenic interface. Our findings can therefore help to create more realistic models of earthquake rupture, aseismic slip and earthquake hazard in subduction zones.

Earthquakes that rupture several faults occur frequently within the shallow lithosphere but are rarely observed for intermediate‐depth events (70–300 km). On 29 November 2007, the Mw7.4 Martinique earthquake struck the Lesser Antilles Island Arc near the deep end of the Wadati‐Benioff‐Zone. The sparse regional seismic network of 2007 previously hampered a detailed examination of this unusually complex event. Here, we combine seismic data from different studies with regional moment tensor inversion results and 3D full‐waveform modeling. We show that the earthquake is a doublet consisting of dip‐slip and strike‐slip motion along two oblique structures, both activated under extensional stress along the strike of the slab. Comparison with tectonic reconstructions suggests that the earthquake ruptured along a re‐activated ridge‐transform segment of the subducted Proto‐Caribbean spreading ridge. The unprecedented resolution of the source process highlights the influence of pre‐existing structures on localizing slab deformation also at intermediate‐depth.

High spatial and temporal resolution of distributed acoustic sensing (DAS) measurements makes them very attractive in different applications in seismology, such as seismic noise analysis (e.g. Bahavar et al 2020, Spica et al 2020) and seismic event detection (e.g. Ajo-Franklin et al 2019, Fernandez Ruiz 2020, Jousset 2020). The quantity measured by a DAS is strain or strain rate of an optic fiber cable, which is related to the spatial gradient of displacement and velocity that is usually measured by single point seismometers. The amplitude (and signal to noise ratio, SNR) and frequency resolutions of DAS recordings depend on spatial and temporal acquisition parameters, such as i.e. gauge-length (GL) and derivative time (DT), the latter being of importance only if the device records the strain rate. In this study, our aims have been to investigate, experimentally, how to adapt the averaging parameters such as GL and DT to gain sensitivity in frequency bands of interests, and to investigate the seismic event detection capability of DAS data under specific set up. We recorded samples of DAS raw data, over a few hours at the German Black Forest Observatory (BFO) and in Sardinia, Italy. We studied the spectral characteristics of strain and strain rate converted from DAS raw data, to analyze the sensitivity of DAS measurements to GL and DT. The power spectral densities are compared with the strain meter recordings at BFO site as a benchmark, which is recorded using the strain-meter arrays measuring horizontal strain in three different directions independently from the DAS (For details about the DAS measurement station at BFO see Azzola et al. EGU 2022). We concluded about the lower limit of the DAS noise level that is achievable with employing different acquisition parameters. Accordingly, we applied suitable parameters for continuous strain-rate data acquisition at another experimental site in Georgia, which is related to the DAMAST (Dams and Seismicity) project. During the acquisition time periods at BFO and in Georgia, the visibility of local, regional and teleseismic events on the DAS data has been investigated. At both sites, a broadband seismometer is continuously operating, and can be considered as a reference to evaluate the event detection capability of the DAS recordings taking into account the monitoring set-up, i.e. cable types, cable coupling to the ground, directional sensitivity and acquisition parameters. In addition, at BFO the DAS seismic event detection capability is evaluated comparing with the strain-meter array. Examples of detected seismic events by DAS are discussed, in terms of achievable SNR for each frequency content and comparison with the seismometers and strain-meter array.

Monitoring crustal movements is essential to volcanic hazard assessment in areas of active volcanism. These surface movements occur on a wide range of time scales and wavelengths. However, the origin of crustal movements is not always associated with volcanic activities, particularly in areas with rigorous human activities (i.e., ground water extraction). It is challenging yet critical to distinguish between the ongoing volcanic and anthropogenic activities. In this study, we focus on the East Eifel Volcanic Field, which consists of multiple active Quaternary volcanoes. We report areas of uplift and subsidence 2-3 km away from each other near the Laacher See volcanic crater (2-3 km distance), and investigate the mechanisms responsible for the reversed deformation in such close proximity. PS-InSAR measurements by the BodenBewegungsdienst Deutschland (BBD) show notable ground displacements in this area for the period between 2014 and 2019. The deformation is clearly mapped by three different tracks of the Sentinel-1 satellite – two ascending and one descending, which confirms the robustness of the signal being detected by PS-InSAR. The main deformation is round in shape, and the rates peak up to 10 mm per year in line-of-sight (LOS) for the uplift area near the village Glees and reach down to -4 mm LOS for the subsidence zone in the vicinity of the village Wehr. To investigate the likely mechanism responsible for the ground displacements, we model the crustal movements with two spherical pressure point sources (i.e., the Mogi sources) simultaneously using a combined global and local optimization scheme. In the inversion, we search for the optimal combinations for a set of four parameters (latitude, longitude, depth and volume) for each Mogi source. The global optimization is achieved by Multi-Level Single-Linkage algorithm and we use the PRAXIS algorithm to find the local minimum. We include all three tracks of data, of which the different satellite viewing geometries help stabilize the inversion. Our results show that the uplift trend in Glees can be explained by an additional volume of 13000 m³ per year at 530 m depth. The subsidence near Wehr can be best fitted by a decrease in volume of 1700 m³ per year at 340 m depth. The modelling results show a trade-off between depth and volume, however, the uncertainties are smaller for the subsidence source near Wehr. Residuals trending in SW-NE direction are observed at the Glees uplift area, and the relatively large parameter uncertainties for Glees uplift zone are likely due to sparse persistent scatters there. Given the shallow depth of the Mogi sources, we interpret the Glees uplift being predominantly associated with fluid refilling in the respective volume caused by former CO2 extraction. The subsidence around Wehr is linked to ongoing industrial CO2 extraction. Our study identifies anthropogenic factors that may cause ground deformation in an active volcanic region, and has implications for future volcanic hazard assessment.

Persistent microseismicity in the East Eifel Volcanic Field occurs along the Ochtendung Fault Zone (OFZ) just SE of Laacher See Volcano. In addition, deep-low-frequency earthquakes close by are a strong indication for active magmatic processes. No surface expression is known for the OFZ, therefore an active seismic study was conducted in the summer 2021 aiming to detect the near-surface structure of the fault. The survey follows a line nearly perpendicular to the assumed fault orientation. The total length of the survey is 4,500 m with 5 m geophone distance and a maximum offset of 1000 m. Additional to these vertical component geophones, 3-component sensors were deployed at several sites along the profile in order to record far offsets. A drop-weight served as a seismic source. 1,022 shots lead to a total of more than 225,000 channels with maximum offsets of up to 1km, if including the 3-component sensors even up to 5km. Standard QC procedures and the stacking of the single shots at each shot point was done. The data set comprises 177 shot gathers with up to 221 receivers active at the same time. On these data the first onset P-wave arrivals were determined resulting in more than 35000 picks. The refraction tomography uses an innovative inversion technique harnessing the power of a transdimensional, hierachical Markov chain Monte Carlo (McMC) algorithm without the need of a priori assumptions. The number of Voronoi cells describing the Earth structure model and the level of data noise is automatically determined during the inversion process. The forward modelling is performed by a fast, finite-difference based eikonal solver. Starting several hundred McMC-chains across multiple CPU-cores leads to the parallelism needed for efficient sampling of the model space, thus computing of a refraction tomography 2-D Earth structure model including its uncertainty. We achieve a good resolution in depth down to about 200 m throughout our model. The thickness of the tephra layer covering the Rhenish shield is increasing from SW (few meters) to NE (80 m) along the profile. Further studies are still needed to illuminate the shallow structure of the OFZ.

We present an original DAS measurement station, equipped with the Febus A1-R interrogator, which has been deployed at the Black Forest Observatory (Schiltach, Germany). The objective of this deployment is twofold. The first is to test the deployed fibre optic cables and to better characterise the recorded signals. The second is to define standards for the processing of these DAS measurements, with a view to using the equipment for passive seismic monitoring in the INSIDE project (supported by the German Federal Ministry for Economic Affairs and Energy, BMWi).Testing sensors involving new acquisition technologies, such as instruments based on Distributed Fiber Optic Sensing (DFOS), is part of the observatory's goals, in order to assess, to maintain and to improve signal quality. Interestingly, reference geophysical instruments are also deployed on a permanent basis in this low seismic-noise environment. Our analyses thus benefit from the records of the observatory's measuring instruments, in particular a set of three strain meters recording along various azimuths. This configuration enables a unique comparison between strain meter and DAS measurements. In addition, an STS-2 seismometer (part of German Regional Seismic Network, GRSN) allows for additional comparisons.These instruments provide a basis for a comparative analysis between the DAS records and the measurements of well-calibrated sensing devices (STS-2 sensor, strain meter array). Such a comparison is indeed essential to physically understand the measurements provided by the Febus A1-R interrogator and to characterise the coupling between the ground and the fiber, in various deployment configurations.We present the experiment where we investigate several Fiber Optic Cable layouts, with currently our most successful setup involving loading a dedicated fiber with sandbags. We discuss different processing approaches, resulting in a considerable improvement of the fit between DAS and strain array acquisitions. The presented comparative analysis is based on the recordings of different earthquakes, including regional and teleseismic events.

Offering a web-interface alongside a back-end for the calculation of probabilistic power-spectral-densities (PPSD) is the main goal of this project. No simple out-of-the-box open-source solution is available so far. Moreover, researchers should get access to the information needed in a straightforward way. The project was initiated to process data routinely acquired at the KIT GPI seismological data center in order to enhance the quality control of the acquired data and station metadata. Furthermore, the PPSD gallery should develop into a starting point for further scientific research. Here we present SeisPPSD, a comprehensive solution for the calculation and inspection of PPSDs building on existing ObsPy codes. Next to an intuitive gallery and archive which offers a fine granularity in time, the user can create PPSDs interactively, e.g. comparing the day and night noise levels. As well, plots with noise levels for distinct frequencies over time can be visualized. The web front-end is mainly written in HTML but using Python and Javascript for the interactive parts. The back-end is implemented in Python and is distributing the PPSD calculation in a scalable fashion to the HPC environment. Apart from an easy to use web-interface, the researcher has access to an archive with the data derived product. This creates the opportunity for the researchers to customize plots with the already (pre-)calculated PPSD files. Furthermore, those files are relatively small and therefore uncomplicated to share with researchers outside. The PPSD-files can be used with the standard ObsPy module opening the possibility for further collaboration.

Structural fault complexity at depth affects seismic hazard, earthquake physics, and regional tectonic behavior, but constraining such complexity is challenging. We present earthquake source models of the February 21, 2008, Mw 6.0 Wells event that occurred in the Basin and Range in the western USA, suggesting the rupture of both the shallow and deep parts of a listric fault. We use a large data set including 150 local seismic waveforms from the USArray combined with high‐quality Interferometric Synthetic Aperture Radar and teleseismic waveforms. Rather than imposing an a priori fault geometry in the source inversions, as is often done in the literature, we use a data‐driven approach whereby all the faulting parameters and number of faults are determined by the data alone. We find a two‐fault normal faulting solution comprising: (i) a shallow (centroid depth ∼4.6 km) sub‐event with Mw 5.3 and fault dip of ∼77°; and (ii) a deeper (centroid depth ∼8.8 km), larger Mw 6.0 sub‐event on a fault with shallower dip angle (∼41°). Our preferred two‐fault model is consistent with aftershocks and with the tectonics of the region. The local USArray waveforms used in the modeling are key to detect the rupture of both shallow and deep parts of the possible listric fault. The lack of such dense and uniform coverage of earthquakes in other regions on Earth may explain why the full seismic rupture of listric faults may have gone undetected in the past. Thus, earthquake slip on whole listric faults may be more common than previously thought.

SUMMARYDetermining the crustal structure of ocean island volcanoes is important to understand the formation and tectonic evolution of the oceanic lithosphere and tectonic swells in marine settings, and to assess seismic hazard in the islands. The Azores Archipelago is located near a triple junction system and is possibly under the influence of a mantle plume, being at the locus of a wide range of geodynamic processes. However, its crustal structure is still poorly constrained and debated due to the limited seismic coverage of the region and the peculiar linear geometry of the islands. To address these limitations, in this study we invert teleseismic Rayleigh wave ellipticity measurements for 1-D shear wave speed (VS) crustal models of the Azores Archipelago. Moreover, we test the reliability of these new models by using them in independent moment tensor inversions of local seismic data and demonstrate that our models improve the waveform fit compared to previous models. We find that data from the westernmost seismic stations used in this study require a shallower Moho depth (∼10 km) than data from stations in the eastern part of the archipelago (∼13–16 km). This apparent increase in the Moho depth with increasing distance from the mid-Atlantic ridge (MAR) is expected. However, the rate at which Moho deepens away from the MAR is greater than that predicted from a half-space cooling model, suggesting that local tectonic perturbations have modified crustal structure. The 1-D VS models obtained beneath the westernmost seismic stations also show higher wave speeds than for the easternmost stations, which correlates well with the ages of the islands except Santa Maria Island. We interpret the relatively low VS profile found beneath Santa Maria Island as resulting from underplating, which agrees with previous geological studies of the island. Compared to a recent receiver function study of the region, the shallow structure (top ∼2 km) in our models shows lower shear wave speed, which may have important implications for future hazard studies of the region. More generally, the new seismic crustal models we present in this study will be useful to better understand the tectonics, seismicity, moment tensors and strong ground motions in the region.

We present a new inversion method for modelling multiple fault sources combining seismic and geodetic data. The technique takes into account 3-D earth structure in the modelling and uses a Monte Carlo inversion scheme that extensively explores the parameter space, which enables the assessment of source parameter uncertainties. Eleven parameters are determined for each subfault: centroid latitude, longitude, depth, time-shift, strike, dip, rake, non-double couple component, average slip, fault length and width. Since multiple fault inversions in the literature are often carried out in an iterative way, we perform synthetic tests comparing simultaneous and iterative synthetic multiple fault inversions for realistic two-fault models including: (i) downdip listricity and (ii) along-strike branching. While the iterative approach leads to substantial errors in the source parameters obtained for the fault model with downdip listricity (e.g. errors of approximate to 22 degrees for fault strike, approximate to 2 degrees for dip, approximate to 41 degrees for rake and approximate to 0.1 in M-w for joint data inversions), the simultaneous inversions recover the input parameters well. This shows that erroneous source models can be obtained if the inversions do not determine all fault parameters simultaneously. Synthetic inversion tests including different levels of real data noise highlight the strong complementarity of five different data types: regional seismic data, teleseismic P and S waves, teleseismic surface waves and interferometric synthetic aperture radar (InSAR). The joint data inversions substantially reducewell-known trade-offs between the fault dimensions, fault depth and slip compared to inversions of InSAR or seismic data alone. The synthetic inversions show that local/regional seismic data are key to constrain the lowest magnitude subevents, notably for the two-fault configuration with downdip listricity, which cannot be well constrained by the InSAR or teleseismic data alone. This suggests that downdip segmentation can be difficult to detect in the absence of local/regional seismic data.

Source models of mid-oceanic earthquakes are often based only on far-field, teleseismic data. The uncertainties of all source parameters are rarely quantified, which restricts our understanding of how these events slip and how oceanic lithosphere is formed. Here, we perform moment tensor inversions for five M-w 4.6-5.9 earthquakes that occurred in the Azores archipelago near theMid-Atlantic Ridge in 2013-2016, taking advantage of the recently expanded seismic network in the region. We assess moment tensor uncertainties due to data and Earthmodel variability as well as the robustness of teleseismic versus local data inversions. We find that for the events studied: (i) existing 1-D Earth models of the region based on receiver function data lead to a slightly improved data fit of local data compared to a widely used regional model based on active seismic surveys; and (ii) using different 1-D Earth models in the local data inversions leads to a variability in the retrieved source parameters of 15 degrees-30 degrees in fault strike, 5 degrees-20 degrees in dip, and 20 degrees-60 degrees in rake, depending on the earthquake's magnitude and location. We study in detail the M-w 5.9 2013 April 30 Povoacao basin earthquake using 1-D and 3-D waveform modelling, for which reported values of strike, dip, and rake in earthquake catalogues differ by 60 degrees, 35 degrees, and 80 degrees. We find that our moment tensor solutions show a lower variability than in the catalogues and exhibit a persistent non-double-couple component of similar to 40-60 per cent, which is not due to a volumetric change. We suggest that it is potentially due to geometrically complex faulting in the Povoacao basin, notably curved faults. We find that the retrieved moment tensor solutions depend strongly on the earthquake's location. If an accurate location is used, joint inversions of local and teleseismic data can help to stabilize moment tensor solutions of oceanic earthquakes and reduce parameter trade-offs, compared to inversions of local data alone.