Enter Zoom Meeting

SM3.2

EDI
New Seismic Data Analysis Methods and Tools for Automatic Characterization of Seismicity

In the last two decades, the number of high quality seismic instruments being installed around the world has grown exponentially and probably will continue to grow in the coming decades. This led to a dramatic increase in the volume of available seismic data and pointed out the limits of the current standard routine seismic analysis, often performed manually by seismologists. Exploiting this massive amount of data is a challenge that can be overcome by using new generation, fully automated and noise-robust seismic processing techniques. In the last years, waveform-based detection, location, and source-parameter estimation methods have grown in popularity and their application have dramatically improved seismic monitoring capability. Moreover, machine learning and deep learning techniques, which are dedicated methods for data-intensive applications, are showing promising results in seismicity characterization applications opening new horizons for the development of innovative, fully automated and noise-robust seismic analysis methods. Such techniques are particularly useful when working with data sets characterized by large numbers of weak events with low signal-to-noise ratio, such as those collected in induced seismicity, seismic swarms and volcanic monitoring operations. This session aims on bringing to light new methods and tools and also optimizations of existing approaches that make use of High Performance Computing resources (CPU, GPU) and can be applied to large data sets, either retro-actively or in (near) real-time, to characterize seismicity (i.e. perform detection, location, magnitude and source-mechanism estimation) at different scales and in different environments. We thus encourage contributions that demonstrate how the proposed methods help improve our understanding of earthquake and/or volcanic processes.

Convener: Nima NooshiriECSECS | Co-conveners: Natalia Poiata, Federica LanzaECSECS, Francesco Grigoli, Simone Cesca
Presentations
| Fri, 27 May, 14:05–16:40 (CEST)
 
Room 0.16

Fri, 27 May, 13:20–14:50

Chairpersons: Nima Nooshiri, Simone Cesca, Francesco Grigoli

14:05–14:08
Introduction

14:08–14:14
|
EGU22-11708
|
ECS
|
|
On-site presentation
Géraldine Zenhäusern et al.

NASA's InSight mission continues to record seismic data over 3 years after landing using its very broadband seismometer. The situation of working with a single station requires efficient back-azimuth determination based on data of available body wave phases in the seismic record.

This study presents an effective way to estimate back azimuths using a comprehensive polarisation analysis. It uses a continuous wavelet transform to transform the seismic signal into time-frequency domain, and then performs an eigenanalysis of the spectral matrix to obtain information on the polarisation of the signal. Non-polarised signals are masked to enhance the seismic signal. We use the polarisation around both the P- and S-wave arrivals in selected frequency bands to estimate the back azimuth. For stronger signals, the P-wave polarisation provides the main information. For weaker signals, the result can be improved significantly based on the orthogonality of the P- and S-wave polarisation vector, which constrains the result for poorly polarised/contaminated P signals. This method is applied to synthetic marsquakes and to well-located earthquakes recorded in Tennant Creek, Australia. We find that the polarisation method reliably estimates the back azimuth for both sets of events.

The Marsquake Service has provided distance estimates for around 35 marsquakes, but only 10 had been assigned back azimuths. Back azimuth estimation – based on the polarisation at narrow-band of initial P-wave energy - is particularly challenging due to the highly scattered signals and noise in the seismic data. Our method, when applied to martian data, obtains results for 30 events in total, significantly improving our understanding of the spatial distribution of seismic activity on Mars. Most of the located events lie in the general Cerberus Fossae region, a large graben structure towards the east of InSight, though we also find quakes in other directions (e.g. north, towards Elysium Mons) that had previously not been expected to be tectonically active. This extended set of located marsquakes will allow for interpretation of martian tectonics, in particular the Cerberus Fossae region. The method could potentially be applied to sparse terrestrial networks, such as ocean bottom seismometers.

How to cite: Zenhäusern, G., Stähler, S., Clinton, J., Giardini, D., Ceylan, S., and Garcia, R.: Low Frequency Marsquakes and Where to Find Them: Automated Event Back Azimuth Determination Using a Multi-Body Wave Polarisation Analysis Approach, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-11708, https://doi.org/10.5194/egusphere-egu22-11708, 2022.

14:14–14:20
|
EGU22-8187
|
ECS
|
|
Virtual presentation
Ayon Ghosh et al.

We study microseismicity in the Kashmir seismic gap from broadband data of the Jammu and Kashmir Seismological NETwork (JAKSNET) using the template-matching technique. Template-matching is done by the Python routine ‘PyMPA’ (Vuan et al. 2018) to detect new events. For templates, we use 189 earthquakes, taken from the ISC reviewed catalogue, which occurred between 2013 and 2018. We use 5 second long waveform templates by taking 2.5 s before and after the S-wave arrival. Normalized cross-correlations are calculated for each template earthquake, recorded at different channels, with continuous data from their respective recording channels. The individual cross-correlation traces are shifted according to the travel times of the template earthquake, calculated using a local 1D velocity model, and stacked to get a Network Stack Function (NSF). A detection is declared if the NSF crosses a threshold value eight times the Median Absolute Deviation. We assign the location of the template, corresponding to the detection, as the location of the newly detected event. If multiple templates detect one event, we consider the one with the maximum NSF value. After running the process, we obtain a catalog of 935 events, an immediate 5-fold increase in the number of events. We also observe two clear sequences of events in the middle of 2013 and in the start of 2016. We intend to perform a probabilistic and relative relocation of all the events to get a detailed seismotectonics of the region.

How to cite: Ghosh, A., Vičič, B., Mitra, S., Priestley, K., and Wanchoo, S. K.: Detecting Microseismicity in Jammu and Kashmir Himalaya Using Template Matching Technique, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-8187, https://doi.org/10.5194/egusphere-egu22-8187, 2022.

14:20–14:26
|
EGU22-2007
|
ECS
|
On-site presentation
Giovanni Diaferia et al.

Improving the capability of seismic networks to detect small-magnitude seismicity, commonly near or below the detectability threshold, is the prerequisite to characterize the seismotectonics of an area in terms of fault geometry, kinematics and mechanics, thus leading to an improved comprehension on the physical mechanisms that generate small and large earthquakes. In this work, we apply template-matching, a cross-correlation based technique for the detection of hidden earthquakes, at the scale of the Southern Apennines (Italy). Here, the ongoing extension of the Mio-Pliocene Apennine thrust-belt poses a major seismic risk, as testified by several Mw~7 earthquakes that struck this area in the past 300 years. No clear consensus exists on the seismotectonic models related to such events, particularly in terms of characterization of the fault structure and crustal rheology that can thus largely benefit from the application of template-matching.

As template events, we use ~9000 earthquakes occurring between 2009 and 2015, recorded by 181 stations from the INGV National Seismic Network. Six years-long (2009-2015) continuous recordings are scanned by the template-matching algorithm. Of about 3 million new detections, around 3% (~88.000 events) comply with the minimum quality thresholds we set (at least four P and S picks, recorded at least at five stations). For determining earthquake locations we used the fully-probabilistic non-linear code NonLinLoc, with an ad-hoc 1D velocity model and corrections for station residuals.

By accounting for the quality of the hypocenter location, the final catalog comprises ~50.000 new seismic events with a mean horizontal and vertical error of 1.4 and 2.5 km, respectively, and a mean RMS of 0.13 s, parameters that are similar to those of the template catalog. Given the small magnitude (Mw<1) of the majority of the newly detected events, the new catalog shows a decrease in magnitude of completeness from 2.5 to 1.9, assessed through the Lilliefors’ goodness-of-fit test.

The spatial and temporal pattern of seismicity unravelled by the enhanced catalog provide new insights especially for those seismogenic structures that are poorly known. For the main seismic sequences that occurred in the analyzed period (i.e., Pollino and Matese Mw5+ sequences) the aftershocks as well as the foreshock phases appear particularly enriched. Main NW-SE trending seismogenic structures of the axial zone of chain are illuminated by abundant microseismicity, with evident gaps delineating the boundary of such structures. In addition, the new catalog unravels distinct E-W oriented clusters in the external zone of the seismic belt, likely related to shear zones developed in the deeper crystalline crust of the Adria plate.

How to cite: Diaferia, G., Valoroso, L., Piccinini, D., and Improta, L.: Earthquake catalogue enhancement through template matching: an application to the Southern Apennines (Italy), EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-2007, https://doi.org/10.5194/egusphere-egu22-2007, 2022.

14:26–14:32
|
EGU22-12443
Daniel Armbruster et al.

Aftershock sequences or earthquake swarms generate a high number of seismic events that are not detected by standard regional network routine processes. Undetected earthquakes are mostly due to low signal to noise ratio, overlapping earthquakes, and a network configuration that targets earthquake detection with a homogeneous magnitude of completeness. Furthermore, the analyst’s workload is increasing dramatically during an intense earthquake sequence, which results in prompt manual review of the largest events, only.

We present a computationally efficient and highly customizable tool (SCDetect) to detect earthquakes in near real-time by applying waveform cross-correlation in the time domain based on a set of template events. SCDetect is a free and open-source SeisComP extension module fully integrated into the SeisComP environment. It may be used to process both archived waveform data, when operated in playback mode, as well as real-time data. In either of the use cases, waveform data is accessed through SeisComP’s standard RecordStream interface. Multiple template event based detectors may be configured. The individual detector configuration is fully stream based which allows for generic multi-stream event detection. Event parameter products for newly detected events (i.e. origins, picks, amplitudes, station magnitudes) may be sent to SeisComP's messaging system for further processing. In addition to earthquake detection, we implement amplitude calculation by measuring amplitudes on the horizontal components. SCDetect offers multiple magnitude estimation methods based on the amplitudes of the template earthquakes and the new detections (i.e regression, amplitude ratios). Magnitude estimation is configurable using SeisComP’s bindings configuration.

We applied SCDetect to recent earthquake sequences in Switzerland between 2019 and 2021. The dense seismic network operated by the Swiss Seismological Service offers a unique opportunity to evaluate the performance of the proposed module. Our first results show that these extended earthquake catalogs contain at least ten times more earthquakes than the national earthquake catalogue.

How to cite: Armbruster, D., Mesimeri, M., Kästli, P., Diehl, T., Massin, F., and Wiemer, S.: SCDetect: Near real-time computationally efficient waveform cross-correlation based earthquake detection during intense earthquake sequences, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-12443, https://doi.org/10.5194/egusphere-egu22-12443, 2022.

14:32–14:38
|
EGU22-2439
|
ECS
|
|
On-site presentation
Roberto Manzo et al.

The detection of low energy seismic events and tremor related to volcanic activity in areas characterized by high background noise represents a crucial challenge for monitoring and surveillance purposes. In the last three years, the seismicity of the Mt. Vesuvius (southern Italy) has been characterized by low-magnitude volcano tectonic earthquakes, the most of which are located at depth shallower than 3 km b.s.l., while very few low-frequency earthquakes and tremor episodes are located at about 6-7 km depth. It is well known that magmatic and hydrothermal systems can play an important role in the generation of low-frequency seismic events, which could be important precursors for assessing the reawakening of a volcano. Therefore, our main objective is to develop a methodology for detecting the presence of low frequency (LF) events hidden in the background noise and not identifiable by classical detection procedures. In particular, we suggest a frequency domain approach based on a joint application of coherence analysis among signals from local network seismic stations and parameterization of the amplitude spectra according to the statistical moments. The proposed methodology has been applied to the analysis of continuous seismic signals recorded over three years at Mt. Vesuvius. Spectral parameters, such as central frequency W, shape factor d and coherence c, were evaluated on 30-s windows signals in the frequency range between 2 and 40 Hz. The selection of the signal windows that could potentially contain low-frequency events or tremor signals was performed according to the following criteria: a) 0.45 < δ < 0.65; b) 3 Hz < W < 6 Hz and c) c greater than 0.5, which are based on the results of preliminary analyses of the seismicity observed at Mt. Vesuvius. The detected signal windows were visually inspected and compared with the seismic catalogues to eliminate those corresponding to earthquakes occurred outside the area of interest. For the three-years of analyzed data, more than 200 episodes of low frequency signals were identified, 120 of which are not present in the seismic catalog. Most of them appear as low-amplitude tremor episodes, with no clear evidence of P and S phases, hidden in the noisy raw signals but visible at the entire seismic network after proper signal filtering. Compared to the few LF events detected and analysed in the past, our findings suggest that the proposed methodology can be an efficient tool for detecting low-amplitude signals not easily identifiable in the background noise and could represent an improvement for the monitoring system of the Mt. Vesuvius volcanic area.

How to cite: Manzo, R., Galluzzo, D., La Rocca, M., and Di Maio, R.: A new frequency-domain based approach for detecting low frequency seismic events: An application to the Mt. Vesuvius seismicity, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-2439, https://doi.org/10.5194/egusphere-egu22-2439, 2022.

14:38–14:44
|
EGU22-3160
|
Juan Porras et al.

Distributed Acoustic Sensing (DAS) is becoming increasingly popular in microseismic monitoring operations. Fiber-optic cables such as conventional telecommunication or built-for-purpose cables can be turned into a dense array of geophones that samples seismic wavefields continuously for several kilometers. DAS is particularly interesting for microseismic monitoring of geothermal systems since it does not have the same temperature limitations as standard electronic equipment. The sensing fiber can therefore be installed at high-temperature reservoir conditions and in the same well that is being stimulated. Because of these advantages, the distance between the detecting sensor and the induced seismicity can be minimized, maximizing the detection capability. Typical DAS acquisition samples the wavefield at about 1 m spacing and sampling frequencies of 1 kHz or higher. Unfortunately, standard seismological techniques are not capable of exploiting this high spatial density of sensors, hence they are ineffective in processing this kind of data. Here we propose a semblance-based seismic event detection method that fully exploits the characteristics of the DAS data. The detection identifies seismic events by looking at waveform coherence along hyperbolas while changing the curvature and position of the vertex. The method returns a time series of coherence values and, if these values are higher than a determined threshold, it catches a seismic event. First we test the detector with synthetic data resembling a realistic setup. Finally, we validate the detector by applying it to real DAS data from the Utah FORGE site in the US. This work is supported by the EU-Geothermica DEEP project.

How to cite: Porras, J., Grigoli, F., Stucchi, E., Tuinstra, K., Tognarelli, A., Lanza, F., Aleardi, M., Mazzotti, A., and Wiemer, S.: A semblance based microseismic event detector for DAS data, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-3160, https://doi.org/10.5194/egusphere-egu22-3160, 2022.

14:44–14:50
Discussion

Fri, 27 May, 15:10–16:40

Chairpersons: Simone Cesca, Natalia Poiata, Federica Lanza

15:10–15:16
|
EGU22-6087
|
ECS
|
Virtual presentation
Peidong Shi et al.

Automatic event detection and location is key to real-time earthquake monitoring. With the increase of computing power and labeled data, automated workflows that utilize machine learning (ML) techniques have become increasingly popular; however, classical workflows using ML as phase pickers still face challenges for seismic events of short inter-event time or low signal-to-noise ratio (SNR). Full waveform methods that do not rely on phase pick and association are suitable for processing these events, but are computationally costly and can lack clear event identification criteria, which is not ideal for real-time processing. To leverage the advantages of both methods, we propose a new workflow, MALMI, which integrates ML and waveform migration to perform automated event detection and location. The new workflow uses a pre-trained ML model to generate continuous phase probabilities that are then back-projected and stacked to locate seismic sources using migration.

We applied the workflow to a microseismic monitoring dataset collected in a borehole at the Utah FORGE geothermal laboratory site. The proposed workflow can automatically detect and locate induced microseismic events from continuous geophone recordings. Different ML models are evaluated for detection capability and phase classification accuracy. We expect that better performance should be possible if a customized ML model re-trained using local dataset would be used in the MALMI workflow. Further comparison with conventional migration methods confirms that MALMI can produce much clearer stacked images with higher resolution and reliability, especially for events with low SNR. The workflow is freely available on GitHub, providing a complementary tool for automated event detection and location from continuous data.

How to cite: Shi, P., Grigoli, F., Lanza, F., and Wiemer, S.: MALMI: towards combining machine learning and waveform migration for fully automated earthquake detection and location, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-6087, https://doi.org/10.5194/egusphere-egu22-6087, 2022.

15:16–15:22
|
EGU22-10588
|
ECS
|
Virtual presentation
Johannes Stampa et al.

In the recent decade, the amount of available seismological broadband data has increased steeply. Picking later arriving phases such as S-phases is difficult, and there are few manual picks available for these phases. Data sets of manual picks can also be problematic, since phase arrival picks are sensitive to the parameters of the filtering, which are often unknown, and the individual picking behavior of the analysts. However, accurate arrival times, especially for these phases, could be used to improve the accuracy of velocity models obtained from seismic tomography. This necessitates the adoption of automatic techniques for determining teleseismic phase arrival times consistently over a large data set.

In this work, a robust automatic picking algorithm based on autoregressive, multi-component, multiple-sample forward prediction is examined with regards to its accuracy. The phase is identified using the Akaike criterion and the onset time is found by evaluating discontinuities in the instantenous period of the signal. This signal analytic approach is tested using synthetic waveforms as well as real data in conjunction with manual picks obtained from the reviewed ISC-catalog.

Picking errors are estimated by comparing the automatic picks with manual picks, automatic picks at the neighboring stations as well as statistical meth- ods. The quality evaluations suggest potential of using these automatically determined phase arrival times for a travel time tomography.

How to cite: Stampa, J., Eckel, F., Kallinich, N., and Meier, T.: Automatic Picking of Teleseismic P- and S-Phases using an Autoregressive Prediction Approach , EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-10588, https://doi.org/10.5194/egusphere-egu22-10588, 2022.

15:22–15:28
|
EGU22-7844
|
ECS
Luis M. Fernandez-Prieto et al.

In recent years there have been a great progress in earthquake detection and picking arrival times of P and S phases using Deep Learning algorithms. However, the general adoption of these methods for the routine processing of monitoring networks has been held back by factors such as the availability of well documented software, computational resources, and a gap in knowledge of these methods. We have analyzed recent available Deep Learning pickers, comparing the results against data picked by a human operartor and against non-Deep Learning programs. We have used data recorded in several locations, with different characteristics and triggering mechanisms, such as volcanic eruptions, induced seismicity and local eartquakes, recorded using different types of instruments. We have found that the Deep Learning algorithms are able to achieve results comparables to a human operator, and several times better than a classical program, specially in data with a low signal to noise ratio. They are very efficient at ignoring large amplitude transient noise and at picking S waves, a task that is often difficult even for experienced analysts, and they require very few parameters to tune (often only the probability threshold) so an in-depth knowledge of neural networks is not required. (This research has been funded by  Spanish Ministry of Science and Innovation MICINN/AEI/10.13039/501100011033 grants CGL2017-88864-R and PRE2018-084986).

How to cite: Fernandez-Prieto, L. M., García, J. E., Villaseñor, A., Sanz, V., Ammirati, J.-B., Díaz, E., and García, C.: Performance of Deep Learning pickers in routine network processing applications, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-7844, https://doi.org/10.5194/egusphere-egu22-7844, 2022.

15:28–15:34
|
EGU22-10541
|
ECS
|
Highlight
|
On-site presentation
Titouan Muzellec et al.

The estimation of spatial and temporal changes in the host medium physical properties is a quest to improve risk evaluation and hazard forecasting application. The space-time evolution of the seismicity gives information about stress variations, fluid content, and pore-pressure changes inside the medium. Thus, the accuracy of arrival-time measurements is crucial for travel-time-based seismological applications, such as earthquake location and travel(delay)-time tomography, especially when double-difference times are used. Standard monitoring networks and tools implements single-station, STA/LTA-based, automatic event detection/location procedures, which may produce inconsistent arrival-times of the same phase among stations. To overcome this problem, refined cross-correlation (CC) techniques for time picking have been recently developed. Their basic approach is to use CC to refine picks of event pairs with high waveform similarity. Similar events are grouped in families, considering the max CC values, the inter-distance and/or the focal mechanism similarity. Two drawbacks of this common approach are (1) the impact of noise from individual receiver levels on the quality of reference trace (RT) and (2) the inability to adjust the systematic shift of automatic picks.

Here we propose a new, fully automatic approach to refine the phase time picks. The CC is used to identify family members with a hierarchical clustering procedure. In each family, after the trace alignment, we build the RT by stacking the events weighted by the signal-to-noise ratio and the polarity. We applied this technique to a catalog of 3574 events of the 2014 MJMA 6.7 sequence occurred at the Northern Nagano prefecture. The results indicate that we can improve the precision of phase picks of similar events and to adjust the systematic shift introduced by the automatic picker with mean differences between refined and automatic picks up to 0,5-1 s.

The high consistency of the phase picks allows to increase the accuracy of absolute location by reducing the mean location error from 0,6 km to 0,1 km and the root-mean-square from 0,15 to 0,075. Consequently, we observe an alignment of the seismicity respect to the main fault plane with an 30°-45° east-dipping angle for the shallow part while the deeper part dips at 50°-65°. Then, the double difference location provides highly resolved hypocenter locations and medium parameters by considering events of the same family as events pair. This improvement allows to use fast-tracking methods, as the Vp/Vs in time and the Coda-Wave Interferometry, to get information about the velocity variations before and during the sequence. By using those methods, we expect to get accurate information of the physical properties evolution and especially about the role of fluid in the triggering of the sequence.

How to cite: Muzellec, T., De Landro, G., Zollo, A., and Russo, G.: Insight on the 2014 MJMA 6.7, northern Nagano earthquake sequence evolution in space and time through high resolved earthquake locations, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-10541, https://doi.org/10.5194/egusphere-egu22-10541, 2022.

15:34–15:40
|
EGU22-11930
Francesco Grigoli et al.

On 2021 March 27th an Mw 5.2 earthquake occurred in the Adriatic Sea, between the Italia and Croatian coast. The earthquake sequence lasted for several months and consisted of more than 150 seismic events with a magnitude above 2. Analyzing offshore seismic sequences is challenging both for the lack of optimal seismic monitoring networks and detailed enough velocity models. These conditions strongly limit the data analysis procedures, leading to inaccurate results that may have severe effects on the identification of the seismogenic structure associated with the seismic sequence, bringing to wrong seismo-tectonic interpretations, with direct consequences in the seismic hazard assessment of an area. In this study, we analyze the March 2021 Mw 5.2 earthquake sequence that occurred in the Adriatic Sea with recently developed location techniques. Our workflow allows achieving a higher location accuracy, even when dealing with suboptimal monitoring conditions. We analyze this dataset using waveform-based location techniques and a recently developed location technique based on Distance Geometry Solvers (DGS). This last approach uses inter-event distances between earthquake pairs estimated at one or two seismic stations to get high-resolution locations of seismicity clusters. The application of such techniques led to different improvements in locating the seismic sequence, which is more clustered and clearly shows an N-S trending compatible with the geological setting of the area.

How to cite: Grigoli, F., Mazzotti, A., Molinari, I., Stucchi, E., Tognarelli, A., Aleardi, M., and Stipcevic, J.: Analysis of the 2021 March 27th Mw 5.2 earthquake sequence in the Adriatic Sea using new workflows for offshore seismicity monitoring, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-11930, https://doi.org/10.5194/egusphere-egu22-11930, 2022.

15:40–15:46
|
EGU22-3800
|
ECS
|
|
On-site presentation
Ioannis Fountoulakis and Christos Evangelidis

We introduce SSA2py, an open-source tool for the implementation of the Source-Scanning Algorithm (SSA) (Honn Kao and Shao-Ju Shan, 2007) in near-real time conditions. In general, Back-Projection methods due to their simplistic but at the same time effective approach provide the circumstances for fast analysis of the seismic rupture with relatively low computational cost and minimum initial assumptions.  In accordance with that and by exploiting local strong motion data, SSA can be used for the detailed imaging of the high frequency seismic radiation after the occurrence of a major earthquake by stacking records based on the predicted arrival times for a specific seismic phase. Areas in a spatiotemporal grid system that produce high brightness values due to constructive stacking, usually point out the radiation of meaningful seismic energy at the examined frequency band. SSA2py is a command line tool, developed in Python high-level programming language and mainly designed to closely work with ‘FDSN Compliant Web Services’  for a real-time seismic event triggering and seismic waveform data provision. After the report of a significant seismic event SSA2py initially calculates the necessary travel time tables, using the optimal velocity model for the study area. The software is intended to offer several travel time calculation alternatives such as the fast marching or the finite difference method together with the possibility to use 1D or 3D velocity models (if it’s applicable).  In a subsequent step, it automatically obtains seismic waveform data and metadata from the user defined data sources (e.g. an FDSN web service) and applies a variety of signal assessment algorithms that examine data-clipping, signal‐to‐noise ratio, long period disturbances, station’s performance based on power spectral density (PSD) of seismic noise etc. Selected data are carefully pre-processed, based on the user given configuration file and back-projected using SSA in an highly efficient way parallelized and adapted to run in GPU and CPU multiprocessing architectures.  An extended configuration file is provided, allowing the user to manipulate in detail SSA settings, ranging from the style and the size of the grid system to the frequencies and the type of the used signals.  Finally the software elucidates the method results by producing a series of plots and other important output info. The robustness of this new software will be presented in case studies from major earthquakes around the world (e.g. Japan, Greece). The program will be open source and freely available to the scientific community, oriented for computers with Linux OS and access to FDSN Web Services.

Honn Kao, Shao‐Ju Shan, Rapid identification of earthquake rupture plane using Source‐Scanning Algorithm, Geophysical Journal International, Volume 168, Issue 3, March 2007, Pages 1011–1020, https://doi.org/10.1111/j.1365-246X.2006.03271.x

The research work was supported by the Hellenic Foundation for Research and Innovation (H.F.R.I.) under the “First Call for H.F.R.I. Research Projects to support Faculty members and Researchers and the procurement of high-cost research equipment grant” (SIREN, Project Number: 910).

How to cite: Fountoulakis, I. and Evangelidis, C.: SSA2py: A seismic source imaging tool in Python based on the Source-Scanning Algorithm, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-3800, https://doi.org/10.5194/egusphere-egu22-3800, 2022.

15:46–15:52
|
EGU22-3826
|
ECS
|
On-site presentation
Angela Carrillo Ponce et al.

Landslides begin with the detachment of a mass and end with its impact at lower altitude. To simultaneously model seismic signals produced by these processes, we consider a double single-force model. We applied this source model to the seismic signals generated by the landslide in Uttarakhand, India, on February 7th, 2021. We model the seismic recordings at 12 seismic stations located at less than 100 km epicentral distance. We perform the source inversion by fitting low-frequency (0.08-0.15 Hz) three component (vertical, radial, transversal) waveforms in the time and simultaneously their amplitude spectra in the frequency domain. We compare our results with those obtained for a single-force model applied to each pulse separately. Our results identify two energetic pulses separated by a time delay of ~1 minute. The amplitude of the second pulse, interpreted as the impact, is ~3 times larger than the first one, and of opposite sign. Together with visual observations of the landslide itself, our results confirm that the first pulse was produced by the detachment of the rock mass and the second one by the impact of the mass in the valley. The orientations of the single forces are consistent with the slope geometry and the direction of the debris flow. We discuss statistical measures of fit of the two different inversion approaches and the possible strengths and weaknesses of the new double single-force model.

How to cite: Carrillo Ponce, A., Cesca, S., Dahm, T., Tilmann, F., Babeyko, A., Rekapalli, R., and Rao, N. P.: Double single-force model to characterize the detachment and impact of a landslide: application to the 07-02-2021 Uttarakhand, India landslide, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-3826, https://doi.org/10.5194/egusphere-egu22-3826, 2022.

15:52–15:58
|
EGU22-11721
|
ECS
Nicolas Luca Celli et al.

The determination of seismic moment tensors (MTs) for microseismicity poses challenges because of both the large number of events that are typically recorded, and their low signal to noise ratio. In recent years, automated moment tensor inversion methods have become more and more accurate, but an objective evaluation of their performance is often problematic due to the absence of site-specific, reference databases for comparison. In this study, we build a database of manually inverted MTs for the recent COSEIMIQ project, using the well-tested FociMT/HybridMT inversion method. COSEISMIQ focussed on microseismic monitoring in the Hellisheiði geothermal field, in the Hengill region, southern Iceland, where a dense network of 33 temporary seismic stations was deployed during 2018-2021, offering an ideal case study for microseismic MT inversion.

As a first step, we test the efficacy and possible pitfalls of the manual MT inversion on both a realistic and a simplified synthetic events waveform database. After careful, repeated manual tests, we observe that the inversion is robust across widely different choices of frequency band, but can be triggered to fail by not including key stations in some rare source-station geometries.

We then analyse the real data from the COSEISMIQ experiment, using previously located events from a large, recently developed microseismic catalog of the area. By running preliminary inversions of a subset of events in the centre of the deployment, we are able to pinpoint pre-processing steps that have a key effect on the MT inversion.  We find that in strong noise conditions such as in the Hengill region, the order and phase of the used frequency filter are fundamental parameters in correctly processing the P-wave onset used later for inversion.

After fine-tuning the event preprocessing, we select a larger subset of 197 events with magnitude > 0.8 from the catalog across the whole COSEISMIQ area, including several seismicity clusters at the edge of the deployment. We then pick all 197 events and invert them first with FociMT, then cluster the events based on their location using K-means clustering, and finally re-invert each cluster using HybridMT. The clustered inversion using HybridMT changes some MT solutions significantly, reducing the intra-cluster MT variance for most clusters. Interestingly, some event clusters show increased variance after the HybridMT inversion, suggesting that these include substantially different source mechanisms within a small area.

This new database of carefully inverted MT solutions can now be used as a test dataset to evaluate the performance of automated inversion tools.

How to cite: Celli, N. L., Nooshiri, N., Bean, C. J., Grigoli, F., Obermann, A., and Wiemer, S.: Manual MT inversions in microseismic areas: good practices and building a reference database for the Hengill region, Iceland, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-11721, https://doi.org/10.5194/egusphere-egu22-11721, 2022.

15:58–16:04
|
EGU22-9220
|
ECS
Marine Menager et al.

In order to rapidly detect, locate and characterize seismic events at regional scale, we are currently improving the parametrization of a continuous grid-search moment tensor inversion tool, called GRiD MT (Grid-based Real-Time Determination of Moment Tensor, Kawakatsu, 1998, Tsuruoka, 2009, Guilhem et al., 2011 and 2013, Menager et al., 2021). Here, we first design the procedure for small to moderate size earthquakes in south-eastern France, but we target its expansion to any potential regions of interest. This rapid and continuous moment tensor approach strongly depends on an adequate pre-selection of inversion parameters such as velocity models, station set, frequency band, grid spacing. In a proof of concept for France, we find source solutions (mechanism and magnitude) for two recent and moderate earthquakes (Mw 4.8 2019 Le Teil and M4.9 2014 Barcelonnette) very similar to those found by other institutes (CEA, GEOAZUR, USGS, INGV, …).

Moreover, instead of using a fixed time window in the inversion, we show that the quality of the solution can be improved by modifying each station’s time window depending on P and S wave travel times. This appears to be particularly true for small magnitude earthquake and potentially noisy data.

However, the previous selected parameters for Le Teil and Barcelonnette events are not the best ones to characterize the lower size events (3.4 < Mw < 4.0). We also propose an approach for automatically defining the best set of stations and frequency band depending on the targeted magnitudes in a region of interest. Based on signal/noise ratio (SNR) analysis to determine the signal quality for each station of a possibly dense seismic network, it helps the analyst in its GRiD MT parametrization. Moreover, here, azimuthal coverage is also taken into account in the station selection.

We illustrate the relevance of these proposed improvements with a selection of events. The ultimate aim is to facilitate the implementation of the GRiD MT method for other regions of interest and for different magnitude ranges. 

How to cite: Menager, M., Guilhem Trilla, A., and Delouis, B.: Tuning the GRID MT method for the moment tensor inversion of small to moderate size earthquakes , EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-9220, https://doi.org/10.5194/egusphere-egu22-9220, 2022.

16:04–16:10
|
EGU22-10020
|
ECS
|
On-site presentation
Nima Nooshiri et al.

In this study, we present a new approach based on recent advances in deep learning for rapid source-parameter estimation of microseismic earthquakes. The seismic inversion is represented in compact form by two convolutional neural networks, with individual feature extraction, and a fully connected neural network, for feature aggregation, to simultaneously obtain moment tensor and spatial location of microseismic sources. The neural network algorithm encapsulates the information about the relationship between seismic waveforms and underlying point-source mechanisms and locations allowing rapid inversion (within a small fraction of a second) once input data are available. A key advantage of the algorithm is that it can be trained using synthesized seismic data only, so it is directly applicable to scenarios where there are insufficient real data for training including temporary seismic networks and hydraulic stimulation experiments, for example. Moreover, we find that the method is robust with respect to perturbations such as observational noise and data incompleteness (missing stations). We apply the new approach on a database of small magnitude (M ≤ 2) earthquakes recorded at the Hellisheiði geothermal field in the Hengill area, Iceland, which is the demonstration site in the EU-GEOTHERMICA project COSEISMIQ (http://www.coseismiq.ethz.ch). For the examined events, the model achieves very good agreement with the inverted solutions determined through standard methodology. The new approach offers great potential for automatic and rapid real-time information on microseismic sources in a deep geothermal context and can be viably used for microseismic monitoring tasks in general.

How to cite: Nooshiri, N., Celli, N., Grigoli, F., Bean, C. J., Dahm, T., Kristjánsdóttir, S., Obermann, A., and Wiemer, S.: Performance evaluation for deep-learning based point-source parameter estimation using a well constrained manual database: examples from the Hengill Geothermal Field, Iceland, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-10020, https://doi.org/10.5194/egusphere-egu22-10020, 2022.

16:10–16:16
|
EGU22-7363
|
ECS
|
|
Virtual presentation
Jack Woollam et al.

Machine Learning (ML) methods have seen widespread adoption in seismology in recent years. The ability of these techniques to efficiently infer the statistical properties of large datasets often provides significant improvements over traditional techniques when the number of data are large (>millions of examples). With the entire spectrum of seismological tasks, e.g., seismic picking and detection, magnitude and source property estimation, ground motion prediction, hypocentre determination; among others, now incorporating ML approaches, numerous models are emerging as these techniques are further adopted within seismology.

To evaluate these algorithms, quality controlled benchmark datasets that contain representative class distributions are vital. In addition to this, models require implementation through a common framework to facilitate comparison. Accessing both benchmark datasets, and integrating models built in such varying frameworks is currently a time-consuming process, hindering further advancement of ML techniques within seismology. These development bottlenecks also affect 'practitioners' seeking to deploy the latest models on seismic data, who may not want to necessarily learn entirely new ML frameworks to perform this task.

We present SeisBench as a software package to tackle these issues. SeisBench is an open-source framework for deploying ML in seismology. SeisBench standardises access to both models and datasets, whilst also integrating a range of common processing and data augmentation operations through the API. Through SeisBench, users can access several seismological ML models and benchmark datasets available in the literature via a single interface. SeisBench is built to be extensible, with community involvement encouraged to expand the package. Having such frameworks available for accessing leading ML models forms an essential tool for seismologists seeking to iterate and apply the next generation of ML techniques to seismic data.

How to cite: Woollam, J., Münchmeyer, J., Tilmann, F., Rietbrock, A., Lange, D., Bornstein, T., Diehl, T., Giunchi, C., Haslinger, F., Jozinovic, D., Michelini, A., Saul, J., and Soto, H.: SeisBench - A Toolbox for Machine Learning in Seismology, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-7363, https://doi.org/10.5194/egusphere-egu22-7363, 2022.

16:16–16:22
|
EGU22-2673
|
ECS
|
|
Virtual presentation
Simone Francesco Fornasari et al.

Real-time monitoring is of primary importance for rapid and targeted emergency operations after potentially destructive earthquakes. A key aspect in determining the impact of an earthquake is the reconstruction of the ground shaking field, usually expressed as a peak ground parameter. Traditional algorithms approach this task by computing the ground shaking field from the punctual data at the stations and relying on ground motion prediction equations (GMPEs) computed on estimates of the earthquake location where the instrumental data are missing. The results of such algorithms are then subordinate to the evaluation of location and magnitude which can take several minutes.
To fill the gap between the arrival of the data and the (first preliminary) estimation (usually computed in a few minutes), we introduce a new data-driven algorithm that exploits the information from the station data only. Such an algorithm, consisting of an ensemble of convolutional neural networks (CNNs) trained with a database of ground shaking maps produced with traditional algorithms, can provide real-time estimates of the ground shaking field and the associated uncertainty. Since CNNs cannot handle sparse data a Voronoi tessellation of a specific peak ground parameter recorded at the stations is computed and used as an input to the CNNs; site effects and network geometry are accounted for using a (normalized) Vs30 map and a station location map, respectively.
The developed method is robust to noise, can handle network geometry changes over time without the need for retraining, and can resolve multiple simultaneous events. Although having a lower resolution, the results obtained are compatible with the ones from traditional methods. A fully-operational version of the algorithm is running on the servers at the Department of Mathematics and Geosciences of the University of Trieste showing real-time capabilities in handling stations from multiple Italian strong-motion networks and outputting results with a resolution of 0.05°x0.05°.

How to cite: Fornasari, S. F., Costa, G., and Pazzi, V.: A machine-learning approach for the reconstruction of ground shaking fields in real-time, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-2673, https://doi.org/10.5194/egusphere-egu22-2673, 2022.

16:22–16:28
|
EGU22-8517
|
ECS
|
Virtual presentation
Camilla Rossi et al.

Microseismic monitoring plays a fundamental role for the risk assessment and management of industrial activities related to the exploitation of georesources. In such application, microseismic monitoring is performed in real-time.

One of the most widely distributed and used tools for seismic monitoring is SeisComP, a software package for automatic data acquisition and processing in real-time or during post-processing developed by the German Research for Geosciences (GFZ).

In this work, we show how SeisComP can be optimized for real-time data-processing for microseismic monitoring of an Underground Gas Storage field in Northern Italy.

We analysed 2-years of continuous seismic data recorded by a network composed of 15 (surface and borehole) stations. In order to improve the accuracy of earthquakes location, after processing seismic data in real-time, we used Joint Hypocentral Inversion techniques to compute a 1D velocity model (both for P and S waves) for the surrounding area of gas storage field. Then, we extracted a P 3D velocity model at reservoir scale, based on the migration velocity from a 3D seismic reflection survey. The Vp model is then converted to Vs by using an average Vp/Vs value extracted from the 1D velocity model and well-logs.

Finally, we compared the different velocities models by analysing earthquakes location obtained with each model.

For the events located in the inner area, our comparison shows a systematic location improvement (both in terms of RMS and waveform coherence) with the 3D model. For events outside that area, the optimized 1D model performs better than the initial model (both in terms of RMS and waveform coherence). Our processing routine for this seismic network is the first application in Italy where a 3D velocity model is fully integrated within the real-time microseismic monitoring operations, as suggested by the Italian Guideline for Microseismicity Monitoring on Industrial activities.

How to cite: Rossi, C., Cocorullo, C., and Grigoli, F.: Monitoring microseismicity with SeisComP and a local 3D velocity model, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-8517, https://doi.org/10.5194/egusphere-egu22-8517, 2022.

16:28–16:40
Discussion