SUMMARY Despite advanced seismological techniques, automatic source characterization for microseismic earthquakes remains difficult and challenging since current inversion and modelling of high-frequency signals are complex and time consuming. For real-time applications such as induced seismicity monitoring, the application of standard methods is often not fast enough for true complete real-time information on seismic sources. In this paper, we present an alternative 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 full moment tensor and spatial location of microseismic sources. Specifically, a multibranch neural network algorithm is trained to encapsulate the information about the relationship between seismic waveforms and underlying point-source mechanisms and locations. The learning-based model allows rapid inversion (within a fraction of second) once input data are available. A key advantage of the algorithm is that it can be trained using synthetic seismic data only, so it is directly applicable to scenarios where there are insufficient real data for training. 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 synthesized and example recorded small magnitude (M ≤ 1.6) earthquakes at the Hellisheiði geothermal field in the Hengill area, Iceland. For the examined events, the model achieves excellent performance and shows very good agreement with the inverted solutions determined through standard methodology. In this study, we seek to demonstrate that this approach is viable for microseismicity real-time estimation of source parameters and can be integrated into advanced decision-support tools for controlling induced seismicity.
Abstract. On September 3rd 2017 official channels of the Democratic People's Republic of Korea announced the successful test of a thermonuclear device. Only seconds to minutes after the alleged nuclear explosion at the Punggye-ri nuclear test site in the mountainous region in the country's northeast at 03:30:02 (UTC) hundreds of seismic stations distributed all around the globe picked up strong and distinct signals associated with an explosion. Different seismological agencies reported body wave magnitudes of well above 6.0, consequently estimating the explosive yield of the device in the order of hundreds of kilotons TNT equivalent. The 2017 event can therefore be assessed being multiple times larger in energy than the two preceding events in January and September 2016. This study provides a multi-technology analysis of the 2017 North Korean event and its aftermath using a wide array of geophysical methods. Seismological investigations locate the event within the test site at a depth of approximately 0.8 km below surface. The radiation and generation of P- and S-wave energy in the source region is significantly influenced by the topography of the Mt. Mantap massif. Inversions for the full moment tensor of the main event reveal a dominant isotropic component accompanied by significant amounts of double couple and compensated linear vector dipole terms, confirming the explosive character of the event. Analysis of the source mechanism of an aftershock that occurred around eight minutes after the test in the direct vicinity suggest a cavity collapse. Measurements at seismic stations of the International Monitoring System result in a body wave magnitude of 6.2, which translates to an yield estimate of around 400 kilotons TNT equivalent. The explosive yield is possibly overestimated, since topography and depth phases both tend to ehance the peak amplitudes of teleseismic P-waves. Interferometric Synthetic-Aperture-Radar analysis using data from the ALOS-2 satellite reveal strong surface deformations in the epicenter region. Additional multispectral optical data from the Pleiades satellite show clear landslide activity at the test site. The strong surface deformations generated large acoustic pressure peaks, which were observed as infrasound signals with distinctive waveforms even in distances of 400 km. In the aftermath of the 2017 event atmospheric traces of the fission product 133Xe have been detected at various locations in the wider region. While for 133Xe measurements in September 2017 the Punggye-ri test site is disfavored as source by means of atmospheric transport modeling, detections in October 2017 at the International Monitoring System station RN58 in Russia indicate a potential delayed leakage of 133Xe at the test site from the 2017 North Korean nuclear test.
<p>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 Hellishei&#240;i 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.</p><p>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.</p><p>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. &#160;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.</p><p>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.</p><p>This new database of carefully inverted MT solutions can now be used as a test dataset to evaluate the performance of automated inversion tools.</p>
'/%3e%3cpath d='M200 752.362h200v300H200z' style='fill:%23fff;fill-opacity:1;stroke:none' transform='translate(0 -752.362)'/%3e%3cpath d='M400 752.362h200v300H400z' style='fill:%23ff883e;fill-opacity:1;stroke:none' transform='translate(0 -752.362)'/%3e%3c/svg%3e)
'%3e%3cpath fill='%23012169' d='M0 0v30h60V0z'/%3e%3cpath stroke='%23fff' stroke-width='6' d='m0 0 60 30m0-30L0 30'/%3e%3cpath stroke='%23C8102E' stroke-width='4' d='m0 0 60 30m0-30L0 30' clip-path='url(%23b)'/%3e%3cpath stroke='%23fff' stroke-width='10' d='M30 0v30M0 15h60'/%3e%3cpath stroke='%23C8102E' stroke-width='6' d='M30 0v30M0 15h60'/%3e%3c/g%3e%3c/svg%3e)