Using Moment Tensor Inversions for Rapid Seismic Source Detection and Characterization: Application to the North Korean Nuclear Tests

doi:10.21203/rs.3.rs-3567205/v1

Download PDF

Research Article

Using Moment Tensor Inversions for Rapid Seismic Source Detection and Characterization: Application to the North Korean Nuclear Tests

https://doi.org/10.21203/rs.3.rs-3567205/v1

This work is licensed under a CC BY 4.0 License

Journal Publication

published 28 Mar, 2024

Read the published version in Pure and Applied Geophysics →

You are reading this latest preprint version

The rapid detection and source characterization of any type of seismic events including nuclear explosions is one of the major objectives for national data centres (NDC) and seismological laboratories. Most often, the used techniques are based on phase picking and amplitude measurements for detecting and locating events, and for estimating magnitudes. From these parameters, event screening is then done empirically, and this may lead to misinterpretations of the source nature. However, it is known that seismic waveform inversion for the determination of the moment tensor has been proven as a reliable source physics-based method for event characterization. Here, we present a technique already used for earthquake monitoring in tectonically active regions, and we test it on the seismicity recorded in the vicinity of the Democratic People’s Republic of Korea (DPRK), and in particular on the DPRK nuclear explosions. From a grid of potential locations and by scanning continuous seismic waveforms, it is possible to implement a rapid detector of seismic events providing the full information of the sources (origin time, location, magnitude, mechanism). We show its overall performances on all past DPRK nuclear tests and regional earthquakes. From such approach fast event screening is achieved and source uncertainties can be estimated.

North Korea

nuclear explosion

moment tensor inversion

event detection

GRiD MT

The Democratic People’s Republic of Korea (DPRK) is the only country to have conducted nuclear tests in the 21st century with six declared nuclear events in 2006, 2009, 2013, 2016 (January and September) and 2017. The size of the explosions has continuously increased to reach in September 2017 a body-wave magnitude (mb) of 6.1 recorded by worldwide distributed seismic stations (Table 1, Fig. 1a). This last event was located similarly to the others in the direct vicinity of the North Korean experimentation test site of Punggye-ri. It was strong enough to trigger a significant collapse of the Mantap mountain within a few minutes after its occurrence, as well as a dense and long-lasting near-by seismic activity (Pabian and Coblentz, 2018; Tian et al., 2018; Wang et al., 2018).

Table 1

Source parameters of the six North Korean nuclear tests (DPRK 1 to 6), and of the collapse that followed the last explosion, reported in the Reviewed Event Bulletin (REB) of the International Data Centre (IDC).
Event	Origin time	Latitude (°N)	Longitude (°E)	mb
DPRK 1	2006/10/09 01:35:27	41.3119	129.0189	4.1
DPRK 2	2009/05/25 00:54:42	41.3110	129.0464	4.5
DPRK 3	2013/02/12 02:57:50	41.3005	129.0652	4.9
DPRK 4	2016/01/06 01:30:00	41.3039	129.0481	4.8
DPRK 5	2016/09/09 00:30:00	41.2992	129.0491	5.1
DPRK 6	2017/09/03 03:30:01	41.3205	129.0349	6.1
Collapse	2017/09/03 03:38:32	41.3206	129.0615	4.1

Figure 1 : The DPRK nuclear explosions: timeline (a) and observed waveforms at station MDJ, China, about 370 km away from the experimentation site (b, c). a) Time occurrence and corresponding body wave magnitudes (mb) reported by the IDC for the six nuclear tests between 2006 and 2017 in North Korea. The size of the stars is proportional to the magnitude mb of the events. b) Observed unfiltered velocity records on the vertical channel of station MDJ. The traces’ colors match those in a). The amplitude scaling for DPRK 1 to 5 is the same, and is different to the one for DPRK 6. c) Filtered (0.033–0.066 Hz) vertical velocity seismograms from b) for the six nuclear tests and for the collapse (purple) that immediately followed DPRK6 (orange). The traces are plotted with the same scale, except for DPRK6 whose amplitudes are divided by a factor of 10.

Figure 1 shows the seismic traces of the six DPRK nuclear tests recorded at the Chinese station MDJ, located at about 370 km to the north of the test site. The unfiltered seismograms (Fig. 1b) and the long period filtered data (Fig. 1c) of the six events show that their waveforms are highly correlated despite their size difference. This observation corroborates their similar source location and type. Moreover, even though the compressional waves (Pn and Pg phases) are very impulsive at station MDJ and there is a relative lack of shear (S) waves for all explosions, significant surface waves emitted by the events are clearly identified, such as the Rayleigh waves observed in Fig. 1b and c.

In the framework of the Comprehensive Test Ban Treaty (CTBT) verification, the national data centre (NDC) role is to rapidly detect and characterize any event of interest by using seismic, infrasound, hydroacoustic, and/or radionuclide records. For the seismic point of view, the need is linked to the rapid detection, location, and magnitude estimation of the event using seismic waveforms recorded at all distance ranges: from local to teleseismic (i.e., > 3000 km) including regional distances (i.e., < 2000 km). Source discrimination, meaning the capability to distinguish a natural tectonic earthquake from an anthropogenic event (i.e., explosion, mining activity, etc.) remains critical for the CTBT organisation and NDCs. Many studies done since the first nuclear tests conducted in the world in the second half of the 20th century proposed a variety of seismic discriminants. Nonetheless, a great majority of them are either empirically derived from, or relative to, other near-by explosions and earthquakes. Among them, one can cite the rapid mb:MS approach that helps differentiate seismic sources based on the specificities of the radiation of body and surface waves recorded at regional and teleseismic distances (Stevens and Day, 1985; Bonner et al., 2006; Russel, 2006). The P to S amplitude and spectral ratio calculated at regional and teleseismic distances, and in several frequency bands, can also be used (Walter et al., 1995; Kim et al., 2017). These very commonly considered methods were proposed based on the interpretation that an explosive mechanism predominantly emits compressional waves (P) and very limited shear (S) and surface waves. However, this representation is oversimplified for several past events, including the DPRK nuclear tests. Indeed, Fig. 1 illustrates that regional seismic records of explosions, especially high-yield and shallow sources, are complex and enriched not only in P-wave energy but also in surface waves. This leads to incorrect event discrimination for the DPRK explosions as their properties may not fit with generalized empirically screening laws (Fig. 2, Selby et al., 2012).

As opposed to these traditionally empirical approaches, it has been shown that seismic event discrimination could be obtained thanks to the use of a physics-based method describing the forces at play at the source. This one, based on seismic waveform inversion for the determination of the moment tensor, can indeed provide information about the sense of fault motion for the analysis of a given earthquake, or about the explosive degree of the origin for a natural or anthropic isotropic event (Vasco and Johnson, 1989; Teyssoneyre et al., 2002; Ford et al., 2009a; Guilhem and Walter, 2015).

In this paper, we show that moment tensor inversions can help NDCs to rapidly and automatically detect, locate, and provide full seismic source information (magnitude, mechanism), and thus be included in their near-real time operational procedures. Such approach is illustrated for the seismic monitoring of North Korea with an emphasis on superficial non-natural sources. With the use of a fixed grid of virtual sources covering a region of interest it becomes possible to propose a rapid tool for the seismic analyst. We show that the results obtained for the DPRK nuclear tests are in agreement with more refined inversions. Moreover, thanks to the thousands of moment tensor inversions computed per time step, this technique also provides some incomparable information relative to the uncertainties of the source derived from rapid waveform analysis.

2.1. Moment tensor

One commonly considered approach for describing the physics of seismic sources uses the approximation of sources by a model of equivalent forces derived from linear wave equations (Jost and Herrmann, 1989). The moment tensor M is a symmetric tensor representing the nine couples of equivalent dipole forces acting at the seismic source in the three spatial dimensions. It is not restricted to the description of earthquakes, but can also be considered for the description of other seismic sources such as explosions (Ford et al., 2009a), implosions, landslides (Dammeier et al., 2015), and complex rupture modes associated with underground fluid and gas circulations such as in volcanic and geothermal environments (Guilhem et al., 2014; Guilhem and Walter, 2015).

The displacement recorded at a given seismic sensor for a given hypocenter is represented as the convolution of the moment tensor and the time derivative Green’s function tensor, G, such as:

d = M * G.

The Green’s functions represent the impulse response of the ground to an excitation. They are calculated from wave propagation assumptions made in velocity models (P and S-wave velocities, attenuation, and medium’s density). In order to retrieve the source information (i.e., the moment tensor elements) from observed data, seismic waveform inversion is performed in time domain or frequency domain such as:

M = (G^TG)⁻¹G^T d

Moment tensor decomposition into more physically interpreting tensors is non-unique. However, the decomposition following Knopoff and Randall (1970) is widely accepted (Jost and Herrmann, 1989; Vavrycuk, 2015). It separates the diagonalized tensor M as the sum of three tensors (M = M_ISO + M_DC + M_CLVD): one describing the isotropic motion (isotropic tensor M_ISO), which can explain any volumetric changes at the source, and two others without any volumetric change called double-couple (M_DC, for dislocation motion) and compensated linear vector dipole (M_CLVD). These last two tensors form the deviatoric tensor.

From the full moment tensor and its decomposition, one can estimate its scalar dimension called seismic moment M₀, which provides a measure of the source strength. It follows the norms proposed by Bowers and Hudson (1989). The moment magnitude Mw is directly derived from the seismic moment.

Multiple approaches are available for estimating the moment tensor solutions of events of interest. Here, we follow the method proposed by Dreger (2003) called Time-Domain Moment Tensor inversion (TDMT_inv) and corrected for the full moment tensor inversion following Minson and Dreger (2008). The quality of the least-square solution from waveform inversion is measured thanks to the variance reduction (VR) between each point of the observed data and synthetic signals for stations i:

$$VR=\left(1-\frac{\sum _{i}{\left({data}_{i}-{synthetics}_{i}\right)}^{2}}{\sum _{i}{\left({data}_{i}\right)}^{2}}\right)\times 100$$

Figure 3 shows the high-quality (i.e., VR = 80.7%) moment tensor solution obtained for the last DPRK nuclear test in 2017 represented as a point source in space and time located at 1 km depth. A total of eight well distributed regional broadband stations are used, and manual temporal shifts are made in order to compensate any small velocity model inaccuracies. The explosive nature of the source is found with a large isotropic component (ISO = 64%) and is represented by a fully-colored beachball diagram.

Similar manual moment tensor solutions for DPRK 2 to 5 also helped to confirm the nature of the sources with high level of goodness fit, and nearly identical mechanisms (Fig. 4, Guilhem Trilla and Cano, 2017; Guilhem Trilla and Menager, 2023).

2.2. The GRiD MT approach

Kawakatsu (1998) proposed to extend the use of moment tensor inversions for earthquake detection and characterization as opposed to their only estimation after detection and location obtained from traditional methods such as phase picking. This is how the approach named GRiD MT, for Grid-based Realtime Determination of Moment Tensors, was born. Since, it has been tested and implemented for earthquake monitoring in several tectonically active regions: Japan (Tsuruoka et al., 2009), California (Guilhem and Dreger, 2011), Taiwan (Lee et al., 2013), and Alaska (MacPherson et al., 2013).

The approach is simple: seismic waveforms are continuously scanned, and inverted for the determination of moment tensor using a grid of virtual sources representing potential hypocenters over a region of interest. In the occurrence of an event within the region, the grid point showing the largest goodness of fit (here, VR) is identified as the event’s hypocenter (latitude, longitude, depth), the origin time of the event is also obtained, and the moment tensor solution informs on the nature (source decomposition) and size of the source (moment magnitude).

GRiD MT presents several major advantages for an operational implementation. First, it is a rapid source inversion technique making use of pre-calculated catalogs of the inverse generalized Green’s functions (i.e., (G^TG)⁻¹G^T) for pre-fixed station configurations and grid points. Hence, the approach only consists in the multiplication of observed data segments with pre-calculated Green’s functions. Second, the complete source information is known thanks to a single algorithm. This is opposed to the traditional procedure used in seismological institutes and NDCs where each source parameter depends on one another. Indeed, while detection, location, and magnitude can be estimated from automated phase pickings and amplitude measurements, moment tensor inversions are not systematically performed, and they most often require the intervention of a senior seismologist or expert. Finally, despite its greater use for tectonic earthquake detection capability, GRiD MT allows the detection of all types of seismic sources including those with volumetric changes (Nayak and Dreger, 2018; Dammeier et al., 2015). For this main reason, it becomes an interesting tool for NDCs.

Inherent to its definition, this approach is highly dependent on the pre-selected grid and station configuration, as well as on the accuracy of the velocity models. Because we simultaneously define an origin time and the other source parameters at once, there is no option for taking into account small time shifts between data and synthetics in the inversion for improving their overall fit, and thus reducing potential velocity model inaccuracies. Perfect time alignments are indeed needed. The configuration of both grid and moment tensor inversion parameters (i.e., stations, velocity models, window length, etc.) requires a significant pre-implementation effort. However, once done, GRiD MT can be run automatically with limited man interaction, and it becomes a simple and easy-to-use tool for a NDC seismic analyst.

North Korea has a relatively low seismic rate (Fig. 5). The natures of the seismic activity, reported by the European Mediterranean Seismological Centre (EMSC), between 2009 and 2022, with magnitude (M) of 4.0 and larger, are contrasted. We count superficial nuclear explosions, deep natural events offshore the northeastern region of the country, shallow mining events in known regions near the Chinese borders, and sequences of induced small magnitude earthquakes following the DPRK events in the vicinity of Punggye-ri (Tian et al., 2018; Yao et al., 2018). The first nuclear explosion in October 2006 with a magnitude 4.0 is also included (Fig. 5).

Seismic records of the events from regional broadband 3-component stations from the International Data Centre (IDC) and Incorporated Research Institutions for Seismology (IRIS) networks are collected and processed. The azimuthal distribution of the seismic stations is chosen such as to recover most of the radiation pattern of the potential sources as possible around the region. From a set of eight well distributed stations we define the optimal distributed station combination composed of only four stations: MDJ, INCN, MAJO and BJT (Fig. 5). A complementary station combination could be composed of the following four stations: USRK, KSRS, MAJO and BJT, stations USRK and KSRS being primary IDC stations. For comparison, others previously mentioned operating systems in Japan, California, Taiwan, and Alaska use from three to five stations per grid. Data processing involves instrumental response removal and integration of the velocity records to displacement traces. In a second time, data segments are filtered at long periods between 0.033–0.066 Hz (i.e., 15–30 s) using a 3-pole one-pass Butterworth filter, and they are decimated to one sample per second.

The grid of virtual sources is designed to monitor shallow and anthropogenic events in North Korea such as nuclear tests. Grid points are spaced every 0.2° in latitude (between 38° N and 44° N) and longitude (between 124° E and 131° E), and a fixed depth is tested (1 km). Considering the wavelength in the 15–30 s period band and the rapid screening objectives of the grid search approach, we do not attempt to automatically extend the search to more depths, including shallower ones. Published source inversion results for the DPRK events had source depths close to 1 km, between 300 m and 1500 m (Dreger et al., 2017; Gaebler et al., 2019; Stevens and O’Brien, 2018). The presented grid setting implies a total of 1190 moment tensor inversions per time step in the waveform scanning approach of GRiD MT.

The Green’s functions required for moment tensor inversions are computed using the frequency-wavenumber approach from the Computer Program in Seismology library (CPS 330, Herrmann, 2003) and three different one-dimensional (1-D) Earth’s velocity models depending on the stations. Signals from stations located in the continent are represented with models published by Ford et al. (2009b) for North Korea, and the wave propagation through the Sea of Japan from North Korea to station MAJO in Japan is represented with an average 1-D cross-section of the global crustal model CRUST2 (Bassin et al., 2000). This last model for MAJO is comparable to the ones derived through path calibration analysis done by Dreger et al. (2021) for a set of Japanese stations. Figure 6 shows the different used models.

In order to determine the 6 independent elements of the full moment tensor for each point of the grid covering North Korea, the generated 10-component Green’s functions are expressed in the vertical (Z), radial (R), and tangential (T) orientations. The horizontal components are then rotated into the North-South and East-West directions according to the station-grid point azimuths, and they are saved into a dedicated Green’s function catalog for the GRiD MT approach. In addition, a second catalog is generated made of the generalized inverse Green’s function tensor ((G^TG)⁻¹G^T). The use of these two catalogs significantly decreases the running time of the grid search moment tensor inversions down to less than one second on a regular workstation for the entire grid for a given time step.

The GRiD MT approach allows the detection and accurate characterization of the DPRK nuclear explosions. Figure 7 shows the results obtained for the September 2016 test. The VR temporal evolution reaches a maximal value 2 seconds past the REB event origin time, and the corresponding defined location agrees with the experimental site location considering the grid spacing uncertainty (12 km distance offset, Table 2). Regarding the source parameters, the dominant isotropic explosive component is well identified (i.e., 64%) and the moment magnitude (Mw = 4.9) is in agreement with published ones (Alvizuri and Tape, 2018; Cesca et al., 2017; Ichinose et al., 2017; Li et al., 2023) despite the very limited number of stations in the GRiD MT approach.

Table 2

GRiD MT results for the Punggye-ri explosions and collapse relative to the REB catalog (origin time and location). Moment magnitude (Mw), isotropic component (ISO) and variance reduction (VR) for the GRiD MT detections are given. No result for DPRK3 given that no data from station INCN were available (see discussion).
Event	Origin time difference (s)	Latitude (°N)	Longitude (°E)	Distance difference (km)	Mw	ISO (%)	VR (%)
DPRK 1	+ 3	41.3	128.9	10	4.0	63	21.38
DPRK 2	+ 4	41.3	128.9	12	4.5	63	69.98
DPRK 3	-	-	-	-	-	-	-
DPRK 4	+ 2	41.3	128.9	12	4.8	64	72.29
DPRK 5	+ 2	41.3	128.9	12	4.9	64	71.50
DPRK 6	+ 3	41.3	128.9	11	5.6	63	72.10
Collapse	+ 1	41.3	128.9	14	4.8	64	29.57

This overall performance found for the DPRK 5 event compares with the ones resulting from the analysis of the other DPRK nuclear explosions with high goodness of fit values (Table 2, Fig. 8).

One particularly compelling outcome of the North Korea GRiD MT configuration is the rapid identification of the first DPRK nuclear explosion in 2006 despite the low signal to noise ratio for this relative lower yield event (Fig. 8 and Fig. 9). The peak in VR (i.e., < 25%) remains limited but the source parameters’ information concurs with the ones determined for the other DPRK explosions (Fig. 10). These results for the DPRK nuclear tests are moreover in remarkable agreement with the solutions from the refined inversions using 8 regional stations (Fig. 4). Manual and automated GRiD MT solutions describe the explosions as dominant explosive sources with some relatively significant CLVD components (i.e., close to crack-opening mechanisms, about 30–35%), and very limited DC (< 5%) motions. This shows that the tested configuration for monitoring North Korea events is well calibrated for the detection of M4 + superficial events in North Korea.

Moreover, sources with different mechanisms can be also spotted such as the collapse that followed the latest nuclear explosion by 8 minutes and 30 seconds (Fig. 11). For this given event observed in the DPRK 6 explosion’s coda, the goodness of fit is lower but the mechanism is well characterized as an implosion. Its moment magnitude Mw compares with the one of the January 2016 explosion when considering long period seismic regional data (Table 2 and Fig. 1c).

Furthermore, the GRiD MT approach for the seismic monitoring in North Korea limits the false detection rate of events whose source depths are greater than the first few kilometers of the crust (Fig. 8) and/or are located out of the grid’s geographic extent. Figure 8 presents no significant high VR peak for earthquakes reported in the EMSC catalog. Such overall performances are awaited for an operational configuration in the CTBT monitoring procedure. However, some shallow M4.6 to 5.7 regional earthquakes exhibit relative large values of VRs, above 25%, but without exceeding 40%. This is particularly true for several moderate earthquakes in South Korea, China, and offshore Japan near the grid’s limits (Fig. 5). Their source solutions (origin time, location, and mechanism) with GRiD MT appear incorrect given that the extent of the grid in latitude, longitude, and more importantly depth, does not correspond to their actual hypocenters. These relative large erroneous VRs are obtained several tens of seconds after their actual origin times.

Thanks to the large number of inversions per time step that are calculated in this approach, estimation of source uncertainties becomes achievable. Indeed, while many moment tensor analyses are lacking discussion about the solidity of resulting mechanisms, others propose some sensitivity analysis – either thanks to Jackknife tests, bootstrapping and/or network sensitivity analysis (Ford et al., 2010). With GRiD MT, the stability of the moment tensor through time and location can be tested, including mechanism and magnitude. Menager et al. (2023) proposed to define an overall GRiD MT solution based on the barycenter solution resulting from an ensemble over time and space. They applied it on several Mediterranean earthquakes as well as for earthquakes in France (Menager, 2023; Menager et al., 2023). They showed that GRiD MT can indeed help to account for the uncertainties in the source characterization with respect to the limited number of used stations and to the overall knowledge in the regional velocity structure.

Figure 12a shows the effects of timing error on the DPRK6 source characterization. Here, only the occurrence time varies, not the location of the grid point, which is identified from the maximal VR value. We find that solutions with time errors lower than 5 s are comparable in term of mechanism. They are located near the explosion section of the lune map (Tape and Tape, 2012). However, for largest timing errors, the solutions move along the opposite section of the lune, near the implosive pole. Such phenomena is known as phase (or cycle) skipping, which is responsible for reversing the mechanism type (Ford et al, 2009a; Alvizuri et al., 2018). Our observation agrees with the good practice’s time shift limit suggested by Ford et al. (2009a) in moment tensor inversion. Indeed, it matches with about one quarter of the high-corner frequency of the inverted signals (here, 0.066 Hz or 15 s period, so about 4 s). P-wave polarities have been shown to help constrain the waveform source inversion, and to avoid such phase skipping (Chiang et al., 2018). However, in an automated run, P-wave polarities and corresponding take-off angles are unknown as no prior information on the event is known (i.e., location, origin time, polarities). Figure 12a also shows that the goodness of fit between observed data and synthetics is strongly dependent on timing and thus, on phase skipping. Indeed, VR values vary with a period depending on the signal wavelength (Fig. 8 and Fig. 12a). In an operational and automatic run, such behavior of cycle skipping has to be correctly calibrated in order to avoid any potential incorrect source detection and misinterpretation.

The location has also a strong impact on the source determination and characterization as shown in Fig. 12b. For the same time occurrence, the solutions obtained at grid points around the GRiD MT optimal solution exhibit a wide diversity of mechanisms, even for the closest points (i.e., apart from only +/-0.2° in latitude and longitude). Once again, the solutions mainly lies on a line in the lune going from the explosive type solution to the implosive type solution. We note however, a significant step in the VRs. The GRiD MT solution of the DPRK 6 nuclear test at this given time is strongly dependent on the location. On the other hand, the moment magnitude Mw remains relatively stable (i.e., 5.2 +/- 0.4), and it is less sensitive to the time or location.

These overall source uncertainties show comparable moment tensor solution dispersion and source determination to those described by Li et al. (2023) for the DPRK 2 to 6 events considering a sophisticated hierarchical Bayesian moment tensor approach used to estimate the sensitivity of the inversions to signal to noise ratio, source depth, and velocity models.

These results show that the GRiD MT algorithm can become an interesting tool for detecting moderate magnitude events of interest for NDCs and other operational seismological institutes. However, such method depends on a very in-depth configuration analysis to identify the stations to be used. The rate of data availability over a relatively long time period may be important to take into account. Also, in the absence of data transmission from one or more stations, the procedure is strongly affected. For a long period in 2013, station INCN was not operating. The missing required data segments from station INCN triggered the obsolescence of the presented configuration, and no detection of the DPRK nuclear test in February 2013 could be obtained (Table 2). In such case, two options are available. The first one is to account for the missing station and to calculate moment tensors using the remaining stations. The other option is to have parallel running systems with complementary station distributions (i.e., using station KS31 in South Korea in replacement of station INCN, Fig. 5). The two cases require important preliminary work in order to constrain their source sensitivities and expected levels of VR for automated event detection in order to limit the false alarm rate.

Similarly, a common remark made about the GRiD MT approach is the very limited number of stations for the inversion. Adding more stations can indeed help better constrain the source radiation. However, we show here that with a well distributed station set, solutions are highly comparable to refined source characterization using two times more stations and time shifts to limit any velocity model inaccuracies (Fig. 4). Also, adding more stations implies larger data storage for the Green’s functions catalogs, and more importantly implies to rely even more on their availability. Furthermore, this requires quasi-perfect knowledge in velocity models for a large number of stations. While it is possible for regions with relatively simple geological properties, this may not be true for more complex tectonically active regions.

Here, large VR peaks are observed for the Mw4.5 + nuclear tests, and these events can then be automatically detected in near-realtime above a given VR threshold. For lower magnitude events, the long period signal to noise ratio appears to be too low in the 15–30 s band period (i.e., DPRK1). In order to better invert and detect lower magnitude events, the option is to consider higher frequency data. However, while theoretically this is possible, our knowledge of the fine Earth’s structure at regional scale is limited.

The proposed approach can be implemented as a rapid and easy to use software, run either automatically or interactively by a seismic analyst or expert. It has been tested on a common Unix workstation, and the 1190 inversions over the grid are computed in less than one second, so less than the considered scanning time step. This demonstrates the opportunity of near-realtime processing that such approach provides. Finally, used in complement to other approaches, it can confirm the occurrence of an event within a region of interest, and more importantly help for event screening. GRiD MT has demonstrated its usefulness for tsunami warning centres that require a quick and reliable moment tensor solution to estimate tsunami impact (Menager et al., 2023). We show here that the use of GRiD MT is not limited to earthquake monitoring, but to any type of seismic event monitoring including events with volumetric changes such as nuclear tests but also complex events in geothermal or volcanic environments.

Acknowledgements:

Great thanks to the IDC and IRIS networks for sharing seismic data. Softwares such as the Generic Mapping Tool (GMT) (https:// www.generic-mapping-tools.org/) and SAC (https:// ds.iris.edu/ds/nodes/dmc/software/downloads/sac/) were used to produce the figures and analyze data. The Computer Program in Seismology (CPS 330) written by Robert B. Herrmann (Department of Earth and Atmospheric Sciences, Saint Louis University) was used for the calculation of Green’s functions. The Matlab and GMT scripts from Carl Tape (https://github.com/carltape/mtbeach) were used for the representation of the source mechanisms on the lune diagram.

Author contributions. Aurelie Guilhem Trilla carried out data processing and analysis. She wrote the manuscript and plotted the results in the published figures. Yoann Cano provided useful comments throughout the data analysis and helped with data acquisition.

No external funding.

Alvizuri, C., and C. Tape (2018). Full moment tensor analysis of nuclear explosions in North Korea. Seismol. Res. Lett. 89, 6, 2139–2151, doi: 10.1785/0220180158.
Bassin, C., Laske, G. and Masters, G. (2000), The Current Limits of Resolution for Surface Wave Tomography in North America. EOS Trans AGU, 81, F897.
Bonner, J. L., Russell, D. R., Harkrider, D. G., Reiter, D. T., and Herrmann, R. B. (2006). Development of a Time-Domain, Variable-Period Surface-Wave Magnitude Measurement Procedure for Application at Regional and Teleseismic Distances, Part II: Application and Ms-mb Performance. Bulletin of the Seismological Society of America, 96, 678–696.
Bowers, D., and Hudson, J. A. (1999). Defining the scalar moment of a seismic source with a general moment tensor. Bull. Seismol. Soc.Am., 89, 5, 1390–1394
Chiang, A., Dreger, D. S., Ford, S. R., Ichinose, G. A., Matzel, E., Myers, S. and Walter, W. R. (2018). Moment tensor source-type analysis for the Democratic People’s Republic of Korea declared nuclear explosions (2006–2017) and 03-SEP 2017 collapse event. Seismol. Res. Lett., 89, 6, 2152–2165, doi: 10.1785/0220180130.
Cesca, S., Heimann, S., Kriegerowski, M., Saul, J., and Dahm, T. (2017). Moment Tensor Inversion for Nuclear Explosions: What Can We Learn from the 6 January and 9 September 2016 Nuclear Tests, North Korea?. Seismological Research Letters, 88, 2A, 300–310, doi: https://doi.org/10.1785/0220160139
Dammeier, F., Guilhem, A., Moore, J. R., Haslinger, F., and Loew, S. (2015). Moment tensor analysis of rockslide seismic signals. Bulletin of the Seismological Society of America, 105, 6, 3001–3014, doi:10.1785/0120150094
Dreger, D. S. (2003). TDMT_INV: Time domain seismic moment tensor INVersion. International Geophysics, 81, Part B, 1627–1627, https://doi.org/10.1016/S0074-6142(03)80290-5
Dreger, D. S., G. Ichinose, and Wang, T. (2017). Source-type inversion of the September 3, 2017 DPRK nuclear test. AGU Fall Annual Meeting Poster, S43H-2392, New Orleans, Louisiana, 14 December 2017.
Dreger, D. S., Gritto, R., and Nelson, O. (2021). Path calibration of the Democratic People’s Republic of Korea 3 September 2017 Nuclear Test. Seismological Research Letters, 92, 3375–3385, doi: 10.1785/0220210105.
Ford, S. R., Dreger, D. S., and Walter, W. R. (2009a). Identifying isotropic events using a regional moment tensor inversion. J. Geophys. Res., 114, B01306, doi: 10.1029/2008JB005743.
Ford, S. R., Dreger, D. S., and Walter, W. R. (2009b). Source analysis of the Memorial day explosion, Kimchaek, North Korea. Geophys. Res. Lett. 36, L21304, doi: 10.1029/2009GL040003.
Ford, S. R., Dreger, D. S., and Walter, W. R (2010). Network sensitivity solutions for regional moment tensor inversions. Bull. Seismol. Soc. Am., 100, 1962–1970.
Gaebler, P. et al. (2019). A multi-technology analysis of the 2017 North Korea nuclear test. Solid Earth, 10, 59–78, https://doi.org/10.5194/se-10-59-2019
Guilhem, A., and Dreger, D. S. (2011). Rapid detection and characterization of large earthquakes using quasi-finite-source Green’s functions in continuous moment tensor inversion. Geophysical Research Letters, 38, L13318, doi: 10.1029/2011GL047550.
Guilhem, A., Hutchings, L., Dreger, D. S., and Johnson, L. R. (2014). Moment tensor inversions of M ~ 3 earthquakes in the Geysers geothermal fields, California. Journal of Geophysical Research: Solid Earth, 119, 3, 2121–2137, doi:10.1002/2013JB010271.
Guilhem, A., and Walter, F. (2015). Full, constrained and stochastic source inversions support evidence for volumetric changes during the Basel earthquake sequence. Swiss Journal of Geoscience, 108, 361–377, doi:10.1007/s00015-015-0197-9
Guilhem, A., and Cano, Y. (2015). MS VMAX: Implementation and developments of an operational tool for event characterization at the French National Data Center. CTBT Science and Technology Conference, oral presentation T2.3-04, Vienna, Austria, June 2015
Guilhem Trilla, A., and Cano, Y. (2017). Constraining the Explosive Nature of the DPRK Nuclear Events at Regional Scale Using Moment Tensor Inversions and MSVMAX. CTBT Science and Technology Conference, oral presentation T2.1-O3, Vienna, Austria, June 2017
Guilhem Trilla, A, and Menager, M. (2023). Using Moment Tensor Inversions for Rapid Seismic Source Characterization: From Nuclear and Regional Seismicity Monitoring to Tsunami Warning. CTBT Science and Technology Conference, oral presentation O2.1-290, Vienna, Austria, June 2023
Herrmann, R. B. (2013). Computer programs in seismology: An evolving tool for instruction and research. Seismol. Res. Lett., 84, 6, 1081–1088
Ichinose, G. A., Myers, S. C., Ford, S. R., Pasyanos, M. E., and Walter, W. R. (2017). Relative surface wave amplitude and phase anomalies from the Democratic People's Republic of Korea announced nuclear tests. 44, 17, 8857–8864, https://doi.org/10.1002/2017GL074577.
Jost, M. L., and Herrmann, R. B. (1989). A student's guide and review to moment tensors. Seismological Research Letters, 37–51.
Kawakatsu, H. (1998). In the realtime monitoring of the long-period seismic wavefield. Bull. Earth. Res. Inst., 73, 267–274.
Kim, W.-Y., Richards, P. G., Schaff, D. P., and Koch, K. (2017). Evaluation of a seismic event, 12 May 2010, in North Korea. Bulletin of the Seismological Society of America, 107, 1, 1–21, doi:10.1785/0120160111.
Knopoff, L., and M. J. Randall (1970). The compensated linear-vector dipole: A possible mechanism for deep earthquakes. J. Geophys. Res., 75, 4957–4963, doi:10.1029/JB075i026p04957.
Lee, S.-J., et al. (2013). Towards real-time regional earthquake simulation I: real-time moment tensor monitoring (RMT) for regional events in Taiwan. Geophysical Journal International, 196, 432–446, doi:10.1093/gji/ggt371.
Li, J., Pham, T.-S., and Tkalcic, H. (2023). Seismic moment tensor inversion with theory errors from 2-D Earth structure: implications for the 2009–2017 DPRK nuclear blasts. Geophysical Journal International, 235, 3, 2035–2054, https://doi.org/10.1093/gji/ggad348.
Liu, J., and Li, L. (2018). North Korea’s 2017 test and its nontectonic aftershock. Geophysical Research Letters, 45, 7, 3017–3025, https://doi.org/10.1002/2018GL077095
MacPherson, K. A., Ruppert, N. A., Dreger, D. S., Lombard, P. N., Freymueller, J. T., Nicolsky, D., and Guilhem, A. (2013). Towards implementation of the GRiD MT algorithm for near real-time calculation of moment tensors at the Alaska earthquake information center. American Geophysical Union, Fall Meeting 2013, abstract id. S51A-2314.
Menager, M. (2023). Étude de l'utilisation de la méthode GRiD MT pour la détection et la caractérisation d'évènements sismiques: de l'échelle régionale à l'échelle locale. PhD dissertation, Université Côte d’Azur, Nice.
Menager, M., Guilhem Trilla, A., Delouis, B. (2023). An automated earthquake detection and characterization tool for rapid earthquake and tsunami response in Western Mediterranean. Geophysical Journal Internation, 234, 2, 998–1011, https://doi.org/10.1093/gji/ggad115
Minson, S. E., and Dreger, D. S. (2008). Stable inversions for complete moment tensors. Geophysical Journal International, 174, 2, 582–592, https://doi.org/10.1111/j.1365-246X.2008.03797.x.
Nayak, A., and Dreger, D. S. (2018). Source inversion of seismic events associated with the sinkhole at Napoleonville sale dome, Louisiana using a 3-D velocity models. Geophysical Journal International, 214, 3, 1808–1829, doi:10.1093/gji/ggy202
Pabian, F., and Coblentz, D. (2018). Observed surface disturbances associated with the DPRK’s 3 September 2017 underground nuclear test. Seismol. Res. Lett. 89, 6, 2017–2024, doi:10.1785/0220180120.
Russell, D. R. (2006). Development of a time-domain, variable-period surface-wave magnitude measurement for application at regional and teleseismic distances, Part I: Theory. BSSA, 96, 2, 665–677, https://doi.org/10.1785/0120050055
Selby, N. D., Marshall, P. D., Bowers, D. (2012). mb:MS event screening revisited. Bulletin of the Seismological Society of America, 102, 1, 88–97, https://doi.org/10.1785/0120100349
Stevens, J. L., and Day, S. M. (1985). The physical basis of mb:Ms and variable frequency magnitude methods for earthquake/explosion discrimination. Journal of Geophysical Research: Solid Earth, 90, B4, 3009–3020, https://doi.org/10.1029/JB090iB04p03009.
Stevens, J. L., and O'Brien, M. (2018). 3D Nonlinear Calculation of the 2017 North Korean Nuclear Test. Seismological Research Letters 2018, 89, 6, 2068–2077, doi: https://doi.org/10.1785/0220180099
Tape, W., and Tape, C. (2012). A geometric setting for moment tensors. Geophysical Journal International, 190, 1, 476–498, doi: 10.1111/j.1365-246X.2012.05491.x
Teysonneyre, V., Feignier, B., Silény, J., and Coutant, O. (2002). Moment tensor inversion of regional phases: application to a mine collapse. Pure and Applied Geophysics, 159, 111–130.
Tian, D., Yao, J., and Wen, L. (2018). Collapse and earthquake swarm after North Korea’s 3 September 2017 Nuclear Test. Geophysical Research Letters, 45, 9, 3976–3983, https://doi.org/10.1029/2018GL077649
Tsuruoka, H., Kawakatsu, H., and Urabe, T. (2009). Grid MT (grid-based real-time determination of moment tensors) monitoring the long-period seismic wavefield. Physics of the Earth and Planetary Interiors, 175, 8–16.
Vasco, D. W., and Johnson, L. R. (1989). Inversion of waveforms for extreme source models with an application to the isotropic moment tensor component. Geophysical Journal International, 97, 1, 1–18, doi:10.1111/j.1365-246X.1989.tb00480.x
Vavrycuk, V. (2015). Moment tensor decompositions revisited. Journal of Seismology, 19, 231–252, https://doi.org/10.1007/s10950-014-9463-y
Walter, W. R., Mayeda, K. M., and Patton, H. J. (1995). Phase and spectral ratio discrimination between earthquakes and explosions. Part I: Empirical observations. Bulletin of the Seismological Society of America, 85, 4, 1050–1067, doi:10.1785/BSSA0850041050.
Wang, T., Q. Shi, M. Nikkhoo, S. Wei, S. Barbot, D. S. Dreger, R. Bürgmann, M. Motagh, and Q. F. Chen (2018). The rise, collapse, and compaction of Mt. Mantap from the 3 September 1 2017, North Korean nuclear test. Science 361, 166–170, doi: 10.1126/science.aar7230.
Yao, J., Tian, D., Lu, Z., Sun, L, and Wen, L. (2018). Triggering seismicity after North Korea’s 3 September 2017 nuclear test. Seismological Research Letters, 89, 6, 2085–2093, doi: 10.1785/0220180135

No competing interests reported.

Download PDF

Journal Publication

published 28 Mar, 2024

Read the published version in Pure and Applied Geophysics →

Editorial decision: Revision requested
14 Dec, 2023
Reviews received at journal
05 Dec, 2023
Reviewers agreed at journal
22 Nov, 2023
Reviewers invited by journal
07 Nov, 2023
Editor assigned by journal
06 Nov, 2023
Submission checks completed at journal
06 Nov, 2023
First submitted to journal
06 Nov, 2023

You are reading this latest preprint version

Using Moment Tensor Inversions for Rapid Seismic Source Detection and Characterization: Application to the North Korean Nuclear Tests

Status:

Journal Publication

Version 1

Abstract

Figures

1. Introduction

2. Moment tensor and grid search approach

2.1. Moment tensor

2.2. The GRiD MT approach

3. Data, grid and velocity models

4. Results for the monitoring of seismic events in North Korea

5. Source uncertainties

6. Discussion and conclusion

Declarations

References

Additional Declarations

Status:

Journal Publication

Version 1