Expert Technical Analysis of the DPRK-2013 seismic event and waveform cross-correlation perspectives at the International Data Centre
Dmitry Bobrov, Ivan Kitov and Mikhail Rozhkov,
International Data Centre, CTBTO
Abstract. We have performed a comparative analysis of the three announced DPRK underground tests using data from the
International Monitoring System (IMS), including the fusion of seismic and infrasound technologies. A unique similarity between the
2009 and 2013 year waveforms and spectra allow one to draw strong conclusions about the similarity of the source mechanism and
the conditions of conduction: depth of burial, geological structure, tectonic stress, and the containment technology. This, however,
presumes that all three events were explosions. For sources close in space and mechanism, waveform cross-correlation (CC) is a
natural technique for continuous monitoring as the joint processing of the 2006, 2009, and 2013 data demonstrates. The CC-based
relative location has a few hundred meters resolution (i.e. by two orders of magnitude more accurate than the IDC absolute location).
The detection capability is enhanced by 0.5 units of magnitude and allows detection of M2.5 aftershocks. The results of the IDC
expedited CC-based location of the 2013 event (14 min after the event) are supported by later reports of different national agencies.
This proves the perspective of the IDC CC monitoring prototype, its state-of-the-art status, and its importance in the CTBT practice.
Conclusions. Relative location based on waveform cross-correlation (CC) approach is a highly demanded technique. In seismology, this demand is rising
from year to year as seismic data accumulated in data centres allows utilization of simple, robust and straightforward technology based on well-known
principle of “matched filter” which is wide–spread in radio industry for decades (radars, pulse compression and image processing). This approach is widely
used in seismology since the early 1980s, mainly in regional studies but for teleseismic records analysis as well. Nevertheless, first near-real-time
implementation, comprising signal detection, association and location was performed in the IDC. No doubt, CC based system is an excellent tool for
monitoring specific area, such as present and former nuclear test sites, and seismically active areas. If just few reporting stations situated within the regional
distance range form the suspected area, the reliable alarm notification with the origin estimate can be obtained within 5 to 15 minutes. Recent DPRK-2013
case confirmed the productivity and perspective of the cross-correlation approach in global CTBT monitoring. Combined with fusion of data from the IMS
network it can also give additional information about the nature of seismic event and it’s features. . The 2009 and 2013 events demonstrate very good
waveform and spectral similarity. Most likely they were conducted very close to each other, for example, in tunnels within different hills or mounts, at interdistance between several hundred meters and a kilometre. Due to high correlation between the events this is not surprising that both events demonstrate
DPRK events waveforms and spectra similarities, and other features
Excellent waveforms similarity between DPRK 2009 and 2013 indicates the same test conditions and very close location. Good examples are stations
KSRS and USRK, presented on figures below. Left: similarity of waveforms for 2009 and 2013 events at KSRS (top) and USRK (bottom). Right: crosscorrelation 2009 and 2013 events at USRK. Correlation window is 23 seconds. Top: raw waveform correlation, bottom: correlation of records filtered at
0.6-4.5 Hz. CC is high even for raw data: 0.9, and 0.96 for 0.6-4.5 band pass filtered. For NVAR station CC achieves 0.99 for 0.5-2Hz band. Different
example is from the set of Soviet Peaceful Nuclear Explosions (PNE), events Vega-5.1 and Vega-5.2. Data recorded at NORSAR, distance to station is
25.25o. Cross-correlation for these closely spaced (0.05o) explosions is 0.68 for raw data and 0.83 for filtered at 1-4 Hz. We can assume that the DPRK
2009 and 2013 events occurred very close to each other (see relative location section on right).
Vega 5.1 and 5.2 PNE as
they go at continuous
record.
5.1 and 5.2
Comparative displacement
spectra for 2013 (green) and
2009
(blue)
events.
Background noise is lower
curves. 2013 event has more
prominent 2-3Hz spectral
maximum while the slope
areas at about 6.0Hz are
similar.
The advantages of the IDC waveform cross-correlation monitoring system prototype
DPRK-2013 event has all earthquake-like and explosion-like
features as DPRK-2009 event has, for example: intensive highfrequency P-wave content for regional stations (EX feature)
and low-frequency measurable P-wave content for all distances
(EQ feature); low-frequency Sn and Lg (EQ feature); low
frequency Rg (EQ feature) and PcP without ScP:
Determination of DPRK-2006 event using 2009 and 2013 location results.
Master1 ORID: 5397597 25-05-2009 00:54:42.8 41.311oN 129.0464oE
Master2 ORID: 9486691 12-02-2013 02:57:50.8 41.3068 oN 129.0421 oE
3 stations used for location (as USRK was not in operation in 2006):
KSRS, MJAR, SONM
DPRK-2009
DPRK-2013
41.305 oN 129.0695 oE
DPRK-2009 relative to DPRK-2006
DIST1: 2.05 km AZ: 108.9 DNORTH1: -0.67 km DEAST1: 1.94 km
DPRK-2013 relative to DPRK-2006
DIST2: 2.31 km AZ: 95.0 DNORTH2: -0.20 km DEAST2: 2.29 km
DPRK-2006
Intensive low-frequency (filter
below 2 seconds) shear waves
signatures
at
broadband
sensors for both 2009 and
2013 events may expose the
presence of DC or CLVD.
Clear PcP without ScP at
wide distance range (filtered
and raw waveforms).
Existence of very low frequency
Rg at regional stations (shallow
earthquake feature). Band-pass
filter band is 100 sec to 20 sec at
broadband channels. Three upper
records correspond to 2013 event,
3 lower records – to 2009 event,
and the record in the middle is
raw 2013 vertical seismogram.
Rg waves for KSRS
array: 2013 (upper
record) and 2009.
Band-pass filter of
0.05-0.1Hz applied.
Spectral Ratio as yield value indication
features of shallow isotropic and double couple, or CLVD sources. The main difference between the events is the existence of I and Pn phases at I45RU
infrasound station collocated with the station USRK (Δ=3.6o), which is is a manifestation of relatively shallow seismic event (for Δ=3.58o and ML=4.2) and
the evidence of higher magnitude of the DPRK-2013 event. Continuous monitoring of the suspected area in DPRK with waveform cross correlation based of
the 2009 DPRK event as a master resulted in the conclusion that there were no aftershocks recorded after the DPRK-2013 event. It means that there was no
major collapse of the cavity created by the explosion. There was a very little chance for radionuclides (both particulates and noble gas) to reach the free
surface and vent into the atmosphere, unless the event was very shallow. From the other hand, seismic and infrasound attributes witness of the shallow event,
with the depth of several hundreds of meters. So the fission products most likely could only penetrate to the atmosphere via the rubble chimney after the
cavity collapse A synthetic approach to the explosion depth determination at teleseismic distances for wide range of yield provide reasonable results
corresponding with the above statement. Similar results obtained at other stations for DPRK events as well as for Soviet PNE program events, so further work
in this direction may contribute into CTBT performance enhancement.
DPRK-2013 relative location. White circle – 2013 event, red circle – 2009 event. Left: solution built with 4 regional IMS stations.
Center: solution built with 22 IMS stations. Right: joint relative location, where initial master event is DPRK-2009, primary slave
is DPRK-2013 event and secondary slave is DPRK-2006 event. Final location is based on reciprocal cross-correlation when the
estimate is based on mutual master-slave permutation. Distance between events is 590 meters for 4 stations, and 570 meters for 22
stations. X and Y distances are 470 and 360 meters for 4 stations, and 400 and 410 meters for 22 stations.
USRK
KSRS
MJAR
SONM
ZALV
MKAR
CMAR
KURK
BVAR
ILAR
ARCES
FINES
WRA
YKA
ASAR
AKASG
HFS
NOA
BRTR
GERES
EKA
NVAR
PDAR
TXAR
2006/2009 2009/2006 2006/2013 2013/2006 2009/2013 2013/2009
0.98
0.98
0.64
0.62
0.62
0.64
0.95
0.95
0.63
0.72
0.73
0.63
0.91
0.91
0.69
0.66
0.66
0.68
0.94
0.94
0.90
0.91
0.51
0.54
0.54
0.50
0.98
0.98
0.89
0.89
0.96
0.97
0.97
0.96
0.99
0.99
0.34
0.37
0.35
0.35
0.91
0.91
0.75
0.78
0.78
0.67
0.97
0.97
0.90
0.91
0.91
0.90
0.97
0.97
0.23
0.93
0.94
0.72
0.68
0.70
0.97
0.97
0.74
0.74
0.74
0.74
0.95
0.95
0.75
0.79
0.69
0.74
0.96
0.96
0.60
0.52
0.79
0.60
0.81
0.81
0.64
0.66
0.96
0.96
0.56
0.57
0.57
0.57
0.97
0.97
0.91
0.92
0.89
0.90
0.90
0.89
0.99
0.99
0.75
0.71
0.73
0.73
0.96
0.97
0.91
0.88
Cross-correlation coefficients for different
master-slave pairs for different stations
nsta
15
Sta
KSRS
MJAR
SONM
MKAR
ARCES
FINES
WRA
ASAR
AKASG
HFS
NOA
BRTR
GERES
NVAR
PDAR
gap
102
CC
0.67
mb+RM
4.92-0.94
Tstd
0.11
TRes
-0.01
0.21
-0.2
-0.11
-0.11
0.01
-0.14
0.14
-0.01
0
0.01
-0.01
-0.01
0.16
0.09
CC
0.64
0.63
0.69
0.51
0.34
0.75
0.9
0.72
0.74
0.75
0.6
0.64
0.56
0.89
0.75
dRM
-0.97
-0.02
-1.07
-1.22
-0.75
-0.89
-0.85
-1.01
-1.05
-0.91
-1.13
-1.12
-1.06
-1.07
cc_snr
8.7
11.2
5.2
3
2.9
8.8
11.2
7.9
9
3.2
8.3
5.3
6.3
7.8
5.9
nsta
15
snr
160.1
19.4
3.7
4.1
2.1
23.2
44.8
8.3
8.9
7.1
4.7
3.4
5.9
18.1
7.2
Sta
KSRS
MJAR
SONM
MKAR
ARCES
FINES
WRA
YKA
AKASG
ASAR
HFS
NOA
GERES
NVAR
PDAR
gap
102
CC
0.65
mb+RM
4.08+0.55
Tstd
0.13
TRes
-0.05
-0.34
0.2
0.1
-0.07
-0.05
0.13
0.2
0
0.1
0
0
0
-0.12
-0.1
CC
0.62
0.72
0.66
0.54
0.37
0.78
0.91
0.23
0.74
0.68
0.79
0.52
0.57
0.9
0.71
dRM
0.52
0.35
0.58
0.8
0.47
0.72
0.41
0.23
0.75
0.34
0.52
0.57
0.75
0.6
0.6
cc_snr
5.4
11.9
5.4
4.1
3.9
5.8
11.7
4.5
9
7.6
4.5
9.1
5.4
8.4
8.2
The relative cross-correlation based location results are in good correspondence
with NORSAR relative location (upper, left figure) and differs from other
agencies: USGS, Stony Brook University, and IRIS
USGS
snr
466.5
33.4
14.8
19.8
15.2
50.8
192.3
9.2
26.7
8.6
35.9
22.3
37.7
56.8
14.5
NORSAR
Stony Brook
Partial XSEL (REB-like cross-correlation) bulletin for relative location of
DPRK-2006 event based on: (1) DPRK-2013 is a master event (left table), and
(2) DPRK -2009 is a master event (right table).
IRIS
Infrasound: epicentral vs. local
DPRK events depth determination with synthetic seismograms
Infrasound detection at I45RU station (USRK-collocated station) associated with the 2013/02/12 02:58:47.525 event is a manifestation of relatively shallow seismic
event (for Δ=3.58o and ML=4.2):
DPRK 2013, 2009 and 2006 waveforms and preceding noise
Sta
I45RU
USRK
Dist
3.58
3.58
EvAz Phase
35.3
I
35.4
Pn
Time
03:16:30.000
02:58:47.525
TRes
-228.0
-0.2
Azim AzRes
212.4 -4.9
213.2 -4.2
Slow
329.2
12.0
SRes Def
-49.9 TA_
-1.8 TAS
SNR
Amp
1.3
1381.0 207.7
Per
Qual Magnitude
a__
0.33 a__ ML 4.2
ArrID
84147154
84146290
The 2009 event did not have infrasonic arrivals at IMS stations, so this is good example of how relatively insignificant magnitude scaling (4.5 mb(IDC) vs 4.9
mb(IDC) of DPRK-2013) of the event occurred at approximately same depth (see Introduction chapter below) may produce qualitatively different wavefield
paradigm. Even more prominent phenomena takes place for infrasound excitations at I45H*/BDF channels corresponding to Pn and Pg seismic phases at USRK
due to good seismic-to-acoustic coupling. Again, 2009 event did not have any evidence of such intercoupling, which is the evidence of higher magnitude of the
DPRK-2013 event. Good time-alignment of the corresponding arrivals at seismic and infrasound stations says about close proximity between the two stations.
DPRK 2013, 2009 and 2006 spectra and preceding noise
Top: Three DPRK events
with noise (bottom of pair)
(left), spectra in linear scale,
and SNR as function of
frequency after applying
homomorphic smoothing.
Note increase of amplitude
up to 6Hz.
Left: Importance of analysis for higher
frequencies. Results represented for near
regional station KSRS recorded DPRK 2006,
2009 and 2013 events. For approximately the
same noise level beyond mark 8 on X-axis,
the spectral ratios demonstrate the spectral
pick drift from higher to lower frequency
going from 2006 event to 2013. This is a
manifestation of high frequency content
increase for lower yield explosions. As it can
be seen from the raw data processing (upper
row) spectral ratios do not help much in
determining this fact. To make it clear we
applied homomorphic filtering passing only
low-quefrency cepstral content according to
optimality criterion established by [Rozhkov,
Raw spectra for pairs of 3 DPRK events and ratios, and smoothed Shpilker, 1986].
spectra for 7Hz and 10Hz cut offs (from top to bottom)
Preparatory Commission for the Comprehensive Nuclear-Test Ban Treaty Organization,
Provisional Technical Secretariat, Vienna International Centre, P.O. Box 1200, A-1400
Vienna, Austria. E-mail: [email protected]
The event location based on master
event approach looks like a promising
way to the improvement of the CTBT
performance (network detectability and
location accuracy). Good correlation
can be demonstrated for these two
events not only for regional but for the
teleseismic distances as well.
First figure demonstrates the
teleseismic P wave for
NVAR for 3 DPRK events:
2006, 2009 and 2013. The
lower trace is the synthetic
seismogram computed for
the origin depth of 700
meters, t*=0.5, teleseismic
model is AK135, receiver
velocity model taken from
CRUST-2
project
and
source model taken from the
Robert Herrmann’s (Univ.
of St. Louis) models
(http://www.eas.slu.edu/eqc)
Note good waveform similarity
especially for the two upper
waveforms: for years 2009 and
2013. We created the synthetic
seismograms for wide range of
depths starting surface to 10 km
depth. It can be seen from the
correlation table, that the best
correlation
between
the
synthetic and real seismogram
is obtained for depths 600 and
700 meters). The same synthetic
approach was applied to both
2006 and 2009 events. In case
of 2009 it demonstrated similar
depth (around 700 meters) but
for 2006 it showed the depth
approximately 200-300 meter
shallower.
Pn and Pg phases at USRK and I45 and their spectra
(top – USRK, bottom – I45) high-pass filtered from
0.4Hz. Note higher frequency content at I45.
Pn and Pg at seismic (USRK) and infrasound (I45) stations
filtered at 0.6-4.5Hz. Note waveforms similarity.
The I-phase was detected by pmcc, and is
not visible at original I45 waveforms. It is
buried in noise and only predicted arrival
is available. So it’s amplitude is not higher
than 0.01-0.02 Pa. The surface wave is too
weak to generate arrival at infrasound
station. In given case acoustic-to-seismic
coupling produces a negligible signal,
while the seismic-to-acoustic coupling is
significant to produce a measurable signal
(normally if the acoustic wave is
significant it can generate the Rayleigh
wave when both velocities approaching by
values).
Seismic (USRK, Δ=3.61o) and
seismo-acoustic (I45, Δ=3.6o) phases
from DPRK-2013 event.
The figures indicate that the prominent signal at I45 station must correspond to the body wave arrivals. The amplitude of the infrasound signal is estimated to 0.02 Pa. Theoretically the overpressure in the infrasound
signal is proportional to the free surface vertical (normal to the surface) particle velocity. By using the relationship ∆P = ρcU, where ∆P is the overpressure, ρ is the air density, c is the sonic speed, and U is the
particle velocity in the direction normal to the surface/air interface, one can estimate the vertical particle velocity for the surface wave (Kitov, 2003). Assuming ρ = 1.29 kg/m3, c=330 m/s, and the overpressure of
0.02 Pa we obtain: Ua = 0.03/(330*1.29) [kg/m2]*[m/s2]/[kg/m3]*[m/s] ~ 0.07 mm/sec. This estimated value corresponds with values obtained at station USRK with certain deviation. According to (Kitov, 2003),
the Geotool scaled displacement amplitude needs to be scaled with the additional factor 2πf, where f =1/calper (calper=0.2 sec for Geotech GS-21 with Nanometrics Europa-T digitizer), in order to obtain an absolute
estimate in terms of particle velocity. Us = 1000*2π*5 = 31400nm/sec ~ 0.0314 mm/sec. So Us/Ua is about 2.2. Ideally, if seismic and infrasound stations are really co-located, the ratio Us/Ua1. There might be
different reasons for this deviation, like significant separation between the stations, or not proper equipment calibration. The coordinates of IS45RU stations is: 44.1999oN, 131.9773oE. The coordinates of USRK
reference channel is: 44.1998oN, 131.9888oE. So distance in longitudinal direction is about 0.01 degree. This is hard to say if the separation like that may be the reason of the ratio deviation, or the station calibration
should be double checked. Nevertheless, seismic-to-acoustic coupling gives reasonable results so it does not look like there is substantial error in station calibration.
Disclaimer
The views expressed on this poster are those of the authors
and do not necessary reflect the views of the CTBTO Preparatory Commission
International Data Centre
http://www.ctbto.org