We use repeating events at the DPRK test site in order to detect the ultra-weak signal at the Mikhnevo array by using waveform cross correlation

1. Detection of the 2006 DPRK explosion by small-aperture array “Mikhnevo” using waveform cross correlation
Ivan Kitov and Irina Sanina
Institute of Geospheres Dynamics, Russian Academy of Sciences
Abstract. Three underground tests conducted by DPRK in 2006,
2009, and 2013 were measured by small aperture seismic array
“Mikhnevo”, which has been operated by the Institute of
Geospheres Dynamics since 2004. This array is designed for the
purposes of regional seismic monitoring. It has aperture of
approximately 1 km and includes ten vertical and two 3-C stations.
Sampling rate is 200 counts per second. Automatic processing
includes beamforming (azimuth and slowness values for the preset
detection beams cover the range of regional and teleseismic body
waves), filtering and detection by standard STA/LTA procedure.
Distinct signals generated by the 2009 and 2013 events were
detected. The 2006 event was not detected by standard procedure
(detection threshold was STA/LTA=3.5) and we applied waveform
cross correlation (matched filter) in order to improve signal-to-
noise ratio (SNR). Multichannel waveforms from the 2009 and
2013 signals filtered in various frequency bands were used as
templates. The 2006 signal was detected with the cross
correlation technique with SNR>4. Therefore, the matched filter
technique improves detection capability of a small-aperture array
even for teleseismic waves.
Corresponding author: Ivan Kitov (IDG RAS)
E-mail: ikitov@mail.ru
IDG RAS
http://idg.chph.ras.ru
Disclaimer: The views expressed on this poster are those of the authors
and do not necessary reflect the views of the Institute of Geospheres Dynamics, RAS
Conclusion
Small aperture seismic array MHVAR has detected the
DPRK 2006 announced underground test using the
technique of waveform cross correlation (matched
filter) and multichannel waveform templates recorded
from the 2009 and 2013 tests. The length and
frequency bands for the templates have to be selected
carefully in order to maximize the performance of
cross correlation.
At the same time, standard method of beamforming
has failed to suppress noise to the level necessary for
detection of the DPRK 2006. In both cases, detection
threshold of STA/LTA=SNR>3.5 was selected as it is
providing a lower rate of false alarms. The STA
window has the length of 0.5 s and the LTA window
was 50 s.
Modelling with changing level of microseismic noise
has demonstrated that the matched filter technique
allows to detect underground tests with magnitude
below 4.0 from the DPRK test site while beamforming
provides detection threshold of 4.1 to 4.3 for the level
of microseismic noise observed on October 9, 2006.
The suppression of not correlated noise by
beamforming could improve SNR by a factor of 3.5
(for 12 channels). The application of waveform cross
correlation at MHVAR has reduced the threshold of
detection of signals from different sources by a factor
of 10 in comparison to a three component station. For
underground explosions, a tenfold decrease in the
amplitude threshold of signal detection means that an
explosion source with seismic energy decreased 10
times can be detected. The use of various signal
detection filters may improve SNR by a factor of 3.
Additional measures such as reduction of the
microseismic noise level (for example, deployment of
sensors in wells at the depth of a few tens of meters),
increase in array aperture, and increment of the
number of seismic sensors lead to further reduction in
the detection threshold.
In the current study we use the
records of three DPRK events at
small-aperture seismic array
Mikhnevo (MHVAR). The Institute of
Geospheres Dynamics (IDG) of the
Russian Academy of Sciences
operates seismic array MHVAR
(54.950N; 37.767E) since 2004.
Small-aperture seismic array
“Mikhnevo” includes ten vertical
stations (solid triangles), with one
station in the geometrical centre of
the array (C00) and other nine
stations distributed over three
circles with radii of 130 m, 320 m,
and 600 m. The array aperture in
approximately 1.1 km. Two 3-C
stations (solid triangles in circles)
were added to the outer circle in
order to improve the overall stations
sensitivity (detection threshold) and
resolution. All stations are equipped
with short-period seismometers
SM3-KV, which are characterized by
flat response between 0.8 Hz and 30
Hz and gain of 180,000 [Vs/m].
Later, a 3-C broad band station (BB)
was installed in the centre of the
array for surface wave
measurements. The array response
function (only for 12 vertical
channels) is similar to that for many
small-aperture arrays. Such arrays
are designed to measure high-
frequency signals from regional and
near-regional sources with
magnitudes above 2.0.
3.0
30.0
0 2 4 6 8 10
SNR
noise factor, C
2013 STA/LTA
2013 CC 1/2
2013 CC 1/1
2009 STA/LTA
2009 CC 2/1
2009 CC 2/2
MHVAR design
Template 2013, Butterworth 3-d
order filter 2-4 Hz
Template 2009, Butterworth 3-d
order filter 3-6 Hz
Beams of MHVAR vertical channels, obtained by summation of waveforms normalized to their respective
peak values. Time delays correspond to the slowness vector with the peak SNR for the explosions of 2009
and 2013. Dashed lines – for the original waveforms, solid lines – the original signals merged with
preceding noise multiplied by factor (C) 3 and 6, respectively.
For signal detection, we applied the method
based on the ratio of the average absolute
amplitudes in the short (STA) and long (LTA)
time window, STA/LTA, which is aslo used to
estimate the signal to noise ratio: SNR =
STA/LTA. The length of the STA window is
0.5 s, while that of the LTA window is 60 s.
To increase the SNR value, we applied
filtering in seven frequency bands (Hz): 0.5–
2, 1–3, 2–4, 3–6, 4–8, 6–12, and 10–20. For
this purpose, we used a third order band-pass
Butterworth filter.
According to the estimates by the
International Data Centre of the
Comprehensive Nuclear-Test-Ban Treaty
Organization (CTBTO), the magnitudes (mb)
of the events in 2006, 2009, and 2013 were
4.1, 4.5, and 4.9, respectively. The estimate of
the 2006 event energy from averaged
spectra of teleseismic P-waves falls within
the range of 0.6–1.0 kilotons of TNT (1 kg
of TNT = 4.184 × 1012 J), while it is 2.0–4.8
kilotons of TNT for the 2009 event. For the
explosion of 2013, the energy estimated
from the magnitudes of Lg waves is from
4.04 to 8.47 kilotons of TNT.
The problem of detecting signals from
sources close in space can be solved with the
help of waveform cross correlation. As a
reference waveform, we chose the records of
the events of 2009 and 2013, having the
highest SNR value. The correlation factor
was calculated for each channel of the
Mikhnevo array.
Date OT Arrival
time
Travel
time
SNR Arrival
time
Travel
time
SNR_CC
09.10.2006 01:35:27 NA NA NA 01:45:24 09:57.0 3.9
25.05.2009 00:54:42 01:04:39 09:57.0 11.6 01:04:38 09:56.0 13.4
12.02.2013 02:57:49 03:07:47 09:58.3 31.2 03:07:47 09:58.0 19.0
WAVEFORM TEMPLATES
WAVEFORMS SPECTRA
signal/noise
SMALL-APERTURE SEISMIC ARRAY MIKHNEVO (MHVAR)
is designed for regional studies
2013
CROSS CORRELATION COEFFICIENTS, SNR, AND FK
2013/
2013
2009/
2013
2013/
2009
2009/
2009
2013/
2006
Template 2013, channel C00
six filters
2009
Array response
Examples of regional signals
Seismometer response
DETECTION USING WAVEFORM CROSS CORRELATION
DETECTION OF DPRK EVENTS WITH BEAMFORMING STUDYING DETECTION THRESHOLD FOR BEAMFORMING AND CROSS CORRELATION
Beamforming and detection with noise factor С=3
SNR as a function of noise factor for
the DPRK 2013 and 2009
Cross correlation and detection with noise factor С=3
2013 2009
2009
2009
2013
2013
Cross correlation with noise factor С=6, 2013
Statistic of detections with beamforming and
cross correlation
BEAMFORMING AND DETECTION
Poster No. T2.2-P7