Waveform cross correlation (WCC) is an optimal detection technique for signals from spatially close seismic sources. Observations at various distances from a multitude of sources in a variety of seismotectonic and geological conditions demonstrate that signals from close events recorded at common stations are characterized by high level of similarity. Signals from remote sources are less similar mainly because of the variations in propagation paths. Different parts of a complete seismic wavetrain have different sensitivity to the propagation path. The initial part retains general characteristics of the source time function. The shape of later seismic phases is chiefly defined by propagation path. Here, we investigate the level of similarity between hundreds of signals generated by chemical blasts within a phosphate mine in Jordan and measured by 5 seismic stations at near-regional distances. We have revealed a much higher similarity of the first 3 s to 5 s of signals from different blasts, also at distances of about 20 km, at the same station as well as at different stations. This observation evidences in favour of high coherency in the initial part of signals at all stations. We also demonstrate that the observed coherency allows the use of very short (say, 3 s) waveform templates for detection and further phase association of signals based on cross correlation. Longer templates are characterized by larger overall signal specificity, which may reduce detection threshold and spatial resolution of the WCC method. However, different propagation paths within the same geological province may have similar transfer functions producing regular seismic phases with similar shapes independent on source position. This may increase the number of false detections from remote sources. We compare the performance of short and long waveform templates using detection statistics and the results of event hypotheses creation and further event location.

1. Abstract
Waveform cross correlation (WCC) is an optimal detection
technique for signals from spatially close seismic sources.
Observations at various distances from a multitude of
sources in a variety of seismotectonic and geological
conditions demonstrate that signals from close events
recorded at common stations are characterized by a high
level of similarity. Signals from remote sources are less
similar mainly because of the variations in propagation
paths. Different parts of a complete seismic wavetrain have
different sensitivities to the propagation path. The initial
part retains general characteristics of the source time
function. The shape of later seismic phases is chiefly
defined by propagation path. Here, we investigate the level
of similarity between hundreds of signals generated by
chemical blasts within a phosphate mine in Jordan and
measured by 5 seismic stations at near-regional distances.
We have revealed a much higher similarity of the first 3 s to
5 s of signals from different blasts, at distances of about 20
km, at the same station as well as at different stations. This
observation favours high coherency in the initial part of
signals at all stations. We also demonstrate that the
observed coherency allows the use of very short (say, 3 s)
waveform templates for detection and further phase
association of signals based on cross correlation. Longer
templates are characterized by larger overall signal
specificity, which may reduce the detection threshold and
spatial resolution of the WCC method. However, different
propagation paths within the same geological province may
have similar transfer functions producing regular seismic
phases with similar shapes independent of source position.
This may increase the number of false detections from
remote sources. We compare the performance of short and
long waveform templates using detection statistics and the
results of event hypotheses creation and further event
location.
WAVEFORM CROSS CORRELATION DEPENDING ON TEMPLATE POSITION
WAVEFORM CROSS CORRELATION DEPENDING ON TEMPLATE LENGTH EFFICIENCY OF TEMPLATE PERFORMANCE IN EVENT BUILDING
The phosphate mine Eshidiya, 70 km away from Ma'an District and 330 km from
Amman, has been extracting fossils using chemical blasts since 1989. It has two
sites with the largest possible distance between events of approximately 20 km, a
challenge for detection using waveform cross correlation
A few thousand blasts were conducted within the quarry. For the study of
waveform cross correlation, we have selected 1654 events (blue circles) with
records at station HRFI. There are other stations measuring most of these events:
PRNI, EIL, ASF, MMAI.
3-C templates (E – east-west, N – north-south, and Z - vertical) as
measured by EIL from a blast conducted within the quarry. Each
waveform has been filtered in six frequency bands (four are shown):
F1: 1-3 Hz, F2: 2-4 Hz, F3: 3-6 Hz, F4: 4-8 Hz, F5: 6-12 Hz, and F :
8-16 Hz in order to obtain the highest signal-to-noise ratio in cross
correlation. All templates include 5-second-long pre-signal segments.
Notice the change in frequency content in P and Lg phases.
BLASTS AT ESHIDIYA MINE
Waveform cross correlation: coherency of seismic signals estimated from repeated mining blasts
Yochai Ben Horin, Soreq Nuclear Research Centre, Israel, Ivan Kitov, PTS CTBTO, Mikhail Rozhkov, PTS CTBTO,
Matthew Yedlin, Department of Electrical and Computer Engineering, University of British Columbia
Hourly distribution of built events within a day. Three versions of
template Length are shown: 3 s, 5 s, and “LONG” corresponding
to the highest CC-detector sensitivity. The number of potentially
false events out of working hours increases with the template
length.
Distribution of the built events over the number of associated
stations depending on the length of templates. The longest
template created more 3-station events and the 3-s templates
create more 5-station events
Distribution of the built events over the number of associated
stations depending on the length of templates. NEW are the
events built by cross correlation in addition to the external
Catalogue.
We randomly select 2000 waveform/template pairs from ~1500
signals measured by EIL from blasts within the same mine and
gradually increase the length of template from 1s to 22 s. The length
with the highest SNR for a given pair is taken as the detection
length and counted in corresponding 1 s bin. The distribution of the
number of detections over these beams allows selection of the most
efficient length – 18 s for EIL. Second run with different random
pairs shows that the result does not change and 2000 is a
representative number.
Same as above for station ASF. The most efficient length is 36 s
with 2 s the second most efficient.
The distribution of the number of detections over template length for
station PRNI. The most efficient length is 1 s with 17 s the second
most efficient.
Distribution of SNRCC for station PRNI over the length of
waveform template. On average, SNRCC are smaller for shorter
templates and larger for the lengths between 15 and 20 s. The latter
is the effect of more effective noise suppression.
ASF EIL HRFI MMAI PRNI DATA DAYS NSTA>2
364 385 399 391 203 402
Nsta #
3 44
4 204
5 154
3-C templates (E – east-west, N – north-south, and Z -
vertical) as measured by ASF from a blast conducted
within the quarry. The ASF waveforms are different from
those at EIL due to larger distance to the quarry
Data availability by station for 415 days with reviewed catalogue. Only 402 days have data
from 3 or more stations available for cross correlation. No events can be found during the
days when less than 3 stations are available.
Data availability the
number of stations. 402
days in total.
Frequency distribution of the number of detections at
station EIL as a function of the position of template in
the signal. The length of template is defined by letter S
(3-5-7). The frequency band is defined by letter F (1-5).
For example, start point 3 s for S7 means the portion of
template between 3 s and 10 s is used for cross
correlation.
Performance of three detection lengths at stations EIL,
HRFI, and PRNI for one filter – F3. For the length of 3 s
(S3) al stations have peak detection rates with 0 start
time, i.e. the start of waveform template. This effect is
most prominent at HRFI.
For station ASF situated at larger distance the
performance of the first portion of the signal is less
effective due to lower coherency in the first phases and
development of a likely more coherent Lg phase, which
becomes more structure dependent with range
The probability distribution function of SNRCC values at five
stations obtained for 1 year period with the length of waveform
template of 3 s. Total number of individual estimates is 365days x
24hours x 3600s x sampling rate. All stations demonstrate very close
distributions with EIL having the largest number of detection and
MMAI the lowest. PRNI is worse than remote ASF.
Same as in above figure for the length of template corresponding to
the highest detection efficiency: 18 s for EIL, HRFI and PRNI; 36
s for ASF. ASF has largest number of detections with higher
SNRCC.. For MMAI, the length of template is 6 s. This length is
called “LONG”.
We selected 22 events with templates at all 5 stations and cross
correlated them through the whole period covered by the external
bulletin. Three versions of template length were used: 3 s, 5 s, and
“LONG”. The upper panel shows the number of events built by
each master. In the middle, we show the number of events left for a
given master after conflict resolution. This bulletin is called
XSEL_CR and the bottom panel shows the number of events from
the bulletin found by each master. Five masters are most prolific.
Input of 5 stations to the final cross correlation bulletin
XSEL_CR. “LONG” creates more events than 3 s and 5 s
templates. PRNI gives lower number of arrivals than ASF and
MMAI likely because of the worst data availability.
Conclusion: The 3-s waveform templates are the most efficient in finding of
similar events with the largest number of associated stations and guarantee a
lower rate of false event hypotheses
The first few seconds demonstrate the highest coherency of signals
The detection efficiency of waveform templates critically depends on their lengths