Global grid of master events for waveform cross-correlation: from testing to real time

Dmitry Bobrov, Ivan Kitov, and Mikhail RozhkovAbstract Seismic monitoring of the Comprehensive Nuclear-Test-Ban Treaty (CTBT) requires a globally uniform detection threshold, which is provided by geographical distribution of the Primary Seismic Network of the International

Monitoring System (IMS). This detection threshold has to be as low as allowed by the entire set of real time and historical data recorded by the IMS. The International Data Centre (IDC) analyzes all relevant data in automatic processing and

interactive review to issue a Reviewed Event Bulletin (REB), which includes all qualified events as obtained for the purpose of nuclear test monitoring. Since 2000, raw data, individual detections, and created events are saved in the IDC

archive currently reaching tens of terabyte. In order to effectively use this archive in global monitoring we introduced the waveform cross correlation (matched filter) technique. Cross correlation between real time records at IMS stations and

template waveforms is calculated for a dense (spacing of ~ 140 km) and regular grid of master events uniformly covering the globe. There are approximately 25,000 master events with 3 to 10 templates at IMS stations. In seismically active

zones, we populate masters with real waveforms. For aseismic zones, we develop an extended set of synthetic templates for virtual master events. For optimal performance of cross correlation, the Principal and Independent Component

Analysis are applied to the historical (from earthquakes and underground nuclear tests) and synthetic waveforms. Real waveform templates and selected PCA/ICA components are used in automatic processing for the production of a tentative

cross-correlation standard event list (XSEL).

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: Mikhail.Rozhkov@ctbto.org

International Data Centrehttp://www.ctbto.org

Disclaimer

The views expressed on this poster are those of the authors

and do not necessary reflect the views of the CTBTO

Preparatory Commission

The International Monitoring System (IMS) network includes 50 primary seismic stations,

which are divided into seismic arrays (circles) and three-component (3-C) seismic stations

(triangles). Auxiliary seismic arrays (circles) are also shown.

Density (# per 1deg x 1 deg cell) of shallow events in the Reviewed Event Bulletin. Waveforms

from many seismic sources can be used as templates for cross correlation. For earthquakes, cross

correlation most effective for close events.

Representations of the 45 dip slip fault. (a) Focal mechanism,

(b) Style of faulting, and (c) Radiation pattern.

Conclusion

• IMS array stations make possible automatic processing based on waveform cross correlation.

• Cross correlation is a powerful technique allowing to reduce the detection threshold and relative location accuracy by an order of magnitude, i.e. to find by 50% to 100% more (smaller) REB events.

• Real and synthetic master events may reduce the magnitude threshold of seismic monitoring by 0.4 units of magnitude.

• The Global Cross Correlation Grid is flexible (e.g. master density, templates, number of stations, thresholds, etc.) to fulfill various tasks including effective monitoring of UNEs.

GLOBAL GRIDCROSS CORRELATION

SOURCES AND WAVEFORM TEMPLATES

SYNTHETICSNUCLEAR EXPLOSIONSEARTHQUAKES

Grand masters

a b c

INTERNATIONAL MONITORING SYSTEM

CC

STA

LTA

SNR>3.0

LOCATION

A segment of Global Grid. Spacing between grid points (masters) ~140 km. P-wave templates from three to ten IMS primary arrays per

master. Distance for P-phase from 6 to 90 degrees. At least three IMS stations are needed to create an REB event.

Global Grid of Master Events is designed for finding and

location of seismic events based on cross-correlation (CC). The

whole globe is subdivided uniformly by cells surrounding the

grid points The IMS array stations consider the hypotheses of

seismic event occurrence within these cells based on matched

filter detection with the pre-established Master Event record

(template). The template is a set of certain data, including array

multi-channel waveform, azimuth and slowness estimations, and

magnitude. As the matched filter detection threshold is exceeded

triggered by the observed array seismogram, the multi-stage

location procedure is started. Adequate template is one of the key

points of the location procedure.

Synthetic waveforms calculated for explosion mechanism with varying depth (H=0.1, 0.3,

0.6, 1.0, 2.0 km), yield (fc= 0.8 Hz to 4.8 Hz), and distances (Δ =30º, 45º, 60º, 90º)

Waveforms (upper panel) and cross correlation

(lower panel) matrix of 100 by 100 waveforms

from the studied UNEs

-80

-60

-40

-20

0

20

40

60

80

-80 -60 -40 -20 0 20 40lat,

deg

lon,deg

AllM…

931 REB events with three and more IMS array stations. All

pairs of events are cross correlated at available stations.

CC=CCi/N

For purposes of signal association and event location, each

grid point is extended by a subnet with five circles of nodes

spaces by ~25 km. Coordinates of all nodes for all masters are

fixed and saved in the database. For each node, the

master/station travel times are corrected for the distance

between the global grid point and the node. These corrected

travel times are used in arrival association by origin times. All

nodes are processed separately but use the same set of arrivals

obtained by cross correlation. The node with the largest

number of associated stations and the lowermost scattering

(RMS) of origin times wins and saved as an event hypothesis.

Other hypotheses rejected. The fifths circle is ~100 km in

radius (i.e. very close to the grid points of adjacent masters)

and all winning hypotheses obtained at this circle are rejected

in further conflict resolution because they must be found by

the adjacent masters in the first place

TEMPLATE AND CONTINUOS WAVEFORM CROSS CORRELATION COEFFICIENT

AND STA/LTA DETECTORDETECTION THRESHOLD

Cross correlation coefficients between 931

events, averaged over detecting stations

DPRK 2013

Global Grid: PROCESSING

The primary purpose of master event processing is to detect all signals

from the relatively small footprints around a dense and continuous

grid of master events called Cross Correlation Global Grid (XGG).

Detection with master events is based on empirical waveforms from

real seismic events representing the masters. We use synthetic

templates where real ones are not available.

Mathematically, the process of signal extraction with cross

correlation using waveform templates is equivalent to matched filter

processing. This is the optimal linear filter maximizing the SNR in the

presence of microseismic noise. Therefore, cross correlation

processing guarantees the weakest possible signals to be detected. The

gain depends on the difference between waveform shapes. For

seismology, the most important reasons of the difference between two

signals are as follows: the spacing between the slave and master

events (both epicentral and depth), their source functions, velocity

structure, and attenuation.

In the current configuration of the IMS, each master

contains from three to ten templates obtained at IMS array stations.

The arrays stations of the IMS enhance the performance of cross

correlation detector and improve its spatial resolution. All advantages

of cross correlation over the current version of network processing

root in the much lower detection thresholds and the proximity of slave

events to the master events. The proximity of events in CC processing

excludes the necessity of network processing, which is designated to

associate arrivals detected and identified. Each master produces an

independent list of detections, create event hypotheses and associate

these detections with these hypotheses, locates the hypotheses

qualified according to the set of IDC event definition criteria

(EDC), and estimates the size of found events relative to the master

event. For a given time interval, the lists obtained by individual

masters are compared and all conflicts between events containing

similar arrivals are resolved in favour of larger events (first number of

associated stations and then scattering of origin times).

Briefly, processing consists of the following steps:

1. Develop homogeneous grid of master events for cross correlation.

2. Populate the XGG with real and synthetic waveform templates.

3. For a given master, process continuous waveforms - convolving

with a conjugated time-reversed version of the template. Use various

frequency bands with different time windows.

4. For a given master, detect signals by appropriate procedure (CC<

STA/LTA, etc.)

5. For a given master, characterize detected signals with defining

parameters: arrival time, azres, slores, RM, semblance, standard for

IDC, etc.

6. For a given master, filter detection list with predefined threshold

for parameters. Only qualified detections are retained.

7. For a given master, create best event hypotheses using origin times

estimated for all nodes of location subnet. The origin times are

obtained by subtraction the master/station travel times from the

relevant arrival times.

8. Resolve conflicts between events having similar arrivals (arrival

time, azimuth, slowness, RM) at the same stations. The largest (nsta)

event with the smallest origin time scattering wins. Other events are

rejected and their arrivals different from those conflicting are used in

further processing.

9. Apply classification based on previous experience (REB or XSEL).

In the cross correlation global grid approach, a grid of master events

covers the entire surface of the earth. (Depth zones at which S/H/I

events are known historically to occur can be also covered but they are

not needed for monitoring). For each grid point, there exists at least

one master event with template waveforms, which is searched for

slave events by examining all cross correlation arrivals at the related

IMS array stations. If three or more stations have phases with origin

times (i.e. arrival times less master/station travel time) within a

predefined time interval (say, 6 s), the event has a higher probability

of being real. This is a preliminary event hypothesis which may

compete with adjacent hypotheses when sharing the same arrival.

Cross correlation

location obtained

with synthetic

templates