g i c a l l a b o r atory m o l o in the region of the ...€¦ · references b e r k e l e y s e i...
TRANSCRIPT
−80 −40 0 40 800
20
40
60
80
100
Time in sec
Varia
nce
redu
ctio
n (%
)
-128 -126 -124 -122
40
42
-128 -126 -124 -122
40
42
ARC
HOPS
JCCWD
YBH
HUMO
GASB
OT - 60 sec
-128 -126 -124 -122
40
42
-128 -126 -124 -122
40
42
ARC
HOPS
JCCWD
YBH
HUMO
GASB
OT
-128 -126 -124 -122
40
42
-128 -126 -124 -122
40
42
ARC
HOPS
JCCWD
YBH
HUMO
GASB
Best VROT + 14 sec
-128 -126 -124 -122
40
42
-128 -126 -124 -122
40
42
ARC
HOPS
JCCWD
YBH
HUMO
GASB
OT + 60 sec
YBH
HUMO
WDC
ORV
-128 -126 -124 -122
40
42
-128 -126 -124 -122
40
42
ARC
HOPS
JCCWD
YBH
HUMO
GASB
OT - 60 sec
-128 -126 -124 -122
40
42
-128 -126 -124 -122
40
42
ARC
HOPS
JCCWD
YBH
HUMO
GASB
OT
-128 -126 -124 -122
40
42
-128 -126 -124 -122
40
42
ARC
HOPS
JCCWD
YBH
HUMO
GASB
Best VROT + 2 sec
-128 -126 -124 -122
40
42
-128 -126 -124 -122
40
42
ARC
HOPS
JCCWD
YBH
HUMO
GASB
OT + 60 sec
WDC
YBH
ORV
HUMO
WDC
YBH
HUMO
ORV
-126 -124 -122
40
42
-126 -124 -122
40
42
ARC
HOPS
JCC
ORV
WDC
YBH
HUMO
GASBOT - 60 sec-126 -124 -122
40
42
-126 -124 -122
40
42
ARC
HOPS
JCC
ORV
WDC
YBH
HUMO
GASBOT
-126 -124 -122
40
42
-126 -124 -122
40
42
ARC
HOPS
JCC
ORV
WDC
YBH
HUMO
GASB
OT + 4 secBest VR
-126 -124 -122
40
42
-126 -124 -122
40
42
ARC
HOPS
JCC
ORV
WDC
YBH
HUMO
GASBOT + 60 sec
WDC
YBH
ORV
HUMO
-126 -124 -122
40
42
-126 -124 -122
40
42
ARC
HOPS
JCC
ORV
WDC
YBH
HUMO
GASBOT - 60 sec
-126 -124 -122
40
42
-126 -124 -122
40
42
ARC
HOPS
JCC
ORV
WDC
YBH
HUMO
GASBOT-126 -124 -122
40
42
-126 -124 -122
40
42
ARC
HOPS
JCC
ORV
WDC
YBH
HUMO
GASB
OT + 60 sec-126 -124 -122
40
42
-126 -124 -122
40
42
ARC
HOPS
JCC
ORV
WDC
YBH
HUMO
GASB
80
Best VROT + 6 sec
WDC
YBH
ORV
HUMO
−80 −40 0 40 800
20
40
60
80
100
Time in sec
Varia
nce
redu
ctio
n (%
)
Magnitude 4.22008/07/30
Magnitude 5.02006/07/19
Magnitude 5.42008/04/30
Magnitude 6.72005/06/17
Magnitude 7.02005/06/15
-126 -124 -122
40
42
-126 -124 -122
40
42
ARC
HOPS
JCC
ORV
WDC
YBH
HUMO
GASB
25.04
OT - 60 sec-126 -124 -122
40
42
-126 -124 -122
40
42
ARC
HOPS
JCC
ORV
WDC
YBH
HUMO
GASBOT
-126 -124 -122
40
42
-126 -124 -122
40
42
ARC
HOPS
JCC
ORV
WDC
YBH
HUMO
GASB
79.97
Best VROT + 4 sec
-126 -124 -122
40
42
-126 -124 -122
40
42
ARC
HOPS
JCC
ORV
WDC
YBH
HUMO
GASB
25.76
OT + 60 sec
−80 −40 0 40 800
20
40
60
80
100
Time in sec
Varia
nce
redu
ctio
n (%
)
78.3
16.168.28.8
31.8 67.9 69.7
78.471.713.6
4.3 60.4 76.2 11.2
−80 −40 0 40 800
20
40
60
80
100
Time in sec
Varia
nce
redu
ctio
n (%
)
−80 −40 0 40 800
20
40
60
80
100
Time in sec
Varia
nce
redu
ctio
n (%
)
6.9
4.4
Continuous Seismic Scanning in the Region of the Mendocino Triple Junction, California
Aurélie Guilhem and Douglas S. Dreger Berkeley Seismological Laboratory, UC Berkeley [email protected]
Abstract The Mendocino Triple Junction (MTJ) region is tectonically complex resulting in a vari-ety of anomalous seismic events that include repeating earthquakes, slow/low-stress-drop earthquakes, and non-volcanic tremors, in addition to typical inter- and intra-plate seismic ac-tivity. The MTJ is the most seismically active region of Northern California with the seismic-ity extending to 40 km depth and including potential large earthquakes (M7+) on the Mendo-cino fault and intra-plate events occurring in the offshore Gorda/Juan-de-Fuca plates (Guilhem et al., 2007). In order to more effectively monitor the offshore region, particularly for slow/low-stress-drop or large possibly tsunamigenic events, we implement an automatic scanning of the continuous long-period (> 10 sec) broadband seismic records to detect, locate and determine moment tensors of events using the method developed by Kawakatsu (1998), independent of current operational applications. The method works by continuously perform-ing the cross-correlation of Green’s functions computed with suitable velocity models and every 2 seconds inverting for the seismic moment tensor for sources distributed over a spatial grid. When a defined level of fit is reported the algorithm has detected, located, and deter-mined the scalar seismic moment and focal mechanism. We plan on inverting for full moment tensors including the isotropic component of the source. Ford et al (2009) demonstrated that regional full moment tensor inversion (deviatoric and isotropic) was successful in identifying isotropic events in western United States. Because the analysis is done continuously every two seconds it offers the possibility of rapidly identifying large damaging and potentially tsu-namigenic events rapidly as the latency is only the time that it takes for the complete low fre-quency wavefield including surface waves to propagate across the network. For stations lo-cated in the 100-300 km distance range it should be possible to detect, locate and determine moment tensors of events ~ 5 minutes after their occurrence. We will present the design per-formance of the method for the Mendocino region.
Method
Preliminary tests Improvements of the method
Conclusions
References
Berk
eley
Sei
smolo
gical Laboratory Moment Tensor Project
www.ncedc.org/ncedc/mt
.htm
l
Figure 2: Preliminary tests for five known earthquakes. For each event we plot the mechanisms obtained with the continuous scanning in four different time windows starting: 60 sec before the catalog origin time, at the catalog origin time, at the time of the best variance reduction (VR), and 60 sec after the origin time respectively. The mechanisms are colored based on the VRs. The best solution for the four windows is plotted in red with the value of the VR. The black beach ball diagram represents the catalog solution. The waveforms (velocity seismograms filtered between 20 and 50 sec) used in the inversion are shown below the corresponding maps. Also we show the variation of the best fit versus time.
40-49.9 % 50-59.9 % 60-69.9 % 70-79.9 % > 80 %
VarianceReduction
(VR)
Event Catalog Time Lat Lon Depth (km) Mw Strike1 Dip1 Rake1
Horizontalmisfit (km)
Verticalmisfit(km)
Magnitude misfit
Time misfit
(s)Berkeley MT
catalog 13:39:34 40.425 -125.33 18 4.2 96 82 169
Our solution 13:39:38 40.6 -125.2 26 4.2 101 84 162 22.5 8 0 4Berkeley MT
catalog 11:41:43 40.281 -124.43 30 5 102 83 173
Our solution 11:41:49 40.4 -124.2 32 4.9 103 86 174 23.2 2 0.1 6Berkeley MT
catalog 3:03:07 40.836 -123.5 30 5.4 359 60 -88
Our solution 3:03:11 40.8 -123.4 32 5.3 6 59 -89 9.1 2 0.1 4Berkeley MT
catalog 6:21:42 40.758 -126.6 21 6.7 105 80 170
Our solution 6:21:44 40.8 -126.6 38 6.7 106 84 151 4.7 17 0 2Berkeley MT
catalog 2:50:54 41.328 -125.87 21 7 225 85 13
Our solution 2:51:08 41.4 -125.8 38 7 229 76 23 9.8 17 0 14
6/17/05
6/15/05
7/30/08
7/19/06
4/30/08
Table 1: Summary table of the results for the test eventsThe table compares the results (location, magnitude, time and mechanism) obtained with our continuous scanning with the Berkeley Moment Tensor Catalog (http://seismo.berkeley.edu/~mike/solutions.new) for the five earthquakes considered in our preliminary tests.
We perform preliminary tests of the continuous seismic scanning using different magnitude earthquakes (from 4.2 to 7.0) located in the study region (Figure 2). We center the grid search around the time of the events (about 90 sec before the origin time and 60 sec after). In order to increase the computational timing of the analy-ses, we reduce the grid (blue boxes in Figure 2). For the five earthquakes we are able to retrieve their location and time as well as their mechanism and magnitude (Figure 2 and Table 1) by searching for the maximum variance reduction (between 69 and 80 %) between data and synthetics in the moment tensor inversion (380 points).
We perform a similar test on seismic noise. The best variance re-duction for this test is 4.8 %. We also try the technique on a large teleseism (the recent M7.6 Tonga earthquake) and our best re-sults are less than 20 % for the variance reduction, which indi-cates that no false alarm would have been triggered by this event using a threshold of more than 65% variance reduction (Figure 3).
Figure 1: Map of the study region.The seismicity from the ANSS catalog between 2000 and 2009 is shown by the gray dots and the broadband seismic stations of the Berkeley network by the black squares. The red and purple stars locate the slow/low-stress-drop and repeating earthquakes, re-spectively. The grid (0.2deg latitude, 0.2deg longitude and 3 km depth) used in the study is shown by the small blue circles.
Ford, S. R., D. S. Dreger, and W. R. Walter (2009), Identifying isotropic events using a regional moment tensor inversion, J. Geophys. Res., 114, B01306, doi:10.1029/2008JB005743.Guilhem, A., D.S. Dreger, and R.M. Nadeau (2007). Scanning of Unusual Seismic Activity in the Mendocino Triple Junction Region, AGU Fall Meeting.Kawakatsu, H. (1998). On the Realtime Monitoring of the Long-Period Seismic Wavefield, Bull. Earthquake Res. Inst., 73, 267-274.Tajima, F., C. Megnin, D.S. Dreger, and B. Romanowicz (2002). Feasibility of Real-Time Broadband Wave-form Inversion for Simultaneous Moment Tensor and Centroid Location Determination, Bull. Seism. Soc. Am., 92, 739-750.Tsuruoka, H., H. Kawakatsu, and T. Urabe (2008), GRiD MT (Grid-based Realtime Determination of Moment Tensors) monitoring the long-period seismic wavefield, Phys. Earth Planet. Int., Special issue: Earthquakes in subduction zones: A multidisciplinary approach , in press
- Optimize the code in order to perform the moment tensor calculations in real-time (Tsuruoka et al., 2008)- Define a VR threshold for the automated detection (probably around 65 %)- Perform more sensitivity tests in order to create a statistical study of the tech-nique for the Mendocino region- Try to improve depth determination by testing other velocity models- Combination of single point-source and multi point-source inversions for the detection of small events to great, potentially tsunamigenic, earthquakes (Figures 2 and 4).
Gisk (t)mi
s = dk (t)i
∑
A jis mi
s = b js
i∑
A jis = A ji
sk
k∑ A ji
sk = G jsk (t)Gi
sk (t)dt∫ bjs = b j
sk
k∑ bj
sk = G jsk (t)dk (t)dt∫
ˆ m = A−1b
Re s = dk (t) − A(t) ˆ m ( )2dt∫
k∑
VR = 1− Re s ÷ dk (t)( )2dt∫
k∑
⎡
⎣ ⎢
⎤
⎦ ⎥ ×100
We follow the algorithm presented by Kawakatsu (1998). One can represent a seismogram d recorded at a station k as the convolution of the response of the medium to an impulsive source G with the moment tensor elements m for a source s: (1)
The normal equation based on (1) is given by:
where: and
A is a square matrix. The least-squares solution for the moment tensor is given by:
The moment tensor vector m can be obtained by a simple matrix multiplication once b is calculated.Then we can obtain the residual between data and synthetic seismo-grams:
And the variance reduction (VR):
evaluates the fit between the data and synthetics for each point of a grid.
- Grid search between:latitude: 40.0 and 43.0 (0.2 degree interval)
longitude: -123 and -128 (0.2 degree interval)
depth: 5 to 38 km (3 km interval)
- One catalog of Green’s functions (Gil7_ model)- Waveforms filtered between 20 and 50 sec (East, North and verti-cal components)- Inversion of 380 points
We show that:- based on the few tests the method is very promising. We obtain very good results in terms of location (latitude, longitude and depth), magnitude, mechanism and time.- the implementation of a continuous scanning of seismic waveforms filtered at low frequency is feasible in Northern California (Tajima et al., 2002).- such scanning will provide complete information on the events in realtime using a single stage of processing, and for this reason it will be faster than the current procedures.- the realtime monitoring may also be useful in the detection of large tsunamigenic earthquakes located along the Cascadia subduction zone.- and finally such technique may also help in the search for unusual seismic activity in the Mendocino Triple Junction and in particular for the slow/low-stress-drop earthquakes (Guilhem et al, 2007).
-128
-128
-126
-126
-124
-124
-122
-122
40 40
42 42
44 44
HOPS
YBH
HUMO
GASB
HOPS
JCCWDC
GASB ORVFigure 3: Test for the M7.6 Tonga earthquake (03/19/2009). The best variance reduction for this test is less than 20 %. The test was performed fixing the depth at 8, 20 and 32 km and for the entire grid, between 500 and 2700 sec (in red). We show the vertical seismograms of this event for the four stations, filtered between 20 and 50 sec.
Figure 4: Synthetic test for a M8.2 re-verse earthquake along the Cascadia subduction zone.Synthetic waveform data was simulated for the four stations assuming a 250 by 80 km square fault with uniform slip yielding a total moment of 2.24e28 dyne cm (Mw 8.2). A constant rupture velocity of 3 km/s and a constant rise time of 5 seconds were assumed. The data were filtered with a causal Butterworth bandpass with corners of 200 and 100 seconds period. The same bandpass filter was applied to the synthetics, which were also convolved with a 80 second source time function. 480 seconds of data were inverted beginning 50 seconds before the origin time to 50 seconds afterward and the best variance reduction for each grid node was determined. The maps show the best variance reduction assuming source depths of 11 and 17 km. In both cases an oblique reverse mechanism is obtained with Mw of 8.0-8.1. The maps also show that a broad region is highlighted with elevated variance reduction due to the finite nature of the source. We are working on developing an algorithm that incorporates some combination of single point-source, multi-point-source inversions together with the spatial distribu-tion of the goodness of fit parameter to identify great potentially tsunamigenic great earthquakes.
11 km depth
17 km depth
VR = 54.6 %
VR = 52.8 %
Origin Time (OT) - 60 sec Origin Time Origin Time (OT) + 60 secBest VR
HUMO
WDC
ORV
YBH
Time (sec)Ve
loci
ty (c
m/s
)Va
rianc
ere
duct
ion
(%)
02468
101214161820
8 km20 km32 km
-128 -126 -124 -122
40
42
44 INPUT-FF INVERSIONINPUT-FF 20-50s
0 10 20 30 40 50 60 Variance Reduction (%)
40
42
44 INPUT-FF INVERSIONINVERSION
-128 -126 -124 -122
Using 20-50 sec, a similar test has been performed and the best variance reduction was 41% two minutes after the occurrence time (brown beach ball diagram with Mw = 6.7).