JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 106,NO. B8, PAGES 16,323-16,337, AUGUST 10, 2001
Seismic
Prieta
waveform
attributes
before
and after
the Loma
earthquake'
Scattering change near the earthquake and temporal recovery Stefan Baisch
Institut fiir Geophysik,Ruhr-Universit/itBochum,Bochum,Germany G6tz
H. R. Bokelmann
Departmentof Geophysics, StanfordUniversity,Stanford,California
Abstract.
During 1987-1995severalclustersof nearly identicalseismicevents
(multipiers) occurred nearthe LomaPrietasource region.Thesemultipiersallow us to investigateand demonstratespatialand temporalchangesin seismicwave character associated with the 1989 Loma Prieta main shock. For seismogram pairs
we usea movingwindowtechniqueto computecoherencies dependingon lapsetime and frequency.Post-LomaPrieta eventshavereducedcoherencies with pre-Loma Prieta eventsin a spatiallylimitedregioncloseto the LomaPrieta hypocenter,while other pathsremainnearlyunaffected.Thesechangesgraduallyrecoverwithin a time intervalof 5 yearsafterthe LomaPrietaearthquake.A possible explanationfor the time dependence is coseismically openedcrackswhichcausescatteringincrease for wavefields after the Loma Prieta event. Postseismicrelaxation processessuch
ascrackhealing,fluiddiffusion, or afterdeformations leadto progressive closure of these crackswith time after the main shock. Thus the scatteringpropertiesof the local crust approachthe pre-main shockstate.
1.
Introduction
Aside from the 1906 San Francisco earthquake, the M=6.9 Loma Prieta quake was the largest and most
to demonstrate such changesby studying attenuation (coda Q) and seismicwave velocities. For this purpose, a number of active experiments were conducted to resolve stress induced velocity changes
damagingquake to strike an American urban area in [e.g., Eisler, 1967, 1969; Aki et al., 1970; DeFazio et this century[Holzer,1999].Althoughdisastrous from a al., 1973] or to monitor temporal changesof the vehazard point of view, Loma Prieta providesan excellent locity field associatedwith big earthquakes[e.g.,Karaopportunity for improvingour understandingof all asgeorgiet al., 1992; Liet al., 1998]. Besidesrepeated pectsof large earthquakessinceit occurredin a densely active experimentswhich are too weak to illuminate the instrumentedregion. In the past 10 yearsmany studies whole crust, researchersconsideredearthquake sources of Loma Prieta have already given a rather detailed picwhich samplethe crust at different times. A fundamenture of the earthquake[e.g.,Spudich,1996;Reasenberg, tal problem with this approach comesfrom the lack in 1997; Simpson,1994]. Nevertheless,there are importhe repeatability of the source. This led to controvertant open questionsrelated to rheologicalpropertiesof the fault zone and the nature of after event deforma-
tion. These questionsmay in principle be addressedby studyingseismicwaves,which carry informationabout the sourceregion and the medium they passthrough. Especially interesting is the study of changesin the characterof seismicwaveswith time (monthsto years) sincethesegive direct cluesto changesof material behaviour with time. In the past, researchersattempted Copyright 2001 by the American GeophysicalUnion.
sies as to whether or not reported velocity changesare
real [e.g., Wessonet al., 1977].Prominentexamplesare studiesof precursoryphenomenain P and S waveveloc-
ities [e.g.,Kanamori and Fuis, 1976]. Other groupsinspectedcoda attenuation for changeswith time. While
Aki and Chouet[1975]suggestthat coda Q showstime variations due to subsurfacemedium changes, other studies indicated that coda Q may not be a stable esti-
mator of attenuation[seeSato, 1988; Got et al., 1990; Got and Fr•chet, 1993].
Paper number 2001JB000151.
More recently, a number of studiesdid not find changes in coda Q attenuation but were able to give upper
0148-0227/01/2001JB000151 $09.00
bounds[Hellweget al., 1995;Antolik et al., 1996;Aster 16,323
16,324
BAISCH
AND BOKELMANN:
CHANGE
OF SEISMIC
WAVEFORM
ATTRIBUTES
10/19/89 10/26/89 10/30/89
11/4/89.•,•
11/9/89
12/29/89 9/18/90 10/29/91
1/6/95 i
i
i
0
5
10
15
time after P [s]
Figure 1. Waveforms of multipletp2 recorded at stationHFP southeast of the rupturezone
(Figure3). Eventdatesaremarked ontheleft. Dataarealigned to the? onset.Notethehigh waveform similarity t,hroughout theentire15stimewindow. et al., 1996]. For the LomaPrieta after shocksan upper different times we can study the temporal evolution of boundwas givenas 5% [Berozaet al., 1995]. waveform similarity, i.e., its gradual recovery after the The existenceof temporal variations in a crustal setting is clearly controversial. That is quite different in
Loma Prieta earthquake.
the smaller-scale problem of reservoir geophysics,where 2. the variation of properties with time is well established
and of considerablepractical importance[Jack, 1998]. In the whole-crust setting, there is a need for showing unambiguousevidencefor temporal changesin seismic properties. That is the focus of the current paper, which presentsexamplesof subsurfacechangesafter Loma
Prieta.
These are documented
in waveform
changesof "similar" events. The technique we use is based on waveform similarities rather than the coda Q technique becausewe find that the latter is subject to a number of rather restrictive assumptions,which are not strictly necessaryfor showingthe existenceof temporal changes. The critical ingredient of our techniqueis the use of
Data
A commonfeatureof seismicity alongmajor faultsis the occurenceof spatially tightly clusteredmicroearthquakes. These clusters often represent stressrelease at the sameasperity,or stressconcentration,alonga
fault surface[GellerandMueller,1980].If all member eventsof suchclustersare approximatelycolocatedand havethe samesourceradiationpattern, the generated waveformsare nearly identical for the entire cluster. In practice, however, each event in a cluster has its own sourcevolume, and it is difficult to decide whether these
volumesare perfectlycolocated.FollowingGeller and Mueller[1980],we take the A/4 criterion,i.e., all events must be separatedby lessthan a quarterwavelength, setsof similarevents[e.g.,Aster et al., 19901,whichwe to definea clusterthat we will call a seismicmultipier will call seismic multiplets. This allows good control of source effects. We perform a multi station analysis which clearly restricts the observedsimilarity decrease to a spatially limited region, thus showingthe robustnessof the technique against data contaminationssuch as potential source variations. From successivecombination of repeating earthquakes sampling the crust at
hereinafter.
At the southernend of the Loma Prieta rupturezone and in the vicinity of the Morgan Hill epicenteron the CalaverasFault, severalmultipierswith intereventseparation of not more than a few tens of meters have been
identified(D. Schaff,personalcommunication, 1999). Thesemultipiersconsistof up to 20 eventswhichpro-
BAISCH
AND BOKELMANN:
CHANGE
OF SEISMIC
WAVEFORM
I
I
• m5 -00
0
Loma Prieta o
•0000
p4
ß• p3
0 0
ß D O0
p2
lED
pl
•DEE•CO0
1984
1986
1988
16,325
I
o
0
ATTRIBUTES
1990
0
0
0
O0
0
0
0
0
0
O0
1992
0
0
0
0
1994
0
1996
year
Figure 2. Time linesof five multipierssamplingthe Loma Prieta earthquake.Event times are shownby circles;the verticalline indicatesthe time of the Loma Prieta earthquake.Multiplets pl-p4 are located at the southernend of the Loma Prieta rupture zone, and multipier m5 is locatedon the CalaverasFault (markedby arrowsin Figure3).
duce nearly identical waveforms. Figure 1 shows an example of the waveforms of multiplet p2 recorded at station
HFP.
The waveforms
are raw data as distributed
by the Northern California Earthquake Data Center
stations to avoid complications from differing instru-
ment types. The short-periodstations (Figure 3) are equipped with Mark L4-C instruments and were operated at a sampling frequency of 100 Hz.
(NCEDC) [Neuhauseret al., 1994]. The five multiplets we used for our analysis sample the Loma Prieta earthquake in time; that is, at least one of their events occurred before the main shock, and
3. Spatial Similarity
Variations
in Doublet
several
event members occurred after the main shock A simple approachwhich clearly demonstrateschan(Figure 2 and Table 1). We will referto theseeventsas gesof crustal propertiesnear the Loma Prieta rupture preshock and aftershocks,respectively. zone is based on seismogramcrosscorrelationsin the
In this paper we focus on the 13-element multipier p2. In order to keep a fixed station network for our investigations we chose the subset of the Northern California
time domain. Figure 4 shows cross-correlation'coefficients between the preshockand the first aftershock
ter than for the other multipiers. We aimed to include as many stations as possible and searched the entire N CSN for stations that satisfy our selection criteria.
rupture zone of the Loma Prieta earthquake, systematically show reduced cross-correlationvalues, whereas
of multiplet p2 (seeTable 1). At each station, cross SeismicNetwork (NCSN) stationsfor whichmostof the correlations are computed in a 15 s time winddw startrecordingspassour selectioncriteria (describedbelow). ing at the P onset. The resulting pattern is clearly The station net used in our analysisis defined on basis divided into two regions. Stations to the northwest, of the station coverageof multiplet p2 which is bet- for which the direct wavestraveled mainly through the
In addition to an excellentsignal-to-noiseratio (SNR)
stations in other azimuths
show correlation
values above
0.9. Basically, the same pattern is reproduced using we required that no alterations of the instrumentation preshock-aftershockcombinationsof the multiplets pl, havebeen undertakenbetweenthe preshockand the af- p3, and p4, which are located closeto p2 (Figure 3). tershocks. This criterion excludes stations which have In principle, severalfactors can explain locally reduced been convertedfrom analogto digital as well as those cross-correlationvalues. Most critical are changesin for which the gain ranginghas been changed. We did source parameters, but these can be excluded by the not use the data from a number of available broadband pattern resulting from the Calaveras Fault multiplet.
16,326
BAISCH AND BOKELMANN:
CHANGE OF SEISMIC WAVEFORM ATTRIBUTES
Table1. Origin Times, Northern California Seismic Network Locations, andCodaMagnitudes ofEvents Used
in This Study a
MultipierNumberYear Month Day Time, LT Latitude, øN Longitude, øE Depth Magnitude pl pl pl pl pl pl pl pl pl pl pl pl pl pl pl pl pl pl pl pl p2 p2 p2 p2 p2
1 2 3 4 5 (6) (7) 8 9 10 11 12 (13) 14 15 16 (17) 18 19 20 I 2 3 4 5
1986 1989 1989 1989 1989 1989 1990 1990 1990 1990 1990 1990 1990 1991 1991 1991 1992 1993 1994 1995 1987 1989 1989 1989 1989
6 10 10 ll 11 12 1 3 4 5 7 9 12 4 8 11 7 6 5 5 3 10 10 10 10
28 20 28 5 12 12 13 20 28 23 5 2 6 2 16 25 16 22 12 17 31 18 19 26 30
064304 171953 112337 041144 124322 215443 003411 195755 070941 152354 023007 190742 205752 060214 201638 020150 050228 101107 020247 041852 061920 231153 173631 024756 214955
36.94 36.94 36.95 36.94 36.94 36.94 36.95 36.94 36.94 36.94 36.94 36.94 36.94 36.94 36.95 36.94 36.94 36.94 36.94 36.94 36.94 36.94 36.94 36.94 36.94
-121.68 -121.68 -121.68 -121.68 -121.68 -121.68 -121.68 -121.68 -121.68 -121.68 -121.68 -121.69 -121.68 -121.68 -121.68 -121.68 -121.68 -121.68 -121.68 -121.69 -121.68 -121.68 -121.68 -121.68 -121.68
10.6 10.9 10.9 10.3 10.8 10.6 10.9 10.5 10.4 10.7 10.2 10.8 11.4 11.1 10.3 10.4 11.0 10.8 11.1 10.8 10.2 10.0 9.9 10.1 10.1
1.1 1.2 1.0 1.1 1.3 1.4 1.5 1.3 1.1 1.1 1.2 1.2 1.5 1.3 1.1 0.9 1.4 0.9 1.2 1.1 1.8 1.6 1.5 1.8 1.6
p2 p2 p2 p2 p2 p2 p2 p2 p3 p3 p3 p3 p3 p3 p3 p3 p3 p3 p3 p3 p3 p3 p4 p4 p4 p4 p4 p4 p4 p4 p4
6 7 8 9 10 11 12 13 I 2 3 4 5 6 7 8 9 10 11 12 13 14 I (2) (3) (4) 5 (6) 7 8 9
1989 1989 1989 1990 1990 1990 1991 1995 1987 1989 1989 1989 1989 1989 1989 1990 1990 1990 1991 1992 1993 1994 1989 1989 1989 1989 1989 1989 1989 1990 1990
11 11 12 2 6 9 10 I 6 10 10 10 10 11 11 I 4 6 4 4 4 12 5 10 10 10 11 11 11 2 4
4 9 29 18 28 18 29 6 22 19 21 23 26 4 18 7 21 28 20 21 11 25 11 22 24 28 8 12 19 6 12
021207 073334 213458 025056 072811 010310 233158 162811 195509 152056 101338 120621 024334 224941 180805 210053 144455 073020 235911 091729 064610 224816 005704 214831 182533 234545 010605 195649 211405 005025 053555
36.94 36.94 36.94 36.94 36.94 36.94 36.94 36.94 36.94 36.94 36.94 36.94 36.94 36.94 36.94 36.94 36.94 36.94 36.94 36.94 36.94 36.94 36.94 36.94 36.94 36.94 36.94 36.94 36.94 36.94 36.94
-121.68 -121.68 -121.68 -121.68 -121.68 -121.68 -121.68 -121.68 -121.68 -121.68 -121.68 -121.68 -121.68 -121.68 -121.68 -121.68 -121.68 -121.68 -121.68 -121.68 -121.68 -121.68 -121.68 -121.68 -121.68 -121.68 -121.68 -121.68 -121.68 -121.68 -121.68
9.7 10.2 10.4 10.1 10.2 10.1 10.1 9.6 10.2 10.5 10.4 9.7 9.8 9.8 10.1 10.4 10.4 9.9 10.5 10.4 10.6 10.4 11.8 12.0 11.8 11.9 11.9 11.3 11.8 11.8 11.9
1.4 1.5 1.9 1.8 1.9 1.7 1.9 2.0 1.7 1.8 2.0 1.7 1.7 1.5 1.8 1.8 2.1 1.7 1.6 1.8 1.7 1.7 1.6 2.3 2.1 1.0 1.5 1.0 1.5 1.6 1.8
8
19
193306
p4
(10) 1990
p4 p4 p4 p4 m5
(12) 13 (14) 15 (1)
p4
11 1990
1990 1991 1991 1992 1984
6
11 3 7 11 4
4
19 4 28 14 25
200618 064803 193206 081236 060854 005201
36.94
36.94
36.94 36.94 36.94 36.94 37.24
-121.68
-121.68
-121.68 -121.68 -121.68 -121.68 -121.63
11.7
1.3
12.1 12.2 12.6 12.5 4.8
0.9 1.8 1.0 1.5 2.6
12.3
1.5
BAISCH AND BOKELMANN'
CHANGE
OF SEISMIC WAVEFORM
ATTRIBUTES
16,327
Table 1. (continued)
MultipierNumberYear Month Day Time,LT Latitude, øN Longitude, øE Depth Magnitude// height m5 m5 m5 m5 m5
2 3 4 5 6
1984 1985 1987 1990 1993
8 4 2 I ll
18 11 26 7 9
115428 103459 144436 150309 191734
37.24 37.24 37.24 37.24 37.24
-121.63 -121.63 -121.63 -121.63 -121.63
4.9 5.2 5.5 5.3 5.1
2.3 2.3 2.3 2.2 2.2
• We useonlythoseeventswhicharesimilarin magnitude.All eventsnot usedfor ouranalysis but usedby Berozaet al.
[1997],Schaffet al. [1998],andD. SchaffandG. C. Beroza(manuscript in preparation, 2000)areshown in parentheses.
can be definedas [e.g.,Jenkinsand Watts, 1968,chap. servedfor the m5 multiplet (Figure3) only for stations 8 and 9]
As will be shownbelow, a similarity decreasecan be ob-
(1)
closeto the Loma Prieta rupture zone. From the limited spatial pattern combinedwith the redundant observations we can also rule out contamination
effects such as
local changesof the backgroundnoiselevel or changes Here"/xy(Y)denotesthe smoothed crossspectrumof of water saturation
due to rainfall.
Thus we conclude
that the decrease of waveform similarity is caused by local changesof crustal properties.
To study the nature and localizationof thesechanges in more detail, we applied a moving-window analysis of seismogramsimilarity implementedin the frequency domain. The analog of the crosscorrelation in the time domain is the coherencyin the frequencydomain, which
seismograms x(t), saythe preshock,and y(t), sayone of the aftershocks,and "/xx(Y) denotesthe smoothed autospectrum of x(t). A propertreatmentof confidence intervalsfor the coherencyis presentedin Appendix A. We computedcoherencies in 2.56 s time windows, which we shift by steps of 1.28 s. As a smoothing functionwe chosea five-pointspectraloperator,defined by the Fouriertransformof a Tukey window. For this
37.6
37.4
37.2
37
36.8
AV
36.6
36.4
-122.5
-122
-121.5
-121
longitude[deg]
Figure 3. Map of the San FranciscoBay area showingthe Loma Prieta rupturezone(shaded area)as definedby the aftershock distribution(D. Schaff,personalcommunication, 1999). Two multipletsusedin this studyare locatednear the southernend of the rupturezone(p2) and on the CalaverasFault (m5). CaliforniaSeismicNetworkstationsusedin this study are shownby triangles. Solid lines showsurfacetracesof mappedfaults.
16,328
BAISCH AND BOKELMANN:
CHANGE
OF SEISMIC WAVEFORM
ATTRIBUTES correlation coefficient
37.6 0.9
37.4 0.8
37.2 __0.7
37 0.6
36.8 0.5
36.6
0.4
36.4
•0.3
-122.5
-122
-121.5
-121
longitude[deg]
Figure 4. Seismogram crosscorrelations between the preshock andthefirstaftershock of multiplet p2 (Table1). Cross-correlation coefficients are computed usingthe 15 s time window startingat the P onset.Notethat owingto theLomaPrietaearthquake, seismogram similarities arestronglydecreased in vicinityof the rupturezone,whiletheyvaryonlylittle at mostother stations.
coherency 37.4
37.3 0.8 37.2
0.7
'0 37.1
"-'
37
0.5
0.4
36.7
-122.2
0.3
-122
-121.8
-121.6
-121.4
Iongilude [deg]
Figure 5. Seismogramcoherenciesbetweenthe preshockand the first aftershockof multiplet
p2. At eachstationcoherencies are shownas a functionof lapsetime (x axis) and frequency(y axis). The time rangeis 11.52s starting 1.28 s after P. The frequencyrangeis 2-10 Hz. Seetext for more details.
BAISCH
AND BOKELMANN-
CHANGE
OF SEISMIC
WAVEFORM
ATTRIBUTES
16,329
coherency 37.4
37.3 0.8 37.2
0.7
37.1 0.6
37
0.5
0.4
36.7 -122.2
0.3 - 122
-121.8
-121.6
-121.4
Iongilude [deg]
Figure 6. Seismogramcoherencies betweenthe preshockand the first aftershockof multiplet p3. Display is as in Figure 5.
the subset of stations used in Figure 4, which have a reasonablecoverageof the eventsin the multipier. This will be especially important for the temporal aspect addressedin section4. As expectedfrom Figure 4, low cofunctionof lapsetime (x axis) and frequency(y axis). herencyvaluesmainly exist at stationsnorthwest of the The frequency range is restricted to 2-10 Hz to avoid multiplet epicenter. At some surroundingstations, mieffectsdue to changesin eventmagnitude(seeTable 1) nor decreasesof coherencyof single seismogramphases at higherfrequenciesand to reducepossiblecontamina- are resolved which were previously masked by the long tions from noise at lower frequencies. We display only time window used for calculating the crosscorrelations.
smoothingoperator the lowest limit to the coherency that may be determined is • 0.06. In Figure 5, coherenciesfor the same data example as in Figure 4 are displayed. At each station, coherenciesare shown as a
coherency 37.4
i to.
37.3
0.8
37.2
0.7
0.6
0.5 36.9
0.4
36.7 -122.2
0.3
- 122
-121.8
-121.6
-121.4
Iongilude [deg]
Figure 7. Seismogramcoherencies betweenthe preshockand the first aftershockof multiplet p l. Display is as in Figure 5.
16,330
BAISCH AND BOKELMANN. CHANGE OF SEISMIC WAVEFORM ATTRIBUTES coherency 37.4
37,3
37.2
36 .g
0.5
36.8
0.4
36,7
-122,2
- 122
-121,8
-121,6
-121,4
0,3
Iongilude[deg]
Figure 8. Seismogram coherencies between thepreshock andeventnumber 5 of multipletp4 (seeTable1). Displayis asin Figure5. At station HAZ southeast of the rupture zone for ex- region where such changesoccurred. Assumingthat ample, a clear onset of a coherency decreasecan be the changesoccur near station JBZ, we identify the coobservedstarting only .-• 6 s after the P onset. Direct herencydecreaseat HAZ as additionally backscattered
travelling phasesat earlier times were not affectedappreciably. Although we do not know the explicit travel path of the seismogramphasesin the coda, we can use the kinematic information to constrain the region where changesin wave propagation properties were caused. Low-coherencyphasesin the early P coda at JBZ and JEC suggestthat these stations are located closeto a
wavesfrom this region. The model of a changein scattering below JBZ also explains the coherencypattern observedat the other stations. For instance, from JRR to JCB the onset of scattered phasesis shifted toward the later seismogramas wouldbe expectedfrom the geometry. Differences appeared at station JSS located in
the forward scatteringdirection. Here the direct waves coherency
37.4
37.3 0.8
37.2
0.7
0.5
0.4
36.8
36,7 -122,2
-122
-121,8
-121,6
-121,4
0,3
Iongilude[deg]
Figure 9. Seismogramcoherenciesbetweentwo multiplet membersof m5 on the Calaveras
Fault (epicenter markedby arrow).EventtimesareAugust18, 1984andJanuary7, 1990for the preshockand aftershock,respectively.Displayis as in Figure 5.
BAISCH AND BOKELMANN: CHANGE OF SEISMIC WAVEFORM ATTRIBUTES
16,331
I24
254
336
742
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0. I
0
Figure 10. Coherenciesfor all preshock-aftershock coxnbinations at station HAZ . Top left showsthe waveformof the preshock.The time windowusedfor the calculationof the coherencies is indicated by the shadedarea. Coherenciesbetweenthe preshockand aftershocksare shown dependingon lapsetime (x axis)andfrequency(y axis),followingthe eventtime of the aftershocks in daysafter the Loma Prieta main shock,whichis indicatedalongthe arrows.Note that patches of anomalouslow coherencies (markedby the boxesin the plot for I day after main shock) progressivelydecreasewith increasingtime afl,er Loma Prieta.
alreadyseemto havesampledthe scatteringregionand note that the coherencydecreaseat JBZ viewed from transfetedinformationabout that regioninto their coda the CalaverasFault starts• I s later in the seismogram waves. than viewedfromthe SanAndreasFault. This mightbe
An almostidenticalcoherency patternto that of Fig-
an indication
that the scatterer is not located in the im-
ure 5 is observedfor multipier p3 (Figure 6), which is mediate vicinity of station JBZ, in which casethe onset located only a few hundred meters away from xnultiplet of scatteredphaseswould be azimuthallyindependent. p2 (D. Schaff,personalcommunication,1999). These independent observationsdemonstrate that reduced co4. Temporal herenciesindeed are due to subsurficialchangesand do Similarities
Variations
of Seismogram
not reflect seismogramcontaminations(note the different event times of the preshockand aftershocksof these
In section3 we focusedon localizing medium changes two multiplets;seeTable 1). Multiplets pl and p4 also whichhavebeengeneratedin the "LomaPrieta time inproducesimilar coherencypatterns (Figures 7 and 8), terval" between the preshock and the first aftershock of although lower magnitudes in multipier p l causeaddi- the multiplets. Now we study the temporal evolution of tional coherencydecreaseat more distant stations(e.g., thesescatterersby successivelycomparingthe preshock HFE) owingto a lower signal-to-noiseratio. to all aftershocksof multipier p2 (listed in Table 1). The coherencypattern of the Calaveras Fault doublet Figure 10 showsthese combinations at station HAZ. As shownin Figure 9 is also compatible with the assumed already demonstrated in Figure 5, a strong coherency location of scatterer. Comparing Figure 5 and 9, we decreaseoccurred for the combination of the preshock
16,332
BAISCH AND BOKELMANN'
CHANGE OF SEISMIC WAVEFORM
ATTRIBUTES
of the last two aftershocks.
with the first aftershock. This decrease is confined to
This
value is still smaller
frequencies below8 Hz and setsin at lapsetimes> 6 s. than for the combination of the preshock with the last For combinationswith subsequentaftershocksthe same aftershock,indicating that the recoveryis not complete lapsetime-frequency range(TFR) is affected,but co- within the observation time of • 5.25 years. In the next step we applied the same averaging proherencieswithin this TFR graduallyincreasewith time after Loma Prieta. The only time interval in which cedure to the other stations used in Figures 5-9. To the medium experiencedstrong changesis one which ensurea reasonabletemporal coverageof eventsat each containsthe Loma Prieta event (betweenthe preshock station, we further restricted our analysisto those staand first aftershock).The faster recoverywithin the tions where a minimum number of multipier events are first daysafter the main shockdemonstrates that the available accordingto our selectioncriteria. For mulcoherencyreductionis indeeddue to the Loma Prieta tipier p2 we chose a minimum number of l0 events. event. Sincethe thresholdtest worksonly at stationswhere coIn orderto quantifythe recoveryprocess we applieda herency values exist below the threshold, we switched simplethresholdtest to the coherencies of the preshock the threshold to a higher value of 0.985 at those staandfirst aftershock (Figure10,left, second panel).We tions where < l0 TFR data points exhibit coherency defined the TFR where coherenciesdropped below a values below 0.75. The higher threshold value is chovalue of 0.75. This marked regionis held constantand sen such that a similar distribution of affected TFR appliedto the subsequent aftershocks (following panels data points existsfor both station sets. Figure 12 maps in Figure10). Further,wedefined •- 1- •, where the resulting time signalsfor multipier p2. Although C(i) denotesthe meancoherency in the markedTFR. there is more scatter in the data at some stations, a In Figurel l, • is plottedagainsteventtime of the similar recovery relation exists at all stations near the aftershockson a semilogarithmicscale. This type of rupture zone. Even at more distant stations, especially presentationprovidesa compactway of visualizingthe in the eastern part, a slight but systematic coherency recoveryprocess. As indicated by the solid line, the increase with time can be observed, suggestingthat coherencyincreaseapproximatelyfollowsa powerlaw backscattered phasesare still measureableat these stadependence.Also shownin Figure l l is an estimateof tions. In contrast, at the northern stations CMH, CSC, theentirely recovered coherency obtained asthe• value and CAO, temporal changesof coherenciesobviously _
years after Loma Prieta ....
I
0.5 I
•
1 '
I
'
2 '
•
,I ß
•
: I
3
456
I
I
I
•
I
........ .........
0.4
._
.......... ..........
......
............ ......
.....
0.35 ...........
............. .......... ........... ..........
.
............
0.3
...................... ........... .......... ß
.....
......
......... .....
0.25
.....
0.2 0.15
0.1
0.05
-Z.. .....
.
i
10o
i
,
,
, ,,I
....
101
I
102
.....
I
103
days after Loma Prieta earthquake Figure 11. Recovery of waveformsimilarity for p2 as observedin a temporal increaseof coheren-
ciesat HAZ.Asterisks denote •(t) - 1-•(t), where•(t) isthemeancoherency in theanomalous regionof Figure 10. Time is shownwith respectto the Loma Prieta earthquake(semilogarithmic scale). Error bars indicate 95% confidenceintervals(see Appendix A for details). The solid
lineshows a fit to a powerlawdependence of theform•(t) - a t-p Thepanelonthe right showsthe C' value for the last two eventsof multiplet p2 as an estimate of the entirely recovered coherency.
BAISCH
AND BOKELMANN:
CHANGE
OF SEISMIC
WAVEFORM
ATTRIBUTES
16,333
37.4
37.3
37.2
-o 37.1
(0
37
36.9
36.8
36.7
-122.2
-122
-121.8
-121.6
- 121.4
longitude[deg]
Figure 12. Coherencyrecoveryat all stationsfor multiplet p2. Station boxesshowmean coherency dependingon time after Loma Prieta. For theseboxes,scalingon the x axis is the sameas in Figure11. On the y axisthe rangebetween0 and 0.5 is shown.Solidline indicates the best fit to a power law curve.
Compared to the other multiplets, the signal-to-noise ratio of multiplet pl events (Table 1) is poor, and we In general,stationsat greater distanceto the assumed had to excludemany noisy recordingsof this multiplet
are caused by noise effects since no systematic pattern is present.
scatterer location show smaller temporal variations of the coherencies. We interpret this as an effect of the
geometricalattenuation of the scatteredphasesmaking them lower in amplitude compared to the near-stationgenerated coda waves at greater distances. The same argument can also explain why the recovery processat the central station JBZ seemsto be lesscomplete within the observation time compared to stations at interme-
diate distances(JUC, JSS, JST, HCR, and HAZ). A
in advance.
5.
Discussion
We find a strong coherencydecreaseassociatedwith the Loma Prieta earthquake which gradually recovers with the years after the event. Different from this data-
driven approach,Beroza et al. [1995] applied a formal coda Q technique and found stable coda Q values
remarkably similar recovery signal is observedfor mul- (upper bound of 5%). However,the coda Q technique tiplet p3 (Figure 13). To obtain a similarstationsubset is commonly(and necessarily)stabilizedby weighting as in Figure 12, we chosea minimum number of nine functionswhichrejectdata of low coherency[Got et al., events required at each station. Except for station JST 1990]. This suppresses exactlythosedata whichin the the recovery signals in Figures 12 and 13 match fairly
present study show the temporal variations.
Using the same multiplet data sets, Beroza et al. well. For multiplet p2 (Figure 12) the recoverysignalat JST is steeperthan for multiplet p3 (Figure 13). This [1997]and D. Schaffand G. C. Beroza(manuscriptin might be causedby slight differencesof the propagation preparation, 2000) find changesin seismicwave velocpath betweenthe p2 and p3 eventsand/or differences ities which show a similar recovery signal to that reported here. The spatial pattern of delayed seismogram phasesas well as the onset of these phasesat each individual station resemble the pattern of reduced coherencies observed in this study, thus indicating that both available for a few stations. At these stations the recovmethods reflect the same crustal changes. The critical ery signal is comparable to that in Figures 12 and 13. difference between the two approachescomesfrom the in frequency content suchthat the p2 and p3 waveforms sample the anomalousregion in a different way. We do not show the recovery signals for multiplets p l and p4 where a reasonablenumber of events is only
16,334
BAISCH AND BOKELMANN:
CHANGE OF SEISMIC WAVEFORM ATTRIBUTES
37.4
37.3
37.2
"0 37.1
•
37
36.9
36.8
36.7 -122.2
-122.1
-122
-121.9
-121.8
-121.7
-121.6
-121.5
-121.4
-121.3
longitude [deg]
Figure13. Coherency recovery at allstations formultiplet p3.Display isasin Figure 12.
fact that velocity delays can be observed in the transmitted wavefieldsof crustal changesonly. Therefore it is difi:icult to interpret the present observationsas pure delay effects. For instance, considerstation HAZ in Figure 5. Explaining the coherencypattern at this station in terms of pure delay effects would require a seismogram phase which travels from the multiplet hypocen-
ter to the anomalousregion (which obviouslyis located northwest to the multiplet hypocenter), samplesthat region (where the phase is delayed), and is reflected back to station HAZ. From kinematic arguments the phase cannot travel much farther than station JBZ to the northwest, even when assuming single scattering. Thus it seems more likely that the anomalous region itself acts as a reflector, and we suggestthat our observationsare causedby additionally scattered phases, rather than by delayed phases. The coherencyanalysis presented here can be regarded as a tool for identifying additional phasesin two otherwise identical seismogramswithout invoking a physicalmodel of the scattering processes.Although the coherencyitself is a statistical property which cannot be related easily to physical parameters, coherency-basedtechniques proved to
Figures 5 and 9 may indicate that the scatterer is not located in the immediate vicinity of station JBZ. Then the scatterer
a vertical
location
would have either
offset to station
JBZ.
From
a horizontal
or
the coherencies
alone it appears difficult to distinguishbetween a shallow or a deep location, but in the latter casethe scatterer might directly coincidewith the region of highest
deformationduringthe Loma Prieta earthquake(Figure 14). Boreholeobservationsat depth suggestthat temporal variationsin the deepercrust exist [Bokelmann and Harjes, 2000]. They are not necessarilyconfined to the near-surface region. However, applying a source array analysis to similar events in the Loma Prieta re-
gion, Dodgeand Beroza[1997]find that most of the coda waves leave the source in direction
of the record-
ing station indicating that the coda consistsprimarily of waves scattered near the stations. Thus they suggest that coseismicvelocity decreaseassociatedwith the Loma Prieta earthquake as observedby Ellsworth et al.
[1992]occurredin the shallowcrust, near the stations. Nevertheless,to explain the coherencypatterns in Figures 5-9 phases are required which leave the source in different directions. Therefore the line of argument of
be usefulfor phaseidentification[e.g.,Ledniakand Ni- Dodgeand Beroza [1997] doesnot apply to the obseritsurea,1998]. By kinematicarguments,theseobserva- vations presented here, and we think that at this point, tions can be related to scattering changesin the vicinity of station JBZ. In principle, the coherencyprovides no information whether the postseismic or the preseismic wave fields experienced enhanced scattering, but it seems more plausible that scattering was enhanced during the rupture processrather than reduced.
neither of the observationscan be used to distinguish between a shallow or a deep location. To relate our observations to a physical model, we suggestthat coseismicdeformation leads to crack open-
ing either by local concentrationsof shear stress(coseismicdilatancy) or by elevated pore fluid pressure.
BAISCH AND BOKELMANN:
CHANGE OF SEISMIC WAVEFORM
ATTRIBUTES
16,335
-5
-10
-15
-20 10
15
10 0
0 -5
-10
-15 -10
1 O0
200
300
400
500
600
slip [cm]
Figure 14. Diagramshowingthe southernpart of the Loma Prieta slip modelof Beroza[1996]. Stars denote the main shockand the multipier p2 hypocenter,respectively. Some station locations
and their projectioninto the multipier depth (griddedplane) are shown.Note that, JBZ is located just above the high-slip area where slip exceeded5 m.
In the postseismic wave fields these cracks are seen scattering potential are characterized by enhanced seis-
as additional seismicreflectors[e.g., Nut and Walder, micity (and earthquakeslip). While he favorsa struc1992]. Immediately after the earthquake,relaxational tural (time independent) nature of the scatterer, the processesstart and causethe cracksto closeagain. Possible candidates for such processesare crack healing
recovery processafter the Loma Prieta earthquake suggeststhat some of this scattering may be due to fluid-
[Smith and Evans, 1984], successive decreaseof pore fluid pressureby diffusion[Booker,19741,and postseis-
filled
mic deformations lowering shear stress concentrations
[Schaffet al., 1998]. The temporalevolutionof all these
6.
cracks.
Conclusions
processesfollows a typical power law dependencecomWe inspected correlationsand lapse time-dependent patible with our observations. Anomalous postseismic coherencies of severalgroupsof similar events(multideformations after Loma Prieta are observed in GPS plets) around the Loma Prieta rupture zone. Reduced
measurements[Savageet al., 1994; B?);rgrnann et al., cohereheyof post-Loma Prieta events with pre-Loma 1997; Segall et al., 2000] and exhibit a compressional Prieta eventsoccurredin a spatiallylimited regionclose componentnormal to the San Andreas Fault. These de- to the Loma Prieta hypocenter. On the basis of multiformations seem to occur in the brittle part of the crust ple redundancy of the data set we were able to exclude and might reflect the postseismicresponseto coseismic contaminating effects,such as sourcevariations, and to
dilatancy[Savageet al., 1994]. However,it is difficult relate our observationsto subsurfacechangesof seismic to directly compare the surface deformations with the healing signal observedin this study. Our resultsindicate that the healing of coherenciesis affectedonly by a small crustal volume, whereasthe surfacedeformations extend over a region of some 10 km. Neverthelessit, is possiblethat the subsurficialchangesreported here also causesome of the surface deformation signal. Our observationsare interesting also in the light, of
wave propagation properties. Successivecombinations
of the preshockand aftershocks, the latter sampling> 5 yearsof the postseismicperiod, revealeda progressive "healing" of these changesfollowing approximately a power law dependence. The spatial pattern of temporally varying phasesis best explainedby local changes of scatteringpropertiesin the vicinity of the Loma Prieta sourcevolume. To interpret these changes,we sugthe suggestion by Revenaugh [1995]that regionsof high gest a model of coseismicallyopened cracksacting as
16,336
BAISCH AND BOKELMANN:
CHANGE OF SEISMIC WAVEFORM ATTRIBUTES
seismic scatterers. Theoretical models of postseismic Aster, R. C., G. Slad, J. Henton, and M. Antolik, Differential analysis of coda Q using similar microearthquakesin relaxation such as crack healing, fluid diffusion, and seismic gaps, part 1, Techniques and application to seisposteventdeformationare possibleexplanationsfor the mogramsrecordedin the Anza seismicgap, Bull. Seismol. observedtime signal. Soc. Am., 86(3), 868-889, 1996. Beroza, G., Rupture history of the earthquake from highfrequency strong-motion data, in The Loma Prieta, California, Earthquake of October 17, 1989-Main-Shock Characteristics, edited by P. Spudich, U.S. Geol. Surv., Pro.]:. Pap., 1550-A, A9-A32, 1996. In order to compute a confidenceinterval on the coBeroza, G. C., A. T. Cole, and W. L. Ellsworth, Stability hereheyC(•) we follow Got and FrSchet[1993]using of coda wave attenuation during the Loma Prieta, Cali-
Appendix A' Confidence Interval on the Coherency
Fisher's
z transformation
fornia, earthquakesequence,J. Geophys.Res., 100(B3),
z- 11n 1qC(•) ' 2 -
3977-3987, 1995.
(A1)Beroza, G. C., D. Schaff, W. L. Ellsworth, A. T.
Cole, and D. A. Dodge, Temporal changesin seismicvelocitiesob-
served
where z is approximately Gaussian distributed with
mean/z and variancecr2 : 1/(B$T). HereB$ denotes the equivalent bandwidth of the smoothing function,
and T is the samplingperiod. The 100(1-a)% lower
in the shear wave coda related
to the 1989 Loma
Prieta, California, earthquake,Eos Trans. A GU, 78(46), Fall Meet. Suppl., F482, 1997. Bokelmann, G. H. R., and H.-P. Harjes, Evidence for temporal variation of seismicvelocity within the upper conti-
nental crust, J. Geophys.Res., 105(B10), 23,879-23,894,
and upper confidenceinterval for z is given by
2000.
Booker, J. R., Time dependent strain following faulting of a porous medium, J. Geophys. Res., 79(14), 2037-2044,
/z+ q[1 - a/2]
(A2)
1974.
Biirgmann, R., P. Segall, M. Lisowski, and J. Svarc, Postseismic strain following the 1989 Loma Prieta earthquake For the assumed normal distribution, probability valuesof •111- a/2] can be derivedfrom tabulatedvalues from GPS and leveling measurements,J. Geophys. Res.,
102(B3), 4933-4955, 1997.
[e.g.,Jenkinsand Watts,1968,p.71]. After backtrans- DeFazio,
formation,the 100(1-c•)%confidence intervalfor the coherency is
T. L., K. Aki, and J. Alba, Solid earth tide and observedchangein the in situ seismicvelocity, J. Geophys.
Res., 78(8), 1319-1322,1973. Dodge, D. A., and G. C. Beroza, Source array analysis of coda waves near the 1989 Loma Prieta, California, mainshock: Implications for the mechanism of coseismicveloc-
(A3) Note
that
this confidence
Acknowledgments.
interval
is not centered on
All data used in this study are
made available through the NCEDC and were contributed
ity changes,J. Geophys.Res., 102(Bll), 24,437-24,458, 1997.
Eisler, J. D., Investigation of a method for determining stress accumulation at depth, Bull. Seismol. Soc. Am., 57(5), 891-911, 1967. Eisler, J. D., Investigation of a method for determining stressaccumulation at depth-II, Bull. Sezsmol. Soc. Am.,
by the NCSN. We thank the staff of the USGS, Menlo Park, 59(5), 43-58, 1969. for providinginformationon instrumentchangesand Greg Ellsworth,W. L., A. T. Cole, G. C. Beroza,and M. C. VerBeroza and David Schaff for valuable discussion and the woerd, Changesin crustalwave velocityassociatedwith compilation of the multipier data sets. Constructive comthe 1989Loma Prieta, California,earthquake,Eos Trans. mentsof Rick Aster and an anonymousreviewerare grateA GU, 73(43), Fall Meet. Suppl.,F360, 1992. fully acknowledged. Geller,R. J., and C. S. Mueller,Foursimilarearthquakes in centralCalifornia,Ceophys.Res. Lett., 7(10), 821-824, References
Aki, K., and B. Chouet, Origin of coda waves,attenuation, and scattering effects, J. Geophys. Res., 80(23), 3322-
1980.
Got, J. L., and J. Frdchet,Originsof amplitudevariationsin seismicdoublets:Sourceor attenuationprocess,Geophys. J. Int., 114, 325-340, 1993.
3342, 1975.
Got, J. L., G. Poupinet,and J. Fr6chet,Changesin source Aki, K., T. DeFazio, P. Reasenberg,and A. Nur, An active and site effectscomparedto coda Q-X-Temporalvariaexperiment with earthquake fault for an estimation of the tionsusingmicroearthquakes doubletsin California,Pure in situ stress,Bull. Scismol. Soc. Am., 60(4), 1315-1336, Appl. Geophys.,134(2), 195-228,1990. 1970. Hellweg,M., P. Spudich,J. B. Fletcher,and L. M. Baker, Antolik, M., R. M. Nadeau, R. C. Aster, and T. V. McEvilly, Stability of coda Q in the regionof Parkfield,California: Differential analysis of coda Q using similar microearthView from the U.S. GeologicalSurveyParkfielddenseseisquakesin seismicgaps, part 2, Application to seismograms mographarray, J. Ceophys. Res., 100(B2), 2089-2102, recorded by the Parkfield High Resolution Seismic Net-
work, Bull. Seismol. Soc. Am., 86(3), 890-910, 1996. Aster, R. C., P.M. Shearer, and J. Berger, Quantitative measurements of shear wave polarizations at the Anza seismic network, southern California: Implications for shear wave splitting and earthquake prediction, J. Geo-
phys. Res., 95(B8), 12,449-12,473,1990.
1995.
Holzer,T. L., Extensiveresearchon Loma Prieta improves understanding of earthquakes, Eos Trans. ACU, 80(41), 482-483, 1999.
Jack,I., Time-lapseseismicin reservoirmanagement, paper presentedat Societyof ExplorationGeophysicists, Tulsa, Okla., 1998.
BAISCH
AND BOKELMANN:
CHANGE
OF SEISMIC
Jenkins, G. M., and D. G. Watts, SpectralAnalysis and Its Applications,525 pp., Holden-Day,Boca Raton, Fla., 1968.
Kanamori, H., and G. Fuis, Variation of P-wave velocity beforeand after the Galway Lake earthquake(ML = 5.2) and the Goat Mountain earthquakes(ML = 4.7, 4.7) in
WAVEFORM
ATTRIBUTES
16,337
studieson codawaves,PureAppl. Geophys., 126(2-4), 465-497, 1988.
Savage, J. C., M. Lisowski, and J. L. Svarc, Postseismicdeformation followingthe 1989 (M-7.1) Loma Prieta, California, earthquake, d. Geophys. Res., 99(B7), 13,75713,765, 1994.
the Mojave desert,California,Bull. Seismol.Soc. Am.., Schaff, D., G. C. Bcroza, and B. E. Shaw, Postseismicresponse of repeating aftershocks, Ueophys. Res. Left., 66(6), 2017-2037,1976. Karageorgi,E., R. Clymer,and T. V. McEvilly, Seismologi- •5(•4), 4•49-4•, •998. cal studies at Parkfield, II, Searchfor temporal variations
Segall,P., R. Biirgmann, and M. Matthews, Time-dependent triggered afterslip following the 1989 Loma Prieta earthquake, d. Geophys.Res., 105(B3), 5615-5634, 2000. Legniak, A., and H. Niitsuma, Time-frequency coherency Simpson,R. W. (Ed.), The Loma Prieta, California, earthanalysisof three-component crosshole seismicdata for arquake of october 17, 1989-Tectonic processesand models, rival detection, Geophysics,63(6), 1847-1857, 1998. U.S. Geol. Surv. Prof. Pap., 1550-F, 1994. Li, Y.-G., J. E. Vidale, K. Aki, F. Xu, and T. Burdette, Ev- Smith, D. L., and B. Evans, Diffusional crack healing in idence of shallow fault zone strengthening after the 1992 quartz, d. Geophys.Res., 89(B6), 4125-4135, 1984. M7.7 Landers, California, earthquake, Science,279, 217- Spudich,P. (Ed.), The Loma Prieta, California, earthquake
in wavepropagationsusingvibroseis,Bull. Seismol.Soc. Am., 82(3), 1388-1415,1992.
219, 1998.
Neuhauscr,D. S., B. Bogaert, and B. Romanowitz,Data access of northern
California
seismic data from the North-
ern California EarthquakeData Center, Eos Trans. A GU, 75(44), Fall Meet. Suppl.,F429, 1994. Nur, A., and J. Walder, Hydraulic pulses in the Earth's crust, in Fault Mechanicsand TransportProperties of Rocks,edited by B. Evansand T.-F. Wong, pp. 461-473, Academic, San Diego, Calif., 1992. Reasenberg,P. A. (Ed.), The LomaPrieta, California,earthquakeof october17, 1989-Aftershocks and postseismic effects, U.S. Geol. Surv. Prof. Pap., 1550-D, 1997. Rcvenaugh,J., Relation of the 1992 Landers, California, earthquakesequenceto seismicscattering,Science,270,
of october 17, 1989-Main-shock characteristics, U.S. Geol. Surv. Prof. Pap., 1550-A, 1996. Wesson, R. L., R. Robinson, C. G. Bufe, W. L. Ellsworth, J. H. Pfiuke, J. A. Steppe, and L. C. Seekins, Search for seismic forerunners to earthquakes in central California, Tectonophysics, •2, 111-126, 1977.
S. Baisch, Ruhr-Universitiit Bochum, Institut fiir Geophysik, Universitiitsstrasse150, D-44780 Bochum,
Germany.(
[email protected]. de) G. H. R. Bokelmann,Department of Geophysics,Stan-
ford University,Stanford,CA 94305-2215.(goetz@pangea. stanford.edu)
1344-1347, 1995.
Sato, H., Temporal changein scatteringand attenuation associatedwith the earthquakeoccurence-Areview of recent
(ReceivedJune 1, 2000; revisedJanuary11, 2001; acceptedMarch 13, 2001.)