You are on page 1of 22

Bulletin of the Seismological Society of America, Vol. 70, No. 2, pp.

447-468, April 1980


Q ESTI MATES USING THE CODA OF LOCAL EARTHQUAKES
BY ROBERT B. HERRMANN
ABSTRACT
An extension of the Aki (1969) coda model takes into account the instrument
and Earth Q-filter effects on the dispersion of the coda. The coda dispersion and
coda shapes are easily derived and plotted as a function of t* = t/Q. Observed
data can be overlaid on master curves to directly estimate shear-wave Q and to
normalize for earthquake moment to estimate the attenuation coefficient of 7.
The method works well when the earthquakes studied have corner frequencies
greater than the combined peak response of the instrument and Earth Q-filters.
Comparing results from central California and the southeastern United States,
there is a slight indication that the excitation of the coda is inversely proportional
to Q.
I NTRODUCTI ON
The coda pf local and regional earthquakes is a very interesting feature of the
seismogram, especially since it is difficult to model it numerically, peak for peak, in
this era of deterministic modeling of P, S, and surface waves of local and regional
earthquakes. This is not surprising since the coda is viewed as a composite of waves
scattered from many places in the Earth' s crust and upper mantle (Aki, 1969; Aki
and Chouet, 1975). The coda does have many nice features that make it useful for
estimating magnitudes (Real and Teng, 1973) and seismic moments (Bakun and
Lindh, 1977) of local earthquakes. The coda has also been used to establish spectral
scaling laws of local earthquakes and estimates of local Q (Aki and Chouet, 1975;
Chouet et at, 1978; Rautian and Khalturin, 1978). The particular reason why coda
studies are performed lies in the fact that deductions about the earthquake process
can be made without a lot of sophisticated calculations, using instead very simple
measurements from the seismogram.
The object of this study is to see if the coda as recorded on broadband instruments
can be used to estimate Q. Aki and Chouet (1975) and Rautian and Khalturin (1978)
were able to estimate Q from studies of coda as recorded by a set of narrow-band
seismographs; however, the coda theory initially proposed by Aki (1969) can be
modified to permit extraction of information on Q from a broadband seismograph.
CODA THEORY
Developing theory for scattering of surface waves, Aki (1969) proposed the
following relation to describe the average peak-to-peak amplitude A(t) observed
around time t as follows
A (t)_~ _ i( f.)Q~/4t_~/2 [ fftf. l -V4e_.f ~/QMoB( f.),
(1)
where I(fp) is the instrument magnification at the predominant frequency observed
at time t, Q is the anelastic attenuation quality factor for the surface waves
comprising the coda, Mo is seismic moment. Following Suteau and Whitcomb (1979),
B(fp) is defined as
447
448 ROBERT B. HERRMANN
B( f p) = [2N(ro)]~/20o( fp, ro), (2)
where N(ro) is t he number of scat t erers wi t hi n a radi us ro from t he eart hquake and
~o(fp, ro) expresses t he excitation of t he secondary scat t ered waves.
Equat i on (1) was devel oped to analyze ear t hquakes in central California for which
t her e is a part i cul ar val ue of Q. To use this relation in ot her areas, it is helpful t o
express equat i on (1) in t er ms of anot her t i me variable, t* = t / Q. Equat i on (1) t hus
becomes
A( t * ) _ Q_~/eMoB ( fp)C( fp, t *), (3)
, / 8
where t he coda shape funct i on is
dfp 1/4
C( f p, t*) = I ( f p ) t *-1/2 dt * e-~fPt*" (4)
An essential r equi r ement for t he use of equat i on (3) is t he specification of t he
change in t he pr edomi nant frequency of t he coda wi t h time, df p/ dt *. Aki (1969)
est abl i shed such a rel at i on empirically by first plotting t he predomi nant frequency
as a funct i on of t i me aft er t he ear t hquake origin time. He obt ai ned t he formul a
t/100 = (f/0.82) -15. (5)
Thi s rel at i on was used l at er by Sut eau and Whi t comb (1979).
The rel at i on found by Aki (1969) is real l y a funct i on of t he i nst rument syst em
used as well as of t he geographical region in whi ch he obt ai ned his observations. It
seems i nt ui t i vel y obvi ous t hat his val ues of d f J d t * shoul d be dependent somehow
upon regional Q. To expand upon this, an assumpt i on as to t he cause of this
f r equency dependence is made, t hat is, t h a t t he p r e d o m i n a n t f r e q u e n c y o b s e r v e d i s
a f u n c t i o n o f t h e e a r t h q u a k e s our c e s pe c t r um, t h e i n s t r u me n t r es pons e, a n d t he Q
f i l t e r o f t h e E a r t h . As a resul t of multiplying t hese t hr ee factors, a peaked, relatively
narrow band-pass frequency response is found. A characteristic of such a filter is
t hat t he t i me series out put for an arbi t rary, r andoml y scat t ered wave i nput woul d
be a random-l i ke signal wi t h predomi nant frequency close t o t hat of t he peak
frequency of t he entire filter response. In ot her words, comparing t he peak frequency
at t wo di fferent t i mes in t he coda, t he effect of t he Q operat or is to decrease t he
pr edomi nant frequency as t i me increases. Likewise, t he pr edomi nant frequency
obser ved at a given t i me is dependent upon t he i nst r ument response.
Set t i ng aside t he effect of t he source-spect rum frequency cont ent and considering
onl y ear t hquakes small enough t hat t hei r corner frequenci es are great er t han t he
peak i nst r ument response, it is rel at i vel y easy t o eval uat e equat i on (4) for C( fp, t*).
To do this, a part i cul ar val ue of t* is assumed. Next t he syst em filter response,
including t he i nst rument and Q-filter effects,
I ( f ) e - ~ r t*
is eval uat ed over a range of frequenci es to det ermi ne t he frequency fp of t he peak
response. Given t hi s frequency at near by val ues of t*, t he deri vat i ve df p/ dt * can be
eval uat ed numerically. Finally, C( fp, t*) is det ermi ned using equat i on (4).
Q ESTI MATES USI NG THE CODA OF LOCAL EARTHQUAKES 449
The pr e domi na nt f r equency fp and t he coda shape f unct i on C(fp, t*) were
eval uat ed as a f unct i on of t* for t wo di f f er ent s ei s mogr aph syst ems, a shor t - per i od
WWSSN s ys t em havi ng a s ei s momet er na t ur a l f r equency of 1 Hz, a gal vanomet er
na t ur a l f r equency of 1.35 Hz, and a magni f i cat i on of 1.0 at 1 Hz, and a USGS
mi c r oe a r t hqua ke s ys t em wi t h a s ei s momet er na t ur a l f r equency of 1 Hz, a gal vanom-
et er na t ur a l f r equency of 15 Hz, and a peak r esponse of uni t y at 10 Hz. Ta bl e 1 gi ves
fp and C( fp, t*) as a f unct i on of t* for bot h syst ems. The r esul t s are al so pl ot t ed i n
Fi gur es 1 and 2. Fi gur e 1 shows fp ver sus t* for t he t wo i ns t r ume nt r esponses whi l e
Fi gur e 2 shows C(fp, t*) as a f unct i on of t*.
I n Fi gur e 1, i t is seen t ha t t he peak f r equenci es of t he t wo i ns t r ume nt s ys t ems are
i ndi s t i ngui s habl e for t* gr eat er t ha n 1.0. Thi s is because t he peak r esponse occur s at
TABLE 1
INSTRUMENT CODA RESPONSES
WWSSN USGS
t*
L (Hz) C([p, t*) /p Clfp, t*)
0.01 1.45 1.25E + 1 10. 67 2.38E + 1
0.02 1.44 1.20E + 1 9.88 1.27E + 1
0.03 1.42 8.72E + 0 8.91 8.11E + 0
0.04 1.41 6.50E + 0 7.73 5.58E + 0
0.05 1.40 5.28E + 0 6.55 3.86E + 0
0.06 1.38 4.52E + 0 5.61 2.72E + 0
0.07 1.37 4.48E + 0 4.90 1.98E + 0
0.08 1.36 3.87E + 0 4.37 1.50E + 0
0.09 1.35 3.33E + 0 3.96 1.17E + 0
0.1 1.34 3.00E + 0 3.63 9.37E - 1
0.2 1.26 1.31E + 0 2.16 1.99E - 1
0.3 1.19 6.97E - 1 1.64 7.56E - 2
0.4 1.13 4.01E - 1 1.36 3.63E - 2
0.5 1.08 2.51E - 1 1.17 2.00E - 2
0.6 1.02 1.72E - 1 1.04 1.20E - 2
0.7 0.97 1.15E - 1 0.94 7.69E - 3
0.8 0.92 7.97E - 2 0.86 5.13E - 3
0.9 0.86 5.70E - 2 0.79 3.55E - 3
1 0.80 4.14E - 2 0.73 2.59E - 3
2 0.43 3.68E - 3 0.43 2.42E - 4
3 0.30 8.29E - 4 0.30 5.35E - 5
4 0.23 2.70E - 4 0.23 1.77E - 5
5 0.18 1.13E - 4 0,19 7.84E - 6
6 0.15 5.73E - 5 0.16 3.73E - 6
f r equenci es l ess t ha n t he s ei s momet er na t ur a l f r equency of 1.0 Hz and because t he
i ns t r ume nt r esponse of bot h s ys t ems var i es as f+a i n t hi s regi on. Bot h Fi gur es 1 and
2 are pl ot t ed wi t h l ogar i t hmi c t* axes si nce t he di fference bet ween r egi ons of
di f f er ent Q woul d be j us t a l i near shi ft of l ogar i t hmi c t i me. A qui ck t est of t he
cor r ect ness of t he ma s t e r cur ves of Fi gur e 1 is t o i nt er pr et t he empi r i cal r el at i on of
equat i on (5), due t o Aki (1969). Fi r st not e t ha t t he s ei s mogr aph s ys t em used by Aki
(1969) had a s ei s momet er na t ur a l f r equency of 1,0 Hz, and a gal vanomet er na t ur a l
f r equency of 5 Hz. Thus , i t s r esponse i n Fi gur e 1 woul d be i nt er medi at e bet ween t he
WWSSN and USGS syst ems. Pl ot t i ng Aki ' s dat a set, equat i on (5), on l ogar i t hmi c
paper , a nd over l ayi ng t he obs er vat i onal dat a on t he mas t er cur ve of Fi gur e 1, a
val ue of t* -- 1.0 ma ps i nt o an obser ved t = 150 sec, whi ch i mpl i es a Q = 150. Thi s
is not unr e a s ona bl e gi ven t he fact t ha t Aki (1969) as s umed Q = 200 for t he regi on.
450 ROB E RT B. HE RRMANN
This supports the assumptions made in generating the master curve of Figure 1, i.e.,
the dispersion is due to Q and to the particular instrument response.
The effect of the source spectrum was ignored in the above derivation by assuming
t hat the source-spectrum corner frequency is greater t han the peak frequency of the
instrument response. This is almost certainly the case for WWSSN instruments
when the earthquake mb is less t han 5.0. A problem arises with the same class of
earthquakes when the higher frequency USGS system is used. Q could still be
determined from the master curves of Figure 1 if data are used only at durations
great enough t hat the fp is less t han the source-spectrum corner frequency. Otherwise
we would conclude t hat the predominant frequency of the coda would be almost
constant, due to the source-spectrum frequency content, until Q began to have an
10
EL
1
I I I I I I I I I I I I I 1 t I I [ I I I t I I I
. 1 1
t
Fro. 1. Predominant coda frequency (Hz) as a function of t* (sec) for the USGS and the short-period
WWSSN instruments.
effect. In the following discussion, care will be taken concerning the effect of the
source-spectrum corner frequency.
METHODS fOR ESTIMATING Q
Assuming t hat the class of earthquakes studied is such t hat the effect of source-
spectrum corner frequency can be ignored, several methods can be used to estimate
Q. These are fp versus t, coda shape, and estimate of y.
f p V e r s u s t . This met hod was indicated above. One just plots the predominant
frequency of the coda versus time after origin. The predominant frequency can be
easily estimated by counting the number of zero crossings of the seismic trace within
a given time interval, 10 sec or so, and dividing the number of zero crossings by
twice the window length. The plot of observations is superimposed upon Figure 1,
and Q is t hat value of t which maps into t* = 1.
Q ESTIMATES USING THE CODA OF LOCAL EARTHQUAKES 451
Coda shape. If the peak-to-peak coda amplitude is plotted as a function of time,
and if data from many earthquakes are superimposed to find a common shape, the
common shape may be used to estimate Q by performing a curve matching with the
theoretical coda shapes of Figure 2. It would seem t hat this would be definitive for
the WWSSN data only if the data were on both sides of the change in slope at t*
= 0 . 5 .
Es t i ma t i o n o f y. Nuttli (1973) proposed t hat the attenuation of the vertical
( . 9
.01
U S G S
WWSSN
.1 1
FIG. 2. Coda shape as a function of t* for the USGS and short-period WWSSN i nst rument s.
component L g wave, a superposition of high-mode surface waves with a group
velocity of about 3.5 km/sec, could be described by the relation
Aoe-Yr
ao= r~/~ , (6)
where ao is the sustained peak amplitude, y is the anelastic attenuation coefficient,
and r is the distance of the observation point from the earthquake. Nuttli {1973)
noted t hat y is related to Q, frequency f, and group velocity U by y = ~f / QU. As
452 ROBERT B. HERRMANN
proposed, this technique may be described as a many station-single earthquake
method. It has been used by Nuttli (1973, 1978) to estimate vertical component Lg
attenuation at frequencies of 1 and 10 Hz, respectively, in the central U.S. and by
Bollinger (1979) for vertical component 1-Hz Lg waves in the southeastern U.S.
The level of the seismic coda at any time, equation (1), is directly proportional to
the seismic moment. Thus, it seems possible to use the coda to correct for seismic
moment and t hen to use a many earthquake-single station approach to estimate y.
To do this, the envelope of the seismic signal, including the Lg maximum and the
coda, is plotted on log-log paper as a function of time after the origin time. The
observed coda is superimposed on the appropriate theoretical curve of Figure 2,
using a value of Q estimated from the "fp versus t " method. A check for the
appropriateness of the Q value is made by comparing the observed and predicted
coda shapes. When the codas are superimposed, it is noted t hat 1 cm peak-to-peak
on the observed seismograms corresponds to a particular value of C( fp, t*) denoted
as Ap. The moment corrected amplitude is just
Ac = Ap X Apeak,
where Apeak is the Lg peak-to-peak maximum amplitude. The coefficient 7 is
determined using the attenuation model of equation (6).
If the assumptions made in the development are correct and if 7 can be used to
estimate Q, the results of the three methods listed for estimating Q should be
internally consistent. To test this, data from several WWSSN and USGS instru-
ments will be used.
WWSSN DATX
Blacksburg, Virginia (BLA)
This station was chosen because Bollinger (1979) estimated a ~/= 0.0009 km -1 for
the attenuation of the vertical component Lg wave for the southeastern United
States, in which location BLA is located. Using the relation between 7 and Q and
the data contained in Bollinger (1979), one obtains Q = 1400. The locations of the
earthquakes used in the study are shown in Figure 3. The date, origin time, and
coordinates of each earthquake are given in Table 2 together with the distance of
the earthquake from BLA, the peak Lg amplitude on the seismogram, the instrument
magnification, the group velocity of the peak motion, the body-wave magnitude rob,
the corrected amplitude Ac, Ap, and the seismic moment Mo. Figure 4 indicates the
manner in which the peak-to-peak coda amplitudes were read.
Figure 5 is the plot of predominant coda frequency (Hz) versus time (seconds)
after the earthquake. Using the master curve of Figure 1, a linear least-squares
determination yielded Q = 1040 _ 111 for a 95 per cent confidence level. Using this
value of Q, a theoretical curve was constructed from Figure 2. Observed data were
adjusted vertically until the codas were tangent to the theoretical curve. From this
curve, Ac and Ap were determined. Sample curves are shown in Figure 6. The fact
t hat the observed codas match the curvature of the theoretical curve indicates the
appropriateness of the Q value estimated.
Finally, the Ae data of Table 2 were plotted as a function of distance in Figure 7.
A least-squares determination of ), using equation (6) yielded
In Ao = 7.718 +_ 0.44
7 = +0.00087 +_ 0.00066 km -1,
Q ESTI MATES USI NG THE CODA OF LOCAL EARTHQUAKES 453
where 95 per cent confidence limits are given. The average group velocity of the
peak amplitudes of Table 2 is 3.10 km/sec, which would imply Q = 1456. The y
value determined here is consistent with t hat determined by Bollinger (1979), given
the broad error limits. Likewise the y-inferred Q is consistent with the Q estimated
from fp versus t*.
A number of the earthquakes have surface-wave estimates of the seismic moment
(Herrmann, 1979). These can be used to estimate the value of the coda excitation
factor B(fp). Equation (3) is used to do this. As an example, for the March 3, 1963
earthquake, equation (3) becomes


I
- 5 , +

e~ BLA
Fro. 3. Locat i on of eart hquakes (solid dots} studied using BLA (solid triangle) data. Crossed marks
indicate latitude-longitude references.
1.0/50,000 = (8/1040) 1/21.1 X 1023B (fp) 0.15,
(7)
from which B (fp) = 1.39 10 -26 cm-sec/dyne-cm. Using ten earthquakes for which
seismic moment is known, the logarithmic average for B(fp) is 2.01 10 -26 cm-sec/
dyne-cm with a 95 per cent confidence limit corresponding to a factor of 2.19. Given
a value for B(fp), seismic moments for the other earthquakes studied can be
determined and are listed in Table 2. A resulting correlation between seismic
moment and rnb is
454 ROBERT B. HERRMANN
~q
<
v
o
o
Q ESTIMATES USING THE CODA OF LOCAL EARTHQUAKES 455
[-,
8
t ~
O
o9
D
ogd
o9..~
~o9
c c ~
o9
, ~o9
,~o9
456 ROB E RT B. HE RRMANN
I I I 1 I I t I } I I I I I I I
- l ~ e
1 0 0 0
i l i I I I I l l l I L i | l l I
t
Fro. 5. Pl ot of fp (Hz) versus t (sec) at BLA. The theoretical curve for Q = 1040 is shown.
1 1 NOV 75
10 SEPTO
30 JUL 70
-%
\ \
\
\
22NOV 74.
3 MAR 63
.1 " \ ~
100
t
Fro. 6. Ex a mp l e o f adj us t i ng f or s e i s mi c mo me n t at BLA. T h e t he o r e t i c a l curve ( das he d l i nes ) i s
appr opr i at e t o Q = 1040. T h e o bs e r v e d dat a ( s ol i d l i nes ) we r e a dj us t e d v e r t i c a l l y unt i l t he i r c odas we r e
t a n g e n t t o t h e t he o r e t i c a l curve. B o t h t he o r e t i c a l a nd obs e r vat i onal c ur ve s are a dj us t e d v e r t i c a l l y f or
cl ari t y.
l o g l o M o = 1 7 . 5 4 + 0 . 4 9 + ( 1 . 0 7 4- 0 . 3 3 ) m b , ( s )
wh e r e 95 p e r c e n t c o n f i d e n c e l i mi t s a r e gi ve n.
Berkeley, California (BKS)
D a t a f r o m B KS we r e c h o s e n f or s t u d y b e c a u s e t h e r e g i o n d e t e r mi n i n g t h e c o d a
h a s h a d a n a l t o g e t h e r d i f f e r e n t r e c e n t t e c t o n i c h i s t o r y t h a n t h e s o u t h e a s t e r n Un i t e d
Q ESTIMATES USING THE CODA OF LOCAL EARTHQUAKES 457
St at es. Whi l e no Y val ue has been publ i s hed i n t he l i t er at ur e, nume r ous es t i mat es
of s hear - wave Q ha ve been ma de f or e a r t hqua ke s al ong t he San Andr eas Faul t
s out he a s t of BKS. Usi ng s hear waves in t he 0- t o 2- Hz band, Kur i t a (1975) es t i mat ed
Q = 20 t o 100. Bu k u n et al. (1976) es t i mat ed Q = 100 t o 150 i n t he 1- t o 30-Hz ba nd
f or e a r t hqua ke s i n Be a r Val l ey, Cal i forni a. Aki and Choue t (1975) es t i mat ed Q = 50
t o 100 f or t he coda waves wi t h f r equenci es of 1 Hz and Q = 1000 t o 2000 f or t he 20-
Hz coda waves f or St one Canyon, Cal i forni a.
Fi gur e 8 shows t he l ocat i on of t he e a r t hqua ke s s t udi ed in r el at i on t o BKS. Th e
speci fi c e ve nt da t a and r eadi ngs ar e gi ven in Ta bl e 3. Fi gur e 9 is t he pl ot of
pr e domi na nt coda f r e que nc y ver sus t i me af t er t he ear t hquake. A l east - squar es
es t i mat e yi el ded Q = 133 __ 14. Usi ng Q = 135, t he mo me n t cor r ect ed ampl i t udes of
Ac wer e de t e r mi ne d i n or der t o de t e r mi ne y. A l eas t - s quar es es t i mat e yi el ded
, , i , , , J ] i , , l ; l , , ]
=I0O\
1 0 " * " "
1 '-/= 0 . 0 0 o 9
100 1000
L i L t l l ~ l , r , , i , , , i
k m
Fro. 7. Plot of BLA moment corrected amplitudes versus distance, together with the least-squares
attenuation curve.
In A0 = 5.097 _ 0.397
~/ = +0. 0048 _+ 0.0013 km -1,
95 pe r cent conf i dence l i mi t s bei ng given. Th e da t a and t he l east - squar es a t t e nua t i on
cur ves ar e s hown i n Fi gur e 10. Usi ng t he aver age gr oup vel oci t y of 2.86 km/ s e c of
t he ma x i mu m ampl i t ude ar r i val s gi ven i n Ta bl e 3, t hi s val ue of ~/would be cons i s t ent
wi t h Q = 229. Usi ng t he da t a cont ai ned in Ta bl e 3, B(fp) was de t e r mi ne d t o be 1.51
10 -25 c m- s e c / dyne - c m t o wi t hi n a f act or of 1.79 f or 95 per cent conf i dence. Aki
(1969) e s t i ma t e d a val ue of 1.5 x 10 -25 c m- s e c / dyne - c m in t he 0.22- t o 1.05-Hz
f r e que nc y r ange f or cent r al Cal i forni a.
Dugway, Utah (DUG)
Da t a f r om DUG wer e s t udi ed wi t h t he ant i ci pat i on t ha t t he Q de t e r mi ne d mi ght
be i nt e r me di a t e be t we e n Ea s t e r n Uni t e d St at es and Cal i f or ni a val ues. Ta bl e 4
458 ROBERT B. HERRMANN
provides data on the earthquakes studied. The locations of the earthquakes in
relation to DUG are shown in Figure 11. Figure 12 shows the plot of fp versus t at
DUG. The least-squares estimate yielded Q = 325 _+ 44 for 95 per cent confidence.
The observed coda shapes were corrected for seismic moment in order to determine
y. The corrected amplitudes are shown in Figure 13 together with the theoretical
curve. It was found t hat
and
In Ao = 5.898 +_ 0.557
y = +0.0025 +_ 0.0010
g
4-

124

+
122 120
Fro. 8. Location of eart hquakes (dots) studied using BLA (triangle) data.
for 95 per cent confidence. Using the average group velocity of 3.17 km/sec of the
peak arrivals from Table 4, this ~( implies a Q = 396.
To estimate B (fp), the following reference earthquakes and seismic moments were
used: June 3, 1965, Mo = 2.1E22 and April 13, 1965, M0 = 6.0E21 (obtained using
BKS coda estimated moments), January 23, 1966, Mo -- 3.5E23, April 10, 1967, Mo
= 7.7E22, August 9, 1967, Mo -- 2.2E23 (Herrmann and Park, personal communica-
tion) and February 15, 1974, Mo = 4.0E22 (Herrmann, 1979). From these data, B (fp)
= 1.74 10 -26 cm-sec/dyne-cm to within a factor of 3.22 for 95 per cent confidence.
Q ESTIMATES USING THE CODA OF LOCAL EARTHQUAKES 459
NMM ~NM~
~ o ~ o o ~ d ~ d ~ o ~ d ~ 4 ~ d ~ 4 ~ d ~ d ~ " ~
I ~ ~ I I I I
~ ~ ~ q ~ m ~ q ~ ~ I I ~ ~ I I I t
m ~ q ~ ~ e ~ ~ ~ q ~ I I ~ m ~ I I I I
v
C~
.<
4 6 0 ROBERT B. HERRMANN
USGS INSTRUMENT DATA
A descri pt i on of t he charact eri st i cs of a mi cr oear t hquake net work in sout heast
Missouri using USGS i nst rument s can be found in St auder et al. (1976). Dat a from
t he st at i on GRT, 36.26N and 89.42W, were used to check t he applicability of t he
i I i i I I I I I i I i I I I I
100
I I I i i i ~ J I , . I I I I I I
t
FIG. 9. Plot offp (Hz) versus t (see) at BKS. The theoretical curve for Q = 140 is shown.
g l I I i I l l , I I t I I a l l l
l O 0 ~ = o . o 0 5 \
I I I L L I I I I I , L L i , i
k m
FIG. 10. Plot of BKS seismic moment corrected amplitudes versus distance, together with the least-
squares attenuation curve.
coda model to i nst rument s ot her t han t he WWSSN short period. Nut t l i (1978) used
dat a from t he ar r ay to obt ai n a y = 0.006 km -1 for 10-Hz vertical component Lg
waves. Fr om this y value one can est i mat e Q -- 1500.
Tabl e 5 gives t he t i mes and coordi nat es of each ear t hquake studied as well as
Q ESTI MATES USI NG THE CODA OF LOCAL EARTHQUAKES
TABLE 4
EARTHQUAKE DATA SET FOR CODA AT DUG
461
Date Origin Time Lat. Long. Distance Peak Mag. Vp~ak A, A~
(UT) (N) (W) (km) (cm) (k) (km/ sec) (sec-') (sec-')
09 Mar 65 06 41 34 39.9 111.3 133 4.0 200 3.09 4 1
27 Mar 65 23 17 39 42.6 111,5 289 2.0 200 3.40 1.9 0.96
30 Mar 65 17 45 48 38.7 115.5 284 1.4 200 2.45 0.56 0.4
02 Apr 65 05 19 26 42.5 111.5 280 2.3 200 3.09 0.92 0.4
06 Apr 65 15 57 05 39.0 118.7 523 2.8 200 3.37 0,92 0.23
13 Apr 65 13 14 22 38.9 117.7 443 11.2 200 3.59 1.9 0.16
15 Apr 65 20 08 33 34.1 117.4 787 5.6 200 3.48 0.26 0.046
27 Apr 65 18 51 37 41.3 112.8 123 3.9 200 2.55 6 1.5
03 May 65 03 30 50 36.0 114.7 495 1.5 200 3.28 0.4 0.265
03 J un 65 16 26 27 38.3 119.2 589 15.1 400 3.33 0,84 0.052
16 J un 65 02 42 06 33.0 115.6 833 3,4 400 3.21 0.115 0.033
17 J un 65 07 30 31 33,0 115.6 834 5.1 400 3.28 0.125 0.025
27 J un 65 19 24 07 39.6 110.4 216 3.8 400 2.86 1.3 0.33
29 J un 65 07 46 28 39.6 110.3 224 4.2 400 3.24 1.8 0.46
05 Jul 65 17 17 09 39.3 111.5 150 9.9 400 2.94 3.5 0.33
15 Jul 65 07 01 42 44.9 111.7 531 2.5 400 3.17 0.66 0.26
25 Sep 65 17 43 44 34.7 116. 5 692 400
01 Nov 65 17 10 15 39.6 118.5 490 15.1 400 3.27 1.51 0.1
23 Jan 66 01 56 39 37.0 107.0 617 51 200 3.12 0.27 0.0052
28 Jan 66 18 00 09 41.6 118.2 479 19 200 3.19 1.6 0.085
10 Apr 67 19 00 26 39.9 104.7 693 200 0.05
09 Aug 67 13 25 06 39.9 104.7 693 400 0.0058
15 Feb 74 13 33 49 36.5 100.7 1133 3.0 400 3.31 0.03 0.01
09 J un 74 00 50 45 44.8 111.1 531 3.4 400 3.36 0.19 0.055
4 . +

D U G
. o-JF- + 4 0
115 110 105
2 .
FIG. 11. Location of eart hquake (dots) studied using DUG (triangle) data.
462 R O B E R T B . H E R R M A N N
c o d a a mp l i t u d e r e a d i n g s . T h e p r e d o mi n a n t c o d a f r e q u e n c y i s p l o t t e d a s a f u n c t i o n
o f t r a v e l t i me i n F i g u r e 14. Us i n g t h e t h e o r e t i c a l c u r v e o f F i g u r e 1, a l e a s t - s q u a r e s
e s t i ma t e y i e l d s Q = 2190 _ 235. Wi t h t h i s v a l u e o f Q, o b s e r v e d a mp l i t u d e s we r e
13.
] i I I l I I I I t I " t I I i I
-1
""2
Q =325
IO0
, , , I I I I I I 1 1 1 L I I ~ ~ t I
t
Fro. 12. Pl ot of fp (Hz} versus t {sec) at DUG. The theoretical curve for Q = 325 is shown.
<~
:10 ~ o
~ o
" ~ Q
100 1000
I L I I I I l l I t I I I I I I I
km
FIG. 13. Plot of DUG seismic moment corrected amplitudes versus distance, together with the least-
squares attenuation curve.
c o r r e c t e d f or s e i s mi c mo me n t a n d a d e t e r mi n a t i o n o f y wa s ma d e . Us i n g f o u r t e e n
d a t a p o i n t s , i t wa s f o u n d t h a t
I n A0 = 7. 093 +_ 0. 307
y = - 0 . 0 0 2 9 -__- 0. 0064.
Q ESTIMATES USING THE CODA OF LOCAL EARTHQUAKES 463
<

~D
~z
I I ~ I I R e ~ R ~ ~ R R
I I ~ I I ~ N ~ M ~ ~ M ~
464 R OB E R T B. HE R R MANN
Surface-wave seismic moment estimates (Herrmann, 1979) were available for four
of the earthquakes studied. From these, B(fp) = 2.25 x 10 -2G cm-sec/dyne-cm to
within a factor of 3.0 for 95 per cent confidence.
The Q value determined from the GRT coda data is in good agreement with the
),-based estimate of Nuttli (1978). Unfortunately, the coda-based y estimate is
meaningless due to the error bounds, so t hat the check on internal consistency of
the coda method cannot be made. Part of the difficulty has to do with the fact t hat
amplitudes were read from a Develocorder on which 20 parallel traces are placed on
a 16-mm film. The problem of trace overlap limits the data available.
The purpose of using GRT data was to see how the source spectrum affects the
coda method. The theoretical curves of Figures 1 and 2 were obtained using the
assumption t hat at the time the coda amplitudes are measured, the corresponding
predominant coda frequency is less t han the source spectrum corner frequency. To
I I I I I I i i I I I I I I I I I
--10
\ -
C1=2200 _
100
I I I I I I I l l I I J ~ I i I J
t
Fr o. 14. Pl o t of fp (Hz) v e r s u s t (sec) f or t h e GR T US GS i n s t r u me n t . T h e t he or e t i c a l c ur ve f or Q =
2200 i s s hown.
estimate the corner frequency of the earthquakes studied, the seismic moment was
estimated from the coda in the manner used for BLA. Street e t al . (1975) found t hat
the relationship between the corner frequency fc and the seismic moment Mo for
central United States earthquakes is
fc(Hz) =
r (3.1E22/Mo) 1/3
1.67
( 3 . 1 E 2 1 / Mo ) 1/3
Mo ->_ 6.8E21
6.8E =< Mo = 6.8E21
Mo -< 6.8E20.
(9)
Thus, the calibration earthquakes with surface-wave seismic moments had corner
frequencies of about 0.7 to 1.6 Hz. Figure 14 t hen implies t hat only coda amplitudes
measured at times greater t han 500 sec be used if the effect of the source spectrum
is to be ignored. This was not the case for the estimates of B (fp) for the June 13,
1975 and Jul y 9, 1975 earthquakes. Using the moments estimated for the smaller
earthquakes, it is found t hat the corresponding comer frequencies must have been
greater t han 2.5 Hz, in which case coda amplitude and predominant frequency
measurements can be interpreted using the master curves of Figures 1 and 2, if the
Q ESTI MATES USI NG THE CODA OF LOCAL EARTHQUAKES 4 6 5
measurements are taken at times greater t han 200 sec. In all cases, the data used to
make Figure 14 had predominant frequencies greater t han the corner frequencies of
the earthquake. Hence, the inference of Q = 2200 is called into question.
As another check, the relation of mb to Mo is compared. The mb values for the
smaller earthquakes in 1979 are taken from the Central Mississippi Valley Earth-
quake Bulletin, distributed quarterly by Saint Louis University. The mb values are
estimates of mb from 10 Hz data from the USGS instruments. There is confidence
t hat these numbers are very close to the true mb. Street and Turcotte (1977)
established a scaling for northeastern United States earthquakes such t hat an
earthquake of mb = 2.0 would have a seismic moment of 3.0E + 19 dyne-cm.
Extrapolating the Mo versus mb relationship derived for BLA in equation (8), an
earthquake with mb = 2 would have a seismic moment of 4.8E + 19 dyne-cm.
Assuming t hat an mb = 2.0 corresponds to a seismic moment of 4.0E19 and t hat
logloMo scales directly as mb for the small earthquakes, it is seen t hat the coda
estimates of seismic moment are about 3 times larger t han the mb estimate.
The surprising thing about this analysis of the USGS instruments is t hat the Q
and seismic moment estimates are almost reasonable. There is a problem in
neglecting the effects of source spectrum scaling. The only mitigating factor is t hat
the very nature of the recording medium, 20 traces on a 16-mm film, forced the use
of relatively long durations, such t hat the observed predominant coda frequencies
were not too very much greater t han the source-spectrum corner frequency.
DI SCUSSI ON
This study indicates a relatively easy way to estimate Q t hat is especially
applicable where fancy data processing facilities are not available. All t hat is
required is the examination of about 20 to 30 seismograms with a good range of
magnitudes to define the coda shape and the predominant frequency curve well,
and with an even distance distribution for 7 determination. Given the seismograms,
about two man days are required per station. A crucial point in the development is
how the instrument response and the Q-filter of the Earth affect the predominant
frequency of the coda. The correctness of this inference was shown by the good
results obtained when the earthquake corner frequency is greater t han the peak
frequency of the instrument-Earth Q-filter. When this is not the case, the waveform
becomes complex and it is difficult to estimate the peak frequency without perform-
ing spectral analysis. The Aki (1969) coda model may have to be modified to take
into account a different coda frequency content.
This paper also demonstrates, for the first time, the correctness of Q inferences
from ~/, as proposed by Nuttli (1973, 1978). This was shown by estimation of both Q
and 7 for central California and for the southeastern United States, by testing the
internal consistency between the Q and ), values determined from the coda and by
comparing these values to other independent estimates.
The coda Q and ~/estimates need not be in complete agreement since the coda
7 values should estimate the Q of the Lg- or S-wave arrival, while the coda Q
represents an average Q over a much wider region t han t hat sampled by the direct
Lg or S waves.
Aki and Chouet (1975) and Rautian and Khalturin (1978) noted a frequency
dependence of Q from their study of coda waves. No such determination was
at t empt ed here. However, the development of the peak frequency versus time is
easily modified to account for the frequency dependence of Q. Assume t hat
466 ROBERT B. t t ERRMANN
Q = Qo( f / f o) P,
t hen the master curves for various values of P are found by searching for the
maximum value of the function
I ( f ) exp(-~rf 1-Pf0Pt*)
where t* now equals t / Qo and fo is the reference frequency. Thus by using the master
curves, the plot of peak frequency versus time could enable one to estimate both Q0
and P.
The Q values estimated show significant regional variations in Q across the
continental United States. Since the method used assumes a fixed Q for each region,
some thought should be given as to how to modify the method for precise region-
alization of Q. For example, the Q at DUG is intermediate between t hat at BKS and
BLA. Is this due to the fact t hat Q = 325 near DUG or because the coda waves are
traveling in regions of varying Q such t hat Q = 325 represents a pat h average? Malin
(1978) studied first-order scattering as a cause of codas. Expanding on the work of
Aki (1969), he realized t hat the locations of the first-order scatterers affecting the
coda amplitude at a given time t would be an ellipse having the earthquake and
seismograph locations as foci. Thus, at times near the direct S-wave arrival, the
ellipse would be very eccentric and only a very narrow region of the Earth' s crust
would affect the coda. At later times, the ellipse would be larger and the coda would
sample a much larger region. By knowing the dimensions of the ellipses as a function
of time, a generalized inversion for regional Q would be possible.
There is a suggestion from the BKS and BLA data, and also from the less
confident DUG and GRT data, t hat the coda source factor B ( f p ) is inversely
proportional to Q, In a qualitative sense this observation is not unreasonable, in
t hat the more efficient the medium is as a wave transmitter, the less efficient it
should be as a wave scatterer. More determinations of B ( f p ) should be made
primarily because this relates to t he excitation of coda amplitudes. If, for example,
if B (fp) is inversely proportional to Q, then a complete broadband estimate of the
coda time history could be made (Herrmann, t977), which may be of interest to
strong-motion seismology.
Aki and Chouet (1975) proposed alternative models for the source of the coda
waves instead of the surface-wave scattering model of Aki (1969). The particular
difference is t hat the exponent of t in equation (1), which is - for surface-wave
scattering, would be - 1 for scattered body waves and - if the coda arose due to
diffusion in three dimensions. The effect of using these exponents instead of the one
used would be to change the coda shape curves of Figure 2 and also the meaning of
B (f~). The coda shapes observed in this study were fit well using the exponent -.
Because of this, no at t empt was made to interpret the data using the other
exponents.
This study also provides some insight into the nature of empirical magnitude-
duration or seismic moment-duration relations. A number of investigators have
noted kinks or curvature in these duration estimates of earthquake size. Herrmann
(1975) noted this effect at larger magnitudes and moments and attributed it to the
source-spectrum corner frequency passing through the peak instrument response.
The curvature in the magnitude-duration relation was also noted by Real and Teng
(1973). In a study of earthquakes near Oroville, California, Bakun and Lindh (1977)
noted a change in curvature in the duration-magnitude relationship at ML = 1.5 and
Q ESTI MATES USI NG THE CODA OF LOCAL EARTHQUAKES 467
in t he dur at i on- moment relationship at Mo = 1.0E + 18 dyne-cm. Bakun and Li ndh
(1977) at t r i but ed this effect to t he corner frequency passing t hrough t he passband
of t he i nst rument . An al t ernat i ve explanation is t hat t he corner frequency is al ready
great er t han t he i nst rument passband, as i ndi cat ed by Bakun and Li ndh (1977), and
t hat t he curvat ure is t he resul t of t he change in coda shape due to t he effect of Q as
seen in Figure 2. Not i ng t hat duration, ~, is usual l y measur ed from t he P arrival
t i me t = r / v p , but t hat for large durations.
loglo( = t - r / v p ) ~ loglo(t)
this curvat ure may be used to infer Q. Figure 7b of Bakun and Li ndh (1977) indicates
t hat t he change in curvat ure occurs at t -- 20 sec. (The effect of cut off level will be
to change t he moment scale and not t he t i me scale.) For t he USGS i nst rument , t he
change in curvat ure occurs at about t* = 0.03 sec. Hence, t he dur at i on- moment
observat i ons yield Q = 600 at 10 Hz. Aki and Chouet (1975) est i mat ed a Q = 300 to
500 at 10 Hz for St one Canyon, California.
One final observat i on has to do wi t h at t empt s to derive a magni t ude-durat i on
rel at i on from Aki' s (1969) coda t heor y (Sut eau and Whi t comb, 1979). As shown in
Fi gures 1 and 2, t he coda shape and coda dispersion are i nt i mat el y tied t oget her by
i nst r ument response and Q. Finding a di screpancy bet ween coda magni t ude and
local magni t ude, Sut eau and Whi t comb (1979) at t r i but ed t he difference t o a Q ot her
t han 200 or to a difference in t he coefficients relating log moment to local magnitude.
However, t hese aut hors did not include t he effect of Q upon t he coda dispersion.
Assumi ng t hat t hei r i nst rument s had a response similar t o t hose used by Aki (1969)
and assuming t hat Aki' s (1969) observed dispersion was due to a Q = 150, equat i on
(4) of Sut eau and Whi t comb (1979) shoul d read
M o = B ( f v ) - ~ ( t / Q ) ~ / ~ 2 e [ ' 6 2 5 " ( t / Q ) ' / ~ ] O . 4 4 Q ~ / 2 A ( t ) I ( f p ) - ~ .
It woul d be i nt erest i ng to r epeat t he Sut eau and Whi t comb {1979) st udy using this
pr oper Q formul at i on t o det ermi ne t he value of Q which woul d bet t er fit t hei r
observat i ons.
ACKNOWLEDGMENTS
Thi s work was performed in part while t he aut hor was a visiting scientist at t he U.S. Geological
Survey, Golden, Colorado. Thi s work was supported in part by t he U.S. Geological Survey Cont ract 14-
08-0001-16708 and by t he Division of Advanced Envi ronment al Research Technology, Nat i onal Science
Foundat i on Grant ENV 76-20875.
REF ERENCES
Aki, K. (1969). Analysis of t he seismic coda of local eart hquakes as scat t ered waves, J. Geophys. Res. 74,
615-631.
Aki, K. and B. Chouet (1975). Origin of coda-waves: source, attenuation, and scattering effects, J.
Geophys. Res. 80, 3322-3342.
Bakun, W. H., C. G. Bufe, and R. M. St ewart (1976). Body-wave spectra of central California earthquakes,
Bul l . Sei sm. Soc. Am. 66, 363-384.
Bakun, W. H. and A. G. Lindh {1977). Local magnitudes, seismic moments, and coda durations for
eart hquakes near Oroville, California, Bul l . Sei sm. Soc. Am. 67, 615-629.
Bollinger, G. A. (1979). At t enuat i on of t he Lg phase and t he det ermi nat i on of mb in t he sout heast ern
Uni t ed States, Bul l . Sei sm. Soc. Am. 69, 343-355.
Chouet, G., K. Aki, and M. Tsuj i ura (1978). Regional variation of t he scaling law of eart hquake source
spectra, Bul l . Sei sm. Soc. Am. 68, 49-79.
468 ROB E RT B. HE RRMANN
Herrmann, R. B. (1975). The use of duration as a measure of seismic moment and magnitude, Bul l .
Sei sm. Soc. Am. 65, 899-913.
Herrmann, R. B. {1977). A met hod for t he synthesis of seismic coda of local earthquakes, J. Geophys. 43,
341-350.
Herrmann, R. B. (1979). Surface wave focal mechani sms for eastern Nor t h Ameri can eart hquakes with
t ect oni c implications, J. Geophys. Res. 84, 3543-3552.
Johnson, L. R. and T. V. McEvi l l y {1974), Near-field observations and source paramet ers of centr[1
California earthquakes, Bul l . Sei sm. Soc. Am. 64, 1855-1886.
Jones, F. B., L. T. Long, and J. M. McKee (1977). St udy of t he at t enuat i on and azimuthal dependence of
seismic-wave propagation in t he sout heast ern Uni t ed States, Bul l . Sei sm. Soc. Am. 67, 1503-1513.
Kurita, T. (1975). At t enuat i on of shear waves along t he San Andreas fault in central California, Bul l .
Sei sm. Soc. Am. 65, 277-292.
Malin, P. E. (1978). A first order scattering solution for modeling linear and terrestrial seismic codas,
Ph. D. Di ssert at i on, Pri ncet on University.
Nut t l i , 0. W. (1973). Seismic wave at t enuat i on and magni t ude relations for eastern Nor t h America, J.
Geophys. Res. 78, 876-885.
Nut t l i , O. W. (1978). A t i me-domai n study of t he at t enuat i on of 10-Hz waves in t he New Madri d Seismic
Zone, Bul l . Sei sm. Soc. Am. 68, 343-355.
Raut i an, T. G. and V. I. Khal t uri n (1978). The use of t he coda for det ermi nat i on of t he eart hquake
source spectrum, Bul l . Sei sm. Soc. Am. 68, 923-948.
Real, C. R. and T. L. Teng (1973). Local Ri cht er magni t ude and t ot al signal duration in sout hern
California, Bul l . Sei sm. Soc. Am. 63, 1809-1827.
Stander, W., M. Kramer, G. Fischer, S. Schaefer, and S. T. Morrissey (1976). Seismic characteristics of
sout heast Missouri as indicated by a regional t el emet er ed mi croeart hquake array, Bul l . Sei sm. Soc.
Am. 66, 1953-1964.
St reet , R. L., R. B. Herrmann, and O. W. Nut t l i (1975). Spect ral characteristics of t he Lg wave generated
by central Uni t ed St at es earthquakes, Geophys. J. 41, 51-63.
Street, R. L. and F. T. Tur cot t e (1977). A study of nort heast ern Nor t h American spectral moments,
magnitudes, and intensities, Bul l . Sei sm. Soc. Am. 67, 599-614.
Suteau, A. M. and J. H. Whi t comb (1979). A local eart hquake coda magni t ude and its rel at i on to duration,
moment M0 and local Ri cht er magni t ude ML, Bul l . Sei sm. Soc. Am. 69, 353-368.
DEPARTMENT OF EARTH AND ATMOSPHERIC SCIENCES
SAINT LOUIS UNIVERSITY
P.O. Box 8099, LACLEDE STATION
ST. LOUIS, MISSOURI 63156
Manuscri pt recei ved October 9, 1979