You are on page 1of 19

Bulletin of t he Seismological Society of America, Vol. 76, No. 6, pp.

1699-1717, December 1986


APPRAI SI NG EARTHQUAKE HYPOCENTER LOCATI ON ERRORS: A
COMPLETE, PRACTI CAL APPROACH FOR SI NGLE- EVENT
LOCATI ONS
BY GARYL. PAVLIS
ABSTRACT
F o r c o n v e n t i o n a l s i n g l e - e v e n t , n o n l i n e a r , l e a s t - s q u a r e s h y p o c e n t r a l e s t i m a t e s ,
I s h o w t h a t t h e t o t a l e r r o r i s e x p r e s s i b l e a s a l i n e a r c o mb i n a t i o n o f t h r e e t e r ms :
( 1 ) m e a s u r e m e n t er r or ; ( 2 ) mo d e l i n g e r r o r s c a u s e d b y i n a d e q u a c y o f t h e t r a v e l -
t i me t a b l e s ; a n d ( 3 ) a n o n l i n e a r t e r m. E r r o r s i n c a l c u l a t i n g t r a v e l - t i m e p a r t i a l
d e r i v a t i v e s a r e s h o w n t o h a v e no e f f e c t , p r o v i d e d a s t a b l e s o l u t i o n c a n b e f o u n d .
T h i s i s in c o n t r a s t t o l i n e a r p r o b l e ms w h e r e e r r o r s in c a l c u l a t i n g ma t r i x e l e m e n t s
c a n d i s t o r t t h e s o l u t i o n d r a s t i c a l l y . T h e e r r o r a p p r a i s a l t e c h n i q u e d e v e l o p e d h e r e
e x a m i n e s e a c h o f t h e t h r e e e r r o r t e r m s i n d e p e n d e n t l y . T h e f i r s t c a n b e a n a l y z e d
b y s t a n d a r d c o n f i d e n c e e l l i p s o i d s wi t h c r i t i c a l v a l u e s b a s e d on m e a s u r e m e n t
e r r o r s t a t i s t i c s . T h e s e c o n d c a n c a u s e c o n v e n t i o n a l e r r o r e l l i p s o i d c a l c u l a t i o n s
t h a t d e r i v e a c r i t i c a l v a l u e f r o m a n e s t i m a t e b a s e d on r ms r e s i d u a l s , t o g i v e
mi s l e a d i n g r e s u l t s . I i n t r o d u c e a n a l t e r n a t i v e e x t r e m a l b o u n d p r o c e d u r e f o r
a p p r a i s i n g s u c h e r r o r s . T r a v e l - t i m e mo d e l i n g e r r o r s a r e b o u n d e d a s t h e p r o d u c t
o f r a y a r c l e n g t h a n d a n e s t i m a t e o f t h e n o mi n a l s c a l e o f s l o w n e s s e r r o r s a l o n g
t h e r a y p a t h . T h e s e a r e u s e d t o d e r i v e a n u p p e r b o u n d on s y s t e m a t i c e r r o r s in
e a c h h y p o c e n t r a l c o o r d i n a t e b a s e d on a n o v e l b o u n d i n g c r i t e r i a . F i n a l l y , I s h o w
t h a t , f o r e r r o r s o f a r e a s o n a b l e s c a l e , t h e n o n l i n e a r e r r o r t e r m c a n b e e s t i m a t e d
a d e q u a t e l y us i ng a s e c o n d - o r d e r a p p r o x i ma t i o n . G i v e n a n u p p e r b o u n d on t h e
t o t a l l o c a t i o n e r r o r , b o u n d s on t h e t r a v e l - t i m e e r r o r i n d u c e d b y n o n l i n e a r i t y c a n
b e c a l c u l a t e d f r o m t h e s p e c t r a l n o r m o f t h e H e s s i a n f o r e a c h m e a s u r e d a r r i v a l
t i me . T h e s y s t e m a t i c e r r o r s in e a c h h y p o c e n t r a l c o o r d i n a t e d u e t o n o n l i n e a r i t y
c a n t h e n b e b o u n d e d u s i n g t h e s a m e c r i t e r i a u s e d f o r c o n s t r u c t i n g mo d e l i n g
e r r o r b o u n d s . T h i s o v e r a l l p r o c e d u r e i s c o m p l e t e b e c a u s e i t a l l o w s o n e t o
i n d e p e n d e n t l y a p p r a i s e t h e r e l a t i v e i m p o r t a n c e o f al l s o u r c e s o f h y p o c e n t r a l
e r r o r s . I t i s p r a c t i c a l b e c a u s e t h e r e q u i r e d c o mp u t a t i o n a l e f f o r t i s s ma l l .
INTRODUCTION
Some of t he most significant advances in eart h science t hat can be at t r i but ed t o
seismology have resul t ed from i nt erpret at i ons of spatial pat t er ns of eart hquake
hypocent ers. For large scale observat i ons, such as t he fact t hat eart hquakes occur
mai nl y along pl at e boundari es, t he issue of eart hquake location errors is of mi nor
i mport ance. As we t r y t o ext ract more i nformat i on from seismicity pat t erns,
however, we soon face t he issue of how precise ear t hquake locations really are.
It is well known t hat ear t hquake l ocat i ons are subj ect t o t wo quite di fferent t ypes
of errors: relative location scat t er and syst emat i c bi ases (e.g., Douglas, 1967; Dewey,
1971, 1972; Jor dan and Sver dmp, 1981). The first is caused by measur ement errors
and short range fl uct uat i ons in t he seismic velocity st ruct ure of t he eart h (Leaver,
1984) and can properl y be t r eat ed from a purel y st at i st i cal poi nt of view. The l at t er
is caused largely by i nadequat e knowledge of even t he long wavel engt h velocity
st ruct ure of t he eart h (e.g., Jor dan and Sverdrup, 1981; Pavl i s and Hokanson, 1985).
I argue below t hat it is not proper t o consider such errors as resulting from a zero
mean r andom process. Convent i onal error ellipsoids (Flinn, 1965) implicitly make
t hi s assumpt i on ( Jor dan and Sverdrup, 1981), whi ch I show here can yield very
misleading results. In t hi s paper, I i nt roduce a new, al t ernat i ve met hod for boundi ng
1699
1 7 0 0 G A R Y L . P A V L I S
syst emat i c mi sl ocat i on errors based on a component -wi se boundi ng t heory. Unl i ke
simple mat ri x vect or norm boundi ng criteria, t he bounds const ruct ed by t hi s new
approach do not t end t o be overl y pessimistic, provi ded one can specify a reasonabl e
upper bound on t he scale of slowness anomal i es wi t hi n t he net work.
Most convent i onal error est i mat es are based on a linear approxi mat i on t o a set
of nonl i near equat i ons [techniques di scussed by Tar ant ol a and Val et t e (1982) and
Rowl et t and For syt h (1984) are exceptions]. The anal ysi s I give here i ndi cat es t hat
nonl i neari t y can be vi ewed as a di fferent t ype of syst emat i c bi as whose effect is
vi rt ual l y i nseparabl e from t hat due t o modeling errors. On t he positive side, however,
calculations pr esent ed here suggest t he influence of nonl i neari t y is pr obabl y small
for most l ocat i ons and is largely cont rol l ed by modeling error biases. In any case, it
is shown t hat such errors can be bounded usi ng a second-order approxi mat i on and
a component -wi se boundi ng t heor y similar t o t hat used for boundi ng t he i nfl uence
of modeling errors.
The net resul t of t hi s paper is a procedure for apprai si ng location errors t hat I
claim is compl et e and practical. It is compl et e because it gives relatively i ndependent
measures for all t hree sources of hypocent ral error: (1) measur ement errors; (2)
velocity model errors; and (3) nonl i neari t y. The procedure is pract i cal because t he
error est i mat es descri bed are relatively easy t o calculate and require few assump-
tions. Thi s is in cont r ast t o t he compl et e error analysis of Tar ant ol a and Val et t e
(1982) which is compl et e but not necessari l y pract i cal because it is comput at i onal l y
demandi ng and requires a p r i o r i assumpt i ons about t he form of t he uncer t ai nt y in
t he velocity st ruct ure of t he ear t h (Thurber, 1986).
FUNDAMENTAL EQUATIONS AND DEFINITIONS
The dat a al ways used for ear t hquake l ocat i ons are a set of m arrival t i mes
measured from one or more phases recorded by a net wor k of sei smomet ers. I will
denot e t hi s vect or of measur ed arrival t i mes as ( t ) ~ R ~. It is rel at ed t o real i t y as
follows
( t ) = t + r l + ~ (1)
where
t -- t rue t ravel -t i me vector;
r = origin t i me (1 E R m denot es a vect or of all ones); and
= measur ement error.
Wi t h act ual data, t is unknowabl e. Inst ead, we must always rely on a mat hemat i cal
model of t cal cul at ed from some est i mat e of t he ear t h' s seismic velocity st ruct ure.
I use tmodol(&) t o symbolize t hi s vect or of t ravel times, tmoael is a funct i on of t he
est i mat ed s pat i al coordi nat es of t he hypocent er, &. A compl et e specification of t he
hypocent er, of course, also requires an est i mat e of r, ~, as well. For conveni ence,
t hi s 4-vect or will be symbol i zed as/~. Wi t h t hese definitions, I define t he residual
funct i on
r ( ] ~ ) ---- ( t ) - - t mo d e l ( ; ~ ) - - T 1 .
(2)
Usi ng equat i on (1), equat i on (2) becomes
(3)
APPRAISING EARTHQUAKE HYPOCENTER LOCATION ERRORS 1701
where emodel(X) = t -- t~odel(X) and Ar = r - ~.
All comput er i zed ear t hquake l ocat i on met hods mi ni mi ze II r ( / ~ ) I I , where I[ II
denot es a vect or nor m. The det ai l s of how t hi s proceeds var y great l y, but all met hods
are t he same i n one respect . All are i t er at i ve pr ocedur es t ha t cal cul at e a series of
correct i ons, 5hh, t o es t i mat e h as
/~ = h o + ~ ' 5 h h ( 4 )
k=l
where niter is t he t ot al numbe r of cor r ect i ons r equi r ed t o obt ai n a st abl e sol ut i on
f r om an i ni t i al guess, ho, of t he hypocent er . By st abl e I mean speci fi cal l y t ha t t he
sequence 5hk is conver gent such t h a t
II II < ( 5 )
where e is some appr opr i at el y smal l number . The 5hk are al ways cal cul at ed as
5hk = A~-~P(/~k-1) (6)
wi t h f t k-1 = ho + F,~---~ 5 h l . A' ~- I E R 4xm is a general i zed i nverse. Wi t h t he except i on
of t he nonl i near me t hod descri bed by Thur ber (1985), A~_~ is al ways cal cul at ed
di r ect l y f r om t he mat r i x of par t i al deri vat i ves z[ wi t h component s
0
X , j = ( t m o e , ) .
A,4 = l (h4 = r ) .
( j = 1, 2, 3), and
(7)
Ther e are essent i al l y as ma ny var i at i ons i n how Ah* is cal cul at ed as t her e are
l ocat i on programs. For t unat el y, we do not have t o pr es ent a di f f er ent error anal ysi s
for every possi bi l i t y t ha t exi st s. The r eason is t h a t every exi st i ng l ocat i on me t hod
can be equat ed t o a wei ght ed l east - squar es probl em. For our purposes, t hi s means
speci fi cal l y t h a t t her e exi st s a posi t i ve def i ni t e mat r i x, W E R mm for whi ch t he
sol ut i on t o t he usual equat i ons of condi t i on
= ( 8 )
sat i sfi es
5 h = A r < ~ ( 9 )
where
A= W~I, r = W~, (10)
is as i n equat i on (5), and A + denot es t he pseudoi nver se of A (e.g., Laws on and
Hans on, 1974, pp. 36-40). Equat i on (9) means t h a t / ~ is a st abl e sol ut i on i n t he
l east - squar es sense t o t he wei ght ed equat i ons of condi t i on, W A 6 h = Wi ' . Wi t h
s t andar d l east - squar es pr ocedur es like HYPO71 (Lee and Lahr , 1975), t he corre-
1702 GARY L. PAVLIS
spondence is obvious. Wi t h procedures using damped least squares (e.g., Her r mann,
1979) or t he recent l y publ i shed appl i cat i on of Newt on' s met hod by Thur ber (1985),
t he correspondence is not so obvious. However, equat i on (9) still holds because t he
purpose of bot h al gori t hms is t o pr omot e convergence of t he sequence defi ned in
equat i on (4). Bot h seek a sol ut i on t hat minimizes I] r II = [ ] ' T W 2 ] ' ] 1/2. As long as
t he l ocat i on is well const rai ned, bot h met hods will converge to t he same sol ut i on as
an equi val ent l east -squares procedure. Furt hermore, even a procedure whi ch seeks
t o minimize L1 ( I] ? ]01 = ~ ?=1 ] r~ I ) can be cast in t hi s form (Anderson, 1982). We
are also forced, on t he ot her hand, t o make t hree f undament al assumpt i ons.
1. h is well const rai ned. That is, we assume A + is not singular.
2. ]~ is t he global mi ni mum of 0] r ]1, not a local minima.
3. ]~ is not enormousl y di fferent from t he t rue hypocent er, h.
The consequences of nos. 2 and 3 are identical, but t he way t hey can arise is
different. What I mean by "enormousl y di fferent " is st at ed specifically below.
ANALYSIS OF HYPOCENTRAL ERRORS
Second- order t heory. /t i nevi t abl y cont ai ns errors t hat arise from a number of
factors. To exami ne this, expand each component of t he residual vect or in a Tayl or
series as follows (Lee and St ewart , 1981, p. 124)
4
~i(/~ + 5h) = ~,(h) + A, j Shj + fii (11)
J=l
where Av is as defi ned in equat i on (7), and ni is t he sum of all second and higher
order t erms of t he expansion. The m equat i ons of (11) can be summari zed in mat ri x
form as
f ( h + 5h) = ?(/~) + ASh + fi (12)
where t he correspondence of di fferent t er ms is obvious. Two fact s about A and !i
need t o be emphasized: (1) t hey are cal cul at ed from a mat hemat i cal model of reality,
and (2) t hey are t hemsel ves nonl i near funct i ons of h. In equat i on (12), bot h are
eval uat ed at/~.
We shall put equat i on (12) t o use by assumi ng 5h is t he difference bet ween h and
real i t y (i.e., h = /t + 5h). I f we t hen appl y t he weighting mat ri ces defi ned in
equat i ons (8) t o (10), we obt ai n t he similar form
r = ~ + ASh + n (13)
where n = W~. r and ~ are a shor t hand used t o denot e r ( h) and r ( / t ) , respectively.
Fr om equat i on (9), we see
A+r - 5h + A+n = A+~ < ~,
( 1 4 )
since we require A+A = I. However, from equat i on (5), II A +~ ]] ~ 0. Hence, using
equat i on (3), equat i on (14) can be wr i t t en as
5 h ~ A + ( emod~l + n + e)
(15)
wi t hi n t he t ol erance par amet er specified by e. (Not e emode~ is eval uat ed at h, not
t l . )
APPRAISING EARTHQUAKE HYPOCENTER LOCATION ERRORS 1703
A convent i onal anal ysi s woul d assume n ~ 0, yielding classical resul t s involving
error ellipsoids (Flinn, 1965). One of t he mai n poi nt s of t hi s paper, however, is t o
consider t he i mport ance of n. To do so, we need a met hod for calculating it.
Application of equat i on (5.68) of Lee and St ewar t (1981) t o equat i on (11) yields
5~ = 5h TH~Sh + . . . (16)
where H, is t he Hessi an matrix. Analytic forms for Hi for a const ant velocity
medi um and a t wo-l ayer model are given in a recent paper by Thur ber (1985). I use
Thur ber ' s results below t o approxi mat e 5i t o second order as
~ ~ l S h T Hi S h = (n2),.
(17)
Equat i on (15) t hen becomes
~h ~ A+(emodel + n2 -b e ) . (18)
There are t wo levels of approxi mat i on in equat i on (18): (1) t he convergence criterion
e, and (2) t he second-order approxi mat i on of n. (The l at t er is emphasi zed wi t h t he
2 subscri pt on n.)
Equat i on (18) is t he focal poi nt of t hi s paper. The first st ep is clearly t o investigate
t he limits of t hi s approxi mat i on. Thi s is done in t he following sect i on using comput er
simulations.
Comput e r si mul at i ons. In t hi s study, I chose t o consi der t he location precision of
eart hquakes in t he vicinity of t he rupt ure zone of t he 1984 Morgan Hill, California,
eart hquake (Cockerham and Eat on, 1984). A crude approxi mat i on t o t he velocity
st ruct ure in t hi s area was used t o generat e t ravel t i mes for a series of synt het i c
events. Thi s velocity model consi st ed of t wo const ant velocity quart er-spaces j oi ned
along a vertical pl ane striking nor t h 31.5 west and passi ng t hrough t he poi nt
3716 ' nor t h l at i t ude by 121o40 , west longitude (Figure 1). Synt het i c arrival t i mes
for every st at i on in t he U.S. Geological Survey Cent ral California Net wor k
( CALNET) and all Uni versi t y of California Berkel ey st at i ons wi t hi n 100 km of t hi s
poi nt (121 st at i ons) were cal cul at ed from t hi s model for t he set of event s shown in
Figure 1 using t wo di fferent quart er-space models. Velocities for t hese t wo model s
are given in Tabl e 1. The measur ement error vect or $ in equat i on (1) was si mul at ed
by using a r andom number generat or t o produce r andom sampl es from a normal
di st ri but i on wi t h zero mean and a vari ance of 0.05 sec. The same ~ was t hen added
t o each synt het i c event arrival t i me vector. The advant age of t hi s is t hat it allows
compari son of errors i nduced by $ as a funct i on of position. On t he ot her hand, it
gives a privileged posi t i on t o a set of random numbers. However, repet i t i on of t hese
results wi t h di fferent r andom vect ors indicates t he resul t s are not very dependent
on t he exact choice of ~, and t he one pr esent ed here is represent at i ve.
All event s were l ocat ed using a simple, damped l east -squares procedure similar
to t hat descri bed by Her r mann (1979), using t ravel t i mes cal cul at ed from a const ant
velocity medi um wi t h a velocity of 5.6 km/ sec. Figures 2 and 3 show t he resulting
location est i mat es. All l ocat i on est i mat es show an east ward bi as caused by approx-
imating t he quart er-space model wi t h a const ant velocity medium. The scale of t hi s
bias is ~10 km for model A event s and ~2 km for model B events. Figure 4 can be
used t o exami ne t he val i di t y of t he second-order approxi mat i on in t he cont ext of
t hese two di fferent error scales. Thi s is summari zed here by exami ni ng onl y t he
1 7 0 4 GARY L. PAVLI S
3 7
2 0 ' "
37
1 0 ' "
CMH CAOz~ / J A
' ~ ~'MHC / /
~csc ~ _..~ o ~ ' ' ' j
JST" ~A~O ~o..~ ~ . . / o ~ C C O ~
0 5 Km \
? J A L . , , ' , ' , ' ' , ' , ' , , , C A D ; , , , ~ . . . . , ,
121 5 0 ' 121 4 0 ' 121 3 0 '
A A '
o
o B
o
o
o
o
o
ooooooooooooooooooo
o
o
o
o
o
o
o
E
.t= - 10
f:k
{D
- 20
10 20 30
Distance (km)
FIG. 1. Act ual l ocat i on o f s ynt he t i c e ve nt s e x a mi ne d i n t hi s st udy. (A) Map vi ew. (B) Cross- sect i on.
St at i on l ocat i ons are marked wi t h t ri angl es and t hei r t hree character na me code. The cros s - s ect i on AA'
i s as s ho wn i n t he map vi ew. The di agonal l i ne r unni ng f rom nor t hwe s t t o s out heas t marks t he boundary
o f t he quarter-spaces us ed to generat e t he s ynt he t i c dat a arrival t i me s ( Tabl e 1).
TABLE 1
SYNTHETI C QUARTER- SPACE MODEL PARAMETERS
Southwest Northeast
Model Quarter-Space Quarter-Space
Velocity Velocity
( km/ s e c ) ( km/ s e c )
A 5. 0 6.2
B 5.5 5.7
n o r m o f t h e f i r s t - a n d s e c o n d - o r d e r p r e d i c t i o n e r r o r
E ~ = ]] ~h - A+(e~od~l + e ) ] ]
a n d
E 2 - - ]] 5h - A+(emod~ + n2 + e) ] l .
T h e 4 - v e c t o r n o r m i s c a l c u l a t e d f o r a n y 4 - v e c t o r a a s
II a I I = [a~ 2 + a22 + a32 + vo2a42] 1/2 (19)
APPRAISING EARTHQUAKE HYPOCENTER LOCATION ERRORS
2137
20'"
37
10'-
~ M H C ~ " ~
JST
0 5 Km \
Z, jA L i t t t I t ~CAD
. . . . . . . ' " ' : . . . . . . . . .
121 50' 121 40' 121 30'
1705
A
x:: -10
121
A t`
i , , ,
B
-2o T
0 10 20 30
Distance (km)
F[G. 2. Estimated locations of synthetic events from model A using 5.6 km/sec constant velocity
medium. For comparison, the map view (A) and cross-section (B) frames are identical to those shown in
Figure 1. Estimated locations are at the centers of crossing lines which are the projections of the major
axes of the conventional 95 per cent confidence ellipsoids. Critical values for these ellipsoids are based
on an F statistic as originally advocated by Flinn (1966).
where v0 is the constant velocity medium velocity. Vo is used to convert origin time
errors into an equivalent length scale so all components of a are at least measured
in the same units.
For model A, the first-order errors are extremely large. This indicates t hat in the
presence of such a large bias, conventional first-order analysis techniques can be
expected to give very misleading results. Inclusion of the second-order term improves
the picture by a factor of about 4, but E2 is not negligibly small. Ee is still almost
two orders of magnitude larger t han the convergence parameter e which was 10 m
for all tests. On the other hand, for model B, Ee is generally of the same size as e.
The only exceptions are the three shallowest sources along the vertical arm of the
jack-shaped test pattern. This failure is to be expected, however, because the depths
of these three events are poorly constrained and required auxiliary constraints to
obtain a stable solution. This is not accounted for in this theory. Furthermore,
Figure 2 shows t hat these events are located very close to station CCO. Since the
second-order term varies as R -1, where R is the hypocentral distance from a given
station (Thurber, 1985), one expects the third-order term to vary as R -2. Thus,
higher order terms would only be important for very short hypocentral distances,
which is consistent with these results.
I conclude from the results presented in Figure 4 t hat the second-order approx-
1706 GARY L. PAVLIS
37
20'"
37
10'
~ ~' MHC a CAO
CSC ~
JST
JAL Oi a I I 5Km ~CAD'~ . .
121 50' 121 40' 121 30'
v
c- -10-
A
0
-20
A f
i i I , i | ,
e
10 20 30~
D,stance (km)
Fro. 3. Same as Figure 2, but here synthetic arrival times were calculated from model B.
2 3 -
v
L~ 2-
"[3-.
A 0 5 1
04-
0 3 -
02-
01-
& O , , ~ , ,
B
J-to strike
//to stnke
D Vertical
- - 1st order errors
- - 2 nd order errors
~ i L i i '
1 3 5 7 9
Distance from Ongm (km) Distance from Origin (kin)
FIG. 4. Prediction errors of location estimates shown in Figures 2 and 3 for: (A) model A and (B)
model B. The vertical axis in each plot is the norm of the total prediction error defined by equation (19).
The first- and second-order prediction errors are as defined in the text. The actual locations of these
synthetic events lie along three mutually perpendicular lines (Figure 1). To allow the results to all be
plotted together, each event is referenced by its true distance from the "origin" defined as the point
where all three lines intersect. The perpendicular to the strike line is defined as positive to the northeast,
the parallel to the strike line is defined as positive to the southeast, and the vertical line is defined as
positive downward.
i ma t i on can gi ve a r e a s ona bl e me a s ur e of t he si ze of t he nonl i ne a r t e r m n. Th e l evel
of t he a p p r o x i ma t i o n is of t he or der of t he c onve r ge nc e p a r a me t e r e f or model B,
wher e t he h y p o c e n t r a l bi as is c o mp a r a b l e t o wha t ma y exi s t i n a c t ua l da t a f r om
t hi s ar ea. For model A, t he l evel of a p p r o x i ma t i o n is not as good, a nd we can onl y
s ay t h a t n2 is a r e a s ona bl e me a s u r e of t he si ze of n.
APPRAISING EARTHQUAKE HYPOCENTER LOCATION ERRORS 1707
ANALYZING DIFFERENT SOURCES OF HYPOCENTRAL ERROR
Int roduct i on. Having seen the limitations of the approximation given by equation
(18), we can approach the practical problem of what to do with this result. The
problem is t hat with real data the vectors emodel, n, and e are fundamentally
unknown. Our only information about them is provided by a projection of the
residual vector, ( I - AA+) i ". Unfortunately, the above analysis shows t hat ~ is a
sum of all three error terms and is evaluated at the wrong place in space-time. The
basic idea here is to use auxiliary information to appraise the relative importance
of each term. This provides a valuable error appraisal tool to provide a more
complete and realistic appraisal of location uncertainties. I now consider methods
for estimating each of these terms. The order in which they are considered is
significant.
Me as ur e me nt error t erm. Of the three terms in equation (18), e is unique in t hat
it is the only truly statistical quantity. Thus, although e may be unknowable, we
can assume we know something about its statistics [see Freedman (1966) or Leaver
(1984) for examples of particularly careful studies]. For impulsive arrivals measured
from analog records, the are approximately normally distributed with zero mean
(Buland, 1976). For emergent arrivals, the e, tend to have a distribution skewed
toward positive numbers due to a tendency to pick weak arrivals too late (Anderson,
1982). Finally, with newer computer picking methods, the distribution of e~ may be
somewhat complicated, but at least the gross details of the distribution are known
(Allen, 1982; Leaver, 1984). In any case, if we know something about the probability
density function of e, it is a standard exercise in regression analysis (see Flinn,
1965; Hoel, 1971; Jordan and Sverdrup, 1981; and their associated references) t hat
the errors induced by e can be appraised by examination of confidence ellipsoids.
For the hypocenter location problem, this amounts to outlining an ellipsoidal region
in space-time characterized by
(h - / t) ~C ~- ~( h - ~) <=~ (20)
Ch = A+C( A+) T where C = cov(e) = E[ e e T] is the data error covariance with
E[ ] denoting expectation value. K~ is a critical value which depends upon the
distribution of e and the confidence level, a, specified. For example, if the elements
of e are independent samples from a normal distribution with zero mean and
variance 2, ~ for a = 95 per cent is given by K~ 2 = 9.488 a2 where 9.488 is the 95
per cent critical level for a x e distribution with 4 degrees of freedom (e.g., Hoel,
1971, p. 392).
Confidence ellipsoids constructed from equation (20) are a classical result, but
there is a major difference between the general concept of equation (20) and what
is conventional in seismology. Since the original work of Flinn (1965), the standard
approach has been to scale the data covariance, C, by the rms residuals. That is,
one sets C = s2L where
s 2 _ II U II
m - 4 " ( 2 1 )
Then, under an assumption of normally distributed errors,
Ka z = 4s2F.(4, m - 4) (22)
1708 GARY L. PAVLIS
where F, (4, m - 4) is an F statistic with 4 and m - 4 degrees of freedom for the
critical level ~ (Flinn, 1965). This almost always leads to error estimates t hat are
considerably larger t han those based on measurement errors alone. The conven-
tional wisdom is t hat this is justified, as it is one way to account for what I have
called emode~ and n. I hope to convince the reader in the following sections t hat the
conventional wisdom is wrong and a different analysis is necessary.
M o d e l e r r o r t e r m. Unlike e, it is not really proper to consider emodo~ as a statistical
quantity. The failure of the statistical model to predict systematic biases in hypo-
center location estimates is well documented (e.g., Bolt e t al . , 1968; Brown and Lee,
1971; Morrison e t al . , 1976; Uhrhammer, 1981). Nonetheless, it is useful to consider
results from the synthetic data tests considered above as they demonstrate at least
part of the reason for the failure of the statistical model to correctly predict errors
due to emodel.
Figure 2 shows the principle axes of the 95 per cent confidence ellipsoids defined
by the standard F statistic given in (22). None of the ellipsoids enclose the true
locations of any of the events examined here. This failure is not due to measurement
errors as the error induced by e is less t han 30 m for every event. The best
explanation is t hat the application of equation (22) ignores a well-known caveat of
the properties of the F distribution summarized in the following quote from an
introductory statistics book
Unfortunately, the preceding test (F test) is not reliable if X and Y do not possess
normal distributions. Just as in the case of applying the x 2 distribution to test a
single variance, the F distribution may be highly unreliable if X and Y possess
distributions whose fourth moments are considerably larger t han for a normal
distribution (Hoel, 1971, p. 273).
For the case considered here, t hat turns out to be precisely the problem. Figure 5
shows the distribution of individual components of ~, emodel, e, and n for all model
B events. (Model A results are similar except the time scale is roughly an order of
magnitude larger. These are not shown for the sake of brevity.) If one did not look
carefully at the residuals (the only information we have in real life), one might be
led to think the residuals were Gaussian and jump to the conclusion t hat e~odel was
also Gaussian. In reality, we see the components of emod~l have a distribution t hat
is basically bimodal due to the fact t hat the data were generated from a two quarter-
space model. A bimodal distribution is an example of a distribution with a large
fourth-order moment. This is obscured in the distribution of ~ by the added influence
of e and n, causing the distribution of ~ to be nearly Gaussian. Nonetheless, the
fourth-order moment of the distribution of ~ was apparently sufficiently large to
make the F distribution error ellipsoids unreliable.
A statistician would say t hat the error ellipsoid problem described above was
caused by the violation of our assumption t hat the residuals were Gaussian, and we
should take steps to account for the actual distribution of ~. I would argue t hat
would accomplish little and misses the basic point. That is, there are two kinds of
errors in hypocenter estimates: (1) systematic biases and (2) scatter in the relative
location of events. The former is dominated by emod~l. The latter is largely controlled
by e and short spatial wavelength structures in emod~l, but relative location scatter
can also be induced by interaction with e~odel through variations in effective array
geometry caused by arrivals not being recorded at all stations (e.g., Pavlis and
Hokanson, 1985). I assert t hat these two kinds of errors should be appraised
differently. Relative location errors can and should be appraised by error ellipsoids
based on known properties of the measurement error statistics. Systematic biases
APPRAISING EARTHQUAKE HYPOCENTER LOCATION ERRORS 1709
1.0
' 0 8 -
(D
0 " 0 6 -
~D
U-
>~ 0 4 -
n" 0 , 2 -
OOl
- 0 3 - 0 ' 1
R e s i d u a l s M e a s u r e m e n t Er r or
011 ' 0 3
1 0 -
0 8 -
0 6 -
0 4 -
0 2 -
O0
- 0 3 - 0 1 0 1 0 3
1 0
o > , 0 8 -
~ 0 6 -
u_
~ 0 4 -
~ 0 2 -
O0
- 0 3 - 0 1
Modeling Error
1 0
0 6
0 4
O2
, 0 0
0 1 3 - 0 3
E r r o r ( sec)
Second Order Errors
- 0 ' 1 0'1 0 3
Error ( sec)
FIG. 5. Frequency distributions of different travel-time error terms from model B synthetic data.
Distributions were derived from all 53 events (5413 total travel times). Each plot is normalized to peak
frequency to facilitate comparisons. For the second-order, nonlinear errors, only two points fell outside
in the first three bins (interval 0.01 sec). The two extremes are beyond the dynamic range of the plot
and occur at times ~0.04 and 0.06, respectively.
s h o u l d be t r e a t e d s e p a r a t e l y . I n o w a d v a n c e a n e w me t h o d o l o g y f o r b o u n d i n g
s y s t e ma t i c mi s l o c a t i o n s .
T o b o u n d emoa~l, we n e e d t o r e c o g n i z e i t s r e l a t i o n t o t h e s e i s mi c v e l o c i t y s t r u c t u r e
o f t h e e a r t h. A s s u m i n g r a y t h e o r y a n d g i v e n t h e d e f i n i t i o n o f emod~l [ e q u a t i o n ( 3) ],
i t f o l l o ws t h a t
I
( e m o d e l ) z = ~ t r u e ds - U m o d e l ds
t ~ ~j F modeL

| ( Ut r u e - - U r n o d e l ) ds
J r
t ~ odel
(23)
wh e r e utru~ i s t h e a c t u a l s l o w n e s s a s a f u n c t i o n o f p o s i t i o n w i t h i n t h e e a r t h a n d F f f "
i s t h e a c t u a l r a y p a t h c o r r e s p o n d i n g t o t h e p h a s e d e f i n e d a s a r r i v a l i. S i mi l a r l y ,
F, md~ i s t h e r a y p a t h c o r r e s p o n d i n g t o t h e s a me ar r i val , b u t b a s e d o n a mo d e l o f
t h e s l o w n e s s f u n c t i o n , U~nodel. I n t h e e a r t h , r ~ t r u e a n d F, rede1 c a n be e x p e c t e d t o n o t
di f f e r e n o r m o u s l y a n d t h e ~ i n ( 23) i s r e a s o n a b l y s a t i s f i e d . S i n c e r o c k s l o w n e s s i s
a l wa y s f i ni t e , t h e i n t e g r a l i n e q u a t i o n ( 23) c a n be b o u n d e d . T h i s y i e l d s t h e b o u n d
I (~modo~),[ =< ~ , Au ( 24)
wh e r e Au i s a n u p p e r b o u n d o n I Utrue - - Umode~ [ a n d ~, i s t h e r a y p a t h l e n g t h , f ds,
a l o n g F, mdel. T h u s , e a c h c o m p o n e n t o f emodel c a n be b o u n d e d i n t e r ms o f a c o m m o n
1710 GARY L. PAVLIS
scale, Au, and t he vari abl e sL. A t heor em proven in t he Appendi x t hen gives t hat
t he error in h due t o emodel, 5hmooel, is bounded by
i ( S h m o d ~ l ) ~ ] ~ A u ~ IA+J si ( 2 5 )
j =l
where s = W~ as in equat i on (10).
The boundi ng criteria given in (25) is appl i ed t o t he dat a from model B
in Figure 6. (Similar resul t s were obt ai ned for model A, but I again omi t t hem
for brevi t y' s sake.) The bounds shown are cal cul at ed from equat i on (25) using
Au = (5.5 -1 - 5.6 -1) = 0.00325 sec/ km. Not e t hat t he est i mat ed bounds cont ai n t he
t rue hypocent ers of t he correspondi ng event s and are not overl y pessi mi st i c for any
of t he spat i al coordi nat es. Thi s is in sharp cont r ast t o bounds based on more
convent i onal mat ri x boundi ng met hods whi ch I also tried. That is, given t he bounds
in equat i on (24), one could also bound ~hmodol as
II ~ h HI --- II A + HIHIs II A u
for any consi st ent mat ri x and vect or norm. However, bounds cal cul at ed t hi s way
using t he same Au as above are ~60 km for L1 and ~10 km for L2 and L~, whi ch
renders t hem essent i al l y useless. On t he ot her hand, one must recognize t hat t he
t est of t he boundi ng criteria given in equat i on (25) is not ent i rel y fair either. The
chosen scale for Au is a nearl y perfect est i mat e, as it is onl y slightly great er t han
t he act ual upper bound (slowness differences bet ween quart er space velocity of 5.5
km/ sec and const ant velocity model vel oci t y of 5.6 km/ sec). Such a precise bound
was available onl y because t hese were synt het i c data. Wi t h real data, how t o
properl y choose Au is an open question. A promi si ng t echni que for est i mat i ng Au
is t o use nonpar amet r i c st at i st i cal t echni ques such as t he boot st rap, j ackkni fe, delta,
or cross-val i dat i on met hods (Efron, 1981; Ef r on and Gong, 1983). Thi s approach
could cast Au as a st at i st i cal quant i t y, but has t he i mpor t ant advant age of not
requiring any a pr i or i assumpt i ons about t he di st ri but i on of Au. The cost, however,
is a very high comput at i onal overhead, whi ch may prove impractical. Thi s or ot her
al t ernat i ve schemes for est i mat i ng Au are subj ect s for fut ure work.
No n l i n e a r error t er m. Even in his original work on t he subject, Fl i nn (1965)
recognized t he pot ent i al pr obl em posed by nonl i neari t y in apprai sal of hypocent ral
location error est i mat es. Fl i nn' s ideas about how t o correct error ellipsoids t o
account for nonl i near effects, however, have been largely ignored as far as I can
judge. Ot her not abl e at t empt s t o provi de more compl et e error cal cul at i ons t hat
account for nonl i neari t y are given by Tar ant ol a and Val et t e (1982) and by Rowl et t
and For syt h (1984). The former is part i cul arl y novel and complete, but its pract i -
cality for rout i ne use is quest i onabl e (Thurber, 1986). In t hi s paper, I propose a
simple, but f undament al l y di fferent met hod for appraising t he pot ent i al influence
of nonl i neari t y based on t he second-order approxi mat i on of equat i on (17) and a
boundi ng cri t eri on similar t o t hat used in equat i on (25) above.
If we knew 5h a pri ori , equat i on (17) provi des a means for est i mat i ng n as shown
by t he resul t s in Figure 4. In real life, however, 5h is unknown. On t he ot her hand,
we usual l y know at least somet hi ng about its scale,
P = HI5x lB. (26)
APPRAISING EARTHQUAKE HYPOCENTER LOCATION ERRORS 1711
36 20' -
I ,
36" 10'
0 5 Km
I I I I t
+ Es t i mat ed
h y p o c e n t e r
o Tr u e
h y p o c e n t e r
A
e- -10
a
-20
121 50'
A
0 . . . . i
o t -
| | i |
4 -
121 = 40'
| s a I t I
121 30'
A'
i | I t |
B
o 4 -
o 4 - 0
+
+ Es t i mat ed
h y p o c e n t e r
o Tr u e
h y p o c e n t e r
, t | t , I ' i | . , w u 1 ' ' t t i w w u
10 20 30
Di s t a n c e ( k i n )
FIG. 6. Extremal bound estimates for model B results. (A) Map view and (B) cross-sections are the
same frame as in Figure 1. Boxes are coordinate bounds on systematic mislocation errors caused by
modeling errors. For clarity of presentation, results are only shown for every fourth event along the
horizontal profiles, and no events along the vertical line are shown. Cross-sections are projections of
extreme edges of these parallel-piped-shaped regions. Projected edges are omitted for clarity.
Then, it follows from equat i on (17) t hat wi t hi n t he limits of t he second-order
approxi mat i on and our i mperfect knowledge of p, t hat each component of n is
bounded as
5, ~ p2 II H, Jl (27)
1712 GARY L. PAVLI S
where II/-/~ I[ implies t he L2 or spect ral norm of H, because t he second-order t erm in
equat i on (17) is quadratic. The scale, p, in equat i on (26) involves onl y t he spatial
coordinates, as t he Hessi an does not depend on t he origin t i me (Thurber, 1985).
We can t hen again appl y t he t heor em in t he Appendi x t o bound t he error, 5h~, due
to nonl i neari t y as
II (Sh~), II < lp2 ~ I A~ III ~ II. (28)
2=1
These bounds, like t hose used for emodel, are based on a scale factor, p, and a set of
variable factors, []//J II, whi ch var y widely for di fferent arrivals. The scale fact or is
especially crucial in t hi s case, however, as t he cal cul at ed bounds depend on p2.
Therefore, before proceeding, it is necessary t o discuss how p can be est i mat ed.
Normal l y, it can be expect ed t hat t he t ot al error is domi nat ed by t he effects of
emoted. In t hi s case, t he most obvious est i mat e of p is ]] (~hmodel I] where ~hmodel is t he
vect or of bounds in equat i on (25). Thi s will cert ai nl y yield a set of absol ut e bounds.
Unf or t unat el y, t he resul t s will t end to be overl y pessimistic. A graphic i l l ust rat i on
of t hi s is shown in Figure 7. The set of bounds labeled "c" were const r uct ed t hi s
way. Figure 7 shows onl y resul t s for t he y coordi nat e of t he series of synt het i c
event s perpendi cul ar to t he st ri ke of t he "fault" shown in Figure 1, but t hese resul t s
are represent at i ve. The bounds are so large t hat t hey are not onl y useless, but also
very misleading. Thi s occurs because we are using t he square of one bound t o
const ruct anot her, whi ch i nevi t abl y gets t oo pessimistic. I experi ment ed wi t h
est i mat i ng p from rms residuals, but t hi s had t he opposi t e problem. We saw above
t hat error ellipsoids scaled using rms under est i mat ed t he syst emat i c bi as in t hese
locations. For t he same reason, any est i mat es of p derived from rms according t o
any convent i onal recipe will under est i mat e p leading t o an under est i mat e of t he
bounds due t o n.
An est i mat e for p t hat proved adequat e, al t hough still not ent i rel y satisfying, was
a crude guess one mi ght be able t o make wi t h t he real dat a t hat t hese synt het i c
dat a model s represent . That is, all event s are offset perpendi cul ar t o t he "fault" by
epi cent ral di st ances of ~9 and ~1. 5 km for model s A and B, respectively. If we
pr esume t hi s is represent at i ve for one degree of freedom of t he solution, t hen a
reasonabl e guess for t he scale p is 9~f3 km and 1.5 ~ km for model s A and B,
respectively, since 5x has t hree degrees of freedom. Usi ng t hese val ues results in
t he bounds shown in Figure 7 labeled "b." These are still somewhat overl y pessi-
mistic, bei ng approxi mat el y an order of magni t ude larger t han t he act ual error. Thi s
is largely caused by t he fact t hat t hese est i mat es of p are not perfect either. Ideal
values of p are val ues of ~10 and - 1. 5 km for model s A and B, respectively. Hence,
if we knew p a p r i o r i , we woul d be able t o reduce t he bounds shown in Figure 7 by
a fact or of 3. In t hat case, t hese bounds woul d do a remarkabl y good j ob of measuri ng
t he pot ent i al i mpact of nonl i neari t y in t he solution. It is i mpor t ant t o recognize
t hat t hi s is a maj or advant age of t he boundi ng criterion based on equat i on (28).
That is, provi ded p can be est i mat ed accurat el y by some auxiliary means, t he bounds
provi ded by (28) can be expect ed t o be reasonable. Thi s is again in cont r ast t o
simpler schemes I fi rst t ri ed using st andar d boundi ng met hods based on mat ri x
norms (i.e., I I 5hn II =< I I A + JRII n ]1 for any pai r of consi st ent mat ri x and vect or norms,
provi ded ]l n ]] is an upper bound on t he t rue nor m of n). These always gave t erri bl e
results even when p was chosen exactly.
APPRAISING EARTHQUAKE HYPOCENTER LOCATION ERRORS 1713
30 O0
20 00-
10 00-
o O0-
LU
--10 O0
- 2 0 O 0
- 30 00
0 60-
0 40-
0.20 -
O.O0-
LU
--0.20-
- 0 4 0 -
- 0 60-
b
_ _ _ _ _ - - - - - - - - - - - - A
, I , I , i , I I , I t ] , l ,
- 7 - 5 - 3 - 1 1 3 5 7 9
B
b
b
C
- 1 1 o 5 7
Di stance f r om Ongm ( km)
FIG. 7. Example of nonlinear error bounding estimates. (A) Model A and (B) model B. Results shown
are for the east-west epicentral coordinate for the line of events oriented perpendicular to the strike
(northeast to southwest) in Figure 1, which are representative. The abscissa is as described in Figure 4.
The actual nonlinear errors are plotted as discrete points as triangles labeled a. The bounds constructed
from the extremal bounds shown in Figure 6 are those labeled as c, Bounds constructed from the guess
described in the text are those labeled as b. Note the extreme difference in the scale between results for
models A and B.
DISCUSSION AND CONCLUSIONS
The first major result of this paper is equation (18). It is significant t hat this was
not the result I originally expected when I began this study. My original intent was
to consider the impact of the fact t hat the matrix of partial derivatives [A defined
by equations (7) and (10)] was calculated from a model of the earth' s velocity
structure in the same way the travel time is. It is well known t hat errors in
calculating coefficients of a matrix cause errors in solutions of linear equations.
Such errors are commonly appraised by matrix peturbation analysis methods (for
least-squares problems, see, e.g., Lawson and Hanson, 1974, pp. 41-52) to bound
the possible influence of computational errors in real computers with a finite
precision. The analysis leading to equation (18) shows t hat even though A is not
calculated perfectly, it makes little difference as long as a stable solution can be
found. The basic reason is t hat in a linear problem, A is fixed; here A is variable.
We try to minimize I[ r [] by a sequence of steps given in equation (6). Each step is
a linear one designed to minimize the norm N r - A S h J] based on the current values
of r and A, which vary from step to step. The net result is t hat when the solution
1714 GARY L. PAVLIS
converges, the only relevant measure of A is the current one. Hence, the fact t hat
it may be wrong due to inadequacies in calculating ray takeoff angles is irrelevant.
Equation (18) shows t hat hypocentral errors are a composite of three terms: (1)
measurement error; (2) what I have called modeling errors; and (3) a nonlinear
term. Of these, I claim only the first should really be viewed as a statistical quantity.
This term can be appraised with confidence ellipsoids, but the scale of the process
should be determined from independent studies of measurement errors such as t hat
by Freedman (1966) or Leaver (1984). Scaling confidence ellipsoids by rms following
Flinn (1965) is a dangerous step t hat can produce misleading results as shown by
the results given in Figure 2. This occurred because the synthetic examples studied
here model a feature of most real data. That is, the errors are dominated by modeling
errors. In this paper, I introduced an alternative means for appraising the influence
of modeling errors using a component-wise bounding criteria based on a theorem
proven in the Appendix. These bounds are based on arrival-dependent ray arc
lengths and a common scale factor, Au, which is a bound on the average slowness
along a given ray segment. The result is a parallelepiped-shaped bounding region
for each earthquake location whose relative dimensions are fixed and whose absolute
scale depends linearly on Au.
The third source of earthquake location errors, nonlinearity, is almost always
ignored in conventional location procedures. The analysis presented here indicates
nonlinearity can be viewed as an additional component of the systematic location
bias superimposed on top of t hat caused by modeling errors. Evidence from the
synthetic examples presented here suggests t hat such errors can be approximated
to adequate precision for any reasonable location estimate using a second-order
approximation. Using a second-order approximation and a component-wise bound-
ing procedure similar to t hat used for modeling errors, the expected size of the
nonlinear error can be bounded. These bounds are constructed from the spectral
norm of the Hessian for each arrival (Thurber, 1985) and a common scale factor, p,
which is a guess of an upper bound on the total hypocentral error for t hat event.
Again, these bounds form a parallelepiped-shaped region of space whose relative
dimensions are fixed. In this case, however, the scale of the bounding region is
determined by the square of the scale factor p. Consequently, the bounds estimated
this way are reasonable only when p is chosen reasonably.
How to best estimate the scale factors Au and p used in the bounding procedures
described here is an open question. Au is probably best estimated by nonparametric
statistical methods (Efron and Gong, 1983) or from a random media viewpoint as
advocated by Leaver (1985). The best approach for p in many cases is probably to
make a reasonable guess of its size based on other information. For example, if a
group of earthquakes appear to be systematically offset a distance d from a mapped
fault trace, p could be estimated from d as described above. Lacking such informa-
tion, p may be estimated from some measure based on rms residuals or the extremal
bounds on systematic biases described here. In doing so, one must recognize,
however, t hat the former may underestimate p, and the latter will always overesti-
mate p. In any case, one should recognize the main advantage of a set of bounds
based on a common scale factor. If one decides the original scale factors used to
calculate the bounds were in error, recalculating them based on the revised scale is
trivial.
I would claim t hat one of the most significant facts about the error appraisal
techniques described here is their practicality. The error estimates are easy to
calculate, and the results are easy to understand at a glance. Furthermore, imple-
APPRAI SI NG EARTHQUAKE HYPOCENTER LOCATI ON ERRORS 1715
ment i ng t hem requires onl y mi nor modi fi cat i ons t o most location programs. Re-
scaling confidence ellipses t o reflect onl y measur ement errors is trivial and is
al ready an opt i on in at least one common location program I am aware of
( HYPOI NVERSE, Klein, 1978). Calculating unscal ed model and nonl i near bounds
requires onl y a mi nor modi fi cat i on t o any program t hat explicitly calculates A +.
One t hen onl y has t o calculate t he vect or of component bounds defi ned in equat i ons
(24) and (27). The model error bound est i mat e requires t he calculation of ray arc
lengths. For local net works, t hi s cal cul at i on can pr obabl y be approxi mat ed ade-
quat el y as t he t ot al source receiver spatial separat i on. For teleseismic locations, it
would pr obabl y require a table. The nonl i near error bound cal cul at i on requires
calculation of t he spectral nor m of t he Hessi an mat ri x for each arrival. Analytic
forms are present l y known onl y for const ant velocity medi a and a layer over a half-
space model (Thurber, 1985). These second-order derivatives could pr esumabl y be
cal cul at ed numeri cal l y for any arbi t rary model, but I expect t hat is pr obabl y
unnecessary in most cases. Thur ber poi nt s out t hat t he Hessi an gives a measure of
local wave front curvat ure. My experience from t hi s work is t hat t hi s causes t he
nonl i near errors t o be domi nat ed by t he one or t wo nearest st at i ons t o t he source
where t he wave front curvat ure is largest. At nearby st at i ons, t he wave front s will
not differ t hat dramat i cal l y from a const ant velocity medium. Therefore, I suspect
t he use of Hessi ans appropri at e t o const ant velocity medi a woul d normal l y give
reasonabl e results.
Finally, it is i mpor t ant t o st ress t he t wo l i mi t at i ons of t he t echni ques di scussed
here. First, t he error anal ysi s pr esent ed here is based on t he f undament al assumpt i on
t hat t he location est i mat e is well const rai ned. That is, I assumed no auxiliary
const rai nt like fixing t he dept h is necessary t o obt ai n a st abl e solution. In t he case
of poorl y const rai ned locations, one pr obabl y shoul d resort t o t he t echni ques of
Tar ant ol a and Val et t e (1982) or Rowl et t and For syt h (1984). Second, t hi s error
analysis is what might be called a single-event t heory. That is, each event is t r eat ed
as an i ndependent ent i t y, whi ch is an unst at ed, implicit assumpt i on in most rout i ne
locations. In fut ure work, I hope t o ext end t hese ideas t o measures of relative
location preci si on of a set of ear t hquake locations. That probl em is compl i cat ed by
t he fact t hat di fferent ear t hquakes will, in general, be recorded by di fferent st at i ons
wi t h di fferent levels of preci si on for correspondi ng arrivals. How t hi s dat a het ero-
geneity i nt eract s wi t h t he biasing influence of modeling errors is a nontrivial
question.
ACKNOWLEDGMENTS
Special t hanks are due to Anne Tur ner for handl i ng t he seemingly endless generations of this paper
t hat emerged from t he word processor and to Ki m Sowder for t urni ng my crude comput er graphics into
works of art. I t hank Robert Uhr hammer and Cliff Thurber for providing very constructive comments.
I t hank Robert Uhr hammer especially for suggesting t he possibilities of nonparamet ri c statistical
methods. Thi s work was sponsored by t he Nat i onal Science Foundat i on under Gr ant EAR83-19418 and
by t he U.S. Geological Survey under I NTER 14-8-1-G1084.
REFERENCES
Allen, R. (1982). Aut omat i c phase pickers: t hei r present use and future prospects, Bull. Sei sm. Soc. Am.
72, $225-$242.
Anderson, K. R. (1982). Robust eart hquake location using M-est i mat es, Phys. Ear t h Planet. Int eri ors
30, 119-130.
Bolt, B. A., C. Lomnitz, and T. V. McEvi l l y (1968). Seismological evidence on t he tectonics of central
and nort hern California and t he Mendoci no escarpment, Bull. Seism. Soc. Am. 58, 1725-1768.
Brown, R. D., Jr. and W. H. K. Lee (1971). Active faults and prel i mi nary eart hquake epicenters (1969-
1970) in t he southern part of t he San Francisco Bay region, U.S. Geol. Surv. Map MF-307.
1716 GARY L. PAVLIS
Buland, R. (1976). The mechanics of locating earthquakes, Bull Seism. Soc. Am. 66, 173-187.
Cockerham, R. S. and J. P. Eaton (1984). The April 24, 1984, Morgan Hill earthquake and its aftershocks:
April 24 through December 30, 1984, in The 1984 Morgan Hill, California Earthquake, California
Division of Mines and Geology, Spec. Publ. 68, J. Bennett and R. Sherburne, Editors, Sacramento,
California.
Dewey, J. W. (1971). Seismicity studies with the method of joint hypocenter determination, Ph.D.
Dissertation, University of California, Berkeley, California.
Dewey, J. W. (1972). Seismicity and tectonics of western Venezuela, Bull. Seism. Soc. Am. 62, 1711-
1751.
Douglas, A. (1967). Joint epicentre determination, Nature 215, 47-48.
Efron, B. (1981). Nonparametric standard errors and confidence intervals, Canad. J. Statist. 9, 139-172.
Efron, B. and G. Gong (1983). A leisurely look at the bootstrap, the jackknife, and cross-validation, Am.
Statist. 37, 36-48.
Flinn, E. A. (1965). Confidence regions and error determinations for seismic event location, Rev. Geophys.
3, 157-185.
Freedman, H. (1966). The "little variable factor": a statistical discussion of the reading of seismograms,
Bull. Seism. Soc. Am. 56, 593-604.
Herrmann, R. B. (1979). FASTHYPO--A hypocenter location program, Earthquake Notes 50, 25-37.
Hoel, P. G. (1971). Introduction to Mathematical Statistics, Wiley, New York, 409 pp.
Jordan, T. H. and K. A. Sverdrup (1981). Teleseismic location techniques and their application to
earthquake clusters in the south-central Pacific, Bull. Seism. Soc. Am. 71, 1105-1130.
Klein, F. W. (1978). Hypocenter location program HYPOINVERSE. Part 1. User's guide to versions 1,
2, 3, and 4, US. Geol. Surv., Open-File Rept. 78-694, 113pp.
Lawson, C. H. and R. J. Hanson (1974). Solving Least Squares Problems, Prentice-Hall, Englewood
Cliffs, New Jersey, 340 pp.
Leaver, D. S. (1984). Mixed stochastic and deterministic modeling of the crustal structure in the vicinity
of Mount Hood, Oregon, Ph.D. Dissertation, University of Washington, Seattle, Washington.
Lee, W. H. K. and J. C. Lahr (1975). HYP071 (revised): a computer program for determining hypocenter,
magnitude, and first motion pattern of local earthquakes, U.S. Geol. Surv., Open-File Rept. 75-311.
Lee, W. H. K. and S. W. Stewart (1981). Principles and applications of microearthquake networks, in
Advances in Geophysics, Academic Press, New York, 293 pp.
Morrison, P. W., B. W. Stump, and R. Uhrhammer (1976). The Oroville earthquake sequence of August
1975, Bull. Seism. Soc. Am. 66, 1065-1084.
Pavlis, G. L. and N. B. Hokanson (1985). Separated earthquake location, J. Geophys. Res. 90, 12777-
12789.
Rowlett, H. and D. W. Forsyth (1984). Recent faulting and microearthquakes at the intersection of the
Vema fracture zone and the Mid-Atlantic Ridge, J. Geophys. Res. 89, 6079-6094.
Stewart, G. W. (1973). Introduction to Matrix Computations, Academic Press, New York, 441 pp.
Tarantola, A. and B. Valette {1982). Inverse problems = quest for information, J. Geophys. 50, 159-170.
Thurber, C. H. (1985). Nonlinear earthquake location: theory and examples, Bull. Se~sm. Soc. Am. 75,
779-790.
Thurber, C. H. (1986). Analysis methods for kinematic data from local earthquakes, Rev. Geophys. 24
(in press).
Uhrhammer, R. A. (1981). Northern and central California earthquakes, 1979, in United States Earth-
quakes, 1979, C. W. Stover, Editor, U.S. Geol. Survey, Golden, Colorado, and National Oceanic
Atmospheric Administration, Boulder, Colorado, 106-108.
DEPARTMENT OF GEOLOGY
INDIANA UNIVERSITY
BLOOMINGTON, INDIANA 47405
Manuscript received 25 March 1986
APPENDIX
The c ompone nt - wi s e er r or boundi ng cr i t er i a devel oped i n t hi s paper depends
upon t he f ol l owi ng t heor em. The pr oof is anal ogous t o t he cons i s t ency pr oof for t he
L1 a nd L~ mat r i x nor ms gi ven i n St ewar t (1973, p. 179). The i nt er es t ed r eader
shoul d compar e t he t wo t o see why c ompone nt - wi s e boundi ng l eads t o a less
pessi mi st i c bond t ha n a si mpl e one based on L1.
APPRAISING EARTHQUAKE HYPOCENTER LOCATION ERRORS 1717
Th e o r e m: Gi v e n A E R m~ a n d a v e c t o r x E R n wh o s e c o mp o n e n t s c a n be
b o u n d e d as
[xj [ =-< bj, (A1)
t h e n t h e c o mp o n e n t s o f t h e v e c t o r y E Rm d e f i n e d as y = A x a r e b o u n d e d b y
l y~l <-- ~: [ A~j [ bj . (A2)
j =l
Pr oof : T h e s t a t e me n t y = A x me a n s
Th u s ,
y , = ~ A , j x j . (A3)
J=l
lYe[ = j =~Ai l x j < j=x ~ [ A~ j [ ] x j [ .
Us i n g (A1), i t f ol l ows d i r e c t l y t h a t
l y~l - - < ~ I A ~ l b j .
J =l