Premonitory seismicity patterns near Vancouver Island, Canada

Premonitory seismicity patterns near Vancouver Island, Canada

Tectonophysics, Elsevier 167 (1989) 299-312 Science Publishers 299 B.V., Amsterdam Premonitory in The Netherlands seismicity D. BROWN pattern...

1MB Sizes 0 Downloads 57 Views

Tectonophysics, Elsevier

167 (1989) 299-312

Science Publishers

299

B.V., Amsterdam

Premonitory

in The Netherlands

seismicity

D. BROWN

patterns near Vancouver Canada

‘, Q. LI ‘, E. NYLAND

Island,

’ and D.H. WEICHERT

of Physics, lJniuersit_v of Alberta Edmonton,

’ Department ’ Geologml

- Printed

2

Alta. T6G 2JI (Canada)

Survey of Canada, Pacific Geoscience Centre, Box 6000, Sidney, B.C. V8L 4B2 (Canada) (Received

November

4, 1987; revised version

accepted

August

18. 1988)

Abstract Brown,

D., Li, Q., Nyland,

Canada.

E. and Weichert,

In: M.J. Berry (Editor),

The search

1989. Premonitory

Hazard

Assessment

seismicity

patterns

for large seismic events requires

and Juan

de Fuca

plates

M z 6.5. The

strongest

earthquakes

three

complete

generalized

for M 2 2.5 are preceded

sense depicted

by the algorithm

been preceded

by activation,

occupy

30% of the total

about

significant

similarity

by activation “M8”

suggested

have

of the earthquake

time-space

domain

the earthquake

of 5 years and 100 km. Such anticipation

could justify

attempts

to make short-term

using patterns

There is evidence

is a natural

consequence

predictions.

Analysis

for the coupling

of seismicity

of the model of a lithosphere

Introduction We seek to diagnose the approach of the strongest earthquakes in that part (Fig. 1) of the Pacific North American plate boundary where the Farallon plate fragments into the Explorer and the Juan de Fuca plates which then subduct under the

them. In the

catalog

of the world

future

strong

some precautionary of quiescence

where

earthquakes measures

restrict

has

been

range, In the 6.0 may have

(TIP) of a strong alarms”.

to be

appears

One event of magnitude

regions

as an assemblage

to define

flow in the lower magnitude probability

I!land,

167: 299-312.

the seismicity

are no “false

in the subduction

Vancouver

this threshold

there

to be tested, of anticipating

accuracy earthquake.

and

flow here and in other

which remains

since

by Soviet scientists.

considered

threshold

Canada)

occurred

near

Tectonophysics,

an a-priori Island,

but two others were not. The times of increased

between

tested. There is a possibility,

(the area of Vancouver

(M z 6.0) which

patterns

and Prediction.

area of the Explorer acceptably

for premonitory

D.H.,

Earthquake

This

algorithm

earthquake implies

a

MX was

in this area with an

and could be usef 11 for

the location

of the anticipated

zone with that above it. Such inter2 ction

of interacting

blocks.

the present state of seismicity analysis as a means of describing the phenomenon of earthquake flow and of suggesting hypothetical regularities. Empirical data are necessary to formulate such hypotheses,

and they are as insufficient

in ..he area of

in-

Vancouver Island as they probably ,ire everywhere. As a consequence we create hypotheses which might apply in diverse regions and use them

northwest of the of the southwest

to study the similarities in the situations before strong earthquakes. The approach se.lrches for

coast of Canada extending to the north end of Vancouver Island. For convenience in this paper we refer to this zone as the Vancouver Island area. Since there is no comprehensive theory which would define a priori an algorithm of earthquake prediction, a-posteriori data fitting is necessary in

those features of seismicity which are I-easonably independent of the tectonic environment. There is a quite surprising number that can be identified world wide. The approach used here considers tectonic information only in so far as seismicity data can be crudely associated with major tectonic

North

American

cludes United

a part States,

0040-1951/89/$03.50

plate.

The region

of the Pacific and that part

considered

0 1989 Elsevier Science

Publishers

B.V

iO(,

Once such a preliminary

features.

the approach

searches

for anomalous

to large earthquakes

without

strictions

related

intended

in this paper

merely comment

to tectonic

expressed

and CN, developed Institute

Moscow

based

specific

on

hypotheses.

re-

It is not

the patterns;

we

and reproducible

a relatively

wide

perceived

subregions patterns

free parameters applications

MS

that

For

world

lower thresholds equally

various

calculated.

This set of components

demands

a

set of criteria

set of premonitory

phenomena and makes effective use of the data available for the Vancouver Island area. Both algorithms recognize that the process of evolution to a large earthquake may be very different in different cases. Both algorithms are open in that the inco~oration of new data constitutes part of the aigorithm. Of the two we have chosen M8 as

a vector

and

describing

are a conse-

in time

window

of the

that

Marly of the

are fixed for all

P with

characteristics

of the catalog is divided

among

four groups.

For every

window

between

a threshold

in-

the components

M8 deals with seven characteristics

for each component

desig-

spaced

groups.

values

than

8 for which it was originally

every

OII

wide. in spite of the fact that

stants

Earth.

no hearing

in these regions. in the algorithm

M8 is now is use for much ned.

have

at the 0. Yu.

et al.. in prep.) reasoning

geographic

the magnitude

in the algorithms

for the Physics

of pragmatic

relatively

prior

artificial

by investigators

(Gabrielov

quence

patterns

making

to explain

is made

that they exist.

The hypotheses Schmidt

selection

is

into

divided

can be chosen

of the vector so that pS of at1

the start of a catalog

and the first

strong earthquake in the window, or the end of the catalog if there are no strong earthquakes in the window, are above this limit. All such values

the simplest.

are designated as extreme. We now examine the vector calculated at time t from t - 3 yrs to t and identify the components that were extreme at least

The M8 algorithm searches for process activation in a seismicity catalog and considers objects

once in this period. A group is extreme if at least one of its characteristics is extreme. We calculate

defined

by time moments

t and windows

centered

the number

of such extreme

characteristics,

on a small number of regularly spaced points in the territory to be analysed. These windows can be

the number of extreme TIP (times of increased

of various shapes and are a type of automatic division of the seismicity into possibly overlapping

(f. I + T) if for two consecutive of such extreme characteristics

Fig. 1. The study area.

h, and

groups, g, and declare a probability) for a period times the number is 2 6 and the

PREMONITORY

number

SEISMICITY.

VANCOUVER

of extreme

these parameters

groups

a pattern

applied

geodynamic

observations

time

record

can

series,

in the Vancouver

that indicate

phenomena, may

whose collective of increased

Although

(Press

et

1979;

Dobrovolsky

al.,

1980a, 1988: Gabrielov

MS is desig-

ers who

the approach

consult

wish

those references,

here. The problem

100 km and

a time accuracy

of a few

identify

for this algorithm

from a study of very large earthquakes

world wide

of a large

toring

We introduce functions

Keilis-Borok this

et

may

the essentials

earthquake

of time

in a region.

s and

in the catalog

consider by calcu-

lating

strong earthquakes, sufficiently complete

time t and all events in the interval [t-s. t]. First we separate the aftershocks from

M, and the start time of a catalog, tO. The algo~thm is

normalized to a fixed rate of production of mainshocks and can be applied in a very wide variety of situations if we accept the hypothesis of the self-similarity, world wide, of the earthquakegenerating process. The results of this study enhance our understanding of the underlying similarity of seismicity world

wide.

previously nate the

If such self-similarity

in fact exists,

unsuspected relationships may domiphysics of earthquake generation. In

addition to the study of self-similarity the analysis clarifies the way seismic volumes such as subducting oceanic slabs and overriding lithosphere may interact. Finally, the results aid in the prediction of destructive seismic events. A suggestion of a region in which there is an increased probability

of the catalog

that depend

than considered in the California and Knopoff. As a consequence

the

maximum

distance

at which

considered an aftershock the mainshock-aftershock

area by Gardner we increased the an event

could

be

by 50% (Table 1). Once separation has been

made by the application of magnitude-~tependent space-time windows (Table 1) we can consider the

TABLE

1

if a mainshock occurring

has magnitude

within

tude less than

T days

after

M or less, any event within the event and having

M is an aftershock

2.5

29

6

result.

3.0

33

11.5

3.5

39

22

Definitions

4.0

45

42

The basic assumption that underlies the pattern recognition approach is the assumption that the earthquake flow includes a large chaotic component. As a result, the specific symptoms of an approaching strong earthquake may be different each time. Accordingly the algorithm uses a set of

on

mainshocks using a modified form of the generally accepted thresholds (Gardner and Knopoff, 1974). Inspection of the events in the data that we used clearly indicated aftershocks at larger distances

of a large seismic event could be used to initiate more detailed studies using either other methods or more time-dependent data to refine the MX

of the functions of M8

to

by moni-

reduce the number of free parameters for the study to two. They are the threshold for defining

characteristics

al.,

and read-

recognition

flow of earthquakes a period

1975;

approach

we discuss

of time defined

is fully

Briggs,

is to use pattern

the TIP of a strong the total

and

et al., in prep.)

to implement

accuracy

derived

(TIP)

inbut

of the

of the algorithm

elsewhere

with a positional

The parameters

can be diagnostic

probability

the basis

described of

behavior

informative

earthquake.

but

Island

each of which taken

be insufficiently

earthquake

of a strong of about

patterns

be

of quantitative

since 1952. The algorithm

ned to identify

years.

longest

premonitory dividually time

approach

that can be used with M8 is the record

seismicity

of

of most of

earthquake.

recognition

to any multicomponent

the chronologically area

activation

In any case no TIP is raised for

3 years after a strong Such

is 2 4. The choice

guarantees

the characte~stics.

301

ISLAND

M

R

T

4.5

52

5.0

60

150 290

X3

5.5

70

6.0

81

510

6.5

91

790

7.0

105

915

7.5

121

960

8.0

141

9X5

R

a magni-

flow of mainshocks { m,

) where

34,.

B,(e))

mainshock.

m,

t is the origin and

that occurred

17,.

time such that I~. , > I,.

parameter

k

those

is the depth.

is the number

of

mainshocks

exercise

that occurred

normalize

in an interval

of

.s before a

greater

than a

set designated

mainshock.

of mainshock

of simply

summing

the events

which we designate

Z, and is as close to 20/yr

aftershocks

measure

as N,.

as possible

I.,.

which we consider that

occur

This function.

depends

on the choice

quakes

of

magnitude

as :V,. I_, . and

and

for another

Zz. The seventh

in M8 is the count

within referred

two

days

of

of

a

to as B below.

of -. M only in that earthless than M ._ are not

mainshocks. We can now consider the earthquake stream to be characterized at a time I by a vector of func-

If ,0 = b/3, then S is a measure of the rate at which the radius of circular fractures is generated, and if /I = 2h/3, then S is the rate at which fracture area is generated. Here. h is the usual coefficient in the energy magnitude relationship log(E) = A + bhf.(The parameter a is inserted to normalize the functions). We carry out this process for all events whose magnitude IV, is between M and an upper limit R (R= M, - 0.1) and the result as S(r; M, a,

s, (Y. p). This pair

of functions can be combined to produce a measure of the way in which seismicity concentrates in space. _M, M, s, (Y. p)

tions

deduced

from the preceding

R, .T)y

defines the ratio of the average radius of a fracture (S/N) to the average event separation for all events between M and R, if they are uniformly distributed in theregion (N”‘). The rate at which mainshocks are produced is not uniform. If we assume that on average changes in a linear fashion. the function -M, s) = N(r: M. t-f,,)

[ - 10 I -

2,) -

s

it

time period

.r.

The problem is to deduce. by pattern recognition, whether such a time is part of a TIP. The duration T of such a TIP. the time within which a strong earthquake is expected. is taken to be the 5 years used in all recent applications of M8 world wide. In order to declare a TIP at time f we require that over a period +, prior to and including I. three of the groups ( N,. N, }, ( L,. L, }. { Z,, Z, ). and { B ) contain functions that have extreme values

and that at least four of the functions N,. Z,. Zz. and B have extreme values. N2, L,. I.?. In this sense extreme values are those in the upper 10% of the range achieved by the function from the start of the catalog

S(K M, R, .s, a, P)

- N(r; M, I-ff,,-s)

by adjust-

set of functions

the number

activ-

I&

on the

to count

ity calculates a weighted form of this sum where the weights are proportional to the size of the

[/v@; ly, s) - A+;

Ilnear trend.

explicitly

the result and this is achieved

A more sophisticated

=

depend

o/’

M. If we expect to compare results with -. in other areas we must find some wav to

function

Z(r:

from a long-term

three functions

lower limit M. We refer to this count as N( I; _M. s).

define

the gencratmn

ing -M so that the average yearly rate of occmrence of mainshocks is in one case IO;‘vr for one

time I and which had a magnitude

earthquake. Instead we calculate:

deviates

to which

in the first e days after

the mainshock. It is a simple

the degree

mainshocks These

H(e)

number

measures

of the

+ is the longitude.

M is the magnitude

of \:ectors

( I,. A,. 9,.

has six components

where i is the sequence

X is the latitude, aftershocks

to bc a sequence

to the time of occurrence

of the first large (M > MO. the prediction threshold) event. Once this condition occurs the interval (I. t + T] is declared to be a TIP. There is no straightforward means of deriving this algorithm from more fundamental statements. It follows from the plausible idea that patterns should exist in the earthquake stream and that these patterns should, as much as possible, be physically realistic and flexible. In the final analysis its justification must rest on the pragmatic statement that it works in many areas of the world. An area such as the southwest coast of Canada can be considered as a collection of perhaps over-

PREMONITORY

lapping define

SEISMICITY,

regions

defined

latitude

is then

such proposed

region

a side of

of latitude.

Each

to check

by the catalog

regions

occur

per

that do not

per year are discarded

and

the M8 algorithm

is applied

determine

a TIP exists in any of them.

whether

to the remainder

to

once

of regionalization

a small

a large

is possiearthquake

If such an area is divided it is possible

or quiescent

event

a quiescent

to occur within similar

can

activation

quiescence

that may precede

of the latter. A

be seen on a 2-dimensional

of event number with

area

area the large event is likely

or on the boundary

process

to classify

in the 6 years prior

(Fig. 2). If an activated

against

time (Fig. 11). A pronounced activation

a large (M

in which

of the event

to the seismic contains

coupled

that contains

area

them as activated

more than one region for each region

an

of location

histogram

and a TIP does not always

to produce

into small cells of size of the order of the accuracy

The regions can overlap, and usually do. As a consequence the same large event can appear in appear

refinement

might occur is identified.

seismic

ten mainshocks

A further

of ble

to have

ten mainshocks

year. Those proposed contain

it may be possible

defined

is then examined

of at least

the regions

area of high probability.

The square region centered

over the time span covered

an average

We

the area

by equal intervals

R = e Mclp5-6 in degrees

whether

seismicity.

by first diving

cells defined

and longitude.

on this event

303

ISLAND

by their

such regionalization

into rectangular

length

VANCOUVER

of smaller scenario

magnitude

and

lack of large events events

is the

(Lamoreaux,

1982)

large shocks.

> M) earthquake. In the restrospective analyses of a catalog such results suggest geographic areas

The seismicity of the Vancouver Island area

that are sensitive to the onset of large earthquakes and may suggest areas that should be monitored with more care. By considering the intersection of

The seismicity of the west coast of Canada has been described by many authors (e.g., Milne et al.,

Elf. ISI

cl

Unknown seismic state Activated seismic state Quiescent seismic state AQ boundary

Fig. 2. A schematic

diagram

to illustrate

the idea of activation

quiescence.

52' N 51' N 50°N 49'N 48' N

46'N 45' N 44'N 32' W30° W 128°W 126' W 124' W 122'W

2o" w 118 W

Fig. 3. The seismicity of the study area. 2 00 !____!__--!----! ‘841

t

1

t86i

1

I

1881

1

1

(931

1

300

I

1

((341

1

1

1951

% 1

1955

1

1959

1 .t

l

1963

I

1

1967

1

1

197t

1

1

1975

1

1

1979

f

t

i9e3

1

1

._ _!_-__a

--_-,_.

-_

1 i

; i?

1

192q

6.00

5.00

4.00 __!____I____!

4 42 45

to

4

1

12 17 22 37 22

10 5 9 9

;

2 2 id 2

i 15

5 6 4 9 3 6 24 14 15 6 13 17 9 11 152 12 3

6 1

37 49 30 14 ; 61 28 24 7 19 33 3 23 17 15 18 R 19 13 3 26 26 20 14 4 39 52 24 16 6 22 13 5 15 4 9 9 3 4 7 9 8 5 25 11 42 15 40 4 18 7 78 43 71 42 23 55 67 7 25 129 72 10 24 5 77 48 2 9 3 1 2 23 ,____!____!____!____!_-__!_---!----!----!----!----!----!----

84 83

Fig. 4. A count table indicating the number of mainshocks

4 ; 5 4 4 i 1 2 1 3 5 2 2 1 4 1 7 1 13 1

3

4 4 8

2 2 2 1 2

;

1 2 I

i

; 1 1 1 4 ;

2

i

2

1

i

i

in a particular magnitude interval during a particular time interval for the entire catalog.

PREMONITORY

SEISIMICITY.

VANCOUVER

4.2 4.5 3.3

2.1

4.3 3.9 3.0

51°N. 50’N. 49-N .

3.8 6.3 5.1 4.6

48’N.

5.0 5.2 4.3 3.3 3.0

47*N -

4.6 4.4

46’N I 49N

2.6 5.1 5.0 3.4 4.8 R

3.6

4.8 3.9 4.6 4.3 4.8 4.3 4.3

4.4

3.7

5kT.6

4.1 4.5

3.7 I I

132; W 130&W 128”W

126&W 124”W

Fig. 5. The largest earthquakes

305

ELAND

122&W 120&W It%” W

in 1 o x 1 o rectangles.

1978; Basham et al., 1982; Basham et al., 1985). Figure 3 shows the geographic distribution of epicenters for events of magnitude 4 or greater in the area which we studied. In this area, from 118” to 130 o W and from 45 o to 51” N, we merged the seismicity catalog of the Geological Survey of Canada with the catalog from the U.S. National Earthquake Information Service. Particularly for the earlier years the Canadian catalog contained many significant earthquakes as far south as 45 o N. The duplicate events were removed manually. Usually, NEIS data were accepted in the United States and GSC data were accepted in Canada. If only one agency reported a magnitude the report of that agency was accepted. Events for which no magnitudes were reported were assigned a zero magnitudes while if a preferred mag~tude was given it was used. For these data this is usually Mr. We also removed those events since May 1980 that were clearly related to the Mt. St. Helen’s Volcanic Eruption. The data are summarized in Fig. 4 as a time-magnitude histogram and in Fig. 5 as a map showing the largest magnitude earthquakes in regions measuring 1” X 1 O. Patterns in space and time can be distorted by changes in the observation networks and associated changes in the degree of completeness of the seismicity catalogs at lower magnitudes. Evolution of the station configuration in the Vancouver Island area, and the associated increase in complete-

ness of the catalog at small magnitudes, has been discussed by Basham et al. (1982). The pattern recognition approach used in this work does not require such a stringent completeness condition as discussed there. Thus Fig. 4 indicates that since 1952 a sufficient number but certainly not all of the magnitude 2.5 earthquakes are present in the catalog for some parts of the test area. The effect of missing small magnitude events seems to predominate in the function that counts af-:ershocks and probably reduced its effectiveness as a diagnostic characteristic in Vancouver Island. SimiIar considerations apply to the time variable location accuracy of the catalog. The spatial resolution of the analysis in this paper is about 0.5 ‘, reflection epicentral accuracy. Focal depths are the least certain, and in some subareas nonexistent. However, the only application of focal depths is as an exercise to determine whether patterns in shallow seismicity precede an intermediate earthquake. Data analysis In the analysis of the subcatalog discussed above we primarily used the standard parameter set developed for M8 by an analysis of a large number of regions world wide. We adjusted the threshold for prediction A+, and the start time of the catalog to to satisfy completeness considerations (Fig. 4) as required to normalize the Vancouver Island earthquake flow to others used in MS and applied local judgment to adjust the windows used to eliminate aftershocks. Table 2 shows the nine strongest earthqu~es of the area and indicates that the threshold for prediction should be M, = 6.5. This choice is by no means unique but is a stable choice; small variations do not seriously affect our conclusions. In order to select earthquakes for prediction we chose, before beginning the prediction process, all mainshocks of magnitude greater than 6 that occurred since 1957. Since a mainshock of magnitude 6.4 that occurred in 1971 near 50 ON and 128O W was followed within 1 year by one of magnitude 6.0 near 49” N and 129O W we noted these events as possibly being predictive. The 6.0 magnitude is substantially lower than our threshold

306

I .4BLF. 2 ‘I he mne strongest earthquakes

m the region h

Year

Month

Da>

I872

12

15

5

1917

12

22

1918

12

1926

~-

min

-_

!.

Lat.

Long.

-._-__-_ I>epth Magmrude _____.__.-...__ 44

.7?

4X.6

121.4

15

4x

50.0

128.0

6.5

6

X

41

S.8

49.6

125.9

7.0

11

1

1

39

1x.0

4X.X

128.5

6.0

1939

2

x

6

39

25.X

49.1

12x.0

6.5

1946

6

23

17

13

25.X

49.8

125.3

7.3

1946

7

18

6

6

5x.5

49.5

129.7

6.5

1965

4

29

15

28

44.0

47.4

122.3

59

6.5

1976

12

20

20

33

12.0 -

49.0

128.7

18

6.7

for prediction so we place no great emphasis on the results for this event. We excluded two large aftershocks of magnitude 6.4 and 6.1 which occurred in 1972 and 1978, and two large events of magnitude 6.3 that occurred in 1954 and 1956 are not likely to be predictable since the preceding seismicity catalog is inadequate. Two events of magnitude 6.0 that occurred in 1958 and 1961 were ignored on the grounds that their magnitudes were both small and their time of occurrence was uncomfortably close to 1957, the beginning of a useful prediction dataset. We therefore set as our goal the prediction of one event near Seattle and

three west of the northern end of Vancouver Island. Figure 6 shows the TIPS that resulted from the analysis of all the seismicity data available to us. The heavy lines under the time scale indicate TIPS, the light lines indicate the center of the region in which the TIP occurred. The three events west of the northern end of Vancouver Island, of magnitude 6.0, 6.4 and 6.7, are associated with three regions in which TIPS occurred. Although the prediction for the magnitudes 6.0 and 6.4 should strictly be considered false alarms, we accept them as being within the uncertainty of

Fig. 6. The centers of the regions analyzed with M8. A heavy line on the pointer to a location indicates the occurrence of a TIP in the time interval read from the scale in the

upper right of

the diagram. In this analysis we used a magnitude

time in the catalog of 1952.

threshold of 6.5 and a start

PREMONITORY

SEISMICITY.

magnitude preceded

VANCOUVER

determinations.

The magnitude

by a TIP in two regions

the event, in another

region.

data

due

from

a region

for both regions The magnitude at a depth ceded

two years before

east of the event.

6.4 event

that contain

resulted

The

was seen

not seen in the region

east. The magnitude

by

It was not seen in the

6.0 event just to the northwest

in three TIPS and

6.7 was

and preceded

a 5 year long TIP, which expired

magnitude

307

ISLAND

to the

in a TIP

it.

mation The

regions

6.5 event near Seattle shallow.

their overlap

activated

at the bottom and

considered.

I

I

I

I

I

I

.6:0’,‘,‘,‘,‘,‘,‘,‘,‘,‘, \\\\\\\\\\ ,,,,,,,,,,,,,

by of

The space-

time volume is area (km2) multiplied

,,,,,,,,,, \\\\\\\\\\

common

by time (yrs).

I

are

fairly

speculation

the

intersections

common This

to all TIPS is shown

but

to explore tectonic

both the 6.5

does not prove

of the separate

the predictions,

In order

in that part of the

domains

it is certainly

more closely

activity

that

the

~~11 in fact sugges-

the potential

as a tool to predict

deeper events we attempted so separate the catalog into essentially crustal earthquakes and events related to subduction. Depth determirations in this area are not adequate for careful separation SO we

chose to discriminate by considering only events of depth known to be less than 25 km. This

I

Fig. 7. The regions covered by overlapping

large

whether

to each TIP. The part

of Fig. 7 and contained

the 6.7 event.

of shallow

/\1

regions

occurred

area. The quality of prediction is estimated space-time volume of all TIPS in percentage

here

ated with the TIPS all occurred

TIP did not appear argues strongly for coupling between the shallow and deep seismicity in this

volume

considered

of the sum of their area defines

that pro-

Nevertheless

esti-

region of high risk. We note that the events associ-

refine tive.

space-time

in Fig. 6 gives a quality

(Fig. 7) and it is an interesting

the existence of four successful TIPS, one almost successful TIP, and only one region in which a

total

shown

of 30%.

of the time domains

of 59 km but the activation

the event was largely

The analysis

TIPS and one time intervals

common

to all TIPS for each event.

308

1960

Fig. 8. The results

of an analysis

using

only

events

whose

depths

explanations

is clearly a set of Iithospheric events but does not include all of them. The results of an analysis of this subset are shown in Fig. 8; there are insufficient events in the remainder of the set to allow application of M8. As expected, since most TABLE

have

1970

been determined

1990

1960

IO be 25 km or less. See Fig. 6 for

of layout.

seismicity near Seattle is shallow, the TIPS around the Seattle event are unchanged. One region to the south of the 6.7 event is lost due to insuf~~ient events in its s&catalog but the others remain viable TIPS.

3

A summary

of the TIPS (times of increased

probability)

for the Vancouver

TIPS

Center of window rp

x

49*N

129ew

Island area

Strong earthquakes

Time

Date

48”N

128” W

128* W

~970,6.20-1977.6.2~

1970,6.20-1978,6.20

i 1971.3.13

6.7 6.4 I

1971.125

6.0

1976,12.20

6.7

i 19?1,3.13

6.4

1971 J2.5

6.0

( 1970,6.20-1975q6.20

72

32

1971,125

6.0

6.0

1983,6.20-1989.6.20

?

?

72

19654.29

6.5

72

48’N

124* W

1959,4.29-1%7,4.29

49ON

123O W

1957,10.29-196410.29

48*N

122OW

1960,4.29-1965‘4.29

49ON

121”W

1957.10.29-1964.10.29

47“N

121°W

( 1962,4.29-1967.4.29 1982,4.29-198924.29

* Time period

M

of

expectations (months)

1976.12.20

SOoN

Tie

from start time of a TIP to a strong

event of M > Me.

84 1%5,4.29

6.5

60 84

19654.29

6.5

36

?

?

84

*

PREMONITORY

SEISMICITY,

VANCOUVER

309

ISLAND

Center of window: 49N 129W .*****A * * * * ~*,***~,~*,“,*,* *I************* I***h*F******* ***********~* .************** .*****~******~************* *****A********* **********~***************** .***********I** I*************** .************** .**********~@t.&************ **************I .*********I**** A************* r*# * A * * * * * *r”*a***** ******** 1

Nl N2 Ll L2 Zl 22 B 1965

1970

1975

1970

1975

Center of window: 50N 128W

1965 Center of window: 48N 124W Nl N2 Ll L2

.* ** * ** **** * **-??**************** .********* * * * * * * * * I) *%.********** .**************** *******n******** .**************** a ‘** * * * * * * * I) * * * .*******n******** ********J******* .***************I **************** .**************** * * * * * * * * *2” A * * * * * ;c**1****** * * * *y&#.&* * * * *************** * * * * A* * **************** ************I*******~* * * **’ ‘* .q#* n * * * * .******** *** 1

Zl 22 B 1955

*

1960

1965

Center of window: 48N 122W Nl N2 Ll L2 Zl 22 B

‘*********3h**L**** .********* *F * .* *I) ** ** ** ******* * * * ***** .*********’ ********* ?* * * * * * * * * ********* ********I ********** .********* ********** ‘******y&+******* .**I** * * * * I * *?,**** .*I******* ************#r ‘L. * * * * * * * *** x

1955

1960

1965

Center of window: 47N 121W Nl N2 Ll L2 Zl 22 B 1955

u.********, * ****** * * * I) I ******. * ********I * * * * * *I * . ********* .*********. ****~*****~*******. ********* ***~*********** * * . I)********. **A****** *********. ******xArte******. ********* *I)*******. ********* ****~*~* ***** * * * * . 1960

1965

Fig. 9. The part played by each function in establishing the TIP. The shaded area are the time intervals in which the activation was sufficient to establish a TIP (referred to here as a “ vote”).

The analysis that exploits the entire dataset is associated with two TIPS for which large events have not yet occurred.

In one case the seismicity

the anticipated earthquake. We and others found it helpful to examine activatu~n on a more detailed

scale and

of

we show some of the regions

subdivided

than 6.5 began in the middle

of

cells. We have examined

1982 and will expire in 1991. If past experience

is

significant

activity

years prior

to the large event.

If no such activity

occurred

the cell was identified

as quiescent.

collection

of quiescent

data

indicate

magnitude a guide

that

greater this TIP

a TIP

is associated

end of Vancouver

Island.

ated with the northwestern States.

It began

expires

in 1991.

Some

traits

of an

earthquake

quiescence

h;t\e and

with the northern

The other TIP is associcorner

in the middle

of the United

of 1983 and

also

of the catalog

contribute

to the the

time intervals that have created TIPS for five windows. The asterisks indicate the occurrence of extreme values for the functions on the left at times

indicated

at the bottom

of the plots.

The

regions are the time intervals in which the activation was sufficient to establish a

TIP. Table

may

strong

TIPS and others do not. In Fig. 9 we illustrate

shaded catalog

cells

3 is a another

means

of summarizing

these results. As indicated in Fig. 6 the M8 algorithm is not particularly effective in specifying the location of

Cantor ot wlndow UN 12oW

Canter of wlndow: SON 128W

the

event

events

on

in it durrng

cells bounded the

will occur. plot

into 256 iI

each cell to determine

occurred

indicate

I-‘& It!

in

the

area

the o A

by activated

in which

the

next

In each of the four cases boundary

between

quiescent and activated areas, but the regions the northwest show a better defined area

the in of

quiescence than those in the southeast. The same process can be used to refine the time of occurrence of the anticipated earthquake. In time, the quiescence when

an earthquake

histogram time for

will help us to refine the time will occur.

Figure

11 is a

of event number against magnitude and the magnitude 6.7 and 6.5 events. A

quiescent period occurs before the magnitude 6.7 event and a courageous analyst would have predicted the event when the activity began to increase. The histogram of magnitude not show such an effect as clearly.

6.5 event does

Illllflllllllll!l

Conclusions

It is scarcely surprising that the seismicity of Vancouver Island bears a strong similarity to that found elsewhere. It would have been surprising if

Centor

of wlndow:

4ON 122W

Cmtw ot wtmlow 48N lZ4W

llllllllI_ccI

q ahsceflt

Vancouver Island prior to at least four large seismic events. This suggests the possibility of

cdl

El Ouioscont boundwy .

this area were radically different from the twenty areas previously tested by Keilis-Borok and his colleagues (Gabrielov et al.. in prep.). Apparently identifiable patterns exist in the seismicity of

Loutton ot strong wrthqwk*

Fig. 10. Plots of activation

and quiescence

events in the catalog.

for the four large

defining time and space intervals that indicate an increased probability of large earthquakes. Three successful retrospective predictions alone are not convincing, but the agreement of this method with many cases world wide suggests that the current TIPS in the northwest and the southeast parts of the study area should be taken seriously. One earthquake of magnitude 6.0 is included in the TIP that precedes a larger event, but two others are not. This indicates that seismicity of

PREMONITORY

SEISMICITY.

VANCOUVER

311

ISLAND

Center of window: 5ON 12’7W 4.00

2.00 3.00 !____!____!____!___ 69

1

70

1

I

0

f

3

0

1

71110 72

1

73

1 IO

74

1

75

1

i I

1 0

1 1 1

10 10

76

1

10

77

1

10

4 5 4 8 8

78110

---

!----!

3 9 166

6.00

5.00

.- !----!----!____!--__!____!

2 7 1 3 1 1

1 3 1 3 2 2

2

; 1 2 1 2 1 6 6 1

2 ; 1 2 2 2 3 2 1

i

; 1

; 2

ii

2 3

2

1

3 1

;

i 2 1

2 2 I

i : i 1 3 1

:

:

:

1 1

;

: I

: 1

i 1 1 2 1 1

:

7.00 _!____!___-

1

i

:

2 1

: 1

:

:

8 13 7 12 6 8 4 2 5 5 6 9 6 15 18 12 14 16 0

-!__--!____!--__!___-I___-!___-!

4 1

3 3

1 2

1 6

2

2

CVENTS

Center of window: 47N 123W 2.00 3.00 !____!____!____!-_ 58

1

1 0

59

1

10

60

1

I

61

1

10

62

1

1 0

63

1

1 0

0

13 10 9 8 8 9 6 6 5 4 2

64110

7 3 4 6 I 2 7 3 1 3 2 1 1

6.00 -!----!

-_-

-_-

5 3 2 6 2 1 1 2 1 1

:

1

0

26 16 15 24 II 13 14 13 7 5 6 5 2 4 2 3 0

1 1

1

1

65

1

1 0

i 1

2

:

8 2

4 4

2 6

66110

166

3

4

3

2

1

0

0

EVENTS

Fig. 11. Magnitude-time

histograms

of events in regions

centered

on two large earthquakes

(see text).

magnitude 6.0 is evidence of activation for events of larger magnitude. but that the activation phenomena required to predict events of magnitude 6.0 by MS do not exist in this catalog. The two events of magnitude 6.0 that are outside the TIP were not followed by large events but the one that was within the TIP was present there because the TIP was initially generated for an event of M = 6.4. Since this M = 6.0 event was foflowed by an event of M = 6.7 and the other two were not followed by large events, these smaller mainshocks were correctly interpreted by the algorithm. The failure to predict them is not a failure of the algorithm; in this area it appears to work only for magnitudes in excess of 6.4. In addition we observe quite clear patterns in shallow seismicity prior to the deep large event at 47O30’N, 122” W depth 59 km). This suggests that the subducting lithosphere may be coupled to shallower structures in some as yet unexplained way. This in turn implies that other near-surface geophysical measurements, such as strain and geochemical studies, may have potential as detectors of precursory patterns to strong deep earthquakes. At this time we make no statement on the potential for a megathrust earthquake in Cascadia. The interval of adequate seisrnicity catalogs may not be sufficiently large to identify this event. The only evidence that may have a bearing is the strangely low number of mainshocks in the area and the anomalous difficulty we encountered in assessing aftershocks. AcknowIe%lgements The analysis reported here was directed by V.I. Keilis-E3orok while he visited the University of Alberta in July and August of 1987. It benefitted greatly from discussions with Garry Rogers of the Pacific Geoscience Centre, Ge&gica.l Survey of Canada. We also acknowledge our debt to a legion of workers who developed the coding for the M8 algorithm.

Keilis-Borok’s visit was supported t’rom the Fund for the Future at the University of Alberta. The Natural Sciences and Engineering Research Council of Canada. the Department of Physics. and the Departments of Computer Services and Computing Sciences at the University of Alberta also aided this research in various ways.

References Basham, P.W., Weichert, D.H., Anghn, F.M. and Berry, M.J., 1982. New probabilistic strong seismic ground motion maps of Canada: a CompiIation of earthquake source zones, metbods and results. Minist. Energy Mines Reaourc., Earth Phys. Branch Gpen File 82-33. Basham, P.W., Weichert, D.H., &$n, FM. and Berry, M.J., 1985. New probabiBstic strong seismic ground motion maps of Canada. BuU. Seismol. Sot. Am., 75: 563-595. Dobrovobky, I.R., Zubkov, S.I. and Miachkin, V.I., 1979. Estimation of the size of earthquakes preparation zones. Pure Appl. Geophys.. 117: 1025-1044. Gabrielov, A.M., Dimitreava, O.E., Keihs-Borok, V.I., Kosobokov, V.G., Kuznetsov, I.V., Levshina, T.A., Mitroev, K.M., Molcban, G.M., NegmatuBaev, SKh., Pisarenko, V.F., Prozeroff, A.G., Rinehart, W., Rotvain, I.M., Shebalin, P.N., Shnirman, M.G. and S&eider, SYu., in prep. Cent. Reg. SismoI. Am. Sud., Lima. Gardner, J.K. and Knopoff, L., 1974. Is the sequence of earthquakes in Southern California, with aftershocks removed, Poissonian? BuII. Seismol. Sot. Am., 64: 1363-1367. Keihs-Borok, V.I., Knopoff, L. and Rotvain, I.M., 1980a. Bursts of aftershocks, long-term precursors of strong earthquakes, Nature, 283: 259-263. KeiIis-Borok, V.I., Knopoff, L., Rotvain, I.M. and Sidorenko, T.M., 1980b. Bursts of seismicity as long-term precursors of strong earthquakes. J. Geophys. Rea., 85: 803-811. Keilis-Borok, V.I., Knopoff, L., Rotvain, LM. and Allen CR., 1988. Intermediate-tetm prediction of occurrence times of strong earthquakes. Nature, 335: 690-694. Lamoreaux, R., 1982. Cluster patterns in seismicity. Ph.D. Thesis, Univ. AIber@ Edmonton. M&e, W.G., Rogers, G.C., Riddihougb, RP., Hyndman, R.D. and McMecban, G.A., 1978. Seismic&y of western Canada. Can. J. Earth Sci., 15: 1170-1193. Press, F. and Briggs, P., 1975. CbandIer wobble, ear&quakes, rotation and geomagnetic changes. Nature, 256: 270-273.