You are on page 1of 50

Lecture 2a: Single-point and

multiple-point source models


Short introduction for ISOLA users

J. Zahradník and E. Sokos

Charles Univ., Prague, Czech Republic


Univ. of Patras, Greece
Part 1 – Hypocenter, centroid

Two different single-point


representations of an earthquake
(should be the same for Mw<6)
Hypocenter (where the rupture starts)
Example: Jan. 2020, M6.8 Turkey
Uncertainty due to
Location from
Source-station geometry
The P and S
Picking accuracy
arrival time by
probabilistic method
NonLinLoc of A. Lomax

Uncertainty due to
velocity model
needs repeated solutions
in several velocity models.

Gallovic_etal_CommsEnv2020.pdf
Location uncertainty due to velocity
model, sequence Turkey Jan 2020

Spurious (false)
concentration
of foci
near shallow
discontinuities

Gallovic_etal_CommsEnv_SUPPL.pdf
Location uncertainty due to velocity
model, Samos Island, Mw7, 2020
Centroid (“center of gravity”
of final slip on fault)

C
H

ISOLA is not calculating slip inversion. Instead, it effectively identifies main slip patch(-es).
Advantage of ISOLA: slip patches can have different focal mechanisms.

Thus ISOLA is useful for initial studies, to indicate the fault(-s) at which slip inversion
should be made. ISOLA (as a low-parametric method) is also more robust, resistant to
possible artifacts than slip inversion.
Concept of the H-C consistency
From MT we know strike and dip of two nodal
planes. Which one is the fault plane? We test both.
We draw planes passing through C, and H should
be in one of them…
Or, we draw planes passing through H, and C should
be in one of them (in fault plane).

Since H and C are determined independently, H-C


C
test is useful. However, uncertainty of H and C is
critical.
H
For M<6, H-C test is almost impossible because H-C
distance and error of H and error of C position are
comparable.
Anyway, even for M<6 it is useful to compare H and Blue symbol
C. If C is too far from H, it can indicate a problem = location of H
with strong 3D heterogeneity of the crust.
[ Importance of smaller ‘calibration’ events] For M>6, H-C testing in ISOLA tool
is always recommended.
Part 2 – Example: a 3-point
model of Elazig earthquake
Example where H-C is less important
Jan. 2020 Turkey, Mw6.8
Both H and C should lie in the same plane (fault
plane). If so, we talk about H-C consistency.

In the example of the Jan 2020


Turkey earthquake, H-C check
is not much important,
because
the fault plane was well
identified by aftershocks.

Gallovic_etal_CommsEnv2020.pdf
Centroid is a low-frequency point-source
approximation
Wavelengts
For λ >> L, the wavefield
is insensitive to L.

Epicentral distances
For D >> L, Green’s functions
for a single source point can be used.
(Here D >> L for ‘outer’ stations
and D > L for ‘inner’ stations.)

By fault length L we mean the ruptured part, here ≈ 40 km.


Wavelength:
Ruptured fault For dominant shear waves, λ = Vs . T, where Vs ≈ 3 km/s,
(small green and period T=20s, λ =60 km, hence λ > L, but not λ >> L.
rectangle)
Gallovic_etal_CommsEnv2020.pdf
Centroid – how to find
Using only “outer-circle” stations (BB)
and frequency range 0.01-0.05
we obtain a point-source model – centroid.

If we try to calculate more than single subevents


in this low-frequency range
their moments are small (sub2<sub1, sub3< sub2).
Search for centroid
Centroid position is searched below epicenter Schematic plot:
(for M<6) or in a plane, or along a line.
The search optimizes fit between real and
synthetic seismograms.

Besides C position, we also grid-search for the


time when the major moment-release takes
place (centroid time). This is equivalent to
optimal waveform shift, common to all stations.

For each space-time grid point, MT is calculated


by LSQ method.
Isolines of the correlation
For a 3D search, we usually use horizontal reflect uncertainty of C-position.
planes at several depths.
Partial summary
(some practical “hints”)
Every earthquake can be either viewed as a single point source, or
a more complex source
(e.g. a smoothly distributed slip on fault in slip inversion codes
or a multiple point source in ISOLA.

It depends on the used frequency range in inversion (Fmin, Fmax)


(higher frequencies and near stations reveal more source detail).

Low-frequency limit Fmin is basically determined by noise, either natural when using BB
or instrumental when using SM sensors. Removal of records with ‘mice’ is a must!

High-frequency limit Fmax: The more accurate velocity model is,


and the shorter is the epicentral distance, the higher Fmax can be used.
For near-regional distances, the rule of 10 MSW sometimes gives a hint:
Maximum distance must be < 10 MSW, where MSW is Minimum Shear Wavelength,
Vs/Fmax, and Vs= 3km/s for the crust.

Simple strategy: In a study of any event we usually start with relatively low Fmax.
Then, in subsequent tests, we increase Fmax to catch more source details
Multiple-point-source (MPS) model
Example: Jan. 2020, M6.8 Turkey
Using the “outer-circle”
stations (BB)
in frequency range
0.01-0.05 Hz
[MSW=60 km]

AND
the “inner-circle”
stations (SM)
in frequency range
0.05-0.10 Hz:
[MSW=30 km]

We fit waveform data with


3 sources (subevents) .

Near stations and shorter wavelengths are sensitive to source size and complexity.
One, two or three “subevents”
station

station
One, two or three “subevents”
station

station
One, two or three “subevents”
station

station
Iterative deconvolution

HOW the subevents were calculated in this example:


Standard in ISOLA is subsequent calculation (iterative deconvolution).
“One-by one”. I.e., full waveforms are partially fitted by one source - sub1
(its MT is either calculated or assumed), synthetics of sub1 are
subtracted from full data. Residual data are inverted for sub2, etc.
MPS - variability

In the previous slide we searched


3 subevents, along a line,
and repeated inversion several times,
always removing one station (jackknifing).
MPS - variability

In the previous slide we searched


3 subevents, along a line,
and repeated inversion several times,
always removing one station (jackknifing).

Besides variation in subevent positions,


jackknifing provides also variation of
focal mechanism.
Another alternative to
Iterative Deconvolution (ID)?
ID – it is subsequent solution, sub1, sub2. Theoretically, there is not much
reason for that. Full problem is non-linear.

Joint MT inversion for several subevents – is a good option (rather rare).


One of our future plans, using a kind of MonteCarlo methods to seek multiple
MTs (unequal focal mechanisms, position, time, moment).

Compromise in ISOLA: Joint inversion of waveforms for position and time function
of two subevents assuming that we know their moment tensors (possibly unequal).
On a space grid, all possible pairs of positions are examined.
Time function of each subevent moment rate is a sum of (equal) shifted
elementary functions , each with its unknown weight (>0).

Shortly NNLS, because it is performed by Non Negative Least Squares method.


NNLS provides an ensemble of solutions within a user-specified threshold of VR.

Theory – see the Appendix in : zahradnik_sokos_gji2014.pdf


Part 3 – Example: A 2-point model of
Mw 5.7 in Sichuan Basin
(strongly non-DC)
Multi-type faulting, possibly related
to water injections in salt mines
The 2019 Mw 5.7 eq. in Sichuan -
a huge non-DC explained as doublet
GCMT reported the DC part as low as 2% !

We calculated full MT and obtained an


almost zero VOL and large (negative) CLVD.

We explained the event as two


different faults, SS and TF,
situated 3-9 km and ~3 s, apart.

ISOLA tool:
NNLS joint inversion
of two subevents
Liu_Zahradnik_GRL2020.pdf
NNLS - example Mw 5.7 doublet,
ID Sichuan, China

NNLS

Liu_Zahradnik_GRL2020.pdf
NNLS - example Mw 5.7 doublet,
ID Sichuan, China
Sum of the two DC parts yields
just a very low DC% (~ 20 %)
of a single-point model.

NNLS

Liu_Zahradnik_GRL2020.pdf
First-motion polarities and MT of the
Sichuan event

Strong disagreement of polarities


with full MT (shading)
and also with its DC part (nodal lines)

Excellent agreement of polarities


with the first subevent .
I
t confirms that the earthquake
started (nucleated) as a thrust-fault.
First-motion polarities and MT of the
Sichuan event

Strong disagreement of polarities


with full MT (shading)
and also with its DC part (nodal lines)

Excellent agreement of polarities


with the first subevent.
It confirms that the earthquake
started (nucleated) as a thrust-fault.
According to regional stress field, SS faults seemed to
be stable, far from Coulomb failure criterion.

Since there are many


injections in salt-mines,
we suggest that
SS fault was activated
due to elevated
pore-water pressure.
vavrycuk_epsl2011.pdf Liu_Zahradnik_GRL2020.pdf
Part 4 – Example: Mw 6.8
Zakynthos Island, Greece
Another “apparently non-DC” event
(composed form two DCs)
The 2018 Mw 6.8 Zakynthos, Greece,
Earthquake: Dominant Strike-Slip Faulting
near Subducting Slab

Sokos_etal_SRL2020.pdf
Two subevents – a strike-slip and a thrust
Strike-slip (blue plane) – oblique to the
slab
Part 5 –
Problems of small earthquakes
Motivation

Waveform inversion for moment-tensor


for small earthquakes and/or
earthquakes recorded in sparse networks
is often difficult (or even impossible).

How to at least partially


overcome these problems?
Need of polarities, envelopes …
Three possible ways how to use
first-motion polarities
MT waveform inversion with posterior polarity check
After calculating MT by waveform inversion, we just check whether the calculated MT
(either its DC part, of full MT) agrees with polarities, and discuss the results.

Pure polarity solution


Waveforms are not inverted. Only DC focal mechanism
is calculated from the polarities.

Combined use of polarities and waveforms


Before calculating MT by waveform inversion, we pre-constrain
focal mechanisms by polarity. Waveforms then just “select” among
the polarity-fitting solutions.
Cyclic Scanning of Polarity Solutions (CSPS method)
MT waveform inversion with posterior
polarity check
This check is always useful.
Special case: sometimes almost all polarities disagree with calculated MT.
We speak about P-T axes flip. We have good strike and dip, but wrong
rake [instead of rake-180°].

Examples: Waveform inversions in narrow frequency bands, and/or long


intervals of time grid search. Then
“stripes“ appear in correlation diagrams.
The stripes have almost
equal correlation, but opposite P-T axes.
One of them is wrong.

A few reliable polarities


help to select the proper stripe.
Pure polarity solution – e.g. software
FOCMEC
Input:
for an assumed source depth
and a velocity model:
Station azimuth take-off
VO09 286. 139. +
VO10 111. 121. -
VO14 200. 115. +
VO04 25. 115. -
VO13 244. 112. + take-off
VO05 67. 108. - angle
VO08 300. 108. +
VO03 340. 102. +
….
….
Pure polarity solution – e.g. software
FOCMEC
Input:
for an assumed source depth
and velocity model
Station azimuth take-off
Output:
VO09 286. 139. + All solutions Plotting some solutions
VO10 111. 121. - with e.g. 5 allowed with polarities
VO14 200. 115. + misfits
VO04 25. 115. -
VO13 244. 112. +
VO05 67. 108. -
VO08 300. 108. +
VO03 340. 102. +

Disadvantages: Often non-unique, needs a preferred source depth, assumes pure shear.
Caution: It refers to initial part of source process. (Is it disadvantage, or advantage? Both …)
Combined use of polarities and
waveforms (CSPS method)
Cyclic Scanning of Polarity Solutions

Pure polarity solution


(strongly non-unique
although only 1 misfit
was allowed)

True solution -shaded


(independent 10-station method)
It is compared with CSPS with
The CSPS solutions
VRropt from 6 single stations.
preferred by waveforms
at a single station.
Plotted are those of VR
from 0.95 Vropt to Vropt.
fojtikova_zahradnik_SRL2014.pdf

Using waveforms from just single station is an extreme ; the more stations, the better.
CSPS and magnitude
When the polarity solutions are “scanned”, i.e. selection among them is made of
those FMs which are fitting a few waveforms, Isola calculates also moment magnitude Mw.
(The most likely Mw values are those corresponding to the solutions with greater VR.)

VR … variance reduction
of waveform fitting
CSPS and magnitude and depth
When the polarity solutions are “scanned”, i.e. selection among them is made of
those FMs which are fitting a few waveforms, Isola calculates also moment magnitude Mw.
(The most likely Mw values are those corresponding to the solutions with greater VR.)
The waveform fitting can be also optimized with respect to the grid-searched source depth.

This procedure is allowed


if the tested depths are near
the depth used in FOCMEC
and/or when the effect of depth
on the polarity solution is small.
CSPS and magnitude and depth
When the polarity solutions are “scanned”, i.e. selection among them is made of
those FMs which are fitting a few waveforms, Isola calculates also moment magnitude Mw.
(The most likely Mw values are those corresponding to the solutions with greater VR.)
The waveform fitting can be also optimized with respect to the grid-searched source depth.

This procedure is allowed


if the tested depths are near
to the depth used in FOCMEC
and/or when the effect of depth
on the polarity solution is small.

However, if another depth


and Mw appears to be also possible,
FOCMEC should be re-considered
at that depth.
More caution is needed...
Part 2 – Envelopes (of
waveforms) and polarities
A tool which is less sensitive to
inaccuracy of velocity model, e.g. in
case of very distant stations
Wrong MT of Mw 4.4 inverted from
waveforms at distant stations
Correct solution from near stations

Zahradnik&Sokos_SRL2018.pdf
Wrong MT of Mw 4.4 inverted from
waveforms at distant stations
Correct solution from near stations

Wrong solution from distant stations


although at low frequencies 0.05-0.1

Zahradnik&Sokos_SRL2018.pdf
Why we obtained a wrong FM?
(Answer: Even waveform forward simulation with
correct FM has a wrong fit  bad velocity model.)
Envelopes (in this case) are indeed
better, providing almost correct
solution in the FM inversion

Here, the envelope solution was constrained by a single polarity (dilatation at station NVR)
Envelopes – practical application
Maranhão eq., 2017, Mw 4.3, except one station
all (3) are at distance > 470 km.
Reference solution
Obtained with path-specific
velocity models from dispersion,
constrained by 10 polarities
(Dias et al., 2017).

Correct solution
Envelopes of
4 stations and
a single polarity
TMAB (D)

Carvalho_etal_BSSA2019.pdf
Envelopes – two possible approaches

Full grid search over source angles strike/dip/rake


(perhaps also over trial depths), constrained by a few polarities.
Time consuming.

Grid search over source angles strike/dip/rake from FOCMEC,


i.e. the s/d/r angles pre-constrained by many polarities,
similarly to CSPS method.
More efficient (but more dependent on quality of polarity solution,
i.e. on quality of reading, quality of velocity model, take-off angles,….)

Both provide also Mw !


Thanks for
your attention.

Regards from
Prague.

http://www.kamvpraze.in

You might also like