My 42 days in Paris 2018/8/3 Ando lab seminar Tomofumi Shimoda - - PowerPoint PPT Presentation

My 42 days in Paris

2018/8/3 Ando lab seminar Tomofumi Shimoda

abstract

• visited Paris from 6/15 to 7/29
• APC (Astroparticule et cosmologie)
• IPGP (Institute de Physique du Globe de Paris)
• mainly worked with Kevin Juhel about earthquake

early warning (EEW) and partly with Donatella about Newtonian noise

• main topic : how TOBA can contribude to EEW?

• gravity perturbation from deformed ground can be

used for earthquake detection

• gravity travels faster than p-wave
• earlier detection especially for off-shore events
• better magnitude estimation of large earthquake
• seismometers can saturate

Earthquake early warning with gravity perturbation

+ +

• compressed

dilated

!

gravity perturbation induced by Tohoku-oki earthquake was detected

• with broadband seismometers
• Mw = 9.1

for smaller earthquakes : Gravity gradiometers

• gravity change cannot be distinguished from

background seismic vibration (equivalence principle)

• gravity gradient change can be separated from it
• Superconducting gravity gradiometer (SGG)
• Torsion bar (TOBA, TorPeDO)
• Atom interferometer (AI)

SGG

research topics

• gravity perturbation signal in different gravity gradient

tensor components

• which component is important for EEW?
• how can different kind of detectors (SGG, TOBA, AI,..)

contribute to EEW?

• EEW with realistic sensitivity of TOBA
• how is it different from model noise (∝ f -2 at low-f, flat at high-f)?
• what kind of design is preferred?

∇⨂ ⃗ % = '( ℎ** ℎ*+ ℎ*, ℎ+* ℎ++ ℎ+, ℎ,* ℎ,+ ℎ,,

gr gravit ity signa ignal l in in dif different gr gradie dient te tensor component nts

• 5 independent components
• each detector measures different set of components
• SGG : all components
• TOBA : only horizontal components
• (AI : only vertical components?)

hij = hji , hxx + hyy + hzz = 0 (in free space) horizontal vertical ∇⨂∇#$ = &' ℎ)) ℎ)* ℎ)+ ℎ*) ℎ** ℎ*+ ℎ+) ℎ+* ℎ++ gravity gradient tensor SGG

S/N with next-stage gravity gradiometers

• sensitivity : ~ 10-15 /rtHz @ 0.1 Hz (“model 2” in Kevin’s definition)
• S/N at t= 10s, 15s → early detection
• S/N at p-wave arrival time → parameter estimation
• three-types of focal mechanism (mechanism of deformation)
• stlike-slip, dip-slip(10o), dip-slip(20o)

E Z 10o, 20o

dip-slip

E N

strike-slip

horizontal

S/N of 5 components

ZZ RZ TZ E N Z vertical

Mw = 7.5 contour : S/N = 3, 8 assuming “model2” noise

(floor=1e-15, cutoff=0.1Hz)

3-types of focal mechanism

E Z 10o, 20o

dip-slip

E N

strike-slip

color=S/N

distance (<300km) azimuth epicenter

E N

t = 10s after onset

PP CC E N

horizontal

S/N of 5 components

ZZ RZ TZ E N Z vertical

Mw = 7.5 contour : S/N = 3, 8 assuming “model2” noise

(floor=1e-15, cutoff=0.1Hz)

3-types of focal mechanism

E Z 10o, 20o

dip-slip

E N

strike-slip

color=S/N

distance (<300km) azimuth epicenter

E N

t = 10s after onset

PP CC E N

PP and CC are large for strike-slip

• nly CC is large at 0o,90o,180o,270o

horizontal

S/N of 5 components

ZZ RZ TZ E N Z vertical

Mw = 7.5 contour : S/N = 3, 8 assuming “model2” noise

(floor=1e-15, cutoff=0.1Hz)

3-types of focal mechanism

E Z 10o, 20o

dip-slip

E N

strike-slip

color=S/N

distance (<300km) azimuth epicenter

E N

t = 10s after onset

PP CC E N

PP and ZZ are large for dip-slip at 180o

• nly RZ is large for dip-slip at 0o

horizontal

S/N of 5 components

ZZ RZ TZ E N Z vertical

Mw = 7.5 contour : S/N = 3, 8 assuming “model2” noise

(floor=1e-15, cutoff=0.1Hz)

3-types of focal mechanism

E Z 10o, 20o

dip-slip

E N

strike-slip

color=S/N

distance (<300km) azimuth epicenter

E N

t = 10s after onset

PP CC E N

CC and TZ can compensate for S/N at 90o, 270o

horizontal

S/N of 5 components

vertical color=S/N

distance (<300km) azimuth epicenter

E N

E N

strike-slip

t = 15s after onset

E Z 10o, 20o

dip-slip

ZZ RZ TZ E N Z PP CC E N

Mw = 7.5 contour : S/N = 3, 8 assuming “model2” noise

(floor=1e-15, cutoff=0.1Hz)

3-types of focal mechanism

things are similar as t=10s

horizontal

S/N of 5 components

vertical

E N

strike-slip

color=S/N

distance (<1000km) azimuth epicenter

E N

t = p-wave arrival time

E Z 10o, 20o

dip-slip

ZZ RZ TZ E N Z PP CC E N

Mw = 7.5 contour : S/N = 100, 300 assuming “model2” noise

(floor=1e-15, cutoff=0.1Hz)

3-types of focal mechanism

total S/N of horizontal(TOBA) and all(SGG?) components

• t=10s

How about atom interferometers? (vertical components only?) almost same S/N S/N of horizontal components are ~70% of all components at 90o ~ 270o (where detectors will be usually placed)

horizontal all

E Z E N

E N

horizontal gradient is small at 0o

total S/N of horizontal(TOBA) and all(SGG?) components

• t=15s

How about atom interferometers? (vertical components only?)

horizontal all

E Z E N

total S/N of horizontal(TOBA) and all(SGG?) components

• t=p-wave

arrival time

How about atom interferometers? (vertical components only?)

horizontal all

E Z E N

RZ component grows

summary of gravity signal in different components (from a viewpoint of TOBA)

• strike-slip : horizontal components are large
• dip-slip (at azimuth = 180o) : horizontal components are

~70% of all components

• dip-slip (at azimuth = 0o) : horizontal components are

much smaller than all components

• but usually azim.=0o is the direction without land

(detectors cannot be placed)

• TOBA seems to have enough contribution to EEW

S/N S/N with realistic sensitivity

• resonant peaks (f0 ~ 8mHz, 40mHz, 0.6Hz)
• ∝ f -2.5 dependence at f < 0.1Hz (thermal noise)
• Newtonian noise (higher at f<0.1Hz)

pendulum rotation translation

phase-III TOBA ∝ f -2 for model noise which has been used so far ∝ f

• 2.5

model

temperature NN @ 100m depth infrasoundNN @ 100m depth

instrument nosie

basic S/N calculation process

• matched filter

! ̃ #(%)' ℎ∗(%) *+(%) ,% = ! ̃ #(%) *+(%) ' ℎ∗(%) *+(%) ,% # . = ℎ . + 0(.)

signal noise data whitened data whitened signal (template)

data (signal+noise) template (signal)

convolve

pre- whitening

S/N

(normalize)

• resonant peak -> apply notch filter for prewhitening
• one resonant peak decreases S/N by 5~15% for Q=5

effect of resonant peaks

f f0 ∝ 1/Q notch filter t=tp t=15s t=10s

peak frequency (f0)

(S/N with peaks) / (S/N without peaks)

Q=1 Q=3 Q=5 Q=10 noise floor peak + ~1/10

assuming floor of peak is 10 times lower than noise floor

Q=50 (Q of notch filter)

total effect of resonant peaks in TOBA design sensitivity

• S/N with design sensitivity with peaks vs without peaks

S/N S/N = 0.78 @ t = 10s

0.74 @ t = 15s 0.73 @ t = p-wave arrival resonant peaks in TOBA senitivity will decrease S/N by 22 ~ 27%

→ notch + HPF → only HPF

f -2.5 noise

S/N is ~2 times smaller than f -2 noise (model2) pre-whitened noise spectrum S/N @ t=10s (x 1000) 3rd order HPF seems to be good pre-whitening Mw = 8.5 distance = 300 km cross-mode what kind of pre-whitening filter should be used? ( realtime f 2.5 filter is not available ) how much does S/N change compared with model noise? d e s i g n s e n s i t i v i t y model noise

f -2.5 noise

S/N is ~4 times smaller than f -2 noise (model2) pre-whitened noise spectrum S/N @ t=tp(38.5s) 2nd order HPF seems to be good pre-whitening Mw = 8.5 distance = 300 km cross-mode what kind of pre-whitening filter should be used? ( realtime f 2.5 filter is not available ) how much does S/N change compared with model noise? d e s i g n s e n s i t i v i t y model noise

effect of NN

S/N @ t=10s (x 1000) S/N @ t=tp (38.5s) S/N is ~20 times smaller than f -2 noise (model2) S/N is ~4000 times smaller than f -2 noise (model2)

temperature NN at 100m underground is assumed (maybe overestimated)

Mw = 8.5 distance = 300 km cross-mode

small!

d e s i g n s e n s i t i v i t y w i t h N N ( v = 1 m / s , d e p t h = 1 m )

summary of realistic noise

• one resonant peak will decrease S/N by 5~15%
• total effect in TOBA is about 22~27%
• f -2.5 thermal noise decreases S/N by ~2 times (t=10s)
• r by ~4 times (t=38s) than f -2 model noise
• proper choice of pre-whitening filter depends on time
• filter bank (multiple filtered signals to monitor)
• or is there any better signal processing?
• temperature NN degrades S/N very much
• but temperature NN is maybe overestimated
• more accurate calculation is necessary

what I’m interested in next

• optimal signal processing
• detectability of actual earthquakes with TOBA vs SGG
• Alaska, Japan, California, etc...
• temperature NN
• focal mechanism estimation using 5 components
• depth dependence

future upgrade possibility of TOBA

• increasing in size is quite effective for EEW

thermal noise ∝ ~(length)-2.5

shot noise ∝ ~(length)-2.5 ( laser power ∝ (mass), sensor response ∝ (length) )

current design (without peaks)

a lit a little w tle work o

• n t

tem emper eratu ture N e NN

temperature NN @ 100m depth

infrasoundNN @ 100m depth

instrument nosie temperature fluctuation (conceptual drawing) airflow Taylor‘s hypothesis: temperature distribution is transported as a frozen pattern maybe overestimated at f < 0.1Hz some assumptions can break at these frequencies

we need to check the model

temperature fluctuation structure function

• measured temperature structure function

correlation of fluctuation between two distant points cT

2 = 0.2 K2m-2/3 was assumed

in previous calculation (Creighton et al., 2008) 3e-4

need to check the paper

that’s all about my work

Siteseeing in France

• I visited many places

Mont Saint-Michel Arc de triomphe Notre-dame Musée d'Orsay tour Effel Seine river

Fête nationale ( 7/14)

smoke from aircraft new national flag fireworks around La tour Eiffel milirary parade in the morning

Foucault’s pendulum @ Panthéon

• resonant frequency ~ 60mHz (!)

suspended from near the top

End