A Conceptual Introduction to Geophysical Inversion€¦ · A Conceptual Introducon to Geophysical...

23
A Conceptual Introduc0on to Geophysical Inversion Andy Ganse ESS & Applied Physics Lab University of Washington 12 Mar 2012 ˆ x 1 )+ N k=1 (z k - h k (x k )) T R -1 k (z + N -1 k=1 (x k+1 - g k (x k )) T Q -1 k (x k+ Ganse

Transcript of A Conceptual Introduction to Geophysical Inversion€¦ · A Conceptual Introducon to Geophysical...

Page 1: A Conceptual Introduction to Geophysical Inversion€¦ · A Conceptual Introducon to Geophysical Inversion Andy Ganse ESS & Applied Physics Lab University of Washington 12 Mar 2012

A Conceptual Introduc0on to Geophysical Inversion 

Andy Ganse ESS & Applied Physics Lab University of Washington 12 Mar 2012 

L(x1, x2, ...xN ) = (x1 ! x̂1)T P!1

1 (x1 ! x̂1) +N!

k=1

(zk ! hk(xk))T R!1k

(zk ! hk(xk))

+N!1!

k=1

(xk+1 ! gk(xk))T Q!1k

(xk+1 ! gk(xk))

Ganse 

Page 2: A Conceptual Introduction to Geophysical Inversion€¦ · A Conceptual Introducon to Geophysical Inversion Andy Ganse ESS & Applied Physics Lab University of Washington 12 Mar 2012

A li7le about APL 

Ganse 

(my stomping ground…) 

Page 3: A Conceptual Introduction to Geophysical Inversion€¦ · A Conceptual Introducon to Geophysical Inversion Andy Ganse ESS & Applied Physics Lab University of Washington 12 Mar 2012

A conceptual / non‐technical talk 

•  Intro via examples •  Data vs. Model •  DeducGon vs. InducGon •  Probability vs. staGsGcs •  FrequenGst vs. Bayesian •  Math vs. Earth Science •  Parameter esGmaGon vs. inversion •  Uncertainty vs. ResoluGon •  Linear vs. Nonlinear •  Recommended reading 

A terminology extravaganza in a sequence of dichotomies… 

Ganse 

Page 4: A Conceptual Introduction to Geophysical Inversion€¦ · A Conceptual Introducon to Geophysical Inversion Andy Ganse ESS & Applied Physics Lab University of Washington 12 Mar 2012

Intro via examples 

Global seismic inversion:  es?mate Earth’s interior wavespeeds & densi?es from EQ seismograms 

Glacier gravimetry:  es?mate glacier cross‐sec?on from gravity measurements 

Ganse 

Page 5: A Conceptual Introduction to Geophysical Inversion€¦ · A Conceptual Introducon to Geophysical Inversion Andy Ganse ESS & Applied Physics Lab University of Washington 12 Mar 2012

Intro via examples 

Groundwater contamina?on:  es?mate source leakage func?on from groundwater samplings 

Computerized Tomography (CT) scans:  es?mate 3D body interior densi?es from Xray aLen 

Ganse 

Page 6: A Conceptual Introduction to Geophysical Inversion€¦ · A Conceptual Introducon to Geophysical Inversion Andy Ganse ESS & Applied Physics Lab University of Washington 12 Mar 2012

Intro via examples 

Radio doppler gravimetry of planetary bodies:  es?mate density of icy moon interiors 

Ocean boLom (“geoacous?c”) inversion:  es?mate seafloor proper?es from sonar in water 

Ganse 

Page 7: A Conceptual Introduction to Geophysical Inversion€¦ · A Conceptual Introducon to Geophysical Inversion Andy Ganse ESS & Applied Physics Lab University of Washington 12 Mar 2012

Deduc0on vs. Induc0on 

cause effect 

deducGon 

“forward problem” 

inducGon 

“inverse problem” 

Note the common theme in those examples (but there are others):  inferring proper?es of interior from measurements on an exterior. 

predicted_data = somefunction( model_of_interest )

density(x,z) wavespeed(z) temperature(z,t) chemsrcleakage(t) etc… 

gravity(xi) travelGme(depthi) waveintensity(xi,tj) dopplerfreq(tj) chemconcentraGon(xi,tj) 

Ganse 

Page 8: A Conceptual Introduction to Geophysical Inversion€¦ · A Conceptual Introducon to Geophysical Inversion Andy Ganse ESS & Applied Physics Lab University of Washington 12 Mar 2012

Probability vs. Sta0s0cs 

cause effect 

deducGon 

“forward problem” 

inducGon 

“inverse problem” 

‐  Leonard Mlodinow The Drunkard’s Walk 

Stemming straight from the difference between deduc?on and induc?on 

Ganse 

(highly recommended!) 

Page 9: A Conceptual Introduction to Geophysical Inversion€¦ · A Conceptual Introducon to Geophysical Inversion Andy Ganse ESS & Applied Physics Lab University of Washington 12 Mar 2012

•  Frequentists define probability in terms of frequency of repeatable events.�So one can’t know anything about model before the event/experiment.�Most common tool – linear (or iteratively linear) approach to problem.

•  Bayesians define probability in terms of degree of belief.�So one can know about the model before the event/experiement.�Common tools – fancy, computationally-heavy MCMC inversion, but can do linear/iteratively-linear problems too, and also can do filters (eg Kalman).

Frequen0st vs. Bayesian A debate raging for 200+ years in the sta?s?cs community 

The ESS class concentrates on frequenGst tools with iteraGvely linear soluGon techniques – you can only fit so much into one quarter… 

Ganse 

Page 10: A Conceptual Introduction to Geophysical Inversion€¦ · A Conceptual Introducon to Geophysical Inversion Andy Ganse ESS & Applied Physics Lab University of Washington 12 Mar 2012

Data vs. Model 

5

10

15

20

25

30

35

40

0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4

rcv

# a

nd

re

lativ

e a

mp

litu

de

time (sec)

0

20

40

60

80

100

1500 1550 1600 1650 1700 1750 1800 1850 1900

de

pth

be

low

se

aflo

or

(m)

Pwave velocity (m/s)

mtrueminit

R10!mtruesoln"10

±#m

0 0.5 1 1.5 2 2.5 3

0

200

400

600

800

1000

1200

cross range x [km]

depth

h(x

) [m

]

Green=initial estimate; Gray=all models on Lcurve; Red=model for red point on Lcurve

meas stns

0 0.5 1 1.5 2 2.5 3 3.5−5

−4

−3

−2

−1x 10−4

cross range x [km]

corre

cted

gra

v an

omal

ies

[m/s

2 ]

data space (measurements & predicGons of them) 

model space (what we really want to know) 

predicted_data = somefunction( model_of_interest )

Note the different sets of X & Y axes in the two spaces. 

Seafloor acous?c example 

Glacier gravimetry example 

Ganse 

Page 11: A Conceptual Introduction to Geophysical Inversion€¦ · A Conceptual Introducon to Geophysical Inversion Andy Ganse ESS & Applied Physics Lab University of Washington 12 Mar 2012

Data vs. Model 

data space (measurements & predicGons of them) 

model space (what we really want to know) 

!1 !0.8 !0.6 !0.4 !0.2 0 0.2 0.4 0.6 0.8 1!50

!40

!30

!20

!10

0

10

20

30

40

50

These two spaces are the same in the special case of curve fiIng. But that’s only a special case. 

predicted_data = somefunction( model_of_interest )

Same set of X & Y axes in the two spaces in this special case. 

curve fi]ng 

Ganse 

Page 12: A Conceptual Introduction to Geophysical Inversion€¦ · A Conceptual Introducon to Geophysical Inversion Andy Ganse ESS & Applied Physics Lab University of Washington 12 Mar 2012

Math vs. Earth Science 

predicted_data = somefunction( model_of_interest )

model_of_interest = somefunction-1( measured_data )

Why not just @#$% flip it around mathema?cally and call it done?! 

d ( s ) 

d ( si ) 

d ( si ) + noise 

m ( x ) 

m ( x ) d ( s ) 

nuclear sca^ering experiments (inverse sca^ering theory) 

CT scans (Radon transform) 

SomeGmes you can, e.g. : 

Ganse 

Page 13: A Conceptual Introduction to Geophysical Inversion€¦ · A Conceptual Introducon to Geophysical Inversion Andy Ganse ESS & Applied Physics Lab University of Washington 12 Mar 2012

Math vs. Earth Science 

Central issues for inverse problem solu0ons: •  existence  •  uniqueness •  stability •  uncertainty 

predicted_data = somefunction( model_of_interest )

But in many Earth science problems, the geometric coverage is lousy and the noise is great enough that somefuncGon‐1() becomes hopelessly unstable.  We must use instead approaches related to opGmizaGon to do the inversion. 

Ganse 

Why not just @#$% flip it around mathema?cally and call it done?! 

Page 14: A Conceptual Introduction to Geophysical Inversion€¦ · A Conceptual Introducon to Geophysical Inversion Andy Ganse ESS & Applied Physics Lab University of Washington 12 Mar 2012

Parameter es0ma0on vs. inversion parameter es0ma0on:  solve for a handful of discrete values 

inversion:  solve for a conGnuous funcGon (be it 1D, 2D, etc.) – much more involved (although it uses parameter esGmaGon) 

0

20

40

60

80

100

1500 1550 1600 1650 1700 1750 1800 1850 1900

depth

belo

w s

eaflo

or

(m)

Pwave velocity (m/s)

mtrueminit

R10!mtruesoln"10

±#m

param est:  find src x,y  inv: find vel(z) 

but actually… Ganse 

Page 15: A Conceptual Introduction to Geophysical Inversion€¦ · A Conceptual Introducon to Geophysical Inversion Andy Ganse ESS & Applied Physics Lab University of Washington 12 Mar 2012

0

20

40

60

80

100

1500 1550 1600 1650 1700 1750 1800 1850 1900

de

pth

be

low

se

aflo

or

(m)

Pwave velocity (m/s)

mtrueminit

soln!1soln!2soln!3soln!4soln!5soln!6soln!7soln!8soln!9soln!10soln!11soln!13soln!14

Parameter es0ma0on vs. inversion parameter es0ma0on:  solve for a handful of discrete values 

inversion:  solve for a conGnuous funcGon (be it 1D, 2D, etc.) – much more involved (although it uses parameter esGmaGon) 

inv: uh‐oh, many curves produce predicGons that 

fit the data param est:  find src x,y 

Ganse 

Page 16: A Conceptual Introduction to Geophysical Inversion€¦ · A Conceptual Introducon to Geophysical Inversion Andy Ganse ESS & Applied Physics Lab University of Washington 12 Mar 2012

Parameter es0ma0on vs. inversion parameter es0ma0on:  solve for a handful of discrete values 

inversion:  solve for a conGnuous funcGon (be it 1D, 2D, etc.) – much more involved (although it uses parameter esGmaGon) 

inv: uh‐oh, many curves produce predicGons that 

fit the data An intuiGve example via curve‐fi]ng problem: 

Ganse 

Page 17: A Conceptual Introduction to Geophysical Inversion€¦ · A Conceptual Introducon to Geophysical Inversion Andy Ganse ESS & Applied Physics Lab University of Washington 12 Mar 2012

Uncertainty vs. Resolu0on 0

20

40

60

80

100

1500 1550 1600 1650 1700 1750 1800 1850 1900

depth

belo

w s

eaflo

or

(m)

Pwave velocity (m/s)

mtrueminit

soln!1soln!2soln!3soln!4soln!5soln!6soln!7soln!8soln!9soln!10soln!11soln!13soln!14

Four of those many curves now shown separately, with their uncertainGes included around them. 

Choose what you want: Higher‐res solu0ons have larger uncertain0es, lower‐res solu0ons have smaller uncertain0es. 

X‐axis “smearing”  Y‐axis “smearing” 

Ganse 

0

20

40

60

80

100

1500 1550 1600 1650 1700 1750 1800 1850 1900

de

pth

be

low

se

aflo

or

(m)

Pwave velocity (m/s)

mtrueminit

R13!mtruesoln"13

±#m

0

20

40

60

80

100

1500 1550 1600 1650 1700 1750 1800 1850 1900

de

pth

be

low

se

aflo

or

(m)

Pwave velocity (m/s)

mtrueminit

R11!mtruesoln"11

±#m

0

20

40

60

80

100

1500 1550 1600 1650 1700 1750 1800 1850 1900

de

pth

be

low

se

aflo

or

(m)

Pwave velocity (m/s)

mtrueminit

R8!mtruesoln"8

±#m

0

20

40

60

80

100

1500 1550 1600 1650 1700 1750 1800 1850 1900

de

pth

be

low

se

aflo

or

(m)

Pwave velocity (m/s)

mtrueminit

R6!mtruesoln"6

±#m

(Uncertain?es are so narrow in top halves only because this was from a highly idealized synthe?c problem.) 

Page 18: A Conceptual Introduction to Geophysical Inversion€¦ · A Conceptual Introducon to Geophysical Inversion Andy Ganse ESS & Applied Physics Lab University of Washington 12 Mar 2012

Linear vs. Nonlinear predicted_data = somefunction( model_of_interest )

Linear problems: scalability and superposiGon; Gaussians map to Gaussians; 

Computes fast –  jump to soluGon in one step; 

d = F m 

Nonlinear problems: more general (and more common!); Uniqueness, stability, uncertainty take MUCH more effort & interpretaGon; Slower – use sequence of linear subproblems, or use many MC samples. 

d = f ( m ) 

d  m 

Ganse 

contained in the vector  parameterized by the vector 

Page 19: A Conceptual Introduction to Geophysical Inversion€¦ · A Conceptual Introducon to Geophysical Inversion Andy Ganse ESS & Applied Physics Lab University of Washington 12 Mar 2012

The Class:  ESS 523 

•  Overall:  learn how to do linear problems, then set up your nonlinear problem as a sequence of linear ones. 

•  Will extensively use Matlab or Octave (free/awesome GNU clone of Matlab) 

•  Recommended Prerequisite background: •  Basic probability & staGsGcs concepts ‐ 

•  e.g. mean, std dev, variance, covariance, correlaGon •  Linear algebra ‐ 

•  e.g. matrix/vector arithmeGc, transpose, inverse, null space, rank, condiGon number, eigenvalues/vectors, under/over‐determined probs 

•  Fourier transforms (Gme/space  frequency) •  Some idea of connecGon between the class and your research 

•  No tests, but weekly labs and a class project based on your research 

Ganse 

Page 20: A Conceptual Introduction to Geophysical Inversion€¦ · A Conceptual Introducon to Geophysical Inversion Andy Ganse ESS & Applied Physics Lab University of Washington 12 Mar 2012

Shameless plug http://staff.washington.edu/aganse

(also linked via ESS and APL directory pages)

•  Recommended textbooks •  Recommended journal papers •  Links to software and other web resources •  Lecture notes and labs from� the inverse theory class I TA’d.

Ganse 

Page 21: A Conceptual Introduction to Geophysical Inversion€¦ · A Conceptual Introducon to Geophysical Inversion Andy Ganse ESS & Applied Physics Lab University of Washington 12 Mar 2012

Another shameless plug http://staff.washington.edu/aganse

(also linked via ESS and APL directory pages)

Ganse 

Page 22: A Conceptual Introduction to Geophysical Inversion€¦ · A Conceptual Introducon to Geophysical Inversion Andy Ganse ESS & Applied Physics Lab University of Washington 12 Mar 2012

Fortunately, not too many shameless plugs… http://staff.washington.edu/aganse

(also linked via ESS and APL directory pages)

•  Enter wave velocity profiles and watch the rays go! •  Spherical geometry one available too…

Ganse 

Page 23: A Conceptual Introduction to Geophysical Inversion€¦ · A Conceptual Introducon to Geophysical Inversion Andy Ganse ESS & Applied Physics Lab University of Washington 12 Mar 2012

Recommended reading 

Really fantasGc popular book re probability and staGsGcs: The Drunkard’s Walk, by Leonard Mlodinow 

My website (of course!)  –  pages on inverse theory resources, linear and nonlinear filter tutorial, ray‐tracing, and much more. h^p://staff.washington.edu/aganse 

The best frequenGst inverse theory textbook: Parameter Es0ma0on and Inverse Theory, by Aster, Borchers, Thurber 

The best Bayesian inverse theory textbook: Inverse Problem Theory and Model Parameter Es0ma0on, by Albert Tarantola  (available free online!) 

Ganse