Polynomial Chaos M.Iskandarani, A.Srinivasan, W.C. Thacker, O. Knio Background.

What? How?

Overview of Polynomial Chaos Methods for Uncertainty Quantication

with Application to HYCOM

Inow example
Setup. Results

Mohamed Iskandarani1 Ashwanth Srinivasan1 Carlisle Thacker12 Omar Knio3

1 University 2 Atlantic

Omar

of Miami

Oceanographic and Meteorological Laboratory

3 Johns

Hopkins University

February 2011

Outline

Polynomial Chaos M.Iskandarani, A.Srinivasan, W.C. Thacker, O. Knio Background.

1 Background.

What? How?

What is polynomial chaos? How does polynomial chaos work?

Inow example
Setup. Results

Omar

2 Uncertain inow through Yucatan Straits.

Setup. Results.

3 Omar: advanced techniques.

Outline

Polynomial Chaos M.Iskandarani, A.Srinivasan, W.C. Thacker, O. Knio Background.

1 Background.

What? How?

What is polynomial chaos? How does polynomial chaos work?

Inow example
Setup. Results

Omar

2 Uncertain inow through Yucatan Straits.

Setup. Results.

3 Omar: advanced techniques.

What is polynomial chaos?

Polynomial Chaos M.Iskandarani, A.Srinivasan, W.C. Thacker, O. Knio Background.

What?

Idea of polynomial chaos originated with Norbert

How?

Wiener in 1938 before computers.

It is being used by engineers to assess how

Inow example
Setup. Results

Omar

uncertainties in a models inputs manifest in its outputs.

It can be much more efcient than Monte Carlo

methods.
Can it be useful to oceanographers?

Why polynomial? Why chaos?

Polynomial Chaos M.Iskandarani, A.Srinivasan, W.C. Thacker, O. Knio Background.

What? How?

Chaos simply refers to uncertainty. Nothing to do

Inow example
Setup. Results

with strange attractors?

Want to compute how uncertainties of a dynamical

Omar

systems inputs manifest in its outputs.

Polynomial refers to use of polynomial expansions

to propagate uncertainties.
Idea is to exploit orthogonality of the polynomials.

HYCOM: uncertain inputs.

Polynomial Chaos M.Iskandarani, A.Srinivasan, W.C. Thacker, O. Knio Background.

Initial conditions. Boundary conditions. Forcing. Parameters. Polynomial Chaos can handle only a limited number

What? How?

Inow example
Setup. Results

Omar

of uncertain inputs.
But it focuses on all likely values of those few

uncertain inputs.

HYCOM: uncertain outputs.

Polynomial Chaos M.Iskandarani, A.Srinivasan, W.C. Thacker, O. Knio

Every eld at every time. Value of a particular eld at a particular point and a

Background.
What? How?

particular time.
Derived quantities,

Inow example
Setup. Results

Omar

e.g. maximum of the Meridional overturning stream function.

Polynomial Chaos allows focus to be on points of

interest.
Not necessary to explore all uncertainties

simultaneously.

Simplest case: one uncertain input, one output of interest.

Call the uncertain input and the output . Uncertainty of is specied via its pdf ( ). Want pdf of or at least information about how it

Polynomial Chaos M.Iskandarani, A.Srinivasan, W.C. Thacker, O. Knio Background.

What? How?

Inow example
Setup. Results

varies as varies.
Basic Idea: Express output as a polynomial series.

Omar

( ) = 0 + 1 P1 ( ) + 2 P2 ( ) + . . .
Orthogonal polynomials Pk are related to pdf .

Pj ( )Pk ( )( )d = j ,k

Simplest case, continued: one uncertain input, one output of interest

( ) = 0 + 1 P1 ( ) + 2 P2 ( ) + . . . Pj ( )Pk ( )( )d = j ,k
Is the series guaranteed to converge? In practice, it must be truncated. How to compute the coefcients 0 , 1 , 2 , . . .?

Polynomial Chaos M.Iskandarani, A.Srinivasan, W.C. Thacker, O. Knio Background.

What? How?

Inow example
Setup. Results

Omar

k =

1 Nk

( )Pk ( )( )d
2 Pk ( )( )d

Nk =

Use Gaussian quadrature to evaluate the integrals.

Simplest case, continued: one uncertain input, one output of interest

Coefcients:

Polynomial Chaos M.Iskandarani, A.Srinivasan, W.C. Thacker, O. Knio Background.

What?

1 k = Nk
Gaussian Quadrature:

How?

( )Pk ( )( )d

Inow example
Setup. Results

Omar

( )Pk ( )( )d
p

(p )Pk (p )wp

Quadrature points: p Quadrature weights: wp Computing outputs (p ) for inputs at quadrature

points p requires multiple model runs.

How many quadrature points (runs) are needed?

Several outputs of interest, one uncertain input.

Two outputs:

Polynomial Chaos M.Iskandarani, A.Srinivasan, W.C. Thacker, O. Knio Background.

What? How?

Inow example

( ) = 0 + 1 P1 ( ) + 2 P2 ( ) + . . . ( ) = 0 + 1 P1 ( ) + 2 P2 ( ) + . . .
Two or more outputs require no more runs than does

Setup. Results

Omar

one output.
Just save values for all outputs of interest, , , . . .. More coefcients, so more quadrature integrals are

needed.
Quadrature integrals are computationally cheap. Can examine uncertainty of an entire eld.

How to use expansion coefcients?

mean:

Polynomial Chaos M.Iskandarani, A.Srinivasan, W.C. Thacker, O. Knio

=
variance:

( )( )d = 0
k max

Background.
What? How?

Inow example
Setup.

( 0 ) =
k =1

2 k

Results

Omar

covariance:
k max

( 0 )( 0 ) =
k =1

k k

Generate a cheap ensemble.

( ) = 0 + 1 P 1 ( ) + 2 P 2 ( ) + . . .

Two uncertain inputs, one output of interest

Call the uncertain inputs and . Uncertainties of are specied via joint pdf (, ). Now have polynomial series in two variables:

Polynomial Chaos M.Iskandarani, A.Srinivasan, W.C. Thacker, O. Knio Background.

What? How?

Inow example
Setup.

(, ) = 0 + 1 P1 (, ) + 2 P2 (, ) + . . .
If uncertain inputs are independent, pdf factors:

Results

Omar

(, ) = ( ) ()
Then 2D quadrature reduces to two 1D quadratures. Number of quadrature points (runs) is squared. Curse of dimensionality. Sparse cubature might provide economy when

exploring consequences of several uncertain inputs.

Outline

Polynomial Chaos M.Iskandarani, A.Srinivasan, W.C. Thacker, O. Knio Background.

1 Background.

What? How?

What is polynomial chaos? How does polynomial chaos work?

Inow example
Setup. Results

Omar

2 Uncertain inow through Yucatan Straits.

Setup. Results.

3 Omar: advanced techniques.

Problem: How do uncertainties of the Yucatan inow manifest within the Gulf of Mexico?

Polynomial Chaos M.Iskandarani, A.Srinivasan, W.C. Thacker, O. Knio Background.

What? How?

Inow example
Setup.

Need to quantify inow uncertainties. Inow is characterized by several 2D time-varying

Results

Omar

elds.
Computational cost increases dramatically with

number of uncertain parameters.

How to characterize uncertainties of inow with only

a few parameters?

How to characterize uncertainties of inow with only a few independent parameters?

Use multivariate EOFs to characterize 2D spatial

Polynomial Chaos M.Iskandarani, A.Srinivasan, W.C. Thacker, O. Knio Background.

What? How?

patterns of inow uncertainty.

Use corresponding principal components to

Inow example
Setup. Results

Omar

characterize their temporal variability.

Each modes amplitude is assumed to have a

Gaussian pdf.
Hermite polynomials Gauss-Hermite quadrature. Quadrature points dictate the required HYCOM runs. Each run is the sum of a "favorite" inow and its

particular EOF contributions.

Meridional velocity component of EOF.

Polynomial Chaos M.Iskandarani, A.Srinivasan, W.C. Thacker, O. Knio Background.

What? How?

Inow example
Setup. Results

Omar

Eigenvalue spectrum.

Polynomial Chaos M.Iskandarani, A.Srinivasan, W.C. Thacker, O. Knio Background.

What? How?

Inow example
Setup. Results

Omar

Contours of 17 cm SSH from quadrature ensemble of 17 cm runs.

17 (cm) SSH Contours Day 090 17 (cm) SSH Contours Day 150
30oN 30oN

Polynomial Chaos M.Iskandarani, A.Srinivasan, W.C. Thacker, O. Knio Background.

What? How?

27oN

27oN

Longitude

Longitude

Inow example
24oN

24oN

Setup. Results

21oN

21oN

Omar

18oN

96oW

92oW

88oW

84oW

80oW

18oN

96oW

92oW

88oW

84oW

80oW

Latitude

Latitude

17 (cm) SSH Contours Day 240

17 (cm) SSH Contours Day 300

30oN

30oN

27oN

27oN

Longitude

24oN

Longitude
96oW 92oW 88oW 84oW 80oW

24oN

21 N

21 N

18oN

18oN

96oW

92oW

88oW

84oW

80oW

Latitude

Latitude

Mean sea-surface height.

Mean SSH(m) Day090 Mean SSH(m) Day150
30 N
o

Polynomial Chaos M.Iskandarani, A.Srinivasan, W.C. Thacker, O. Knio Background.

What? How?

30 N

27 N

27 N

Longitude

24 N

Longitude

24 N

Inow example
Setup. Results

21 N

21 N

Omar
18oN 96oW 92oW 88oW 84oW 80oW 18oN 96oW 92oW 88oW 84oW 80oW

Latitude
0.4 0.2 0 0.2 0.4 0.6 0.4 0.2

Latitude
0 0.2 0.4 0.6

Mean SSH(m) Day240

Mean SSH(m) Day300

30 N

30 N

27 N

27 N

Longitude

24oN

Longitude
96oW 92oW 88oW 84oW 80oW

24oN

21oN

21oN

18oN

18oN

96oW

92oW

88oW

84oW

80oW

Latitude
0.4 0.2 0 0.2 0.4 0.6 0.4 0.2

Latitude
0 0.2 0.4 0.6

Standard deviation of sea-surface height.

Std dev. SSH(m) Day 090 Std dev. SSH(m) Day 150
30 N
o

Polynomial Chaos M.Iskandarani, A.Srinivasan, W.C. Thacker, O. Knio Background.

What? How?

30 N

27 N

27 N

Longitude

24 N

Longitude

24 N

Inow example
Setup. Results

21 N

21 N

Omar
18oN 96oW 92oW 88oW 84oW 80oW 18oN 96oW 92oW 88oW 84oW 80oW

Latitude
0 0.2 0.4 0.6 0.8 1 0 0.2

Latitude
0.4 0.6 0.8 1

Std dev. SSH(m) Day 240

Std dev. SSH(m) Day 300

30 N

30 N

27 N

27 N

Longitude

24oN

Longitude
96oW 92oW 88oW 84oW 80oW

24oN

21oN

21oN

18oN

18oN

96oW

92oW

88oW

84oW

80oW

Latitude
0 0.2 0.4 0.6 0.8 1 0 0.2

Latitude
0.4 0.6 0.8 1

Convergence of series for SSH standard deviation for day 90.

Std dev. SSH(m) Day 090 Std dev. SSH(m) Day 090
30oN 30oN

Polynomial Chaos M.Iskandarani, A.Srinivasan, W.C. Thacker, O. Knio Background.

What? How?

27oN

27oN

Longitude

Longitude

Inow example
24oN

24oN

Setup. Results

21oN

21oN

Omar
96 W
o

18 N

96 W

92 W

88 W

84 W

80 W

18 N

92 W

88 W

84 W

80 W

Latitude
0.05 0.1 0.15 0.2 0.05

Latitude
0.1 0.15 0.2

Std dev. SSH(m) Day 090

Std dev. SSH(m) Day 090

30 N

30 N

27oN

27oN

Longitude

24 N

Longitude
o o o o o

24 N

21 N

21 N

18 N

96 W

92 W

88 W

84 W

80 W

18 N

96 W

92 W

88 W

84 W

80 W

Latitude
0.05 0.1 0.15 0.2 0.05

Latitude
0.1 0.15 0.2

Convergence of series for SSH standard deviation for day 300.

Std dev. SSH(m) Day 300 Std dev. SSH(m) Day 300
30oN 30oN

Polynomial Chaos M.Iskandarani, A.Srinivasan, W.C. Thacker, O. Knio Background.

What? How?

27oN

27oN

Longitude

Longitude

Inow example
24oN

24oN

Setup. Results

21oN

21oN

Omar
96 W
o

18 N

96 W

92 W

88 W

84 W

80 W

18 N

92 W

88 W

84 W

80 W

Latitude
0.05 0.1 0.15 0.2 0.25 0.3 0.05

Latitude
0.1 0.15 0.2

Std dev. SSH(m) Day 300

Std dev. SSH(m) Day 300

30 N

30 N

27oN

27oN

Longitude

24 N

Longitude
o o o o o

24 N

21 N

21 N

18 N

96 W

92 W

88 W

84 W

80 W

18 N

96 W

92 W

88 W

84 W

80 W

Latitude
0.05 0.1 0.15 0.2 0.02 0.04

Latitude
0.06 0.08 0.1 0.12 0.14

Covariance of SSH with SSH at one point for days 90, 150, 240, and 300.
Covariance SSHSSH Day 090 Covariance SSHSSH Day 150
30oN 30oN

Polynomial Chaos M.Iskandarani, A.Srinivasan, W.C. Thacker, O. Knio Background.

What? How?

27oN

27oN

Longitude

Longitude

Inow example
24oN

24oN

Setup. Results

21oN

21oN

Omar
96 W
o

18 N

96 W

92 W

88 W

84 W

80 W

18 N

92 W

88 W

84 W

80 W

Latitude
0.3 0.2 0.1 0 0.1 0.2 0.3 0.2 0.1

Latitude
0 0.1 0.2 0.3

Covariance SSHSSH Day 240

Covariance SSHSSH Day 300

30 N

30 N

27oN

27oN

Longitude

24 N

Longitude

24 N

21 N

21 N

18 N

96 W

92 W

88 W

84 W

80 W

18 N

96 W

92 W

88 W

84 W

80 W

Latitude
2 1 0 1 2 1 0

Latitude
1 2 3

Covariance of u,v-velocity with SSH at one point for day 90.

Polynomial Chaos M.Iskandarani, A.Srinivasan, W.C. Thacker, O. Knio Background.

What? How?

Inow example
Covariance sshvvel Day 300 Covariance sshuvel Day 300

Setup. Results

30 N

30 N

Omar

27oN

27oN

Longitude

24oN

Longitude

24oN

21oN

21oN

18oN

96oW

92oW

88oW

84oW

80oW

18oN

96oW

92oW

88oW

84oW

80oW

Latitude
0.05 0 0.05 0.1

Latitude
0.06 0.04 0.02 0 0.02 0.04 0.06

Kernel density estimates for mixed-layer depth at day 90 from articial ensemble.
PDF of Mixed Layer depth[m] at 86W/24N Day 90
0.35 1 0.9 0.3 0.8 0.25

Polynomial Chaos M.Iskandarani, A.Srinivasan, W.C. Thacker, O. Knio Background.

What? How?

PDF of Mixed Layer depth[m] at 86W/24N Day 150

0.7

0.6

Density

Density

0.2

Inow example
Setup. Results

0.5

0.15

0.4

0.1

0.3

Omar

0.2 0.05 0.1

0 55

56

57

58

59

60

61

62

63

0 39

40

41

42

43

44

45

46

Mixed Layer Depth [m] PDF of Mixed Layer depth[m] at 86W/24N Day 240
0.8 0.14

Mixed Layer Depth [m] PDF of Mixed Layer depth[m] at 86W/24N Day 300

0.7

0.12

0.6 0.1 0.5

Density

0.4

Density
14.5 15 15.5 16 16.5 17 17.5

0.08

0.06

0.3 0.04 0.2

0.1

0.02

0 14

0 42

44

46

48

50

52

54

56

58

Mixed Layer Depth [m]

Mixed Layer Depth [m]

Outline

Polynomial Chaos M.Iskandarani, A.Srinivasan, W.C. Thacker, O. Knio Background.

1 Background.

What? How?

What is polynomial chaos? How does polynomial chaos work?

Inow example
Setup. Results

Omar

2 Uncertain inow through Yucatan Straits.

Setup. Results.

3 Omar: advanced techniques.