application of normal mode functions for the improved

19
(e) 0 days (f) 5 days (g) 10 days CSIRO Application of normal mode functions for the improved balance in the CAFE data assimilation system and characterisation of modes of variability Vassili Kitsios & Terry O’Kane 2 nd International Conference on Seasonal to Decadal Prediction (S2D), 17 th -21 st , Boulder, CO, USA OCEANS AND ATMOSPHERE www.csiro.au

Upload: others

Post on 27-Dec-2021

1 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Application of normal mode functions for the improved

15◦S7.5◦S

0◦7.5◦N15◦N

(a) -20 days

15◦S7.5◦S

0◦7.5◦N15◦N

(b) -15 days

15◦S7.5◦S

0◦7.5◦N15◦N

(c) -10 days

15◦S7.5◦S

0◦7.5◦N15◦N

(d) -5 days

15◦S7.5◦S

0◦7.5◦N15◦N

(e) 0 days

15◦S7.5◦S

0◦7.5◦N15◦N

(f) 5 days

15◦S7.5◦S

0◦7.5◦N15◦N

(g) 10 days

15◦S7.5◦S

0◦7.5◦N15◦N

(h) 15 days

15◦S7.5◦S

0◦7.5◦N15◦N

0◦ 90◦E 180◦ 90◦W0◦

(i) 20 days

CSIRO

Application of normal mode functions for theimproved balance in the CAFE data assimilation

system and characterisation of modes of variabilityVassili Kitsios & Terry O’Kane

2nd International Conference on Seasonal to Decadal Prediction (S2D), 17th-21st, Boulder, CO, USA

OCEANS AND ATMOSPHEREwww.csiro.au

Page 2: Application of normal mode functions for the improved

Application of normal mode functions for the improved balance in the CAFE data assimilation system and characterisation of modes of variability : Slide 2 of 19

Approach

• Will focus on the characterisation of the Madden Julian Oscillation (MJO)via a Normal Mode Functions (NMF) decomposition of JRA55, and dis-cuss implications for Normal Mode Inialisation (NMI).• MJO is a mode of variability resulting from coupled tropical deep convec-

tion and atmospheric dynamics.• Use key MJO properties to identify representatives NMFs:

– Eastward propagating– Dominant variance over intra-seasonal timescales: 30-90 days.– Tropics centric dynamics.– Horizontal velocity field has a dominant longitudinal wave of k = 1.– Dominant component of the zonal velocity is symmetric about the

equator.• Using one such mode we produce phase and conditional averages of:

– velocity potential to illustrate atmospheric dynamics– outgoing longwave radiation to illustrate convection

• Acknowledge Zagar for sharing the NCAR NMF code, MODES.

Page 3: Application of normal mode functions for the improved

Application of normal mode functions for the improved balance in the CAFE data assimilation system and characterisation of modes of variability : Slide 3 of 19

What are Normal Mode Functions ?

• Decompose 3D (λ, φ, σ) velocity (u, v) and geopotential height (h) fieldsinto horizonal and vertical scales, and mode type, using the eigensolu-tion of the linearised primitive equations on a sphere.• Each scale decomposed into: Balanced Component (BAL) ; Eastward

Inertial Gravity Wave (EIG) ; Westward Inertial Gravity Wave (WIG).• Vertical Structure Functions (VSF), Gm in σ coordinates.

• Longitudinal (λ) waves eikλ

• Meridional (φ) Horizonal Structure Functions (HSF) of wind and height(U, V,H) for EIG, WIG and BAL modes.

[u(σ, λ, φ, t)v(σ, λ, φ, t)h(σ, λ, φ, t)

]=

M∑

m=1

Gm(σ)

[ √gDm 0 00

√gDm 0

0 0 gDm

]K∑

k=−Keikλ

N∑

n=0

EIG,WIG,BAL∑

p

[U(φ)−iV (φ)H(φ)

]p

knm

χpknm(t)

• Complex coefficients, χpknm(t), represent contributions of each compo-nent.

Page 4: Application of normal mode functions for the improved

Application of normal mode functions for the improved balance in the CAFE data assimilation system and characterisation of modes of variability : Slide 4 of 19

Vertical Structure Functions

• VSF given by solution of the VSF eigenvalue problem (EVP).• VSF EVP requires only a time and horizontally averaged static stability

profile in σ coordinates.• Equivalent heights Dm (eigenvalues) are indicative of vertical scale.• VSFGm(σ) (eigenvectors) havem−1 zero crossings. G1 is the barotropic,G2 the first baroclinic, G3 the second baroclinic, ...• For large m, the VSF represent boundary layer processes.

101 102 103 104 105

Γ0 [K]

100

101

102

103

σ×p 0

[hPa]

(a) static stability ≡ Γ0

0 10 20 30 40vertical mode number

10−1

100

101

102

103

104

Dm[m

etres]

(b) equivalent height ≡ Dm

−0.5 0.0 0.5amplitude [unitless]

100

101

102

103

σ×p 0

[hPa]

(c) vertical structure functions

m=1

m=2

m=8

m=15

m=28

Page 5: Application of normal mode functions for the improved

Application of normal mode functions for the improved balance in the CAFE data assimilation system and characterisation of modes of variability : Slide 5 of 19

Horizontal Structure Functions

• HSF given by solution of a HSF EVP for each equivalent height (Dm).• Eigenvectors give meridional dependence, with mode types (EIG, WIG,

BAL) defined by symmetry properties.• Frequency ν (eigenvalue) is indicative of temporal scale.

−30 −20 −10 0 10 20 30longitudinal wavenumber (k)

100

101

102

103

104

1/νq knm[days]

(a) timescales for m = 28

BAL

EIG

WIG

MRG

KW

0 10 20 30 40vertical mode number (m)

100

101

102

103

1/νq knm[days]

90days

60days

30days

(b) timescales

BAL(1,1,m)

WIG(1,0,m)

EIG(1,0,m)

−1 0 1amplitude

−75

−50

−25

0

25

50

75

latitude(φ)

(c) U qknm(φ)

EIG(1,0,28)

EIG(1,0,8)

BAL(1,1,15)

BAL(1,1,8)

• Recall for MJO: k = 1; tropics centric; U is symmetric• Only the EIG n = 0, WIG n = 0, and BAL n = 1 are tropics centric with

the appropriate symmetries.• EIG m = 23− 32 ; BAL m = 11− 22 have intra-seasonal timescales.

Page 6: Application of normal mode functions for the improved

Application of normal mode functions for the improved balance in the CAFE data assimilation system and characterisation of modes of variability : Slide 6 of 19

Energy Contribution χpknm(t)χp∗knm(t) of NMFs in JRA55

• BAL dominates for low k, IG dominate for high k• BAL dominates for all vertical modes m• For (k, n) = (1, 1) BAL HSF eigenvectors have MJO-like properties:

– BAL (k, n,m) = (1, 1, 2) and (1, 1, 2) local peaks in energy, but toofast

– BAL (k, n,m) = (1, 1, 15) has HSF eigenvalue of 46 days.

• For (k, n) = (1, 0) EIG HSF eigenvectors also have MJO-like properties:

– EIG (k, n,m) = (1, 0, 8) has most energy, but timescale too fast

– EIG (k, n,m) = (1, 0, 28) has HSF eigenvalue of 56 days.

100 101 102

zonal wavenumber (k)

10−4

10−3

10−2

10−1

100

101

102

103

104

nonzonal

energy

[J/kg]

(a)

BAL

EIG

WIG

total

0 10 20 30vertical mode (m)

(b)

0 10 20 30vertical mode (m)

(c) (k, n) = (1, 1)

0 10 20 30vertical mode (m)

(d) (k, n) = (1, 0)

Page 7: Application of normal mode functions for the improved

Application of normal mode functions for the improved balance in the CAFE data assimilation system and characterisation of modes of variability : Slide 7 of 19

Cross-spectral Analysis of Candidate NMFs

• All candidate modes are tropics centric and have the appropriate symme-tries.• Only EIG (k, n,m) = (1, 0, 28) has an intra-seasonal timescale, and

propagates eastward, but has low energy.• Cross-spectral analysis identifies only slow intra-seasonal timescales are

coherent between EIG (k, n,m) = (1, 0, 28) and the more energeticmodes.• Fast Kelvin wave removed from energetic EIG (k, n,m) = (1, 0, 8).

2 4 6 8 10k

0.0

0.1

0.2

0.3

0.4

0.5

f[1/days]

30days

(a) PSD of EIG(k,0,8) [J/kg]

10−4

10−3

10−2

10−1

100

101

2 4 6 8 10k

(b) PSD of EIG(k,0,28) [J/kg]

10−4

10−3

10−2

10−1

2 4 6 8 10k

(c) coherence

0.0

0.2

0.4

0.6

0.8

1.0

2 4 6 8 10k

(d) coherence output [J/kg]

10−4

10−3

10−2

10−1

100

101

5 10 15 20 25 30 35m

0.0

0.1

0.2

0.3

0.4

0.5

f[1/days]

(e) coherence output of EIG(1,0,m) [J/kg]

10−4

10−3

10−2

10−1

100

101

102

5 10 15 20 25 30 35m

(f) coherence output of BAL(1,1,m) [J/kg]

10−4

10−3

10−2

10−1

100

101

102

Page 8: Application of normal mode functions for the improved

Application of normal mode functions for the improved balance in the CAFE data assimilation system and characterisation of modes of variability : Slide 8 of 19

Cross-spectral Analysis of Candidate NMFs

• All candidate modes are tropics centric and have the appropriate symme-tries.• Only EIG (k, n,m) = (1, 0, 28) has an intra-seasonal timescale, and

propagates eastward, but has low energy.• Cross-spectral analysis identifies only slow intra-seasonal timescales are

coherent between EIG (k, n,m) = (1, 0, 28) and the more energeticmodes.• Repeated for all vertical scales (m) with k = 1 for BAL and EIG.

2 4 6 8 10k

0.0

0.1

0.2

0.3

0.4

0.5

f[1/days]

30days

(a) PSD of EIG(k,0,8) [J/kg]

10−4

10−3

10−2

10−1

100

101

2 4 6 8 10k

(b) PSD of EIG(k,0,28) [J/kg]

10−4

10−3

10−2

10−1

2 4 6 8 10k

(c) coherence

0.0

0.2

0.4

0.6

0.8

1.0

2 4 6 8 10k

(d) coherence output [J/kg]

10−4

10−3

10−2

10−1

100

101

5 10 15 20 25 30 35m

0.0

0.1

0.2

0.3

0.4

0.5

f[1/days]

(e) coherence output of EIG(1,0,m) [J/kg]

10−4

10−3

10−2

10−1

100

101

102

5 10 15 20 25 30 35m

(f) coherence output of BAL(1,1,m) [J/kg]

10−4

10−3

10−2

10−1

100

101

102

• Other candidate modes highlighted in coherence output clusters.

Page 9: Application of normal mode functions for the improved

Application of normal mode functions for the improved balance in the CAFE data assimilation system and characterisation of modes of variability : Slide 9 of 19

Phase Average on Basis of EIG (k, n,m) = (1, 0, 28)

• Phase angle calculated from complex χpnmk. Dates associated with eachphase angle octant averaged. All modes contribute to phase averages.

• Velocity potential is a propagating longitudinal wave, with a vertical signchange representing upper level divergence and lower level convergence.

Page 10: Application of normal mode functions for the improved

Application of normal mode functions for the improved balance in the CAFE data assimilation system and characterisation of modes of variability : Slide 10 of 19

Phase Average on Basis of EIG (k, n,m) = (1, 0, 28)

• Phase angle calculated from complex χpnmk. Dates associated with eachphase angle octant averaged. All modes contribute to phase averages.

• OLR has a dipole pattern over the maritime continent, tracking with veloc-ity potential of like sign.

Page 11: Application of normal mode functions for the improved

Application of normal mode functions for the improved balance in the CAFE data assimilation system and characterisation of modes of variability : Slide 11 of 19

Large and Persistent MJO Events

• Start and end of each event defined as discontinuities in phase angle ofEIG (k, n,m) = (1, 0, 28).• Persistent events have a continuous phase for longer than 270◦.• Large events have a magnitude in the upper quartile.• Composite average magnitude is greater than background for 20 days be-

fore and after day 0.• Composite average phase angle indicates eastward propagation.

• Dates associated with each phase shift identified and averaged to producecomposite fields of velocity potential and outgoing longwave radiation.

Page 12: Application of normal mode functions for the improved

Application of normal mode functions for the improved balance in the CAFE data assimilation system and characterisation of modes of variability : Slide 12 of 19

Velocity Potential at 200hPa - Wave Like

15◦S7.5◦S

0◦7.5◦N15◦N

(a) -20 days

15◦S7.5◦S

0◦7.5◦N15◦N

(b) -15 days

15◦S7.5◦S

0◦7.5◦N15◦N

(c) -10 days

15◦S7.5◦S

0◦7.5◦N15◦N

(d) -5 days

15◦S7.5◦S

0◦7.5◦N15◦N

(e) 0 days

15◦S7.5◦S

0◦7.5◦N15◦N

(f) 5 days

15◦S7.5◦S

0◦7.5◦N15◦N

(g) 10 days

15◦S7.5◦S

0◦7.5◦N15◦N

(h) 15 days

15◦S7.5◦S

0◦7.5◦N15◦N

0◦ 90◦E 180◦ 90◦W0◦

(i) 20 days

Page 13: Application of normal mode functions for the improved

Application of normal mode functions for the improved balance in the CAFE data assimilation system and characterisation of modes of variability : Slide 13 of 19

Outgoing Longwave Radiation at 200hPa - Dipole Like

15◦S7.5◦S

0◦7.5◦N15◦N

(a) -20 days

15◦S7.5◦S

0◦7.5◦N15◦N

(b) -15 days

15◦S7.5◦S

0◦7.5◦N15◦N

(c) -10 days

15◦S7.5◦S

0◦7.5◦N15◦N

(d) -5 days

15◦S7.5◦S

0◦7.5◦N15◦N

(e) 0 days

15◦S7.5◦S

0◦7.5◦N15◦N

(f) 5 days

15◦S7.5◦S

0◦7.5◦N15◦N

(g) 10 days

15◦S7.5◦S

0◦7.5◦N15◦N

(h) 15 days

15◦S7.5◦S

0◦7.5◦N15◦N

0◦ 90◦E 180◦ 90◦W0◦

(i) 20 days

Page 14: Application of normal mode functions for the improved

Application of normal mode functions for the improved balance in the CAFE data assimilation system and characterisation of modes of variability : Slide 14 of 19

Instantaneous Comparison of NMFs and WH

• Wheeler & Hendon index based on first two PCs of meridionally averagedu at 200hPa, 850hPa and OLR within 15◦S and 15◦N .• Compare to tropics centric NMFs, with MJO-like symmetries:

– BAL(k, n,m) = (1, 1, 8) : energetic, but westward, timescale too fast.

– EIG(k, n,m) = (1, 0, 8) : energetic, but timescale too fast.

– BAL(k, n,m) = (1, 1, 15) : intra-seasonal timescale, but westward.

– EIG(k, n,m) = (1, 0, 28) : intra-seasonal timescale, eastward, notenergetic.

• Correlation of BAL (k, n,m) = (1, 1, 8) with WH when filtered to retaintemporal scales coherent with EIG (k, n,m) = (1, 0, 28) is 0.78.

2001 2003 2005 2007 2009 2011 2013 20150.00

0.25

0.50

0.75

1.00

1.25

1.50

1.75

2.00

√〈W

H〉 M

A

0.0

0.5

1.0

1.5

2.0

2.5

3.0

√〈E

q knm〉 M

A

(a) 91 day moving average

BAL(1,1,8) unfiltered BAL(1,1,8) EIG(1,0,8) BAL(1,1,15) EIG(1,0,28) unfiltered WH

2012-0

2-05

2012-0

2-19

2012-0

3-04

2012-0

3-18

2012-0

4-01

2012-0

4-15

−150

−100

−50

0

50

100

150

phaseangle

(b) phase angle

−2 −1 0 1 2RMM1

−2

−1

0

1

2

RMM2

(c) phase space WH

−2 −1 0 1 2 3real(χq

knm)√Dmg/2

−2

−1

0

1

2

imaginary(χq knm)√

Dmg/2

(d) phase space BAL(1,1,8)

Page 15: Application of normal mode functions for the improved

Application of normal mode functions for the improved balance in the CAFE data assimilation system and characterisation of modes of variability : Slide 15 of 19

Instantaneous Comparison of NMFs and WH

• Wheeler & Hendon index based on first two PCs of meridionally averagedu at 200hPa, 850hPa and OLR within 15◦S and 15◦N .• In comparision to WH, coherence filtered BAL(k, n,m) = (1, 1, 8) mode

exhibits:– high correlation in magnitude over the entire time series (1958-2016)– consistent phase propagation for a specific event (Jan 2012)– consistent spiralling in for a specific event in phase space

2001 2003 2005 2007 2009 2011 2013 20150.00

0.25

0.50

0.75

1.00

1.25

1.50

1.75

2.00

√〈W

H〉 M

A

0.0

0.5

1.0

1.5

2.0

2.5

3.0

√〈E

q knm〉 M

A

(a) 91 day moving average

BAL(1,1,8) unfiltered BAL(1,1,8) EIG(1,0,8) BAL(1,1,15) EIG(1,0,28) unfiltered WH

2012-0

2-05

2012-0

2-19

2012-0

3-04

2012-0

3-18

2012-0

4-01

2012-0

4-15

−150

−100

−50

0

50

100

150

phaseangle

(b) phase angle

−2 −1 0 1 2RMM1

−2

−1

0

1

2

RMM2

(c) phase space WH

−2 −1 0 1 2 3real(χq

knm)√Dmg/2

−2

−1

0

1

2

imaginary(χq knm)√

Dmg/2

(d) phase space BAL(1,1,8)

Page 16: Application of normal mode functions for the improved

Application of normal mode functions for the improved balance in the CAFE data assimilation system and characterisation of modes of variability : Slide 16 of 19

Proposed MJO Skeleton

• Nonlinear interactions of these modes generates an energetic MJO ofeastward propagation and correct phase period.• All modes have MJO-like longitudinal and meridional structure.

�������� ����������������������������

��������� ��������������������� ������� ��

��������������

������������������������������������������� ��������

������������ ������� �������������

��������� ���������

�������������� ����

������� ����

����������

����������

• Interaction strength inferred from cross-spectral analysis.• In the future will calculate the nonlinear transfer terms explicitly.

Page 17: Application of normal mode functions for the improved

Application of normal mode functions for the improved balance in the CAFE data assimilation system and characterisation of modes of variability : Slide 17 of 19

Concluding Remarks

• NMFs decompose a 3-D atmospheric geopotential height and horizon-tal velocity field into scale (zonal, meridional, vertical) and mode class(BAL, EIG, WIG).• MJO-like NMF modes and their interactions were isolated in the JRA-55.• A skeleton physical model of the MJO was proposed.

• Implications for Normal Modes Initialisation:– Since the IG waves have shorter timescales, they are potentially less

predictable over a multi-year period.– Naively one would think that filtering the IG waves would improve pre-

dictability.– However, the EIG waves (even of small vertical scale and low energy)

are shown here to be dynamically important for the MJO.

Page 18: Application of normal mode functions for the improved

Application of normal mode functions for the improved balance in the CAFE data assimilation system and characterisation of modes of variability : Slide 18 of 19

QuestionsVertical Structure Functions Horizontal Structure Functions

101 102 103 104 105

Γ0 [K]

100

101

102

103

σ×p 0

[hPa]

(a) static stability ≡ Γ0

0 10 20 30 40vertical mode number

10−1

100

101

102

103

104

Dm[m

etres]

(b) equivalent height ≡ Dm

−0.5 0.0 0.5amplitude [unitless]

100

101

102

103

σ×p 0

[hPa]

(c) vertical structure functions

m=1

m=2

m=8

m=15

m=28

−30 −20 −10 0 10 20 30longitudinal wavenumber (k)

100

101

102

103

104

1/νq knm[days]

(a) timescales for m = 28

BAL

EIG

WIG

MRG

KW

0 10 20 30 40vertical mode number (m)

100

101

102

103

1/νq knm[days]

90days

60days

30days

(b) timescales

BAL(1,1,m)

WIG(1,0,m)

EIG(1,0,m)

−1 0 1amplitude

−75

−50

−25

0

25

50

75

latitude(φ)

(c) U qknm(φ)

EIG(1,0,28)

EIG(1,0,8)

BAL(1,1,15)

BAL(1,1,8)

Cross-Spectral Analysis Phase Averages2 4 6 8 10k

0.0

0.1

0.2

0.3

0.4

0.5

f[1/days]

30days

(a) PSD of EIG(k,0,8) [J/kg]

10−4

10−3

10−2

10−1

100

101

2 4 6 8 10k

(b) PSD of EIG(k,0,28) [J/kg]

10−4

10−3

10−2

10−1

2 4 6 8 10k

(c) coherence

0.0

0.2

0.4

0.6

0.8

1.0

2 4 6 8 10k

(d) coherence output [J/kg]

10−4

10−3

10−2

10−1

100

101

5 10 15 20 25 30 35m

0.0

0.1

0.2

0.3

0.4

0.5

f[1/days]

(e) coherence output of EIG(1,0,m) [J/kg]

10−4

10−3

10−2

10−1

100

101

102

5 10 15 20 25 30 35m

(f) coherence output of BAL(1,1,m) [J/kg]

10−4

10−3

10−2

10−1

100

101

102

Composite Averages Skeleton Physical Model

15◦S7.5◦S

0◦7.5◦N15◦N

(a) -20 days

15◦S7.5◦S

0◦7.5◦N15◦N

(b) -15 days

15◦S7.5◦S

0◦7.5◦N15◦N

(c) -10 days

15◦S7.5◦S

0◦7.5◦N15◦N

(d) -5 days

15◦S7.5◦S

0◦7.5◦N15◦N

(e) 0 days

15◦S7.5◦S

0◦7.5◦N15◦N

(f) 5 days

15◦S7.5◦S

0◦7.5◦N15◦N

(g) 10 days

15◦S7.5◦S

0◦7.5◦N15◦N

(h) 15 days

15◦S7.5◦S

0◦7.5◦N15◦N

0◦ 90◦E 180◦ 90◦W0◦

(i) 20 days

�������� ����������������������������

��������� ��������������������� ������� ��

��������������

������������������������������������������� ��������

������������ ������� �������������

��������� ���������

�������������� ����

������� ����

����������

����������

Page 19: Application of normal mode functions for the improved

CSIRO

OCEANS AND ATMOSPHEREwww.csiro.au

CSIRO Oceans and Atmosphere

Vassili Kitsiost 0362325312e [email protected] research.csiro.au/dfp/

CSIRO Oceans and Atmosphere

Terry O’Kanete Terrence.O’[email protected] research.csiro.au/dfp/