CLIVAR ENSO workshop, Paris, Nov. 2010 ENSO in GCMs: overview, progress and challenges Eric Guilyardi IPSL/LOCEAN, Paris, France & NCAS-Climate, Univ

CLIVAR ENSO workshop, Paris, Nov. 2010

ENSO in GCMs: overview, progress and challenges

Eric Guilyardi IPSL/LOCEAN, Paris, France & NCAS-Climate, Univ. Reading, UK

1. ENSO in coupled GCMs

2. Atmosphere feedbacks during ENSO• Dynamical (Bjerknes) feedback• Heat flux feedback

3. Strategies to improve ENSO in models

CLIVAR ENSO workshop, Paris, Nov. 2010 2

El Niño in coupled GCMs – mean state

Trade winds too strong

Both mean and annual cycle

Guilyardi et al. (BAMS 2009)

Tau

x N

iño

4 (

Pa)

IPCC-class models

OBS.

Zonal wind stress in central Pacific (mean and annual cycle)


El Niño in coupled GCMs - amplitude

ENSO amplitude in IPCC AR4 : much too large diversity !

Standard deviation SSTA (C)

EN

SO

Am

pli

tud

e

picntrl2xco2


El Niño in coupled GCMs - frequency

Maximum power of Niño3 SSTA spectra

AchutaRao & Sperber (2006)

IPCC TAR: to high frequency

CMIP3: improved towards low frequency but still large diversity


Westward zonal extension

Too small meridional extension

El Niño in coupled GCMs - structure and timing

With impacts on periodicity

(Capotondi et al. 2007)

SST standard deviation

e.g.: Model events terminate in West rather than in East Pacific

Leloup et al. (2008)

Time sequence of El Niño/La Niña also has errors

El Niño termination spatial characteristics


El Niño in coupled GCMs - summary

Clear improvement since ~15 years

• some models get Mean and Annual cycle and ENSO right !

but:

• Amplitude: models diversity much larger than (recent) observed diversity

• Frequency: progress towards low frequency/wider spectra but still errors

• Seasonal phase lock: very few models have the spring relaxation and the

winter variability maximum

• Structure and timing: westward extension and narrowing around

equator, issues with time sequence (onset, termination)

• Modes: very few model exhibits the diversity of observed ENSO modes;

most are locked into a S-mode (coherent with too strong trade winds)

• Teleconnections: ENSO influence over-dominantGuilyardi et al. (BAMS 2009)


Ocean response to τ and HF anomalies•Upwelling, mixing, ("thermocline feedback", "cold tongue dynamics") (Meehl al. 2001, Burgers & van Oldenborgh 2003)

•Zonal advection (Picaut al. 1997)

•Wave dynamics •Energy Dissipation (Fedorov 2006)

van Oldenborgh al. 2005

Atmosphere response to SSTA•Bjerknes wind stress feedback (van Oldenborgh al. 2005, Guilyardi 2006)

•Meridional response of wind stress (An & Wang 2000, Capotondi al. 2006, Merryfield 2006)

•Radiative and cloud feedbacks (Sun al. 2006, Bony al. 2006, Sun et al. 2009)

Other processes:

• NL dynamical heating (∇xT + U in phase, An & Jin 2004)

• "Multiplicative noise" - MJO (Lengaigne et al. 2004, Perez et al. 2005, Philip & van Oldenborgh 2007)

Attributing ENSO errors: physical mechanisms


• Other studies confirm this result (e.g. Schneider 2002, Kim et al. 2008, Neale et

al. 2008, Sun et al. 2009, Shin et al. 2010,...)

Atmosphere GCM has a dominant role

Guilyardi et al. (2004)

Q1: Why this dominant role ? (ENSO theories focus on ocean role)

Q2: How to attribute ENSO errors to model systematic errors ?

Attributing ENSO errors: the role of the atmosphere


The BJ coupled-stability index for ENSO IBJ

QuickTime™ and aTIFF (LZW) decompressor

are needed to see this picture.

α: atmosphere heat flux feedback

μa: Bjerknes feedback or “coupling strength”

Mean advection and upwelling (damping)

Zonal advection

feedback

Ekman pumping feedback

Thermocline feedback

αis a negative feedback (damping)

μ is a positive feedback (amplification)

Jin et al. (2006), Kim et al. (2010a,b)


Atmosphere feedbacks during ENSO

Dynamical: Bjerknes feedback

East-west SST gradient

Trade winds

Equatorial upwelling in the east

Heat flux feedback

SST increase in the east

Modified heat fluxes (SHF, LHF)


Evaluating the Bjerknes feedback

• Monthly variability = measure of seasonal phase lock• Bjerknes amplification stronger in July-December

• Obs. values of vary from 10 to 15 (10-3 N.m-2/C)

Niño 3 SST anomaly

Niñ

o 4

Tau

X a

no

mal

y

Seasonal evolution of


Evaluating the heat flux feedback α

• Defined as slope of heat flux QA = F(SSTA)• α varies from -10 W.m-2/C to -30 W.m-2/C• Damping stronger in January-May

Niño 3 SST anomaly

Niñ

o 3

Hea

t F

lux

ano

mal

y


IPSL/Tiedke (TI)(0.3 C) – old scheme

IPSL (KE)Kerry Emanuel(1.0 C) - in IPCC

Impact of atmosphere convection scheme on ENSO

Observations(0.9 C) - HadiSST1.1

ENSO has disappeared !

What role for α and μ?

Hourdin et al. (2006) Braconnot et al. (2007)

Guilyardi et al. J.Clim (2009b)


BJ index for KE and TI

Linear theory: α dominant factor in TI/KE difference

Guilyardi et al. J.Clim (2009b)


~10 -18 0.9

4 -5 1.0

4 -20 0.3

El Niño

Amplitude

KE

TI

Obs

10-3 N.m-2/C W.m-2/C oC

Compute and in the KE and TI simulations

Error compensation !

Can we relate this change to physical processes ?

Getting the right ENSO amplitude for the wrong reasons !


Seasonal evolution of feedbacks

• Shortwave HF feedback SW in second half of year explains most of the difference

-25 W.m-2/C

<=>

1oC/month in SST cooling !!!

(MXL 50 m,

SSTA of 2oC)


SW feedback distribution

• Point-wise regression of SHF anomaly vs. SSTA (correl. less than 0.2 blanked out)

• Negative feedback (blue) = convective/ascent regime

• Positive feedback (red/orange) = subsidence regime

• ERA40 has large errors in East Pacific (Cronin et al. 2006)

• AMIP KE closer to ISCCP

• AMIP TI has too strong convection

• In KE, subsidence/+ve SW invades central Pacific

• In TI, convection/-ve SW invades east Pacific

• Coupled vs. forced (Yu & Kirtman 2007)


Composite analysis of large-scale regimes during El Niño

Subsidence

Convection

in far east Pacific• Positive SW has the potential to amplify El Niño

during growing phase, i.e. before convective/ascent threshold is reached

• AMIP KE potential much larger than that of AMIP TI

• KE and TI have opposite regimes during ENSO in far east Pacific

• TI ascent threshold 2oC lower than KE

TI physics triggers convection too easily which prevents ENSO from developping


Can we test this, i.e. suppress ENSO in KE ?• Perform KE run with increased SW

• Interannual Flux Correction:

• SHFO= SHFSCKE + SW

mod (SSTO-SSTSCKE)

• SWmod = -15 W.m

-2

• Mean state (SC) unchanged

~10

-180.9

El NiñoAmplitude

4 -51.0

4 -200.3

5 -210.4

Obs

KE

TI

KE mod TI

10-3 N/m2/C W/m2/C oC

ENSO gone as well !

KE

TI

KE

mod TI ?


Summary

• El Niño in IPCC-class GCMs: progress but still major errors (too much diversity) and atmosphere GCM is a dominant contributor

• Dynamical positive () and heat flux negative (α) feedbacks both likely to control El Niño properties in CGCMs

• Both feedbacks are usually too weak in models (poster by James Lloyd)

• In the IPSL-CM4 KE model, weak and α compensate each other

• When the convection scheme is changed to Tiedke (TI), α is increased and ENSO is damped (convective regime stronger)


Summary cont’d• Errors in these feedbacks:

• can be indentified in AMIP mode (good for model development) • have an impact on the mean as well (improve the mean and you’ll

improve ENSO)

• Improving ENSO in coupled GCMs requires:• Getting the right amplitude for the right reasons !• Increased atmosphere GCM horizontal resolution to improve

bjerknes feedback • Improved atmosphere convective physics (value of αSHF-)• Improved low clouds feedbacks (value of αSHF+)• Reduced cold tongue bias (spatial extent of αSHF+)• Validate ENSO feedbacks during model development phases

• Analysis method clearly attributes model systematic biases to errors in ENSO atmosphere feedbacks


How can we improve ENSO in models ?

• Improve the quality and utility of historical records

• Maintain present ENSO observing system into the future

• Continue promoting intercomparison studies (ENSO metrics)

• Isolate the main sources of model error, guided by theory (BJ), observations, and rigourous evaluation of the models, including tests in seasonal forecast mode (talk by Benoît Vannière)

A few challenges lying ahead:


Amplitude error in Niño 3 and Niño 4 ENSO Frequency RMS

Zonal wind RMS error in Equatorial Pacific SST annual cycle amplitude error in Niño 3

ENSO performance metricsDevised by CLIVAR Pacific Panel WG for CMIP5

a few examples...


How can we improve ENSO in models ?

• Improve the quality and utility of historical records

• Maintain present ENSO observing system into the future

• Continue promoting intercomparison studies (ENSO metrics)

• Isolate the main sources of model error, guided by theory (BJ), observations, and rigourous evaluation of the models, including tests in seasonal forecast mode (talk by Benoît Vannière)

A few challenges lying ahead:

• Should we revisit the role of atmosphere processes in ENSO theory (non-linear feedbacks, WWB, ...) ?

• Are observational records long/precise enough ?

Thank you !

Other challenges:



ENSO atmosphere feedbacks: what about

12 -18 0.9

3.9 -4.9 0.8

4.4 -5.1 0.95

6.2 -5.7 1.15

7.5 -4.3 1.13

7.8 -5 1.18

El Niño

Amplitude

Varying the horizontal atmosphere resolution in IPSL-CM4

Obs

10-3 N/m2/C W/m2/C oC

R97E (3.75 x 2.5)

R99A (3.75 x 1.8)

R149A (2.5 x 1.8)

R1414A (2.5 x 1.2)

R1914E (1.8 x 1.2)

AGCM resolution affects (but not ):• Atmosphere grid can “see” ocean equatorial wave guide

• Added non-linearities in AGCM: better circulation (on/off convection behavior reduced,...)

(Zonal x Merid.)

• SW (blue bars) and LH (green bars) components dominate.• The SW feedback shows the greatest model diversity – linked to cloud

uncertainties in East Pacific region (e.g. Sun et al. 2009).

• Split net feedback into four components: shortwave (SW), longwave (LW), latent heat (LH) and sensible heat (SH) flux feedbacks.

Lloyd et al. (2009)

Heat flux feedback components


SST threshold for ascent/convection

• Bin vertical velocity at 500 hPa in SST bins

• Ascent/Convective threshold when regime switches from subsidence to convection (Bony et al. 2004)

Subsidence

Convection

• AMIP KE threshold larger by 1oC / AMIP TI (same SST !)

• KE threshold unchanged

• TI threshold even lower: 2oC difference with KE !

TI physics triggers convection too easily which prevents ENSO from developping


Bjerknes feedback μ

• Bjerknes feedback in AMIP KE and AMIP TI similar and within re-analysis estimates

• Bjerknes feedback in KE = 1/3rd of that of AMIP KE


Heat flux feedback α

• Heat flux feedback α in AMIP TI is double that of AMIP KE (and closer to re-analysis estimates)

• Heat flux feedback in KE = half of that of AMIP KE

• Value unchanged in TI


Evolution of El Nino composite Wind Stress

• Wind stress anomaly (shading)• SST anomaly (solid contours)• 3 year composite at equator

• Bjerknes feedback in AMIP KE and AMIP TI similar

• Bjerknes feedback in KE much weaker than in AMIP KE


Evolution of El Nino composite Heat Flux

• Heat flux anomaly (shading)• SST anomaly (solid contours)• 3 year composite at equator

• Negative Heat Flux anomaly in AMIP TI much larger than in AMIP KE

• Negative Heat Flux anomaly during positive SST anomaly

• Negative Heat Flux anomaly in KE too far west and weak

• SHF main contributor to differences between AMIP TI and AMIP KE (LHF also contributes)


Mean seasonal cycle at Eq.

• AMIP KE performs rather well• Convection in AMIP TI too strong

and triggered too soon

• Biases amplified in coupled mode

• Semi-annual cycle in TI

• Equinoctial Central American monsoon too strong in TI (Braconnot et al. 2007)

• Wind stress (shading)• SST (solid contours)• Precipitation (3 and 8 mm/day dashed)

Documents

CLIVAR ENSO workshop, Paris, Nov. 2010 ENSO in GCMs: overview, progress and challenges Eric Guilyardi IPSL/LOCEAN, Paris, France & NCAS-Climate, Univ