A Journey to ENSO Simulation at COLA

1
A Journey to ENSO Simulation at COLA

• Vasu Misra, Larry Marx, Zhichang Guo, Jim Kinter,
  Ben Kirtman, Dughong Min, David Straus, Paul
  Dirmeyer, Mike Fennessey
• Acknowledgements Ramesh Kallumal, Ben Cash,
  Byron Boville, M. Kanamitsu, Song Hong, S.
  Moorthi, J. Bacmeister, Kathy Pegion

2
Why is this exercise important?
As end users of model we could complain MJO/ISO
is bad, ENSO is bad, split ITCZ is a problem, no
monsoon, mid-latitude response is bad, fluxes are
bad, clouds are bad, MODEL IS BAD-Symptomatic
analysis Sometimes Generous!
Suggest from incremental (documented) changes
what reduced/increased the bias,
variability-attribution of model errors. But
change in one model does not translate to similar
response in another? Yes, but does provide a
motivation to pursue a testable hypothesis.

R D of Center xyz
Scientists outside xyz
Stake holders
3
ENSO Metrics to evaluate a simulation

• Mean state errors
• Spectrum of SST in the Nino3 region (power, width
  of the peak, frequency)
• Evolution of ENSO (asymmetry in cold and warm
  phase sub surface ocean anomalies)
• Duration of ENSO event
• ENSO forcing (correlations) in other ocean basins
• Seasonal phase locking of ENSO Variability

• 7. Relationship of
• wind stress with SST
• Precipitation with SST
• 8. Mid-latitude response

4
Starting from.
3.92
2
Symm
10 months
Erroneous
No
Un-verifiable
Un-verifiable

• Nino3 root mean square SST errors.
• Spectrum of Nino3 SST
• Asymmetry of ENSO warm/cold events
• Duration of ENSO event
• Nino3 SST correlations with other ocean basins

6. Seasonal phase locking of ENSO 7. a)
windstress-SST relationship b) precip-sst
relationship 8. Mid-latitude response
5
What is new
6
Philosophy for improving simulationmoving
towards more physically based schemes

• PBL Local K-theory which parameterizes turbulent
  mixing with an eddy diffusivity based on local
  gradients of wind and temperature may fail in
  unstable boundary layers because influence of
  large eddy transports is not accounted for.
• Long wave Developed from water vapor line and
  continuum treatments-uses line-byline radiative
  transfer model GENLEN2-an improvement over
  broad-band absorptance method.
• Convection Determination of fraction of
  detrained cloud liquid water was through an
  empirical profile. Now a budget for cloud liquid
  water is included in the convection scheme.
• SSiB Going from 1 layer in root zone to 4
  layers.
• Horizontal Diffusion Way too strong.
• Consistency Saturation vapor pressure and
  variation of Lv with T
• Vertical Resolution Skewed.

7
Profile of vertical resolution of the AGCM
8
Anecdote

• ..implementing the CAM long wave scheme
  produced excessive cold bias in the upper
  troposphere. I seek your advice to tune the long
  wave scheme
• .I would not suggest adjusting the scheme
  itself. The new scheme is based upon much more
  recent water vapor line and continuum
  treatmentsProblems in other parts of the model
  may be getting reflected.-William Collins, NCAR

9
Experiment Design
Observational verification 1955-2000 ODA
1980-98 IC of coupled integrations Length of
model experiments are not the same. Showing the
last 45 years. At a minimum the first 20 years
have been removed in the analysis. Ocean model
MOM3 -1.50 (zonal resolution), 0.50 from 10 S to
10 N and 1.50 in the extra-tropics. 25 vertical
levels with 17 in the upper 450 m. Will be
looking at annual mean quantities
10
Downwelling Shortwave flux at surface
11
Annual Mean SST Errors
12
Annual Cycle of Equatorial Pacific SST
13
Small changes can lead to significant change in
model variability
14
ERSST-V2
Seasonal phase locking of ENSO to the annual cycle
15
Nino3 SST regression on observed and simulated SST
ERSST-V2
16
Lead/Lag regression of the Nino3 SST with
equatorial Pacific SST
17
Joseph and Nigam, 2005
18
Nino3 SST regression on sub-surface ocean
anomalies over equatorial Pacific
19
e-10.368
ERSST-V2
20
ERSST-V2
Contemporaneous correlation of annual mean Nino3
SST with global tropical SST
21
Wind Stress-Nino3 Regression (dynes/cm2)
22
Precipitation-Nino3 SST regression (mm/day)
23
Summary
0.93
3
12months
Improvement
Improvement
Improvement
Improvement
Assym

• Nino3 Mean SST errors.
• Spectrum of Nino3 SST
• Asymmetry of ENSO warm/cold events
• Duration of ENSO event
• Nino3 SST correlation with other ocean basins

6. Seasonal phase locking of ENSO 7. a)
windstress-SST relationship b) precip-sst
relationship 8. Mid-latitude response
Spectrum has the largest peak between 2.5-7
years and falls within the 95 confidence
interval of the observed spectrum
24
Where we stand (Thanks to PCMDI)

• Nino3 Root Mean square SST errors.
• Spectrum of Nino3 SST
• Asymmetry of ENSO warm/cold events
• Duration of ENSO event
• Nino3 SST correlations in other ocean basins

6. Seasonal phase locking of ENSO 7. a)
windstress-SST relationship b) precip-sst
relationship 8. Mid-latitude response
Spectrum has the largest peak between 2.5-7
years and falls within the 95 confidence
interval of the observed spectrum
25
Concluding Remarks

• It is easy to abandon models that dont simulate
  ENSO. But it will be a great learning experience
  if we make an attempt to change these models.
• From our exercise in COLA we are learning
• Development of climate models are best achieved
  in a coupled framework.
• All eight metrics by themselves are necessary but
  not sufficient conditions for verifiable
  seasonal-interannual simulation. To get every
  metric of ENSO right even in ball park is
  important for at least seasonal to inter-annual
  prediction.
• Wind stress simulation is important in the
  eastern Pacific to get the bulk of the annual
  cycle right besides the stratus clouds. We got
  that to a large part by having a bottom heavy
  convective heating profile. We are investigating
  the asymmetry of ENSO causality from 3.1 to 101
  to 102.
• Small changes can lead to significant change in
  the model variability. The coupled model has to
  be integrated for long periods to determine the
  efficacy of a change.
• Not flux correction but improved models is the
  way to move forward. Flux correction, in the
  short term may help and could be given as a
  testable magic wand for operational RD teams.