Understanding the Tropical Biases in GCMs: DoubleITCZ, ENSO, MJO and Convectively Coupled Equatorial

1
Understanding the Tropical Biases in
GCMsDouble-ITCZ, ENSO, MJO and Convectively
Coupled Equatorial Waves
2
More than half of the earths precipitation falls
in the tropics Heating sources for tropical
large-scale circulations and teleconnections to
extratropics
22N
22S
3
The tropical precipitation does not occur
randomly, but is organized by a number of
dominant tropical modes, leading to useful
predictability in both tropics and extratropics
4
Unfortunately, these dominant tropical modes are
not well simulated by the state-of-the-art
GCMsTropical biases

• The double-ITCZ problem
• The ENSO problem
• The MJO and BSIO (boreal summer intraseasonal
  oscillation) problem
• The convectively coupled equatorial wave problem

These problems have been persisting in the last
several generations of GCMs, in spite of the
significant increase of model resolution and more
and more complicated model physics.
5
The major difficulties for understanding and
alleviating these tropical biases

• They all involve some forms of feedback, such as
  the ocean-atmosphere feedback and the
  wave-heating feedback, making it difficult to
  determine the real cause of the bias
• The biases need to be traced back to specific
  model characteristics, such as certain aspect of
  the physical parameterizations, in order to
  provide useful guidance on how to improve the
  model simulations.

6
Multiple-model intercomparison, when combined
with (1) feedback analysis and (2)
model physics evaluation, provide a good way to
overcome these difficulties

• It can help us to
• understand the feedback mechanisms of the
  tropical biases
• relate the tropical biases to specific model
  characteristics and
• transfer the success of individual models to
  other models.
• Hypothesis
• The tropical biases are caused by some
  missing physics in current GCMs.

7
GCMs analyzed 27 models including almost all the
major GCMs used for predictions and projections

• 22 IPCC AR4 coupled GCMs (IPCC Fourth Assessment
  Report to be released in 2007 from PCMDI data
  archive)
• GFDL2.0, GFDL2.1, NCAR-CCSM3, NCAR-PCM,
  GISS-AOM, GISS-ER, GISS-EH
• IAP, CCC-CGCM, MIROC-hires, MIROC-medres,
  MRI, HadCM3, HadGEM1, CSIRO,
• MPI, ECHO-G, CNRM, BCCR, INM, IPSL
• NCEP operational Global Forecasting System (GFS)
  and Climate Forecasting System (CFS) (in
  collaboration with Wanqiu Wang of NCEP)
• ECMWF model (from DEMETER archive)
• NASA GMAO GEOS5 GCM currently under development
  (in collaboration with Siegfried Schubert, Max
  Suarez, Julio Bacmeister of NASA GMAO)
• GFDL next generation GCM currently under
  development (in collaboration with Leo Donner of
  GFDL)
• Seoul National University (SNU) GCM (in
  collaboration with Myong-In Lee of NASA GMAO)

8
Observational data
Monthly (10-100yrs), Daily (1-10 yrs)
Precipitation TRMM, GPCP Cloud/radiation
ISCCP, ERBE, AVHRR Temperature MSU
SSH TOPEX
Hourly (10-100days) TOGA COARE JASMINE
SCSMEX KWAJEX TEPPS EPIC
LBA
NCEP reanalysis ECMWF reanalysis Raw sounding
ERSST HADISST TAO buoy
Ocean analysis (White)
Every second (1-10hrs) TOGA COARE
9
Part I The double-ITCZ problem
10
The double-ITCZ problem has been persisting in
the last several generations of coupled GCMs

• Models in early 1990s (Mechoso et al. 1995
  first generation without flux correction)
• Models in late 1990s (Latif et al. 2001 Davey
  et al. 2002)
• Latest models (Lin 2006a)
• GCM sensitivity experiments

• Schneider (2002) the problem is mainly caused
  by the AGCM rather than the OGCM.
• Mechoso (2006) higher AGCM resolution
• Zhang and Wang (2006) convection scheme
• However, as pointed out by Mechoso (2006)
• A synthetic view of the double-ITCZ problem is
  still elusive.

11
Annual mean SST/precipitation of 22 IPCC AR4
CGCMs About half of the models have significant
double-ITCZ problem
Obs
NCAR
GFDL
Double-ITCZ
Shading SST Contours precipitation
From Lin (2006a, J. Climate)
12
Zonal profiles of SST Significant cold bias not
only along the equator, but also off the equator
5N-15N
5N-5S
SST cold bias
5S-15S
13
Precipitation Excessive tropical precipitation
in spite of the SST cold bias, except over
western Pacific
5N-15N
Excessive precipitation
5N-5S
5S-15S
14
Surface fluxes Overly strong trade winds,
excessive LHF, insufficient SWF
?x
Overly strong trade winds
Excessive LHF
LHF
SWF
Insufficient SWF
15
Theories of tropical mean climate
Ocean-atmosphere feedback mechanisms
SST gradient - trade wind (Bjerknes) feedback
(e.g. Bjerknes 1969, Neelin and Dijkstra 1995
Pierrehumbert 1995 Sun and Liu 1996 Jin 1996
Clement et al. 1996 Liu 1997 Cai 2003)
SST - LHF feedback (e.g. Wallace 1992 Liu et al
1994 Zhang et al. 1995)
SST - SWF feedback (e.g. Ramanathan and Collins
1991)
Neelin and Dijkstra (1995) shows that any
excessive positive feedback (or insufficient
negative feedback) tends to shift the whole
system westward, leading to a double-ITCZ
pattern. However, few previous studies have
evaluated quantitatively the feedback parameters
in GCMs.
From Lin (2006a)
16
Quantitative evaluation of feedback parameters in
AGCMs Bjerknes feedback Overly strong in
several models, which is caused by insufficient
boundary layer mechanical damping
Using area averaged data gives similar results
Feedback parameter ?x vs dSST/dx
Overly strong
dSLP/dx vs dSST/dx
?x vs dSLP/dx
Insufficient mechanical damping
Linear regression of raw monthly data
Linear correlation 99 level0.2
17
SST-LHF feedback Incorrectly positive in
several models, which is caused by excessive
sensitivity of surface air humidity to SST
Feedback parameter LHF vs SST
qair vs SST
Incorrectly positive
Excessive sensitivity
??? vs SST
(qsurf-qair) vs SST
Insufficient sensitivity
18
SST-SWF feedback Overly weak in several models,
which is caused by insufficient sensitivity of
cloud amount to precipitation
Feedback parameter SWF vs SST
Precip vs SST
Excessive sensitivity
Overly weak
SWF vs Cloud amount
Cloud amount vs Precip
Insufficient sensitivity
19
Summary of the double-ITCZ problem

• Mean climate biases
• Overly cold SST
• Excessive tropical precipitation
• Overly strong trade winds
• Excessive LHF
• Insufficient SWF Most of these biases
  already exist in the AGCMs.

Incorrect ocean-atmosphere feedback parameters
Suggeted ways to alleviate the double-ITCZ problem
From Lin (2006a)
20
Future work (1) Apply the analysis to NCEP
CFS/GFS and NASA GEOS5 models (2) Apply the
analysis to OGCMs to assess the feedback
parameters
CFS climatology shares the double-ITCZ problem
From Mechoso (2006)
21
Part II The ENSO problem
22
Spectrum of Nino3 SST in 22 IPCC AR4 coupled
GCMs Large scatter in variance. Many models
display too-short period.
(From Lin 2006b)
23
Existing ENSO theories
(6) Stochastic forcing theory (McWilliams and
Gent 1978, Lau 1985, Penland and Sardeshmukh
1995, Blanke et al. 1997, Kleeman and Moore 1997,
Eckert and Latif 1997)
(1) Slow coupled mode theory (Philander et al.
1984, Gill 1985, Hirst 1986, Neelin 1991, Jin and
Neelin 1993, Wang and Weisberg 1996)
(2) Delayer oscillator theory (Suarez and Schopf
1988, Battisti and Hirst 1989)
(3) Advective-reflective oscillator theory
(Picaut et al 1997)
(4) Western Pacific oscillator theory (Weisberg
and Wang 1997)
Common characteristics (1) Confined to Pacific
(2) Many treat ENSO as a quasi-standing
oscillation (3) Many use free oceanic waves
(Kelvin, ER)
(5) Recharge oscillator theory (Jin 1997a,b)
From Lin (2006c)
24
Observed life cycle of ENSO along the equator
(8N-8S)Inconsistent with the existing ENSO
theories
SST (zonal mean removed)
Sea surface height (SSH)
Characteristics (1) Planetary
scale (wavenumbers 1-2), not confined to Pacific
(2) Dominated by eastward propagation,
although standing or westward propagation at some
longitudes (3) Coupled
signals with very slow phase speed (0.2 m/s),
not free oceanic waves
?x
Precipitation
SLP
u200
Z200 (zonal mean removed)
ltTgttroposphere (zonal mean removed)
(Lag-regression of interannual anomaly vs Nino3
SST anomaly. Shading denotes regions where
lag-correlation is above 95 confidence level.
From Lin 2006d.)
25
Wavenumber-frequency spectrumAll variables
display two modes (1) eastward wavenumbers 1-2,
3-7 years, and (2) westeard wavenumbers 1-2, 3-7
years
Ocean heat content
SST
Troposphere temperature
Precipitation
SLP
Z200
?x
u200
26
Are the dominant modes physically meaningful?
Yes, variances consistent with those of Kelvin
and ER waves
Sea surface height variance of the eastward
component (all wavenumbers, all
frequencies after seasonal cycle is removed)
Consistent with Kelvin wave
Sea surface height variance of the westward
component
Consistent with ER wave
27
Interactions among the coupled Kelvin, coupled
ER, and off-equatorial Rossby waves Boundary
reflection
Kelvin wave
ER wave
Eastern boundary WSC forcing
Eastern boundary reflection
Off-equatorial Rossby wave at 15N
Western boundary reflection
28
(1) Dominant equatorial wave modes coupled
Kelvin and ER waves, not free waves(2) These
waves are of planetary-scale (wavenumbers 1-2),
not confined to the Pacific(3) It takes gt3 years
for them to circle the equator, leading to the gt3
years ENSO period(4) They interact w/ each
other, and w/ fast Kelvin wave and off-equatorial
Rossby wave (5) Intraseasonal variances propagate
eastward with Kelvin wave, multiplicative forcing
(6) Discharge/recharge of Pacific-basin-averaged
OHC is associated w/ passage of waves
A new mechanism for ENSO The coupled wave
oscillator
From Lin (2006c)
29
Evaluation of two coupled GCMs in this
frameworkThe too-short ENSO period in some
models is associated with too-fast phase speed of
the coupled waves
Too-fast phase speed
Realistic phase speed
SSH
SSH
?x
?x
CCSM3 ENSO Period2.5 yrs
MPI ENSO Period4 yrs
30
Future works

• Apply the analysis to all 22 IPCC AR4 coupled
  GCMs
• Apply the analysis to NCEP CFS and NASA GEOS5
  models
• Evaluate the feedback parameters to understand
  the too-fast phase speed and unrealistic variance
  in many models
• Trace the errors in feedback parameters back to
  deficiencies in model physics

31
Part III The MJO, BSIO and convectively coupled
equatorial wave problems
32
The MJO and convectively coupled equatorial waves
are important for both weather prediction and
climate prediction
(Lin et al. 2006a,b,c)
33
But these waves have not been well simulated by
GCMs (e.g. Poor MJO simulation has been a
well-known long-standing problem)

• Pioneering studies in 1980s (e.g. Hayashi and
  Golder 1986, 1988, Lau et al. 1988)
• Eastward Kelvin-Rossby or Kelvin waves but
  with too fast phase speeds (10-18 m/s)
• AMIP models in early 1990s (Slingo et al.
  1996)
• Simulated signals are generally too weak and
  too fast

Models in late 1990s (Schubert et al. 2002,
Waliser et al. 2003) Several models are
getting stronger MJO variance and/or better
eastward propagation. Latest models (Lin et
al. 2006a, b, c) MJO and convectively coupled
equatorial waves (Lin, J.L., G.N. Kiladis, B.E.
Mapes, K.M. Weickmann, K.R. Sperber, W.Y. Lin, M.
Wheeler, S.D. Schubert, A. Del Genio, L.J.
Donner, S. Emori, J.-F. Gueremy, F. Hourdin, P.J.
Rasch, E. Roeckner, and J.F. Scinocca, 2006a)
Asian monsoon and associated intraseaonal
variability (Lin et al. 2006b) North American
monsoon and associated intraseasonal variability
(Lin et al. 2006c)
34
Wavenumber-frequency spectra (15N-15S,
raw/background)Only half of the models have
convectively coupled equatorial waves, and their
phase speeds are too fast (same equivalent depth
within each model)
Obs
GFDL
NCAR
35
Variances of the convectively coupled
wavesGenerally too weak except for the EIG wave
MRG
Kelvin
EIG
ER
WIG
36
Variance of the MJOApproaches the observed
value in two models (MPI,CNRM), but is less than
half of the observed value in the other 12 models
Interestingly, the two models producing the
largest MJO variance are the only models having
convective closure/trigger linked to moisture
convergence.
37
Ratio between variance of the MJO and that of its
westward counterpartToo small in most of the
models (lack of eastward propagation)
38
Raw spectra of eastward wavenumbers 1-6 at
0N85EThe MJO variance in 13 of the 14 models
does not come from a pronounced spectral peak,
but from a too red spectrum.
39
Intraseasonal variability associated with Asian
summer monsoon Overly weak variance for
both modes, poor BSIO eastward propagation, good
BSIO northward propagation, good 12-24 day mode
westward propagation
BSIO eastward component
BSIO northward component
Westward 12-24 day mode
(Lin et al. 2006b, first coupled GCM
intercomparison in literature)
Variance
Propagation
40
North American monsoon Seasonal cycle of
precipitation (100W-115W)Models display a large
scatter in onset time, often excessive
precipitation
Obs
July onset
NCAR
May onset
GFDL
No obvious onset
(Lin et al. 2006c, first GCM intercomparison in
literature)
41
Intraseasonal variability associated with North
American monsoon Overly weak variance for both
modes, poor eastward propagation of MJO, good
westward propagation of easterly waves
MJO
Easterly waves
Variance
Propagation
42
Summary of model evaluation for tropical
intraseasonal variability
The current state-of-the-art climate models still
have significant problems and display a wide
range of skill in simulating the tropical
intraseasonal variability, including the MJO,
BSIO and other convectively coupled equatorial
waves. The common characteristics are

• Convectively coupled equatorial waves (Kelvin,
  ER, MRG, EIG, WIG)
• Only half of the models have significant
  signals
• Overly weak variance for most of the waves
• Overly fast phase speed
• MJO/BSIO
• Overly weak variance
• Lack of spectral peak
• Poor eastward propagation
• Relatively good northward propagation of BSIO

43
Theories of the tropical intraseasonal
variability Wave-heating feedback mechanisms
Vertical heating profile
Mesoscale effect suppress all waves (Lin, Donner
et al.)
Convection scheme and moisture trigger
significantly affect the waves (Lin, Lee et al.)
Column-integrated diabatic heating has six major
components (Mean state and higher-frequency modes
affect the MJO through the nonlinear terms)
Cloud-radiative heating enhances all waves (Lin,
Lee et al.)
Model resolution
Higher vertical resolution enhances some waves
but not MJO(Lin, Wang et al.)
Air-sea coupling has little effect on the waves
(Lin et al.)
44
Observational background Convective updrafts
are generally diluted and thus sensitive to lower
troposphere moistureFew updrafts are wider than
8 km, suggesting significant entrainment rate
(inversely proportional to updraft width)
NOAA P3 aircraft penetration of deep convection
during TOGA COARE (Lin, unpublished)
45
A common problem in GCMs Undiluted updrafts
which are insensitive to lower troposphere
moisture
A solution Moisture trigger Previous studies
have examined the impacts of moisture trigger on
MJO (Wang and Schlesinger 1999, Lee et al.
2003). However, no previous study has examined
the impacts of moisture trigger on other
convectively coupled equatorial waves.
46
Effect of convection scheme and moisture trigger
Moisture trigger significantly enhances spectral
power for Kuo and SAS (Simplified
Arakawa-Schubert) schemes (Lin, Lee, and others)
Obs
Weak trigger
Medium trigger
Strong trigger
No convection
Kuo
SAS
MCA
47
Effect of convection scheme and moisture trigger
Moisture trigger slows down the waves for SAS
scheme, but not Kuo or MCA schemes. No
convection produces too slow speed.
Obs
Weak trigger
Medium trigger
Strong trigger
No convection
Kuo
SAS
MCA
48
Effect of convection scheme and moisture trigger
(1) MCA scheme produces better wave signals than
Kuo and SAS schemes (2) Adding moisture trigger
to Kuo and SAS schemes significantly enhances all
waves except MJO (3) No convection experiment
always produces one of the largest variance
Kelvin
EIG
ER
WIG
MRG
MJO
49
The no convection experiment also produces a
good MJO spectral peak
50
(No Transcript)