Tropicallyforced Decadal Variability in the North Pacific

1
Tropically-forced Decadal Variability in the
North Pacific

• Matt Newman
• NOAA-CIRES CDC

2
Newman, Compo, and Alexander (2003)

• Null hypothesis of the PDO
• North Pacific integrates effects of ENSO via the
  atmospheric bridge
• Re-emergence brings back ENSO-induced anomalies
  in succeeding winters
• (no summer/fall PDO)
• Pn aPn-1 bEn hn
• PDO should be redder than ENSO, e.g.

3
North Pacific SST anomaly should be greater than
Tropical Pacific SST anomaly on decadal
timescales
4
However

• North Pacific SST not necessarily equally
  sensitive to all parts of the Tropics
• Univariate (single index) approach limited
• Null hypothesis in Newman et al assumed perfect
  prediction of ENSO
• So try multivariate model
• continuing to avoid seasonal cycle

5
Data

• Hadley Ice Sea Surface Temperature Analysis
  (1900-2002), averaged in 5 deg grid boxes
• repeated with Kaplan SST, not shown
• Annual mean (July-June) SST anomalies determined
  from (annual mean) 1950-2002 climatology

6
EOF analysis

• North Pacific
• Defined as 24N-60N
• Retain first 3 PCs (64 variance explained)
• Tropical IndoPacific
• Defined as 20S-20N, 60E-60W
• Retain first 7 PCs (91 variance explained)
• EOFs determined from 1950/51-2001/02 earlier
  data projected on these EOFs

7

Linear inverse model (LIM)
x(t) 10-component vector whose components are
the time-varying coefficients of 7 EOFs of
Annual mean IndoPacific SST and 3 EOFs
of Annual mean North Pacific SST L is thus a
10x10 matrix
8

Potential predictability within the LIM framework
9
Forecast skill, 1900-2001
AC score of 1- and 2-year forecasts
All forecasts are cross-validated in 10-year
blocks
10
Empirical normal modes
eft2.2 yr
eft1.6 yr
eft15 yr
eft6 yr
11
Normal mode timeseries
12
Constructing the PDO from a sum of three red
noise processes
13
Forecast skill by normal mode
AC score of cross-validated 1- and 2-year
forecasts
Modes ordered by decreasing decorrelation time
14
Power spectra

• Given L and Fs, we run this model for 20000
  years (takes about 6 min.) to get time series of
  all 10 PCs.
• Break time series up into 2000 100-yr chunks, do
  spectrum of each (Matlab Thomson MTM), get mean
  and confidence intervals

15
Power spectra (Tropics)
LIM mean
95 confidence interval
16
Power spectra (North Pacific)
17
Turn off coupling
LIM can be written in its components parts
as dx BNN BNT xN ---
noise dt BTN BTT
xT So we can set submatrices BNT and BTN to
zero and recompute power spectra.
18
Power spectra (North Pacific)no Tropical-North
Pacificcoupling
19
TROPICSComparison of observed power spectra and
LIM confidence interval (gray shading) with
CMIP2power spectraannual means,
usingobserved EOFs
TROPICS
20
NORTH PACIFICComparison of observed power
spectra and LIM confidence interval (gray
shading) with CMIP2power spectraannual
means, usingobserved EOFs
21
CMIP data projected on normal modes
22
Conclusions

• Decadal varability null hypothesis one year
  lagged autocorrelation
• PDO not really a single phenomenon
• Reflects a mix of tropical-extratropical
  interactions (modes) on different timescales
  and with different patterns
• Univariate approach is fundamentally flawed
• None of these modes appear to have potential
  predictability gt 6 yrs
• Except for leading mode, which may partly reflect
  a trend of some kind
• Regime shifts are essentially unpredictable

23
Conclusions

• CMIP2 coupled models poorly simulate leading
  modes
• Tropical decadal variability usually too weak
• North Pacific variability usually too strong
• North Pacific too independent of Tropics?
• Coupled GCMS must reproduce not only ENSO but
  also its power spectrum and seasonal cycle of a
  and b

24
ENSO-PDO relationship in CMIP2 coupled models
25
Null hypothesis applied to CMIP2 output