Welcome to Scribd, the world's digital library. Read, publish, and share books and documents. See more
Download
Standard view
Full view
of .
Look up keyword
Like this
2Activity
0 of .
Results for:
No results containing your search query
P. 1
Coupled Pattern Analysis of Sea Surface Temperature and TOPEX/Poseidon Sea Surface Height

Coupled Pattern Analysis of Sea Surface Temperature and TOPEX/Poseidon Sea Surface Height

Ratings: (0)|Views: 11|Likes:
Published by GoesCroops

More info:

Categories:Types, Research, Science
Published by: GoesCroops on Dec 03, 2011
Copyright:Attribution Non-commercial

Availability:

Read on Scribd mobile: iPhone, iPad and Android.
download as PDF, TXT or read online from Scribd
See more
See less

08/27/2013

pdf

text

original

 
A
PRIL
1999
599
L E U L I E T T E A N D W A H R
1999 American Meteorological Society
Coupled Pattern Analysis of Sea Surface Temperature and TOPEX/Poseidon SeaSurface Height
E
RIC
W. L
EULIETTE
*
AND
J
OHN
M. W
AHR
 Department of Physics and CIRES, University of Colorado, Boulder, Colorado
(Manuscript received 10 April 1997, in final form 14 May 1998)ABSTRACTThough thermal effects dominate steric changes in sea level, the long-period contribution of thermal expansionto sea level is uncertain. Nerem et al. found that a global map of sea surface temperature (SST) trends and acorresponding map of TOPEX/Poseidon-derived sea surface height (SSH) trends were strongly correlated. Thisresult is explored with a coupled pattern analysis (CPA) between five years of global SST and SSH, whichallows for matching of modes of common temporal variability.The dominant mode found is an annual cycle that accounts for nearly all (95.3%) of the covariance betweenthe fields and has a strong SST/SSH spatial correlation (0.68). The spatial correlation is strong in both theAtlantic (0.80) and the Pacific (0.70). Good temporal and spatial agreement between the SSH and SST fieldsfor the primary seasonal mode suggests that a robust regression between fields may have some physical sig-nificance with respect to thermal expansion and that the regression coefficient might be a proxy for the mixingdepth of the mode. The value of the regression coefficient,
H,
scaled by a thermal expansion coefficient of 2
10
4
C
1
is 40 m for this mode, and ranges from 33 to 47 m among the basins.The primary mode of a nonseasonal CPA is an interannual mode that captures 38.0% of the covariance andhas significant spatial correlations (0.54) between SSH and SST spatial patterns. The spatial pattern and temporalcoefficients of this mode are correlated with ENSO events. A robust regression between fields finds that thenonseasonal modes have a regression coefficient 2–4 times that of the seasonal modes, indicative of deeperthermal mixing. The secondary nonseasonal mode captures most of the secular trend in both fields during theperiod examined. The temporal coefficients of this mode lag those of primary mode. Evidence is presented thatthis mode is consistent with the behavior expected from secular trends that are dominantly forced by thermalexpansion.
1. Introduction
Steric changes in sea level at global scales are dom-inated by thermal effects. Freshwater and salinity fluxesare less important, though they can be significant locally(Pattullo et al. 1955). At long wavelengths sea surfacetemperature is a reasonable proxy for the thermal con-tent of the mixed layer, with regions of notable excep-tions such as the Southern Ocean and the intertropicalconvergence zone. A secular increasein seasurfacetem-perature should be accompanied by a rise in global stericheight. For example, Wigley and Raper(1987)estimatedthat while the global mean temperature increased by
0.5
from 1880 to 1985, the thermal expansion con-tribution to sea level rise was 2–5 cm. And by using a
* Current affiliation: Center for Space Research, Austin, Texas.
Corresponding author address:
Dr. E. W. Leuliette, Center forSpace Research, 3925 W. Braker Lane, Suite 200, Austin, TX 78759-5321.E-mail: leuliette@csr.utexas.edu
model based on the dynamics of subduction and Sver-drup balance, Church et al. (1991) estimated that a glob-al-mean 3.0
C temperature rise by 2050 would produce20–30 cm of sea level rise.Operational satellite oceanography now providesglobal datasets for sea surface height (SSH) and seasurface temperature (SST), which can be used to studythermal effects on sea level at periods of a few yearsand less. Results from global SST analyses have beenavailable for many years. But continuous and accurateSSH results from satellite altimetry have become avail-able only relatively recently. The TOPEX/Poseidon al-timetric satellite, launched in August 1992, has beenachieving extraordinary accuracies of better than 5 cmat all spatial scales (Fu et al. 1994; Shum et al. 1995).In particular, the precise determination of the orbit of the TOPEX/Poseidon altimetry mission has allowed forthe accurate study of long-wavelength oceanographicfeatures (e.g., Tapley et al. 1994).Previous studies (Cheney 1982; Khedouri and Szcze-chowski 1983; Carnes et al. 1990) have found high cor-relations between altimeter-derived sea surface heightand subsurface temperatures. Nerem et al. (1997) report
 
600
V
OLUME
29J O U R N A L O F P H Y S I C A L O C E A N O G R A P H Y
that the correlation between a global map of SST trendsand a corresponding map of TOPEX/Poseidon-derivedSSH trends during 1993–96 is 0.6. They suggest thatthis correlation may be evidence of sea level rise dueto heat storage near the surface of the ocean. However,they note that some of the correlation may be attrib-utable to indirect causes, primarily water displacementsdue to wind stress, presumably from ENSO-related var-iability.Here, we extend this type of comparison by consid-ering the spatial structure of the SSH/SST relationshipat seasonal and interannual periods. We use a globalcoupled pattern analysis to compare five years of TO-PEX/Poseidon-derived SSH and optimally interpolatedSST. For intercomparison, the global analysis was sam-pled for each of the three ocean basins (Pacific, Atlantic,and Indian) to calculate key parameters.
2. Coupled pattern analysis
The enormous complexity of global meteorologicaland oceanographic datasets has lead to the increasinguse of objective analysis methods to help in their in-terpretation. Probably the most familiar methods usedfor analyzing a single field involve empirical orthogonalfunctions (EOFs).In an EOF analysis, a temporally and spatially vary-ing field,
s
(
 x, t 
) (such as either SSH or SST), is expandedas a sum of orthogonal, spatially dependent functions,
(
 x
):
 N 
s
(
 x, t 
)
a
(
)
(
 x
), (1)
k
1
where the coefficients
a
(
) are time-varying functionsand
is the number of spatial points
x.
If 
s
(
 x, t 
) is SSH,for example, then (1) is analogous to a standing waveexpansion of the ocean’s surface. The shape of a‘‘wave’’ is described by
(
 x
), and the ‘‘wave’s’’ timedependence is
a
(
). There are many possible ways of constructing an orthogonal set of 
. In EOF analyses,the
are chosen to be eigenfunctions of the
temporal covariance matrix:
S
(
 x, x
)
s
(
 x, t 
)
s
(
 x
,
). (2)
The eigenvalue associated with
turns out to be equalto the fraction of the total variance of 
s
(
 x, t 
) [the sumover space and time of 
s
(
 x, t 
)
2
] contributed by the term
a
(
)
(
 x
) in the expansion (1).For real data, most of the terms
a
(
)
(
 x
) in (1) areusually mathematical artifacts with little,ifany,physicalsignificance. But there are usually a few terms where
has a particularly large eigenvalue, indicating that anunusually large fraction of the field is organized co-herently into that spatial pattern. Those terms are thenextracted for further analysis and interpretation.EOF and related analyses of SST at regionaland glob-al scales have been described in numerous studies (e.g.,see Weare et al. 1976; Weare 1977, 1979; Hsiung andNewell 1983; Hu et al. 1994; Kawamura 1994). Similaranalyses have been applied to altimetry data from boththe Geosat Exact Repeat Mission and the TOPEX/Po-seidon satellite (White and Tai 1992; Zou and Lati1994; Hendricks et al. 1996). All of these analysis meth-ods have drawbacks, which can be mostly traced to thelack of a physical basis for the decomposition (Mitchum1993). On the other hand, the total independence of anya priori prejudice of what to expect is perhaps theirgreatest strength.One approach that has been used to investigate ther-mal effects on sea level is to compare EOFs for SSHand SST. However, when multiple fields are analyzedseparately, it can be difficult to identify related modes.This problem can be reduced by using singular valuedecomposition of the covariance of two geophysicalfields. This technique provides an objective method of isolating coupled modes of variability between thefields, and has been in use in meteorology for two de-cades (Prohaska 1976; Lanzante 1984). Here we adoptthe coupled pattern analysis method and notation of Bretherton et al. (1992).Coupled pattern analysis is similar in spirit to EOFanalysis. Two fields,
s
(
 x, t 
) and
z
(
 x, t 
), are expandedinto sums of orthonormal modes as in (1):
 N 
s
(
 x, t 
)
a
(
)
(
 x
) (3a)
k
1
 N 
z
(
 x, t 
)
b
(
)
R  
(
 x
), (3b)
k
1
where we will assume here that both fields are definedon the same spatial grid of 
points (this assumption isnot necessary). In our applications,
s
(
 x, t 
) will representSSH and
z
(
 x, t 
) will represent SST. In coupled patternanalysis, the orthonormal, spatially dependent functions
(
 x
) and
R  
(
 x
) are chosen to be the right and left sin-gular vectors of the cross-covariance matrix:
(
 x, x
)
w
(
 x
)
w
(
 x
)
s
(
 x, t 
)
z
(
 x
,
)
 
, (4)
where
 
is the number of observations and
w
(
 x
) is alatitude-dependent weighting suggested by North et al.(1982) to produce an equal-area weighting. Then
(
 x, x
) can be expanded in terms of these singular vec-tors as
 N 
T
(
 x, x
)
 
(
 x
)
R  
(
 x
), (5)
k k
1
where
 
is the singular value associated with
,
R  
and is nonnegative. Note from (5) that, since both
and
R  
are orthonormal sets, then
R  
 
(6a)
k k
T
 
R  
. (6b)
k k
Equations (6) imply that
and
R  
are eigenvectors of 
 
A
PRIL
1999
601
L E U L I E T T E A N D W A H R
CC 
T
and
T
C,
respectively, and that is theeigenvalue
2
 
for both these eigenvectors. It can be shown that thereare at most
 
1 values of that are nonzero.
2
 
The intent of coupled pattern analysis is to find pairsof ‘‘standing wave’’ patterns, one for
s
(
 x, t 
) and one for
z
(
 x, t 
), that have similar temporal dependence and thatcontribute significantly to the data. When the methoddescribed above is used to find these pairs of patterns,
and
R  
, the covariance between their temporal co-efficients is
a
(
)
b
(
), which reduces to
 
. Thus if a pair of patterns corresponds to a large singular value,the covariance between their temporal coefficients islarge. There is no enforced criterion, however, that thetwo spatial patterns must look alike. If a mode with alarge singular value has patterns for
s
(
 x, t 
) and
z
(
 x, t 
)that are similar, it tends to reinforce the suggestion thatthe mode truly does reflect some sort of linear physicalconnection between the two fields. The spatial corre-lation between the two patterns,
(
 x
)
R  
(
 x
)
k  x
, (7)
space2 2
(
 x
)
R  
(
 x
)
k
is a measure of their similarity. However, there couldalso be a physical connection, even if the two patternsare different since cause and effect need not be local.The statistical significance of the individual modeshas proven difficult to quantify (Wallace 1992). Wehaveadopted an ad hoc method to provide a rough estimateof the significance of the spatial correlations
space
. Theconfidence levels for
space
depend on the number of degrees of freedom
 
.
Because the fields of SSH andSST tend to be coherent over long periods and longwavelengths, using the standard assumption that
 
isrelated to the number of data points
N,
as
 
2,would greatly overstate the case. To ameliorate this de-ficiency, we sought to reduce the degrees of freedom toaccount for the spatial coherence of the fields. We as-sumed that the first modes of maximum covariancefound by the coupled pattern analysis (CPA) includemost of the spatially coherent signal. The fields werereconstructed using (3), successively eliminating thefirst few modes. To be conservative, the more coherentfield, SST, was used to calculate the average spatialcorrelations versus distance in both longitude and lati-tude.For each reconstruction, the average
e
-folding dis-tance, the distance required forthe amplitudetodecreaseby 1/ 
e,
was calculated along lines of longitude and lat-itude. As expected, the average
e
-folding distancedropped as the dominant modes were eliminated fromthe reconstructions. Eliminating the first six modes(91%–95% of the covariance) excluded the significantcoherent modes. Using this reconstruction, weestimatedthe degrees of freedom, to be the number of data
 
˜ 
 ,
points divided by the
e
-folding distances in number of grid points in longitude and latitude. To estimate the99% confidence levels for
space
of the first five modes,the reconstructed values were used in a Student’st-dis-
 
˜ 
tribution with the following results: global
209,
 
˜ 
space
(99%)
0.16; Pacific
128,
space
(99%)
0.20;
 
˜ 
Atlantic
85.7,
space
(99%)
0.25; and Indian
 
˜ 
 
˜ 
80.1,
space
(99%)
0.26. For the results with the sea-sonal terms removed from both datasets (section 3a),the confidence levels were estimated to be global
 
˜ 
234,
space
(99%)
0.15; Pacific
141,
space
(99%)
 
˜ 
0.19; Atlantic
87.4,
space
(99%)
0.25; and Indian
 
˜ 
90.9,
space
(99%)
0.24.
 
˜ 
A number of other functions have been defined tocharacterize the significance of each mode.Forexample,the total squared covariance is
 x,x
(
 x, x
)
(
 x, x
),which can be shown to equal . Thus, the squared
 
2
 
1
covariance fraction,
2
 
SCF
, (8)
 
12
 
ll
1
is a measure of the fraction of the total squared co-variance explained by the
th pair of patterns, and willbe our primary means of identifying significant modes.However, there can be some ambiguity in the inter-pretation of SCF values. A mode described by
(
 x
) and
R  
(
 x
) can make a large contribution to the total squaredcovariance if the temporal coefficients
a
(
) and
b
(
)are well correlated, but also if either
a
(
) or
b
(
) isespecially large. For example, suppose we are compar-ing SST and TOPEX/Poseidon SSH data. And supposewe do not remove an adequate tide model from the SSHresults so that the SSH residual field,
s
(
 x, t 
), has a largeremaining tidal signal. In such cases we have found thatthe coupled pattern analysis can produce modes that aredominated by the tidal signal and have large SCF values.Clearly these modes do not describe any causal physicalrelationship because the SST data do not contain anyobvious tidal signals. In this case the analysis will con-struct a mode by extracting a spurious time dependencefrom the SST data that has a small amplitude but hap-pens to look reasonably similar to the tidal SSH timedependence. Because the tidal variance in the SSH sig-nal is large, the mode provides a substantialcontributionto the total squared covariance even though the corre-sponding SST signal is small.One way of helping to identify problems of this sortis to consider the temporal correlation coefficient
a
(
),
b
(
)
k
. (9)
time2 2
a
(
)
b
(
)
k
Because of its denominator
(
)
(
),
time
is not
2 2
a b
k
inflated as much by large, spurious modes as is the SCF.The tidal modes described above are still apt to giveseemingly large values of 
time
since the analysis tech-nique does extract a tidelike time dependence from the

You're Reading a Free Preview

Download
scribd
/*********** DO NOT ALTER ANYTHING BELOW THIS LINE ! ************/ var s_code=s.t();if(s_code)document.write(s_code)//-->