This article was downloaded by: [1.9.46.

41]
On: 12 November 2014, At: 20:22
Publisher: Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered
office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Marine Geodesy
Publication details, including instructions for authors and
subscription information:
http://www.tandfonline.com/loi/umgd20

Regional Sea Level Reconstruction in the

Pacific Ocean
a

B. D. Hamlington , R. R. Leben , L. A. Wright & K. -Y. Kim

Colorado Center for Astrodynamics Research , University of

Colorado at Boulder , Colorado , USA
Published online: 18 Dec 2012.

To cite this article: B. D. Hamlington , R. R. Leben , L. A. Wright & K. -Y. Kim (2012) Regional
Sea Level Reconstruction in the Pacific Ocean, Marine Geodesy, 35:sup1, 98-117, DOI:
10.1080/01490419.2012.718210
To link to this article: http://dx.doi.org/10.1080/01490419.2012.718210

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the
Content) contained in the publications on our platform. However, Taylor & Francis,
our agents, and our licensors make no representations or warranties whatsoever as to
the accuracy, completeness, or suitability for any purpose of the Content. Any opinions
and views expressed in this publication are the opinions and views of the authors,
and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content
should not be relied upon and should be independently verified with primary sources
of information. Taylor and Francis shall not be liable for any losses, actions, claims,
proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or
howsoever caused arising directly or indirectly in connection with, in relation to or arising
out of the use of the Content.
This article may be used for research, teaching, and private study purposes. Any
substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing,
systematic supply, or distribution in any form to anyone is expressly forbidden. Terms &
Conditions of access and use can be found at http://www.tandfonline.com/page/termsand-conditions

Marine Geodesy, 35(S1):98117, 2012

Copyright Taylor & Francis Group, LLC
ISSN: 0149-0419 print / 1521-060X online
DOI: 10.1080/01490419.2012.718210

Regional Sea Level Reconstruction

in the Pacific Ocean
B. D. HAMLINGTON, R. R. LEBEN, L. A. WRIGHT,
AND K.-Y. KIM

Downloaded by [1.9.46.41] at 20:22 12 November 2014

Colorado Center for Astrodynamics Research, University of Colorado at

Boulder, Colorado, USA
As a result of the spatial and temporal sampling shortcomings of tide gauge and satellite
altimetry data, forming a sea level record that has the quality and duration necessary to
compare past to present sea level is a challenge. Here, we build on the recently developed
method for reconstructing sea level involving cyclostationary empirical orthogonal
functions (CSEOFs) and, in particular, examine how the reconstruction varies as a
result of the domain used to compute the CSEOF basis functions. While not affecting
the ability to capture the dominant climate signals, the choice of basis function is found
to have a significant effect on the regional distribution of the trend in sea level. We also
demonstrate the use of a CSEOF reconstructed sea level dataset for climate monitoring,
focusing on climate signals in the Pacific Ocean.
Keywords CSEOF, sea level reconstruction, climate monitoring, ENSO, PDO

1. Introduction
Studying past climate variations is vital to improving the understanding of current and future
climate change. To compare past and present climate variations, a long and consistent data
record is necessary. Obtaining such a data record is difficult, however, due to changes in
observing systems over the past century. For instance, past measurements of sea surface
temperature (SST) were heavily affected by economic and political changes that altered
shipping routes. As technology advanced, the method for measuring SST shifted from
ships to buoys and then subsequently to satellites. Satellites provide near global coverage
of the ocean but have only been in use over the last couple of decades. In an attempt to
create a data record with the spatial coverage of satellites and the temporal span of the
historical SST measurements from ships and buoys, reconstruction techniques have been
developed in recent years. Using basis functions computed from a short, spatially dense
dataset to interpolate a long time series of spatially sparse observations was a technique
first implemented by Smith et al. (1996). Empirical orthogonal functions (EOFs) were
computed from 12 years of satellite-derived SST data and used as basis functions to estimate
global SST temperature fields from 1950 to 1992. This technique (known commonly as
EOF reconstruction) has been modified and improved upon in several studies since and

Received 16 January 2012; accepted 2 May 2012.

Address correspondence to Dr. Benjamin Hamlington, 431 UCB, ECNT 320, University of
Colorado at Boulder, Boulder, CO 80309-0431. E-mail: hamlingt@colorado.edu

98

Downloaded by [1.9.46.41] at 20:22 12 November 2014

Pacific Ocean CSEOF Sea Level Reconstruction

99

its application has been extended to other ocean observations such as sea level pressure
(Kaplan et al. 1998, 2000; Smith et al. 2008).
Similar to SST, the techniques for measuring sea level have changed over the past
century. Tide gauges have measured sea level over the past 200 years, with some records
extending back to 1807. While providing long records, the spatial resolution of tide gauge
sampling is poor, making studies of the large-scale patterns of ocean variability and estimates of global mean sea level (GMSL) difficult. The vast majority of tide gauges are
located in the northern hemisphere with comparatively fewer tide gauges in the southern hemisphere, particularly before 1950. Large clusters of tide gauges are found around
heavily populated areas in North America, western and northern Europe, and Japan. Since
1993, however, satellite altimetry has provided accurate measurements of sea surface height
(SSH) with near-global coverage. The measurements taken by satellite altimeters have led
to the first definitive estimates of GMSL rise and have improved understanding of how
sea levels are changing regionally (e.g., Beckley et al. 2007; Cazenave and Nerem 2004;
Leuliette et al. 2004; Miller and Douglas 2007; Nerem et al. 2010). Nevertheless, these
relatively short records provide no information about the state of the ocean prior to 1993,
and the lower-frequency signals that are known to be present in the ocean are difficult or
impossible to resolve. As a result of the respective shortcomings of tide gauge and satellite
altimetry records, combining the shorter but essentially complete global coverage afforded
by satellite altimetry with the longer but sparsely distributed tide gauge data set is a research
area of interest.
Chambers et al. (2002) made one of the first attempts to combine tide gauge data and
satellite altimeter data. Performing an EOF reconstruction, Chambers et al. (2002) focused
on capturing the interannual-scale signals such as El Nino-Southern Oscillation (ENSO)
and the Pacific Decadal Oscillation (PDO). Building on this work and that of Kaplan et al.
(1998, 2000), Church et al. (2004) produced the most comprehensive and widely cited sea
level reconstruction to date, spanning 1950 through 2001. This reconstruction has since
been updated to extend back to 1870 and forward to the present (Church and White 2006,
2011), although only the portion of the full dataset from 1950 to 2001 has been publicly
released. Other studies on EOF sea level reconstructions have been published, albeit using
data sources other than tide gauges and satellite altimetry or with a more limited scope
or region of interest (Berge-Nguyen et al. 2008; Calafat et al. 2009; Llovel et al. 2009;
Christiansen et al. 2010; Meyssignac et al. 2012). More recently, Hamlington et al. (2011b)
attempted to improve on the EOF reconstruction of sea level by introducing the use of
cyclostationary empirical orthogonal functions (CSEOFs) to serve as basis functions for
the reconstruction. The use of CSEOFs proved advantageous for capturing large-scale ocean
variability such as ENSO.
Techniques for reconstructing sea level are only a decade old, and several questions
remain regarding how sea level reconstructions should be computed. One important issue
arises when considering the best way to reconstruct sea level variability for a specific area
of the ocean. Policy makers are not as interested in globally averaged sea level variability
as much as they are in changes that are taking place on a regional scale. The spatial
variability of the secular trend in sea level varies significantly across the globe with some
areas experiencing trends on the order of 1 cm/year over the past two decades. Accurate
assessments of both past and present regional sea level variability are important when setting
policy for the future. To obtain the most accurate representation of sea level variability,
should one globally reconstruct sea level and then subset the resulting dataset, or is it better
to reconstruct only the specific area of interest? Such a question has important implications
for how sea level reconstructions can and should be used for climate monitoring. The ability

100

B. D. Hamlington et al.

Downloaded by [1.9.46.41] at 20:22 12 November 2014

to use sea level reconstructions for monitoring climate signals has not been discussed at
length in literature, since past studies were mainly focused on reconstructing the secular
trend at both global and regional scales. In this article, we address these issues by focusing
on reconstructing sea level in the Pacific Ocean from 1950 to present. Section 2 of this paper
discusses the satellite altimetry data and tide gauge data that are used for the reconstruction.
Section 3 provides an overview of CSEOFs and describes how a reconstruction is performed
using CSEOFs. Section 4 compares globally computed basis functions to basis functions
computed using only data in the Pacific Ocean and the effect each set of basis functions
would have on a subsequent sea level reconstruction. Also in Section 4, the ability to use a
CSEOF sea level reconstruction for the purpose of climate monitoring in the Pacific Ocean
is discussed. Finally, Section 5 provides an evaluation of what the presented results suggest
about reconstructing sea level on a regional basis and how such reconstructions can be used
to improve long-term monitoring of sea level in the Pacific Ocean.

2. Data
Both tide gauge and satellite altimeter data records are needed to reconstruct sea level. The
satellite altimetry is used to find patterns of spatial variability that will be used as basis
functions for the reconstruction. These basis functions are fit in a least-squares sense to the
tide gauge record extending back over the time period of interest. The satellite altimetry
and tide gauge data records used in this paper are similar to those used in Hamlington
et al. (2011b), and for a complete description the reader should refer to that study. The data
used for this study and the applicable editing choices are briefly outlined below, and any
differences from Hamlington et al. (2011b) are highlighted.
2.1. Tide Gauge Data
The tide gauge dataset used for the period from 1950 to 2011 is monthly mean sea level
records obtained from the data archive at the Permanent Service for Mean Sea Level
(PSMSL). Only the Revised Local Reference (RLR) data are used, which are measured sea
levels at each site relative to a constant local datum over the complete record. The dataset
used from PSMSL was downloaded in July 2011 and contains tide gauge records extending
into 2011. Unlike Hamlington et al. (2011b), tide gauges are rejected from use only if the
nearest satellite altimeter grid point containing data was more than 250 km away from
the tide gauge location. As seen in Figure 1, this resulted in a total of 1081 tide gauges
available for the global reconstruction and 496 available for the regional reconstruction
of the Pacific Ocean only. In the future, some of the included tide gauges may warrant
more careful consideration and exclusion from analysis (e.g., short records, records with
unphysical datum shifts), but for the purposes of this article we choose to adopt lenient
selection criteria.
We linearly interpolated the monthly tide gauge data to weekly intervals to match the
1-week temporal resolution of the altimeter-derived CSEOF basis functions. Prior to this,
we filled gaps of 12 months in the tide gauge records using cubic spline interpolation.
Times when a tide gauge underwent month-to-month change of greater than 250 mm were
edited and removed. The available tide gauge records were averaged to produce a single
time series if there were multiple tide gauges associated with a single spatial grid point of
the basis functions. As in Hamlington et al. (2011b), the annual and semi-annual cycles
were not removed from the tide gauge time series prior to performing the reconstruction,

Pacific Ocean CSEOF Sea Level Reconstruction

101

600
Global
Pacific Only

# of Available Tide Gauges

500

400

300

200

100

Downloaded by [1.9.46.41] at 20:22 12 November 2014

0
1950

1960

1970

1980
Year

1990

2000

2010

Figure 1. Top: Locations of the tide gauges used for the global reconstruction (black and red) and
Pacific reconstruction (red). Bottom: The number of tide gauges available for both reconstructions
over the time period from 1950 to 2010. (Color figure available online.)

and we fit our basis functions to the tide gauge sea level measurements, not their first
derivative. The process of fitting CSEOF basis functions is found to be insensitive to small
datum shifts or biases, particularly when only reconstructing the time period from 1950 to
present during which tide gauge coverage is relatively consistent.

2.2. Satellite Altimetry Dataset

CSEOF basis functions for our reconstruction were estimated from the Archiving, Validation, and Interpretation of Satellite Oceanographic (AVISO) quarter-degree-resolution,
multiple altimeter product based on satellite altimeter measurements spanning 19922011
collected by the TOPEX/Poseidon, ERS-1&2, Geosat Follow-On, Envisat, Jason-1, and
OSTM satellites. This updated and reprocessed gridded data product, which was released
in 2011, was created using the delayed time Ssalto/DUACS multi-mission altimeter data
processing system with improved homogenous inter-calibrated corrections applied to the
entire data record. Global crossover minimizations and local inverse methods were used
to derive inter-calibrated highly accurate along-track data that are referenced to a consistent mean. The along-track data were then merged through a global space-time objective
mapping technique that takes into account correlated noise (LeTraon et al. 1998).
We applied little additional processing other than removing the mean and a linear
trend estimated by least-squares fit of the time series at each spatial grid point. A CSEOF
decomposition of the satellite altimetry data is not able to extract the change in mean sea
level into a single mode. It is therefore necessary to remove mean sea level from the satellite
altimetry data before computing the basis functions to avoid putting low-frequency power
into each CSEOF mode. Although our technique does result in the removal of the spatial
pattern of sea level trends, it is unlikely that the regional distribution of sea level trends
over the past two decades is the same as that since 1950, and it is therefore unwise to
force this stationary trend pattern on the reconstruction. Removing the trend from each grid
point does not significantly affect the ability of the reconstruction to capture regional trend
variability or decadal time-scale signals in the reconstruction. Using the study of Tai (1989),

102

B. D. Hamlington et al.

the percentage of signal reduction at various periods caused by removing the linear trend
from a 17-year record can be computed. Signals with periods of approximately 10 years
undergo an RMS signal reduction of less than 5%, and even signals at 20-year periods are
reduced only 30%. Linear detrending of the altimeter data before computing the CSEOFs
should have little effect on decadal-scale variations and our ability to capture them in the
reconstruction.

3. Methods

Downloaded by [1.9.46.41] at 20:22 12 November 2014

3.1. Basis Functions: CSEOFs vs. EOFs

When compared with CSEOFs, EOFs have characteristics that make them suboptimal for
use as basis functions for sea level reconstruction. EOFs enforce a stationarity on the
spatial variability. A single spatial map defines the basis function, and the reconstruction
procedure simply computes the amplitude modulation of this map through time. Given
the evidence that many signals in geophysical data are cyclostationary, CSEOFs provide
significant advantages over EOFs when dealing with signals such as the modulated annual
cycle (MAC) and ENSO signals (Hamlington et al. 2011a, 2011b).
The decomposition of data in terms of a set of basis functions is often useful in
understanding the complicated response of a physical system. By decomposing into less
complicated patterns, it may be easier to understand and shed light into the nature of the
variability in a dataset. While theoretical basis functions have been studied extensively,
exact theoretical basis functions are very difficult to find and in general, computational
basis functions are sought instead. Perhaps the simplest and most common computational
basis functions are EOFs. Consider a simple system defined by:
T(r, t) =

LVi (r)PCi (t)

(1)

where LV(r) is a physical process (termed to be the loading vector, as above) with amplitude
defined by a stochastic time series PC(t), which is called the principal component time series.
Each loading vector and principal component time series pair represents a single EOF mode.
As mentioned above, however, physical processes and the corresponding statistics are timedependent, and representing the data with stationary loading vectors can lead to erroneous
interpretation of the data.
Kim et al. (1996; Kim and North 1997; Kim and Wu 1999; Kim and Chung 2001)
introduced the concept of cyclostationary empirical orthogonal function (CSEOF) analysis to capture the time-varying spatial patterns and longer-timescale fluctuations present
in geophysical signals. The significant difference between CSEOF and EOF analysis is
the LVs time dependence, which allows the spatial pattern of each CSEOF mode to
vary in time, with the temporal evolution of the spatial pattern of the CSEOF LVs constrained to be periodic with a selected nested period. In other words, the system is
defined as:
T(r, t) =

LVi (r, t)PCi (t)

LV(r, t) = LV (r, t + d)

(2)

Pacific Ocean CSEOF Sea Level Reconstruction

103

where the loading vectors are now time dependent, and are periodic with the nested period,
d. The CSEOF LVs and corresponding PC time series are obtained by solving:
Cov(r, t; r , t )LVn (r , t ) = n LVn (r, t)

(3)

with r and t representing another point in space and time and under the assumption that:

Downloaded by [1.9.46.41] at 20:22 12 November 2014

Cov(r, t; r , t ) = Cov(r, t + d; r , t + d)

(4)

In other words, the space-time covariance function is periodic in time with the nested
period, d. Since the covariance matrix cannot be written as a square matrix, Eq. (3) cannot
be solved in the same manner as EOFs. Instead, Eq. (3) is Fourier transformed twice with
respect to t and t , making use of the assumption that the covariance matrix is periodic.
Because of the periodicity of the covariance function, Eq. (3) could be written in the same
form as EOFs in Fourier space. The PC time series in Fourier space are easy to obtain, and
finally, both the LVs and PC time series are transformed back to physical space. For further
details, the reader should refer to Kim and North (1997) in which a detailed description of
the computation of CSEOFs is provided.
3.2. Reconstruction Procedure
To test the dependence of the reconstruction on the basis functions, a CSEOF decomposition
was conducted on the satellite altimetry data both globally and on the Pacific Ocean region.
In both cases, the nested period is defined to be one year. Hamlington et al. (2011a, 2011b)
have demonstrated the ability of CSEOF analysis with a nested period of one year to extract
the modulated annual cycle (MAC) and ENSO signals from the AVISO satellite altimetry
data. Since the MAC and ENSO signals explain much of the variance in the AVISO data,
we select a one-year nested period for our analysis. Further information on the selection of
the nested period can be found in the references (Kim et al. 1997; Hamlington et al. 2011a,
2011b). For the weekly AVISO data, 52 separate LV patterns are obtained for each CSEOF
mode, with the corresponding PC time series representing the amplitude modulation of
these LV patterns over time.
The reconstruction procedure used here follows closely that of Hamlington et al.
(2011a). The mathematical details of the computation of the reconstruction are discussed
at length in Kaplan et al. (2000) and Church et al. (2004) and will not be repeated here. The
significant difference between those previous reconstructions and the one presented here
is the use of CSEOFs as basis functions instead of EOFs. When performing the sea level
reconstruction, every LV space-time pattern associated with an individual CSEOF mode
is fit simultaneously. In other words, rather than fitting a single spatial pattern at a single
point in time as in an EOF reconstruction, we fit a window of 52 spatial patterns to 52
points in time. The window of 52 spatial patterns shifts by one week for the duration of
the tide gauge data. This leads to a loss of a half-year at each end of the reconstruction but
does reduce the sensitivity to the sampling error associated with the limited number of tide
gauge measurements and to erroneous tide gauge measurements at a single point.
Accounting for GMSL using reconstruction techniques is not a trivial task. Christiansen
et al. (2010) discussed the difficulties of estimating GMSL using EOF reconstruction
techniques, albeit using model data instead of satellite altimetry and tide gauge data. While
CSEOF basis functions describe cyclostationary variability in sea level, as a result of the
short length of the satellite altimetry record, no single CSEOF mode captures the secular
trend. Similar difficulties are encountered when using EOF analysis. Church et al. (2004)

Downloaded by [1.9.46.41] at 20:22 12 November 2014

104

B. D. Hamlington et al.

approximated the trend in the reconstruction by introducing a constant basis function that
is fit along with the other EOF basis functions. This spatially uniform basis function
distributes the same time series at every spatial point in the reconstruction. Hamlington
et al. (2011a) took a different approach, separating the computation of the secular trend
from the reconstruction procedure. This method, however, still has the effect of adding
the same time series to every location in the reconstruction. Rather than adopt either of
these techniques, we make no attempt to account for the secular trend in GMSL. The focus
of this article is on reconstructing regional variability, and requiring the reconstruction to
reproduce GMSL is not necessary. Note that the regional trends shown below are computed
relative to the trend in GMSL over the same time period.
Estimating the error in a sea level reconstruction is not a trivial matter. While we do
not explicitly address the errors in our reconstruction in this paper, we treat the reconstruction error in a similar way to previous studies (e.g., Church et al. 2004; Hamlington
et al. 2011b; Meyssignac et al. 2012). Errors can be estimated as a combination of instrument errors and the error resulting from truncating the basis functions estimated from
the AVISO dataset. Similar to previously published research, we do not provide values
for the error estimated in this way as such a discussion does not significantly impact the
results of this study, but emphasize that such an analysis is possible and available. More
exhaustive error analysis of the estimate of GMSL is generally included in published studies on sea level reconstructions, but since GMSL is not addressed here, such a discussion
is omitted.

4. Results
4.1. Global vs. Regional Sea Level Reconstruction
To study the effect of using globally computed basis functions versus regionally computed
basis functions, a CSEOF decomposition was performed on the global AVISO dataset and
also on the subsetted Pacific Ocean region. In both cases, a nested period of one year was
used, and the CSEOF modes containing 95% of the variability in each region were retained.
This resulted in 17 CSEOF modes for both the global and Pacific decompositions. Before
using these modes to perform the reconstruction, the LVs and PC time series from both
decompositions were compared. As seen in Figure 2, the agreement between the PC time
series was very good for the first 5 modes, with correlations of 0.79, 0.99, 0.90, 0.83, and
0.82 for modes 1 through 5, respectively. As described in Hamlington et al. (2011a, 2011b),
mode 1 represents the modulated annual cycle, mode 2 represents the eastern-Pacific (EP)
ENSO signal, and mode 3 represents the central-Pacific (CP) ENSO signal. In particular,
modes 2 and 3 are dominated by climate signals in the Pacific Ocean, so the high correlation
with the globally computed PC time series is expected. Although the higher order modes
do not correlate as highly, the majority of the variance is contained in the first 5 modes.
When comparing the LVs, correlations among the corresponding weeks of the first 5 modes
are all greater than 0.8. In short, when focusing on the Pacific Ocean region, the CSEOF
analysis does not vary significantly when performed globally or specifically in the Pacific
Ocean basin.
As described in Section 3, CSEOF sea level reconstructions for 19502010 have been
computed using basis functions computed from both the full dataset and from the subset
containing the Pacific Ocean. The annual cycle is not removed from the tide gauges and is
fit by the first CSEOF mode. All 17 modes, describing 95% of the variability in the dataset,
are fit in both reconstructions. To study the difference between the two reconstructions, we

Pacific Ocean CSEOF Sea Level Reconstruction

105

1994

1996

1998

2000

2002

2004

2006

2008

2010

2012

1994

1996

1998

2000

2002

2004

2006

2008

2010

2012

1994

1996

1998

2000

2002

2004

2006

2008

2010

2012

1994

1996

1998

2000

2002

2004

2006

2008

2010

2012

1994

1996

1998

2000

2002

2004

2006

2008

2010

2012

10

10

Downloaded by [1.9.46.41] at 20:22 12 November 2014

5
0
5
10
5

5
5

Figure 2. The PC time series for the first five modes obtained from a global CSEOF decomposition
(blue) and a Pacific CSEOF decomposition (red) of the AVISO satellite altimetry data over the period
from 1993 to 2010. Mode 1 represents the modulated annual cycle while modes 2 and 3 represent the
EP and CP ENSO signals, respectively. (Color figure available online.)

compare the reconstructed PC time series from the global and Pacific reconstructions. As
seen in Figure 3, the reconstructed PC time series agree well for the first 5 modes, with
correlations of 0.83, 0.99, 0.76, 0.67 and 0.76, respectively. As a result of the dominance
of the Pacific Ocean signal in the global CSEOF mode, mode 2 (representing the EP
ENSO) shows an almost identical reconstructed PC time series for the global and Pacific
reconstructions. More surprisingly, however, is the disagreement of the reconstruction of
mode 3, the CP ENSO mode, before 1970. Prior to 1970, the correlation between the global
and Pacific reconstructions is only 0.25, but from 1970 to 2011, the correlation rises to
0.90. After 1970, the reconstructed PC time series shows several large CP La Nina events

106

B. D. Hamlington et al.

1960

1970

1980

1990

2000

2010

1960

1970

1980

1990

2000

2010

1960

1970

1980

1990

2000

2010

1960

1970

1980

1990

2000

2010

1960

1970

1980

1990

2000

2010

10

10

Downloaded by [1.9.46.41] at 20:22 12 November 2014

5
0
5
10
10

10
10
5
0
5

Figure 3. The reconstructed PC time series for the first 5 modes using CSEOF basis functions
computed globally (blue) and from the Pacific (red). Mode 1 represents the modulated annual cycle
while modes 2 and 3 represent the EP and CP ENSO signals, respectively. (Color figure available
online.)

that are not present in the prior years, perhaps signaling a shift in the nature of the mode
after 1970.
To further compare the two reconstructions, we compute correlations between the time
series from the global reconstruction and the Pacific reconstructions. The annual cycle is
removed by a least-squares fit from both reconstructions to avoid demonstrating a simple
agreement in the dominant seasonal signal. Figure 4 shows the spatial variation of the correlation between the two reconstructions. In the tropical Pacific region, which is dominated
by ENSO variability, the correlations are greater than 0.9. Based on the reconstructed PC
time series shown in Figure 3, this is not unexpected as both reconstructions capture the

Downloaded by [1.9.46.41] at 20:22 12 November 2014

Pacific Ocean CSEOF Sea Level Reconstruction

107

Figure 4. The spatial variation of the correlation between the global reconstruction and Pacific
reconstruction over the time period from 1950 to 2010. The seasonal signal is removed before
computing the correlations. (Color figure available online.)

variability associated with ENSO over the majority of the time period considered. Outside
of this region, correlations are generally lower although areas of better agreement are still
visible.
While the results shown above compare the two reconstructions, the question of which
is more accurate remains. To test which reconstruction agrees better with the tide gauge
data used to reconstruct sea level in the Pacific Ocean region, correlations between each
reconstruction and the 496 available tide gauges are computed. Again, the seasonal signal
is removed from both the sea level reconstructions and tide gauges. Figure 5 shows a
comparison of the correlations computed between the tide gauges and both the global and the
Pacific reconstructions. In general, the Pacific reconstruction yields better agreement with
the tide gauges, producing higher correlations for 58% of the tide gauges when compared
to the correlations obtained with the global reconstruction. The median correlation for
the global reconstruction is found to be 0.44 while the mean correlation is 0.42. For the
correlation between the tide gauges and Pacific reconstruction, the median correlation is
0.49, with a mean correlation of 0.48. In both cases very short tide gauge records and tide
gauge records that would not pass the editing criteria used in past reconstructions result in the
majority of the lower correlations seen in Figure 5 and serve to lower the median correlation.
Nonetheless, the Pacific Ocean reconstruction appears to more accurately represent the tide
gauge data when compared to the global reconstruction, although the improvement is
relatively small.
Finally, we compare the regional variation of the secular trend from both the global
and Pacific reconstructions. An important validation for any sea level reconstruction is the
ability to accurately reproduce the trend over the satellite altimeter time period. As seen in

108

B. D. Hamlington et al.
1

0.9

Pacific Recon. TG Correlation

0.8
0.7
0.6
0.5
0.4
0.3

Downloaded by [1.9.46.41] at 20:22 12 November 2014

0.2
0.1
0

0.1

0.2

0.3

0.4
0.5
0.6
Global Recon. TG Correlation

0.7

0.8

0.9

Figure 5. Comparison of the correlation of each reconstruction with the tide gauges in the Pacific
Ocean. Both the global and Pacific reconstructions are subsampled at each tide gauge location and
correlated with the tide gauge records. The seasonal signal is removed from both the reconstruction
and tide gauge records prior to computing the correlation. Tide gauges above the red line indicate a
higher correlation found in the Pacific reconstruction. (Color figure available online.)

Figure 6, the linear trend is computed from 1993 through 2010 for the global reconstruction
(6A), the Pacific reconstruction (6B), and the AVISO data (6C). Both the global and Pacific
reconstructions do well in reproducing the spatial trend pattern computed from AVISO,
although the Pacific reconstruction appears to more accurately capture the trends in the
tropical Pacific. Note that the regional trends shown are computed relative to the trend in
GMSL over the same time period. When computing the trends from 1950 to 2010, however,
there is less agreement between the two reconstructions. Figure 7 shows the linear trend
estimated from the global reconstruction (7A), the Pacific reconstruction (7B) and the
difference in the estimated trends between the global and Pacific reconstructions. In many
locations, the difference between the two reconstructions is as large as the estimated trends.
The best agreement is found in the tropical Pacific region while areas at higher latitudes
show little correspondence.

4.2. Pacific Ocean Climate Monitoring

Determining the best way to reconstruct sea level for a specific region has important
implications for climate monitoring. If a global reconstruction is not able to adequately
capture the dominant signals in the Pacific Ocean, for instance, tracking of the evolution
of ENSO over time would be difficult. To demonstrate the potential to use sea level
reconstructions for the purpose of climate monitoring, we look at three different signals
in the Pacific Ocean: the eastern Pacific ENSO, the central Pacific ENSO, and the Pacific
Decadal Oscillation (PDO). For the results presented below, we use only the reconstruction
created with the basis functions computed from the Pacific Ocean.

109

Downloaded by [1.9.46.41] at 20:22 12 November 2014

Pacific Ocean CSEOF Sea Level Reconstruction

Figure 6. Top: Estimates of the linear trend over the period from 1993 to 2010 for the global
reconstruction. Bottom: Pacific reconstruction (middle) and AVISO satellite altimetry data. (Color
figure available online.)

B. D. Hamlington et al.

Downloaded by [1.9.46.41] at 20:22 12 November 2014

110

Figure 7. Estimates of the linear trend over the period from 1950 to 2010 for the global reconstruction
(top) and Pacific reconstruction (middle). The difference between the global reconstruction trends
and the Pacific reconstruction trends are shown in the bottom panel. (Color figure available online.)

Downloaded by [1.9.46.41] at 20:22 12 November 2014

Pacific Ocean CSEOF Sea Level Reconstruction

111

The EP ENSO signal is described by the second CSEOF mode in both the satellite
altimetry data (Figure 2) and the reconstructed sea level (Figure 3). The LV associated with
this mode is shown in the top panels of Figure 8. While there are 52 patterns associated
with this LV, only four corresponding to the seasons are shown in the interest of space.
As expected, the peak signal within this mode occurs during the boreal winter. Several
indices are used to monitor ENSO. One commonly cited index is the Multivariate ENSO
Index (MEI). The MEI is computed using six different variables but does not include SSH
measurements (Wolter 2010; Wolter and Timlin 1998). In the bottom panel of Figure 8, the
reconstructed EP ENSO mode is shown along with the MEI for the period 19502010. The
correlation between the MEI and the reconstructed EP ENSO is 0.90. The reconstructed
PC time series could potentially be used as a climate index for monitoring the EP ENSO,
providing an alternative to other indices by relying solely on sea level measurements.
The LVs and reconstructed PC time series for CSEOF mode 3 are shown in Figure 8.
When compared to the analysis by Ashok et al. (2007), the spatial pattern seen in the LVs
is representative of the CP ENSO or ENSO Modoki signal with warming in the central
Pacific and cooling in the eastern and western regions of the Pacific Ocean. While there is
no commonly used index for monitoring the CP ENSO, we compute the ENSO Modoki
Index (EMI) as defined by Ashok et al. (2007) from the Hadley SST dataset. As seen in
the bottom panels of Figure 9, the agreement between the EMI and the reconstructed mode
3 PC time series over the period 19502009 is good, showing excellent agreement for the
large events in 1983 and 1998 in particular. The correlation before 1970, however, is only
0.33 while the correlation from 1970 to 2010 increases to 0.69. As discussed with regards to
the difference in the global and Pacific reconstructions of this mode, the behaviour of the CP
ENSO signal appears to change after 1970, with larger events occurring more frequently.
The changing nature of the CP ENSO signal has been the subject of several recent studies,
and a more detailed examination of the results shown here are left for future consideration.
With the satellite altimetry record spanning less than two decades, there are questions
regarding how well basis functions computed from such a short record can capture decadal
variability. To test the ability of the reconstruction to explain decadal variability, we attempt
to extract the signal associated with the PDO from the Pacific Ocean reconstruction. Using
the region of the northeast Pacific from 25 N to 60 N and 180 W to the North American
coast, we perform a CSEOF decomposition of the reconstruction, using a nested period
of four years. A longer and shorter nested period has also been used with similar results
and the selection of the nested period in this case does not have a significant impact on the
result other than to produce a smoother time series for comparison. Past studies of both
SST and SSH in this region show that the PDO signal is contained in the first EOF mode,
and similarly we find that the PDO signal is explained by the first CSEOF mode. The top
panel of Figure 10 shows one of the typical winter-month patterns contained in the first
LV (in total, there are 208 weekly patterns spanning the 4-year nested period). This spatial
pattern agrees well with the results shown in Cummins et al. (2005) for SSH and Mantua
et al. (1997) for SST. The CSEOF PC time series is compared to the PDO index derived
from a reconstructed SST dataset (www.jisao.washington.edu/pdo) in the bottom panel of
Figure 10. While the CSEOF PC time series is much smoother than the SST PDO index, the
agreement over the full-time period is excellent, with a correlation between the noisier SST
index and CSEOF PC time series of 0.52, which increases significantly with smoothing of
the SST index. While a more thorough study of the ability to explain decadal variability
may be warranted in the future, the CSEOF reconstructed sea level does capture the PDO
signal and could thus potentially be used for monitoring climate signals varying on decadal
timescales.

B. D. Hamlington et al.

Downloaded by [1.9.46.41] at 20:22 12 November 2014

112

Figure 8. Top: LVs for mode 2 of the CSEOF decomposition of the AVISO satellite altimetry data
representing the EP ENSO pattern. Bottom: Pacific reconstruction PC time series (blue) for mode 2
compared to the MEI (green). The correlation between the reconstructed PC time series and MEI is
found to be 0.90. (Color figure available online.)

113

Downloaded by [1.9.46.41] at 20:22 12 November 2014

Pacific Ocean CSEOF Sea Level Reconstruction

Figure 9. Top: LVs for mode 3 of the CSEOF decomposition of the AVISO satellite altimetry
data representing the CP ENSO pattern. Bottom: Reconstructed PC time series (blue) for mode 3
compared to the EMI computed from the Hadley SST reconstruction (green). The correlation between
the reconstructed PC time series and EMI is found to be 0.59. (Color figure available online.)

B. D. Hamlington et al.

Downloaded by [1.9.46.41] at 20:22 12 November 2014

114

Figure 10. Top: The spatial pattern associated with the PDO obtained by performing a CSEOF
decomposition of the reconstructed sea level in the northeast Pacific. Bottom: The corresponding
PC time series is compared to the PDO index computed by performing an EOF decomposition of
reconstructed SST. (Color figure available online.)

5. Discussion and Conclusions

While sea level reconstructions have been performed at both regional and global scales,
there has been little discussion regarding how the computation of basis functions and the
subsequent fitting to tide gauges affects the resulting reconstruction. In this article, we have
studied the ability to capture sea level variability in the Pacific Ocean by computing two
separate sea level reconstructions, one using globally computed basis functions fitted to
every available tide gauge, and one using basis functions computed from just the data in

Downloaded by [1.9.46.41] at 20:22 12 November 2014

Pacific Ocean CSEOF Sea Level Reconstruction

115

the Pacific Ocean that are then fit to Pacific Ocean tide gauges. The global and Pacific
CSEOF decompositions do not show significant differences for the first few modes in terms
of both the LVs and corresponding PC time series. Furthermore, when performing the
sea level reconstruction, there is only small disagreement between the reconstructed PC
time series. The most significant discrepancy occurs in the reconstructed CP ENSO mode
(Figure 3) prior to 1970, although the correlation between the two reconstructions after
this time is found to be 0.90. When comparing the reconstructed sea level data, the two
reconstructions match closely in the tropical Pacific where the ENSO signal dominates the
sea level variability (Figure 4). Elsewhere, other areas of high correlation are visible with
no negative correlations anywhere in the reconstruction.
Despite the similarities between the global and Pacific reconstructions described above,
there is significant discrepancy between the spatial variability of the secular trend produced
by both reconstructions. In many locations, the difference between the trends estimated by
the two reconstructions is as large as the estimated trends (Figure 7). Policymakers require
accurate estimates of the secular trend at the regional level to make informed decisions
about both the present and future. This disagreement between the trends requires further
investigation and reinforces some of the results and statements made in Hamlington et al.
(2011b) regarding the difference between the global CSEOF reconstruction and the global
EOF reconstruction of Church et al. (2004). It is difficult to determine definitively whether
the global or Pacific CSEOF sea level reconstruction presented here is more accurate,
although as seen in Figure 5 the Pacific reconstruction more accurately reproduces the tide
gauge records in the Pacific Ocean. Further work is required to definitively determine if sea
level should be reconstructed on a regional level rather than on a global scale. However,
the analysis done here suggests that the issue needs to be examined closely if sea level
reconstructions are to be useful for studies of sea level, particularly with respect to secular
trends, in small regions of the ocean.
Determining the best way to reconstruct sea level for a specific region has important
implications for climate monitoring. The CSEOF sea level reconstruction technique has
been proven to capture the large-scale climate signals in the Pacific Ocean. The EP ENSO,
CP ENSO and PDO are all well represented in the CSEOF sea level reconstruction presented
here over the time period from 1950 to 2010. Most commonly cited climate indices rely on
SST measurements, so climate monitoring based solely on sea level measurements would
provide an important comparison and verification of these indices. Furthermore, SSH is
perhaps a more appropriate variable for climate monitoring than SST, providing a more
complete representation of the state and structure of the ocean.
Techniques for reconstructing sea level have been developed only in the last decade,
and several questions regarding the best way to perform sea level reconstructions remain.
As the focus of both the public and scientific community shifts from global to regional
sea level change, it is important not only to explore how sea level reconstructions can be
applied regionally but also to understand how sea level reconstructions should be computed
for use in regional studies.

Acknowledgements
This work was supported by NASA Ocean Surface Topography Mission Science Team
grant NNX08AR60G and NASA ROSES grant NNX11AE26G. Ocean Topography Project
Scientist support for R.R.L. by the NASA/JPL PO.DAAC is gratefully acknowledged.
K.Y.K. acknowledges support by the Ministry of Land, Transport, and Maritime Affairs (Ocean Climate Variability Program). The altimeter products used in this study were

116

B. D. Hamlington et al.

produced by Ssalto/Duacs and distributed by AVISO, with support from the Centre National
dEtudesSpatiales (CNES).

Downloaded by [1.9.46.41] at 20:22 12 November 2014

References
Ashok, K., S. K. Behera, S. A. Rao, H. Wang, and T. Yamagata. 2007. El Nino Modoki and its
possible teleconnection. J. Geophys. Res. 112:C11007.
Beckley, B. D., F. G. Lemoine, S. B. Lutchke, R. D. Ray, and N. P. Zelensky. 2007. A reassessment of
global and regional mean sea level trends from TOPEX and Jason-1 altimetry based on revised
reference frame and orbits. Geophys. Res. Lett. 34:L14607.
Berge-Nguyen, M., A. Cazenave, A. Lombard, W. Llovel, J. Viarre, and J. F. Cretaux. 2008. Reconstruction of past decades sea level using thermosteric sea level, tide gauge, satellite altimetry,
and ocean reanalysis data. Global Planet. Change 62:113.
Calafat, F. M., D. Gomis, and M. Marcos. 2009. Comparison of Mediterranean Sea level fields for
the period 19612000 as given by a data reconstruction and a 3D model. Global Planet. Change
68(3):175184.
Cazenave, A., and R. S. Nerem. 2004. Present day sea level change: Observations and causes. Rev.
Geophys. 42:RG3001.
Chambers, D. P., C. A. Melhaff, T. J. Urban, D. Fuji, and R. S. Nerem. 2002. Low-frequency variations
in global mean sea level: 19502000. J. Geophys. Res. 107:3026.
Christiansen, B., T. Schmith, and P. Thejll. 2010. A surrogate ensemble study of sea level reconstructions. J. Climate 23:43064326.
Church, J. A., N. J. White, R. Coleman, K. Layback, and J. X. Mitrovica. 2004. Estimates of the
regional distribution of sea level rise over the 19502000 period. J. Climate 17:26092625.
Church, J. A., and N. J. White. 2006. A 20th century acceleration in global sea level rise. Geophys.
Res. Lett. 36:L040608.
Church, J. A., and N. J. White. 2011. Sea-level rise from the late 19th to the early 21st century.
Surveys in Geophysics 32(4):585602.
Hamlington, B. D., R. R. Leben, R. S. Nerem, and K.-Y. Kim. 2011a. The effect of signal-to-noise
ratio on the study of sea level trends. J. Climate 24:13961408.
Hamlington, B. D., R. R. Leben, R. S. Nerem, W. Han, and K.-Y. Kim. 2011b. Reconstruction sea
level using cyclostationary empirical orthogonal functions. J. Geophys. Res. 116:C12015.
Kaplan, A. M., M. A. Cane, Y. Kushnir, A. C. Clement, M. B. Blumenthal, and B. Rajagopalan. 1998.
Analyses of global sea surface temperature 18561991. J. Geophys. Res. 103:1856717589.
Kaplan, A. M., Y. Kushnir, and M. A. Cane. 2000. Reduced space optimal interpolation of historical
marine sea level pressure 18541992. J. Climate 13:29873002.
Kim, K.-Y., G. R. North, and J. Huang. 1996. EOFs of one-dimensional cyclostationary time series:
Computations, examples, and stochastic modeling. J. Atmos. Sci. 53:10071017.
Kim, K.-Y., and G. R. North. 1997. EOFs of harmonizable cyclostationary processes. J. Atmos. Sci.
54:24162427.
Kim, K.-Y., and Q. Wu. 1999. A comparison of study of EOF techniques: Analysis of nonstationary
data with periodic statistics. J. Climate 12:185199.
Kim, K.-Y., and C. Chung. 2001. On the evolution of the annual cycle in the tropical Pacific.
J. Climate 14:991994.
Le Traon, P.-Y., F. Nadal, and N. Ducet. 1998. An improved mapping method of multi-satellite
altimeter data. J. Atmos. Oceanic Technol. 15:522534.
Leuliette, E. W., R. S. Nerem, and G. T. Mitchum. 2004. Results of TOPEX/Poseidon and Jason
calibration to construct a continuous record of mean sea level. Marine Geodesy 27:7994.
Llovel, W., A. Cazenave, P. Rogel, A. Lombard, and M. B. Nguyen. 2009. Two-dimensional reconstruction of past sea level (19502003) from tide gauge data and an Ocean General Circulation
Model. Clim. Past 5:217227.
Meyssignac, B., M. Becker, W. Llovel, and A. Cazenave. 2012. Assessment of two-dimensional past
sea level reconstructions over 19502009 based on tide-gauge data and different input sea level
grids. Surv. Geophys. 33(5):945972.

Downloaded by [1.9.46.41] at 20:22 12 November 2014

Pacific Ocean CSEOF Sea Level Reconstruction

117

Miller, L., and B. C. Douglas. 2007. Gyre-scale atmospheric pressure variations and their relation to
19th and 20th century sea level rise. Geophys. Res. Lett. 34:L16602.
Nerem, R. S. 1995. Global mean sea level variations from TOPEX/POSEIDON altimeter data. Science
268:708710.
Nerem, R. S., D. P. Chambers, B. D. Hamlington, R. R. Leben, G. T. Mitchum, T. Phillips, and J. K.
Willis. 2010. Observations of recent sea level change, IPCC workshop on sea level rise & ice
sheet instabilities, Malaysia.
Smith, T. M., R. W. Reynolds, R. E. Livezey, and D. C. Stokes. 1996. Reconstruction of historical
sea surface temperatures using empirical orthogonal functions. J. Climate 9:14031420.
Smith, T. M., R. W. Reynolds, T. C. Peterson, and J. Lawrimore. 2008. Improvements to NOAAs
historical merger land-ocean surface temperature analysis (18802006). Journal of Climate
21:22832296.
Tai, C. K. 1989. Accuracy assessment of widely used orbit error approximations in satellite altimetry.
J. Atmos. Oceanic Technol. 6:147150.
Wolter, K., and M. S. Timlin. 1998. Measuring the strength of ENSO eventshow does 1997/98
rank? Weather 53:315324.
Wolter, K. 2010. NOAA Earth System Research Laboratory, Mulitvariate ENSO Index. http://www.
esrl.noaa.gov/psd/people/klaus.wolter/MEI/