Advertisement

Climate Dynamics

, Volume 45, Issue 11–12, pp 3077–3090 | Cite as

A Tripole Index for the Interdecadal Pacific Oscillation

  • Benjamin J. Henley
  • Joelle Gergis
  • David J. Karoly
  • Scott Power
  • John Kennedy
  • Chris K. Folland
Article

Abstract

A new index is developed for the Interdecadal Pacific Oscillation, termed the IPO Tripole Index (TPI). The IPO is associated with a distinct ‘tripole’ pattern of sea surface temperature anomalies (SSTA), with three large centres of action and variations on decadal timescales, evident in the second principal component (PC) of low-pass filtered global SST. The new index is based on the difference between the SSTA averaged over the central equatorial Pacific and the average of the SSTA in the Northwest and Southwest Pacific. The TPI is an easily calculated, non-PC-based index for tracking decadal SST variability associated with the IPO. The TPI time series bears a close resemblance to previously published PC-based indices and has the advantages of being simpler to compute and more consistent with indices used to track the El Niño–Southern Oscillation (ENSO), such as Niño 3.4. The TPI also provides a simple metric in physical units of °C for evaluating decadal and interdecadal variability of SST fields in a straightforward manner, and can be used to evaluate the skill of dynamical decadal prediction systems. Composites of SST and mean sea level pressure anomalies reveal that the IPO has maintained a broadly stable structure across the seven most recent positive and negative epochs that occurred during 1870–2013. The TPI is shown to be a robust and stable representation of the IPO phenomenon in instrumental records, with relatively more variance in decadal than shorter timescales compared to Niño 3.4, due to the explicit inclusion of off-equatorial SST variability associated with the IPO.

Keywords

Interdecadal Pacific Oscillation Pacific Decadal Oscillation Pacific Decadal Variability IPO PDO PDV TPI 

1 Introduction

The Interdecadal Pacific Oscillation (IPO, Power et al. 1999; Folland et al. 2002; Arblaster et al. 2002; Christensen et al. 2013; Hartmann et al. 2013; Kirtman et al. 2013) and the Pacific Decadal Oscillation (PDO, Mantua et al. 1997; Christensen et al. 2013) are closely related modes of decadal to interdecadal climate variability in the Pacific Ocean (Power et al. 1999; Folland et al. 2002; Parker et al. 2007; Christensen et al. 2013). The PDO can be regarded as the North Pacific node of the Pacific-wide IPO (Power et al. 1999; Folland et al. 2002). The IPO was originally defined by Power et al. (1999) using the second principal component (PC) of low frequency global SST of Folland et al. (1999) over the 1911–1995 period. Folland et al. (1999) also used global night marine air temperature (NMAT) using the same analysis methods to confirm that the NMAT IPO was broadly similar. A more recent analysis of the IPO by Parker et al. (2007) has been widely used. This differs only slightly from the IPO of Folland et al. (1999), but was based on the more recent HadSST2 data set (Rayner et al. 2006) and used the longer period 1891–2005 to define the IPO.

The PDO and IPO have statistical links with decadal changes in the strength of El Niño–Southern Oscillation (ENSO) teleconnections globally (Wang et al. 2014), with significant regional impacts in North America (Gershunov and Barnett 1998; McCabe and Dettinger 1999; Gershunov and Cayan 2003), East Asia (Wang et al. 2008, 2014), South America (Andreoli and Kayano 2005) and Australia (Power et al. 1998, 1999). In Australia the IPO influences rainfall (Power et al. 1998), streamflow (Power et al. 1998; Verdon et al. 2004), ENSO event frequency (Kiem et al. 2003), flood risk in eastern Australia (Micevski et al. 2006) and both surface temperature and agricultural production (Power et al. 1998).

In the IPO negative phase, La Niña intensity is more strongly related to rainfall extremes in Australia than during IPO positive phases (Power et al. 1998; Cai and van Rensch 2012; King et al. 2013). Statistical links with the IPO also extend to catchment antecedent wetness (Pui et al. 2011) and water supply drought risk (Henley et al. 2011; Henley et al. 2013). The IPO also influences the South Pacific Convergence Zone (SPCZ, Folland et al. 2002; Salinger et al. 2014; Salinger et al. 2001). Folland et al. (2002) showed that the IPO and ENSO have fairly similar (but independent) influences on the SPCZ using a sensitive two-way analysis of variance methodology, with the location of the SPCZ convergence maximum shifting southwest in negative IPO and La Niña episodes. The IPO also modulates New Zealand climate, including ENSO influences (Salinger et al. 2001). PDO influences on Northern Hemisphere climate have been studied widely, including drought in China (Qian and Zhou 2014), variability in North America (Hamlet and Lettenmaier 2007; Stewart et al. 2005), and ecological impacts such as salmon production (Mantua et al. 1997).

The phases of the IPO are thought to influence global surface air temperature (GSAT). The accelerated warming of 1976–2000 and the ‘hiatus’ periods of 1945–1975 and since 2000 coincide reasonably well with positive (warm) and negative (cool) phases in the IPO (Meehl et al. 2013; England et al. 2014). Positive and negative phases of the Atlantic Multidecadal Oscillation (AMO) also likely influence GSAT (Folland et al. 2013). Deep ocean heat content is influenced by the strength of trade winds in the equatorial central Pacific, which appear to be related to the IPO (England et al. 2014). The influence of Indian (Luo et al. 2012) and Atlantic basin (McGregor et al. 2014) SSTs on global circulation changes and GSAT have also been explored.

A number of studies have investigated the links between the IPO and ENSO variability. Newman et al. (2003) proposed that the North Pacific Ocean integrates the effects of ENSO, and that the PDO can be explained by stochastic, low frequency variations in ENSO. Schneider and Cornuelle (2005) identified the relative influences of a range of physical climate drivers on the PDO and their frequency dependence, proposing that the PDO is a response to sea level pressure fluctuations of the Aleutian low, ENSO variability, and ocean circulation anomalies in the Kuroshio-Oyashio Extension region.

Power and Colman (2006) showed that some IPO variability can arise from random fluctuations in ENSO activity from decade to decade. Power and Colman (2006) demonstrated that the integration of stochastic variations in the effects of ENSO occurred in the South Pacific in their climate model. They also showed that the ocean integrates ENSO-driven signals in both surface heat-flux and wind-stress. Power and Colman (2006) argued that the structure of decadal variability (in SST and upper ocean heat content) is meridionally broader than the structure of interannual variability, and that this provides evidence that some of the IPO variability in the off-equatorial upper ocean is a low-frequency response to multi-year and decadal changes in preceding ENSO activity.

Meehl and Hu (2006) have also described plausible mechanisms for the IPO. These involve decadally varying wind-forced ocean Rossby waves in both hemispheres that modulate the strength of a pair of tropical Pacific oceanic overturning cells symmetrical about the equator, influencing near equatorial upwelling and SST in the equatorial Pacific. For further discussion, see the review of SST modes and variability provided by Deser et al. (2010). Given that decadal to interdecadal Pacific variability and its associated teleconnections have societal impacts irrespective of the causal mechanisms of the interactions between the IPO and ENSO, an index for the IPO is considered a worthwhile pursuit.

Previous studies have calculated time series and spatial patterns of the IPO from principal component analysis (PCA) of gridded datasets of sea surface temperature (SST) and night marine air temperature (NMAT). Details of previously published definitions of the IPO and PDO are summarised in Table 1. These studies used a variety of datasets, data pre-treatments, base periods and filters, as well as temporal and spatial resolutions. PCA itself is a collection of related methods, with various algorithms and variants (e.g. rotated EOFs) (Jolliffe 2002). As the underlying dataset or the pre-processing of the dataset can vary, so can the order of the principal components, which can switch places from PC2 to PC3 (Cai and Whetton 2001; Folland et al. 2002; Parker et al. 2007), the polarity of the PCs can reverse, or the PCs can be confounded if they explain similar amounts of variance. Similarly, the order and separation of the internal PCs in climate models are not guaranteed to align with observed spatio-temporal structures, hindering direct comparisons of PC timeseries and loading patterns. Consequently, estimating uncertainty in EOF patterns and loadings is a challenging exercise, particularly with noisy and incomplete data. In contrast, indices calculated to track other modes of variability such as ENSO, the Indian Ocean Dipole (IOD), Southern Annular Mode (SAM) and the North Atlantic Oscillation (NAO) are simpler, and are based on box-averaged SST (e.g. Niño 3.4, TNI, DMI) or differences in station-based sea level pressure between two points on the globe (e.g. SOI, SAM index, NAO index). Such indices allow for simple comparison between observational and model simulated datasets.
Table 1

Summary of definitions, datasets and analysis methods of IPO/PDO indices

References

Primary base dataset(s)

Spatial resolution

Domain

Time resolution

Period of analysis

Filtering

Index definition

Mantua et al. (1997)

HSSTD/OISST

5° × 5°/1° × 1°

Poleward of 20°N

Monthly

1900–1993

None

PC1 of SST over Pacific domain

Zhang et al. (1997)

COADS/HSSTD

2° × 2°, aggregated to 4° × 6°

Poleward of 20°N

Monthly

1900–1993

6-year low pass

PC1 of low pass filtered SST over Pacific domain

Folland et al. (1999)

MOHSST6C/NMAT (Parker et al. 1995)

Equal areas, 10° × 12° at equator

Global

Seasonal

1861–1996

13-year low pass

Projection of unfiltered MOHSST6C onto low-pass filtered EOF2 for 1861–1996, then filtered to obtain low frequency version

Power et al. (1999)

HadSST/NMAT

As for Folland et al. (1999)

Global and North Pacific

Seasonal and annual

1856–1998

13-year low pass

PC1

Cai and Whetton (2001)

GISST3

5° × 5°

Global

Annual

1880–1997

11-year low pass Butterworth

PC2

Folland et al. (2002)

As for Folland et al. (1999)

As for Folland et al. (1999)

Global

Annual

1911–1995

13-year low pass

As for Folland et al. (1999)

Parker et al. (2007)

HadCRUT3 (Rayner et al. 2006)

Equal areas, 10° × 12° at equator

Global

Seasonal

1850–2006

11-year low pass Chebyshev

Projection of unfiltered HadSST2 onto low-pass filtered EOF2 for 1891–2001, then filtered to obtain low frequency version

Henley et al. (This study)

HadISST1, HadISST2.1, ERSSTv3b, HadSST3.1

1° × 1°

Global

Monthly

1870–2007

13-year low pass Chebyshev

(Given in Sect. 6)

It is recognised that PCA is necessary to help identify patterns of variability. However, indices with physical units, such as sea surface temperature anomalies in °C, provide more physical context than the standardised units that are output from PCA. Accordingly, this study aims to develop an alternative SST-based index for the IPO that is:
  1. 1.

    definitive, reproducible and simpler to update than a PCA-based index;

     
  2. 2.

    consistent with methods used to compute indices of other modes of climate variability;

     
  3. 3.

    in native units of °C; and

     
  4. 4.

    a robust and computationally stable representation of the phenomenon.

     

This study is organised as follows: Sect. 2 describes the data used, Sect. 3 describes the evaluation of filters applied to identify low-frequency SST variability, and Sect. 4 describes the development of the new TPI. Section 5 compares the variance within and cross-correlation between the mean SSTA in the chosen TPI regions. Section 6 summarises the method used to calculate the index. Section 7 describes the comparison of IPO indices and an uncertainty analysis of the TPI. Section 8 uses the new index to investigate the evolution of the IPO in SST and SLP data in epochs of phases of the IPO during the 1870–2007 period. The study concludes in Sect. 9 with a discussion of how the new index can be applied to climate research.

2 Data

The primary SST data used in this study are HadISST2.1.0.0, which include 10 realisations of globally complete SST at a resolution of 1° × 1° at a monthly resolution covering 1850–2007 (Kennedy et al. 2015, in preparation; Rayner et al. 2015, in preparation). HadISST.2.1.0.0 is based on in situ and satellite data reconstructed using a two-step process. Step one is an iterative PCA-based reconstruction (Ilin and Kaplan 2009) that captures large-scale features of the SST fields. Step two is a local optimal interpolation that adds small scale detail in well-observed regions (Karspeck et al. 2011). Uncertainty is expressed as an ensemble with representative samples drawn from the bias adjustment scheme and both steps of the interpolation process. SST fields were blended with sea ice concentrations to produce a globally complete data set.

Results from the HadISST.2.1.0.0 analysis are compared to the (a) HadISST1 dataset (1° × 1°, monthly, 1870–2014, globally complete, Rayner et al. 2003), (b) the HadSST.3.1.0.0 dataset (5° × 5°, monthly, 1850–2014, uninterpolated, Kennedy et al. (2011a, b) and (c) the extended reconstructed SST analysis (ERSSTv3b) dataset (2° × 2° grid, monthly, 1854–2013, globally complete, Smith et al. (2008)) from the United States National Oceanic and Atmospheric Administration (NOAA). Multiple datasets are used here to investigate the sensitivity of the results to different bias correction and uncertainty estimation methods. Mean sea level pressure (MSLP) data used in this study are from version 2 of the Twentieth Century Reanalysis project (20CR) (Compo et al. 2011). This is provided on a 2° × 2° grid with monthly resolution and covers the period 1871–2013. The previously published IPO index of Parker et al. (2007) is available from the Climate of the 20th Century (C20C) project site (www.iges.org/c20c/IPO_v2.doc), and the PDO index of Mantua and Hare (2002) is available from the University of Washington, USA (www.jisao.washington.edu/pdo).

3 Low pass filter selection

Previous studies (summarised in Table 1) have used a variety of low-pass filters to remove sub-decadal variability and noise from SST data prior to identifying the decadal-scale component of variability in the IPO signal. Filters are sometimes chosen somewhat subjectively, and without full specification of the filter design parameters or filter coefficients. To objectively identify an appropriate low-pass filter for use in this study, we perform principal component analyses with a range of Chebyshev filters and select the filter that results in the highest proportion of explained variance in the IPO PC. This is performed by varying the filter cutoff period between 11 and 21 years and the filter order between 4 and 7 filter coefficients. The IPO PC is identified as the PC with the maximum correlation with the IPO index of Parker et al. (2007). In the UKMO IPO index method, the IPO pattern is first obtained by performing PCA on low-pass filtered SST data. This pattern is then projected onto unfiltered spatially varying SST data to obtain an unfiltered IPO index (IPOUKMO). The resulting index can then be filtered to obtain the low frequency component of the IPO signal. In our PCA, the IPO emerged in PC2 of low-pass filtered global SST in HadISST1, the 10 HadISST2.1 realisations and the ERSSTv3b dataset; the global warming signal emerged as PC1. A local maxima in the variance explained by PC2 (not shown) was evident at a cutoff period of 13 years and a filter order of 6, so this filter is used to extract the low frequency portion of the IPO in both the development of the TPI and in computing the filtered IPOUKMO index.

4 Development of an IPO Tripole Index

The approach we use in the development of an alternative SST-based index for the IPO is to:
  1. 1.

    perform PCA on ten realisations of low-pass filtered global HadISST2.1 using the filter identified in Sect. 3;

     
  2. 2.

    calculate a composite of the ten PC2 patterns and correlations between PC2 time series and SST;

     
  3. 3.

    identify the locations of consistently high SST anomaly amplitude from the composite in (2);

     
  4. 4.

    define geographic regions, by visual inspection of SST anomalies, that best align with these centres of high amplitude; analyse the time series, variance and correlations between SST anomalies in these regions; and

     
  5. 5.

    develop an index as a linear combination of the mean SST in these regions and compare the index to previous indices and the IPOUKMO timeseries calculated from the same dataset.

     
Figure 1 shows a composite of the correlation pattern between the filtered IPOUKMO (PC2) time series and SST data of 10 realisations of HadISST2.1. Four broad but distinct regions of high modal variability are evident in (1) the central–west of the North Pacific, (2) the eastern equatorial Pacific, (3) the central–west South Pacific and (4) the Southern Ocean midway between New Zealand and southern South America. We only use the first three regions because region (4) has poorer observational data in many earlier decades and is not very well observed until the satellite era around 1980. Region (4) is nevertheless an expected part of the IPO SST pattern because during El Niño events, SST tends to be warm here due to the tendency for anomalous anticyclonic conditions (Folland and Salinger 1995; Folland et al. 2003, Fig. 4). The sign of the composite (Fig. 1) in the equatorial region is opposite to that in the mid-latitude regions, agreeing well with the IPO spatial loading patterns identified by Folland et al. (2002), Parker et al. (2007) and Meehl et al. (2010). The IPO pattern is somewhat similar to that of ENSO, except that the Fig. 1 composite is meridionally broader, as noted by Power and Colman (2006), with a larger off-equatorial modal amplitude, particularly in the North Pacific, and lower amplitude in the far east tropical Pacific. The three geographic regions used to calculate the new index are shown in Fig. 1, with their spatial coordinates given in the caption. Correlation maps for the unfiltered IPOUKMO and both filtered and unfiltered TPI show similar results (not shown).
Fig. 1

Correlation between SST and IPOUKMO (unfiltered), composite of 10 realisations of HadISST2.1. Black boxes indicate TPI regions: region 1 25°N–45°N, 140°E–145°W; region 2 10°S–10°N, 170°E–90°W; region 3 50°S–15°S, 150°E–160°W

5 TPI region SST cross-correlation and variance comparison

Examination of the temporal variability and co-evolution of the average SSTA in each of the TPI regions provides insight into the temporal evolution of the IPO and the relative contribution of the three regions used to represent the IPO in the TPI. This is assessed by comparing the correlation between, and the variance within, the SSTA in each of the TPI regions.

Cross-correlation coefficients (ρMon and ρDec) between the average SSTA (with the seasonal cycle removed) in the TPI regions for both the unfiltered monthly series and low-pass filtered series are shown in Table 2. SST anomalies in Region 1 (in the central–west North Pacific) and SST anomalies in Region 3 (in the central–west South Pacific) are positively correlated at both monthly and decadal timescales (r = 0.28–0.41 monthly, r = 0.40–0.68 decadal). SST anomalies in Regions 1 and 3 are negatively correlated with SST anomalies in Region 2 (in the eastern equatorial Pacific) on both timescales ranging from r = −0.28 to −0.46 on monthly, and r = −0.05 to −0.61 on decadal timescales. Note that the data was detrended (to remove the global warming signal from the individual timeseries) and filter end effects were removed prior to calculating correlations. Inspection of the anomaly time series in Fig. 2 (for one realisation of HadISST2.1, others similar) confirms this tripole behaviour, with Region 2 frequently out of phase with the other two regions. The filtered and unfiltered time series also reveal significant interdecadal variability in SST, superimposed on a longer-term warming trend.
Table 2

Cross correlations between TPI region monthly SST anomalies for (unfiltered) monthly (ρ Mon) and low-pass filtered (ρ Dec ) data

 

HadISST1

HadISST2.1

ERSSTv3b

HadSST3.1

ρMon

ρDec

ρMon

ρDec

ρMon

ρDec

ρMon

ρDec

Region 1–2

−0.46

−0.59

−0.34

−0.34

−0.32

−0.19

−0.28

−0.35

Region 1–3

0.28

0.40

0.36

0.57

0.41

0.67

0.31

0.68

Region 2–3

−0.54

−0.61

−0.46

−0.35

−0.36

−0.05

−0.32

−0.44

Fig. 2

SST anomaly time series for TPI regions in HadISST2.1 (one realisation)

Table 3 presents a comparison of the mean standard deviation in 20-year moving windows of monthly SST indices. This window was used to reduce the influence of interdecadal trends in SST in the variance comparison, a simple high pass filter. The results are reported as the high-pass filtered monthly standard deviation (σMon), low-pass filtered decadal standard deviation (σDec) and the ratio of the two (σDecMon). SST variability in Region 3 (in the Southwest Pacific) has a lower monthly variance (0.26–0.38) than does SST variability in the other two regions (0.35–0.66), but a similar variance at decadal timescales (0.11–0.18). This results in a higher relative proportion of decadal to monthly variance in Region 3. This in turn leads to a higher proportion of decadal to monthly variance in both the TPI and IPOUKMO indices (0.37–0.41), than there is in the Niño 3.4 index (0.24–0.27). These results are qualitatively consistent across the three SST datasets and for moving windows of varying length (10–40 years, not shown). Autocorrelograms of the seasonal TPI and Niño 3.4 indices given in Fig. 3 indicate relatively higher autocorrelation in the TPI, with statistically significant (compared to white noise 95 % confidence intervals of ±1.96/√, where n is the sample length) autocorrelation out to a lag of 5 seasons for the TPI and 3 seasons for Niño 3.4.
Table 3

SST variance comparison

 

HadISST1

HadISST2.1

ERSSTv3b

HadSST3.1

σMon

σDec

σDecMon

σMon

σDec

σDecMon

σMon

σDec

σDecMon

σMon

σDec

σDecMon

Region 1

0.35

0.13

0.37

0.40

0.15

0.38

0.38

0.15

0.38

0.48

0.18

0.37

Region 2

0.53

0.17

0.31

0.58

0.16

0.28

0.49

0.15

0.30

0.66

0.14

0.21

Region 3

0.26

0.12

0.46

0.28

0.11

0.40

0.28

0.13

0.48

0.38

0.15

0.39

TPI

0.71

0.25

0.36

0.76

0.25

0.32

0.67

0.23

0.35

0.92

0.26

0.28

IPOUKMO

0.96

0.39

0.41

0.95

0.39

0.41

0.95

0.41

0.43

0.89

0.44

0.49

Niño 3.4

0.75

0.21

0.27

0.87

0.21

0.24

0.78

0.21

0.27

0.94

0.21

0.22

Mean of the running (20 year) standard deviation of monthly SST indices for three SST datasets. Results are reported as the (high pass filtered) monthly standard deviation (σMon), low-pass filtered decadal standard deviation (σDec) and the ratio σDecMon. Note that the HadISST2.1 results are the mean over the 10 realisations. Units are °C except for IPOUKMO, which has standardized units

Fig. 3

Autocorrelograms of TPI and Niño 3.4 index (seasonal resolution) for one realisation of HadISST2.1, white noise 95 % confidence intervals shown at \(\pm 1.96/\surd n\), where n is the sample length

6 The Tripole Index (TPI) computational method

The following method is used to compute the Tripole Index, based on monthly global SST data:
  1. 1.

    Subtract the monthly climatology from each SST grid cell to remove the seasonal cycle and compute the monthly mean SST anomalies (SSTAi) in each of the three TPI regions using a chosen base period (1971–2000 used here).

     
  2. 2.
    Compute the unfiltered TPI as:
    $$TPI = SSTA_{2} - \frac{{SSTA_{1} + SSTA_{3} }}{2}$$
    (1)
     
  3. 3.

    Apply a 13-year Chebyshev low-pass filter to obtain the filtered version of the index (filtered TPI).

     

7 IPO and TPI time series and uncertainty estimation

Figure 4a compares the IPO index of (Parker et al. 2007) and the 13-year Chebyshev low-pass filtered (North Pacific) PDO index of (Mantua et al. 1997), showing strong agreement between the two (standardized) indices (r = 0.83). Figure 4b shows the timeseries of filtered IPOUKMO, obtained by first projecting the unfiltered SST field onto the spatial pattern of PC2 from low-pass filtered SST data (to obtain IPOUKMO), then low-pass filtering the resulting time series. The filtered IPOUKMO series are shown for the three complete datasets described in Sect. 2. Figure 4c shows the filtered TPI calculated from all four SST datasets. The correlations between the TPI and IPOUKMO with the different datasets range between 0.92 and 0.97; for the filtered TPI and IPOUKMO the correlations range between 0.85 and 0.97. The uncertainty in the filtered TPI is presented for the HadSST3.1 dataset, taking into account the following sources of uncertainty:
Fig. 4

Low pass filtered IPO/PDO index time series. a Previously published IPO and PDO indices b IPOUKMO (HadISST1, 10 realisations of HadISST2.1 and ERSSTv3b) and c TPI for the three datasets. Dotted vertical lines indicate approximate timing of IPO phase shifts, based on IPO index of Parker et al. (2007)

  1. 1.

    Uncertainty in the bias adjustments in HadSST3.1

     
  2. 2.

    Measurement and within (HadSST3.1) gridbox sampling uncertainty

     
  3. 3.

    Undersampling of the TPI region averages due to some HadSST3.1 gridboxes containing no data

     
Uncertainty in the bias adjustments is assessed using the HadSST.3.1.0.0 ensemble. The TPI is calculated for each of the 100 ensemble members. The spread arising from bias uncertainty was small because the errors associated with bias adjustments are correlated in space. Taking the difference of two regional averages therefore effectively removes any systematic error. Measurement and within-gridbox sampling uncertainty are calculated using the HadSST.3.1.0.0 error covariance matrices. The TPI for a particular month can be expressed as the weighted sum of populated gridboxes.
$${\text{TPI }} = {\text{ wa}}^{\text{T}}$$
(2)
where a is a vector containing the gridbox anomalies and w is a vector of weights. Using the regular error propagation formula, the uncertainty in the TPI can be written as:
$$\upsigma^{ 2}_{\text{TPI}} = {\text{ wCw}}^{\text{T}}$$
(3)
where C is the error covariance matrix for a given month.
Uncertainty due to under-sampling for each of the regional averages is estimated using the formula:
$$\upsigma^{ 2}_{\text{undersampling}} = \left( { 1/{\text{m }}{-}{ 1}/{\text{n}}} \right)\upsigma^{ 2} \left( { 1- {\text{r}}} \right)$$
(4)

In which n is the maximum number of gridboxes in one of the regions, m (≤n) is the number of gridboxes sampled in a particular month, σ is the standard deviation of the true gridbox SST and r is the average correlation between two points within the region. Gridbox values of σ are estimated as described in Rayner et al. (2006). An average standard deviation is calculated for each region. For region 1, n was equal to 60, for region 2, n was 80 and for region 3, n was 70. The undersampling uncertainties for the three regions were combined under the assumption that they were uncorrelated. At most times, the undersampling uncertainty was smaller than the measurement and within gridbox sampling uncertainty even when r was assumed to be zero, the most conservative assumption. Consequently, r was set to zero.

An ensemble approach was taken to estimate the overall uncertainty in the TPI from HadSST3.1. Representative samples of measurement and sampling error, and of undersampling error were added to each of the one hundred HadSST.3.1.0.0 ensemble members to give an ensemble of estimates of pre-smoothing monthly average TPI. The undersampling uncertainty is assumed to be uncorrelated from month to month (i.e white noise). The measurement and within-gridbox sampling uncertainty is assumed to be an AR(1) process with a lag-1 correlation of 0.77 (Morice et al. 2012).

Table 4 lists the proportion of variance explained in global and Pacific-only SST data for the IPOUKMO and TPI indices, for both unfiltered and filtered SST data. In all cases, the two indices explain similar proportions of variance across the HadISST1, HadISST2.1 and ERSSTv3b datasets, showing that the TPI is a good complement or alternative to other IPO indices.
Table 4

SST Variance explained by IPO and TPI indices for Global and Pacific domains

Dataset

Proportion of variance explained of global SSTs

Proportion of variance explained of Pacific SSTs (120°E–70°W)

IPOUKMO

TPI

IPOUKMO (filt)

TPI (filt)

IPOUKMO

TPI

IPOUKMO (filt)

TPI (filt)

HadISST1

0.10

0.10

0.11

0.12

0.18

0.18

0.20

0.20

HadISST2.1

0.06

0.06

0.10

0.09

0.11

0.11

0.16

0.15

ERSSTv3b

0.09

0.08

0.09

0.08

0.16

0.15

0.15

0.14

8 Multi-decadal SST and MSLP evolution

Although the correlation pattern of PC2 in Fig. 1 provides a representation of the spatial extent of the SST pattern associated with the IPO, the temporal evolution of the SST anomaly patterns through the decadal to multi-decadal epochs of the IPO provide insight into the stability and symmetry of the IPO pattern through time. Deser et al. (2004) found coherent patterns of interdecadal variability by investigating differences between SLP, SST, precipitation, cloudiness and surface air temperature in IPO/PDO epochs in the North Pacific region for 1900–1997.

Figure 5 shows the temporal composite (mean) Pacific SST anomalies (relative to 1890–2007) for HadISST2.1 and ERSSTv3b during the seven most recent IPO epochs as determined from the HadISST2.1 TPI index: 1870–1895, 1896–1910, 1911–1923, 1924–1944, 1945–1976, 1977–1999, and 2000 to the most recent data available (2007 for HadISST2.1, and 2013 for ERSSTv3b). The SST composites for the IPO epochs in both HadISST2.1 and ERSST3vb datasets in Fig. 5 display strong similarities with the spatial mode structure of the IPO in the correlation pattern of Fig. 1. Overall, stronger warming is most apparent in the period since 1977, evident in both negative and positive IPO epochs. Figure 6 shows the composite of all negative and positive IPO epochs in the observational period. Once again, the tripole mode pattern of the IPO is evident. Interestingly, SST during the negative IPO phase exhibits greater symmetry between the north and south extratropical Pacific regions than during the positive phase, with relatively stronger north Pacific SST anomalies evident in IPO positive phases.
Fig. 5

Pacific SST composites during the most recent seven IPO epochs, left column composite of 10 realisations from HadISST2.1 (most recent epoch 1999–2007), right column ERSSTv3b (most recent epoch 1999–2013). Significance at the 5 % level is indicated by stippling, based on Monte Carlo block sampling

Fig. 6

Composites of SST anomalies during IPO positive and negative phases for HadISST2.1 (left) for 1870–2007 and ERSSTv3b (right) for 1870–2013

Corresponding mean sea level pressure anomaly composites for the IPO epochs are shown in Fig. 7. The four most recent IPO epochs indicate a tendency for relatively higher mean pressure in the central and eastern Pacific during IPO negative epochs and lower pressure in the western Pacific, with the opposite influence apparent during IPO positive epochs.
Fig. 7

Pacific MSLP anomaly composites from 20CR2 from the seven most recent IPO epochs, units are hPa

9 Summary and discussion

This study introduced a Tripole Index (TPI) for the IPO, based on the SST anomalies in three large geographic regions of the Pacific (Fig. 1). The TPI is shown to be complementary to PCA-based indices such as the IPOUKMO. As with PCA-based indices, the TPI is computed without making explicit assumptions about the temporal character of global warming. It also provides a simple metric for IPO evolution that is expressed in physical units of °C and allows for direct comparison between observational and model simulated datasets. Both the unfiltered TPI and IPOUKMO indices show higher variance on decadal timescales (relative to shorter timescales) than the Niño 3.4 index, showing a distinction between IPO and ENSO indices.

Considering the brevity of the instrumental record in relation to the timescale of the phase shifts of the IPO, skilful multi-century palaeoclimate reconstructions of the IPO are needed. The TPI is an IPO index that does not require a near-global SST record, a property that could be exploited given that palaeoclimate archives have significantly sparser spatial coverage than instrumental datasets. Unlike other IPO indices, the TPI is comprised of three separable regional average SSTA indices. The TPI and the individual regional average SSTAs that contribute to the TPI are therefore useful target variables for regional palaeoclimate reconstructions. In particular, South Pacific coral-based SST proxies, which were used in the IPO reconstruction by Linsley et al. (2008), are likely to be skilful proxies for SSTA in TPI region 3. Similarly, precipitation/temperature proxies from Northern Hemisphere tree ring chronologies used in PDO reconstruction studies such as D’Arrigo and Wilson (2006) and Biondi et al. (2001), are likely to be skilful proxies for SSTA in TPI regions 1 and 2.

An analysis of SST and MSLP across IPO epochs was presented for the periods 1870–2007 (HadISST2.1) and 1870–2013 (ERSSTv3b). Indices that describe climate modes using a fixed method (e.g. Nino3.4, SOI, TPI, IPOUKMO) assume a degree of consistency in the spatial structure of the mode, however, variability in spatial structures is likely with any natural phenomenon (Gallant et al. 2013). A consistent SST pattern is evident in each IPO epoch presented in Fig. 5, and provides evidence for a stable IPO signature in SST since 1870. This stability, along with the use of large regional domains in the TPI computation, ensures that the TPI is robust in tracking the phases of the IPO.

The IPO negative phase in 1945–1976 and the IPO positive phase in 1977–1999 (Fig. 5) exhibit particularly strong resemblance to the IPOUKMO correlation pattern in Fig. 1. The positive phase in 1924–1944 is more pronounced in HadISST2.1 than in ERSSTv3b, with some negative SST anomalies apparent in the equatorial Pacific present in ERSSTv3b but not in HadISST2.1. The IPO timeseries (IPOUKMO and TPI) indicate a weak negative or neutral IPO for the period 1911–1923. However, the epoch analysis shows a spatial structure that resembles a positive phase. This appears to be due to the cool North Pacific SST “horseshoe” occupying a larger area in a more northward location than the encompassed warm SST pool.

Interestingly, this epoch analysis shows very strong negative IPO phase in 1870–1895, similar in magnitude to 1945–1976 phase. Little has been published about early IPO epochs, however, wet conditions in precipitation reconstructions from SE Australia, and the well-known ‘Federation Drought’ in Australia are consistent with a shift from negative to positive IPO phase in around 1895 (Gergis et al. 2012). The most recent period since 1999 appears to be a strong negative IPO phase in both datasets, with stronger warm SSTs in the west Pacific in ERSSTv3b for 1999–2013 in comparison to HadISST2.1 for the slightly shorter period of 1999–2007. Given the ERSSTv3b data extends until 2013, this gives greater confidence to the presence of an IPO negative period since around 1999, likely to be related to the rebounding of the meridional overturning circulation in the tropical Pacific Ocean (McPhaden and Zhang 2004). The multiple-epoch composites in Fig. 6 provide a long-term view of the average structure of all IPO positive and negative phases (in the available data), with an apparent asymmetry in the SST structure between positive and negative IPO phases.

Given that the TPI is computed using a simple and computationally fixed method that is reported as native temperature anomalies in °C, the index could simplify the evaluation of IPO SST variability in climate model simulations. Evaluating a model’s internal principal components in SST and comparing modelled and observed PC time series does not necessarily imply that the modelled spatial structures are realistic. In contrast, the TPI simply measures a model’s ability to simulate the IPO’s spatial and temporal variability, since it is a metric that is explicitly aligned with the spatial pattern present in observed data. Model bias can also be assessed in a physically meaningful manner using the TPI. This is not generally possible with normalised timeseries that result from PCA, without a possible additional step of projecting a physical field onto the PC pattern and performing pattern correlation analysis to investigate the similarity between spatial fields. It is therefore proposed that the TPI is useful for: (1) assessing climate models’ ability to reproduce natural Pacific Decadal Variability, (2) examining the predictive skill in dynamical decadal predictions systems and (3) examining the future behaviour of the IPO. While the assessment of Pacific Decadal Variability in climate models using the TPI is beyond the scope of the current study, it will be the focus of forthcoming work.

Notes

Acknowledgments

BH is supported by the Australian Research Council under a Cooperative Research Network (CRN) fellowship. JG is supported by an Australian Research Council DECRA fellowship DE130100668, and DK is supported by the Australian Research Council Centre of Excellence for Climate System Science. SBP is supported by the Australian Department of the Environment through the Australian Climate Change Science Program. JK and CKF are supported by the Joint UK DECC/Defra Met Office Hadley Centre Climate Programme (GA01101). The TPI is available at: www.esrl.noaa.gov/psd/data/climateindices/list/.

Supplementary material

382_2015_2525_MOESM1_ESM.xlsx (698 kb)
Supplementary material 1 (XLSX 698 kb)

References

  1. Andreoli RV, Kayano MT (2005) ENSO-related rainfall anomalies in South America and associated circulation features during warm and cold Pacific decadal oscillation regimes. Int J Climatol 25(15):2017–2030CrossRefGoogle Scholar
  2. Arblaster JM, Meehl GA, Moore AM (2002) Interdecadal modulation of Australian rainfall. Clim Dyn 18(6):519–531CrossRefGoogle Scholar
  3. Biondi F, Gershunov A, Cayan DR (2001) North Pacific decadal climate variability since 1661. J Clim 14:5–10CrossRefGoogle Scholar
  4. Cai W, van Rensch P (2012) The 2011 southeast Queensland extreme summer rainfall: a confirmation of a negative Pacific Decadal Oscillation phase? Geophys Res Lett 39(8):L08702Google Scholar
  5. Cai W, Whetton P (2001) Modes of SST variability and the fluctuation of global mean temperature. Clim Dyn 17(11):889–901CrossRefGoogle Scholar
  6. Christensen JH et al (2013) Climate phenomena and their relevance for future regional climate change. In: Stocker TF et al (eds) Climate change 2013: the physical science basis. Contribution of working group I to the fifth assessment report of the intergovernmental panel on climate change. Cambridge University Press, Cambridge, pp 1217–1308Google Scholar
  7. Compo GP et al (2011) The twentieth century reanalysis project. Q J R Meteorol Soc 137(654):1–28CrossRefGoogle Scholar
  8. D’Arrigo R, Wilson R (2006) On the Asian expression of the PDO. Int J Climatol 26(12):1607–1617CrossRefGoogle Scholar
  9. Deser C, Phillips AS, Hurrell JW (2004) Pacific interdecadal climate variability: linkages between the tropics and the North Pacific during boreal winter since 1900. J Clim 1900:3109–3124CrossRefGoogle Scholar
  10. Deser C et al (2010) Sea surface temperature variability: patterns and mechanisms. Ann Rev Mar Sci 2:115–143CrossRefGoogle Scholar
  11. England MH et al (2014) Recent intensification of wind-driven circulation in the Pacific and the ongoing warming hiatus. Nat Clim Change 4(3):222–227CrossRefGoogle Scholar
  12. Folland CK et al (2002) Relative influences of the Interdecadal Pacific Oscillation and ENSO on the South Pacific Convergence Zone. Geophys Res Lett 29(13)Google Scholar
  13. Folland CK, Salinger MJ (1995) Surface temperature trends and variations in New Zealand and the surrounding ocean, 1871–1993. Int J Climatol 15:1195–1218CrossRefGoogle Scholar
  14. Folland CK et al (1999) Large scale modes of ocean surface temperature since the late nineteenth century. In: Navarra A (ed) Beyond El Nino: decadal and interdecadal climate variability. Springer, New York, pp 73–102CrossRefGoogle Scholar
  15. Folland CK et al (2003) Trends and variations in South Pacific island and ocean surface temperatures. J Clim 16(1995):2859–2874CrossRefGoogle Scholar
  16. Folland CK et al (2013) High predictive skill of global surface temperature a year ahead. Geophys Res Lett 40(4):761–767CrossRefGoogle Scholar
  17. Gallant AJE et al (2013) Nonstationary Australasian teleconnections and implications for paleoclimate reconstructions. J Clim 26(22):8827–8849CrossRefGoogle Scholar
  18. Gergis J et al (2012) On the long-term context of the 1997–2009 “Big Dry” in South-Eastern Australia: insights from a 206-year multi-proxy rainfall reconstruction. Clim Change 111:923–944CrossRefGoogle Scholar
  19. Gershunov A, Barnett TP (1998) Interdecadal modulation of ENSO teleconnections. Bull Am Meteorol Soc 79(12):2715–2725CrossRefGoogle Scholar
  20. Gershunov A, Cayan D (2003) Heavy daily precipitation frequency over the contiguous United States: sources of climatic variability and seasonal predictability. J Clim 16:2752–2765CrossRefGoogle Scholar
  21. Hamlet AF, Lettenmaier DP (2007) Effects of 20th century warming and climate variability on flood risk in the western U.S. Water Resour Res 43(6):W06427Google Scholar
  22. Hartmann DL et al (2013) Observations: atmosphere and surface. In: Stocker TF et al (eds) Climate change 2013: the physical science basis. Contribution of working group I to the fifth assessment report of the intergovernmental panel on climate change. Cambridge University Press, Cambridge, pp 159–254Google Scholar
  23. Henley BJ et al (2011) Climate-informed stochastic hydrological modeling: incorporating decadal-scale variability using paleo data. Water Resour Res 47(11):W11509Google Scholar
  24. Henley BJ, Thyer MA, Kuczera G (2013) Climate driver informed short-term drought risk evaluation. Water Resour Res 49(5):2317–2326Google Scholar
  25. Ilin A, Kaplan A (2009) Bayesian PCA for reconstruction of historical sea surface temperatures. In: Proceedings of the international joint conference on neural networks (IJCNN 2009). Atlanta, USA, pp 1322–1327Google Scholar
  26. Jolliffe IT (2002) Principal component analysis, 2nd edn. Springer, New YorkGoogle Scholar
  27. Karspeck AR, Kaplan A, Sain SR (2011) Bayesian modelling and ensemble reconstruction of mid-scale spatial variability in North Atlantic sea-surface temperatures for 1850–2008. Q J R Meteorol Soc 138:234–248CrossRefGoogle Scholar
  28. Kennedy JJ et al (2011a) Reassessing biases and other uncertainties in sea surface temperature observations measured in situ since 1850: 1. Measurement and sampling uncertainties. J Geophys Res 116(D14):D14103CrossRefGoogle Scholar
  29. Kennedy JJ et al (2011b) Reassessing biases and other uncertainties in sea surface temperature observations measured in situ since 1850: 2. Biases and homogenization. J Geophys Res 116(D14):D14104CrossRefGoogle Scholar
  30. Kennedy JJ et al (2015) The Met Office Hadley Centre sea ice and sea-surface temperature data set, version 2: 2. Sea-surface temperature analysis (in preparation)Google Scholar
  31. Kiem AS, Franks SW, Kuczera G (2003) Multi-decadal variability of flood risk. Geophys Res Lett 30(2):1035CrossRefGoogle Scholar
  32. King AD, Alexander LV, Donat MG (2013) Asymmetry in the response of eastern Australia extreme rainfall to low-frequency Pacific variability. Geophys Res Lett 40(10):2271–2277CrossRefGoogle Scholar
  33. Kirtman B et al (2013) Near-term climate change: projections and predictability. In: Stocker TF et al (eds) Climate change 2013: the physical science basis. Contribution of working group I to the fifth assessment report of the intergovernmental panel on climate change. Cambridge University Press, Cambridge, pp 953–1028Google Scholar
  34. Linsley BK et al (2008) Interdecadal-decadal climate variability from multicoral oxygen isotope records in the South Pacific Convergence Zone region since 1650 AD. Paleoceanography 23(2):PA2219CrossRefGoogle Scholar
  35. Luo J-J, Sasaki W, Masumoto Y (2012) Indian Ocean warming modulates Pacific climate change. Proc Natl Acad Sci USA 109(46):18701–18706CrossRefGoogle Scholar
  36. Mantua NJ, Hare SR (2002) The pacific decadal oscillation. J Oceanogr 58(1):35–44CrossRefGoogle Scholar
  37. Mantua NJ et al (1997) A Pacific interdecadal climate oscillation with impacts on salmon production. Bull Am Meteorol Soc 78(January):1069–1079CrossRefGoogle Scholar
  38. McCabe GJ, Dettinger MD (1999) Decadal variations in the strength of ENSO teleconnections with precipitation in the western United States. Int J Climatol 19(13):1399–1410CrossRefGoogle Scholar
  39. McGregor S et al (2014) Recent Walker circulation strengthening and Pacific cooling amplified by Atlantic warming. Nat Clim Change 4(10):888–892CrossRefGoogle Scholar
  40. McPhaden MJ, Zhang D (2004) Pacific Ocean circulation rebounds. Geophys Res Lett 31(18):L18301CrossRefGoogle Scholar
  41. Meehl G, Hu A (2006) Megadroughts in the Indian monsoon region and southwest North America and a mechanism for associated multidecadal Pacific sea surface temperature anomalies. J Clim 19(9):1605–1623CrossRefGoogle Scholar
  42. Meehl GA, Hu A, Tebaldi C (2010) Decadal prediction in the Pacific region. J Clim 23(11):2959–2973CrossRefGoogle Scholar
  43. Meehl G et al (2013) Externally forced and internally generated decadal climate variability associated with the Interdecadal Pacific Oscillation. J Clim 26(18):7298–7310  CrossRefGoogle Scholar
  44. Micevski T, Franks SW, Kuczera G (2006) Multidecadal variability in coastal eastern Australian flood data. J Hydrol 327(1–2):219–225Google Scholar
  45. Morice CP et al (2012) Quantifying uncertainties in global and regional temperature change using an ensemble of observational estimates: the HadCRUT4 data set. J Geophys Res 117(D8):D08101Google Scholar
  46. Newman M, Compo GP, Alexander MA (2003) ENSO-forced variability of the Pacific decadal oscillation. J Clim 16(23):3853–3857  CrossRefGoogle Scholar
  47. Parker DE, Folland CK, Jackson M (1995) Marine surface temperature: observed variations and data requirements. Clim Change 31(2–4):559–600CrossRefGoogle Scholar
  48. Parker D et al (2007) Decadal to multidecadal variability and the climate change background. J Geophys Res 112(D18):D18115Google Scholar
  49. Power S, Colman R (2006) Multi-year predictability in a coupled general circulation model. Clim Dyn 26(2–3):247–272CrossRefGoogle Scholar
  50. Power S et al (1998) Australian temperature, Australian rainfall and the Southern Oscillation, 1910–1992: coherent variability and recent changes. Aust Meteorol Mag 47(2):85–101Google Scholar
  51. Power S et al (1999) Inter-decadal modulation of the impact of ENSO on Australia. Clim Dyn 15(5):319–324CrossRefGoogle Scholar
  52. Pui A, Lal A, Sharma A (2011) How does the Interdecadal Pacific Oscillation affect design floods in Australia? Water Resour Res 47:W05554CrossRefGoogle Scholar
  53. Qian C, Zhou T (2014) Multidecadal variability of North China aridity and its relationship to PDO during 1900–2010. J Clim 27(3):1210–1222CrossRefGoogle Scholar
  54. Rayner NA et al (2003) Global analyses of sea surface temperature, sea ice, and night marine air temperature since the late nineteenth century. J Geophys Res Atmos 108(D14):4407CrossRefGoogle Scholar
  55. Rayner NA et al (2006) Improved analyses of changes and uncertainties in sea surface temperature measured in situ sice the mid-nineteenth century: the HadSST2 dataset. J Clim 19(3):446–469CrossRefGoogle Scholar
  56. Rayner NA et al (2015) The Met Office Hadley Centre sea ice and sea-surface temperature data set, version 2: 3. The combined analysis (in preparation)Google Scholar
  57. Salinger MJ, Renwick JA, Mullan AB (2001) Interdecadal Pacific Oscillation and South Pacific climate. Int J Climatol 21:1705–1721CrossRefGoogle Scholar
  58. Salinger MJ et al (2014) A new index for variations in the position of the South Pacific convergence zone 1910/11–2011/2012. Clim Dyn 43(3–4):881–892Google Scholar
  59. Schneider N, Cornuelle B (2005) The forcing of the Pacific decadal oscillation. J Clim 18:4355–4373CrossRefGoogle Scholar
  60. Smith TM et al (2008) Improvements to NOAA’s historical merged land-ocean surface temperature analysis (1880–2006). J Clim 21(10):2283–2296CrossRefGoogle Scholar
  61. Stewart I, Cayan D, Dettinger M (2005) Changes toward earlier streamflow timing across western North America. J Clim 18:1136–1155CrossRefGoogle Scholar
  62. Verdon DC et al (2004) Multidecadal variability of rainfall and streamflow: eastern Australia. Water Resour Res 40(10):W10201CrossRefGoogle Scholar
  63. Wang L, Chen W, Huang R (2008) Interdecadal modulation of PDO on the impact of ENSO on the east Asian winter monsoon. Geophys Res Lett 35(20):L20702CrossRefGoogle Scholar
  64. Wang S et al (2014) Combined effects of the Pacific decadal oscillation and El Niño–Southern oscillation on global land dry-wet changes. Sci Rep 4:6651CrossRefGoogle Scholar
  65. Zhang Y, Wallace JM, Battisti DS (1997) ENSO-like interdecadal variability: 1900-93. J Clim 10(5):1004–1020Google Scholar

Copyright information

© Springer-Verlag Berlin Heidelberg 2015

Authors and Affiliations

  • Benjamin J. Henley
    • 1
  • Joelle Gergis
    • 1
  • David J. Karoly
    • 1
  • Scott Power
    • 2
  • John Kennedy
    • 3
  • Chris K. Folland
    • 3
    • 4
  1. 1.School of Earth SciencesUniversity of MelbourneParkvilleAustralia
  2. 2.Centre for Australian Weather and Climate ResearchBureau of MeteorologyMelbourneAustralia
  3. 3.Met Office Hadley CentreExeterUK
  4. 4.Department of Earth SciencesUniversity of GothenburgGothenburgSweden

Personalised recommendations