An Observational Analysis of the Oceanic and Atmospheric Structure of Global-Scale Multi-decadal Variability

Since the beginning of the industrial revolution in the 1850s, scientists like John Tyndall and Svante Arrhenius had already begun to understand the possibility of warming of the atmosphere as a result of the burning of coal. Over the years since that time, measured increases in the concentrations of greenhouse gases and progress in theoretical and model estimates of global warming due to greenhouse gases have provided strong evidence for global warming (IPCC, 2007). In addition to a warming trend evident in the observed global mean surface temperature (ST _gm) record in the most recent 100 years, the data also show periods of decadal cooling trends, signaling multi-decadal climate oscillations in both the Pacific and Atlantic oceans (e.g. Chen et al., 2008; Schubert et al., 2009; Deser et al., 2010).

In the Pacific, analyses of SST have revealed decadal to interdecadal oscillations in the eastern and northern basin that are ENSO-like (e.g., Zhang et al., 1997). Similar patterns have also been obtained through Empirical Orthogonal Function (EOF) analysis of detrended SST north of 20^°N (Mantua et al., 1997) or SST in the Pacific basin (Power et al., 1999). These similar oscillations are referred to as the Pacific Decadal Oscillation/Inter-decadal Pacific Oscillation (PDO/IPO). Although EOF analyses of SST in different regions of the Pacific have led to different regional decadal signals like those identified by (Barlow et al., 2001) and (Nakamura et al., 1997), more and more studies are revealing that the PDO/IPO is likely a low-frequency climate signal associated with ENSO (Meehl et al., 2009; Schubert et al., 2009), and is the most dominant interdecadal signal in the Pacific.

Studies have also found multi-decadal oscillations in the Atlantic (Kushnir, 1994; Delworth and Mann, 2000; Enfield et al., 2001; Knight et al., 2005). These studies analyzed detrended SST to extract the leading mode in either the averaged value or the EOF analysis of SST over the Atlantic north of the Equator. The dominant climate oscillation is defined as the Atlantic Multi-decadal Oscillation (AMO) or Atlantic Multi-decadal Variability (AMV). Analyses of Atlantic SST in both the Southern and Northern Hemisphere have identified an Atlantic Meridional Overturning Circulation (AMOC) (Delworth et al., 1993; Dong and Sutton, 2005; Danabasoglu, 2008). Multi-decadal oscillations of Atlantic SST are hypothesized to be related to the AMOC, the Atlantic lobe of the global thermohaline circulation that may affect the global climate (Zhou et al., 2000; Zhou, 2003; Bentsen et al., 2004; Wang et al., 2004; Cheng et al., 2013).

While the AMO is generally regarded as a signature of the AMOC in the Atlantic and has been simulated in climate models (e.g., Knight et al., 2005), the exact cause of it is not well understood. A climate modeling study by (Park and Latif, 2010) simulated both the PDO and AMO, and argued that the two oscillations are independent. This is different from the view that the PDO and AMO are related. (Zhang and Delworth, 2007) used a hybrid coupled model to show that the AMO can contribute to the PDO. The study of (d'Orgeville and Peltier, 2007) analyzed SST to demonstrate the phase relation between the PDO and AMO. They suggested that the AMO and PDO are signatures of the same oscillation cycle.

Besides the greenhouse effect and climate dynamical processes, aerosols have been recognized to influence long-term changes in SSTs. For example, it is suggested that aerosols from anthropogenic and natural sources (such as volcanic activity and dust) play an important role in regulating Atlantic multi-decadal variability (Stenchikov et al., 2009; Chang et al., 2011; Booth et al., 2012) and Pacific decadal variability (Wang et al., 2012). All of these studies have relied on the modeling approach, which is a challenging task when it comes to prescribing the temporal and spatial distributions of aerosols, as well as representing aerosol-cloud-radiative processes. The viability of such model results remain to be fully evaluated.

Because of the difficulty in identifying and attributing the effect of aerosols on climate change, it is important to understand the structure of multi-decadal variability (MDV), and to explore the extent that climate change is due to internal climate dynamics. Such knowledge can be useful in our attempts to separate climate variability into internal and forced components. The goal of the present study is to systematically analyze the climate signals in available data, including global-scale SST, ocean temperature, and atmospheric variables, and to synthesize the signals to identify consistent features in the different data. Previously, many studies have used the regional EOF method to study Pacific or Atlantic MDV (Liu, 2012). However, with regional EOF analysis it is difficult to compare which MDV is more important on the global scale. In the present study, we apply a global-scale EOF analysis of available data, enabling us to identify the dominant MDV on the global scale and compare it with previously identified MDV in the Pacific and Atlantic basins.

Identifying consistent structure of the climate change modes is important for improving our knowledge of climate change, including causes and maintenance mechanisms for the AMO and PDO. It also provides a reference to evaluate climate model simulations. The results also provide an essential background for assessing modeling results of climate warming and the associated changes in extreme weather events.

The primary datasets used for the present study are the Hadley Centre's Sea Ice and Sea Surface Temperature (HadISST) dataset (Rayner et al., 2003) and the National Oceanic and Atmospheric Administration's (NOAA) Extended Reconstructed Sea Surface Temperature dataset, version 3 (ERSST.v3) (Smith et al., 2008). The two datasets are analysis products of ship measurements, but use different optimal statistical procedures to smooth the data and fill in missing values for the period 1880-2009 at different spatial resolutions: HadISST at 1^°× 1^° and ERSST.v3 at 2^°×2^°.

We also use two sets of ST _gm data that are combinations of land surface air temperature (SAT) observed by meteorological stations and SST observed by ships. One is the National Climatic Data Center's (NCDC) ST _gm, which is a global merged land SAT dataset available from the Global Historical Climate Network (GHCN) and ERSST analysis (Smith and Reynolds, 2005). The other is the National Aeronautics and Space Administration's (NASA) Goddard Institute for Space Studies (GISS) ST _gm dataset (Hansen et al., 1999). In the GISS ST _gm analysis, SAT is based on the GHCN dataset, and SST is based on HadISST1 from 1880 to November 1981 and the NOAA Optimum Interpolation SST dataset, version 2 (NOAA OISST V2; Reynolds et al., 2002) from December 1981 to present.

The atmospheric signal was extracted from the two available reanalysis datasets: the European Centre for Medium-Range Weather Forecasts (ECMWF) Reanalysis (Uppala et al., 2005) for the period 1958-2001 (ERA-40) and the National Centers for Environmental Prediction-National Center for Atmospheric Research (NCEP-NCAR) reanalysis product (Kalnay et al., 1996) for the period 1958-2009. Both datasets have the same spatial resolution at 2.5^°× 2.5^°. The low frequency variability in circulation and thermodynamic fields were derived by an analysis of six basic variables (wind U, V; vertical velocity W; height Z; air temperature T; and specific humidity Q) at eight pressure levels (1000, 925, 850, 700, 600, 500, 400, and 300 hPa).

The upper ocean thermal structure was obtained by analyzing ocean temperature from the World Ocean Database 2009 (WOD09) (Levitus et al., 2005; Boyer et al., 2009; Johnson et al., 2009). The variable has a 1^°× 1^° resolution at 16 height levels below the sea surface (0, 10, 20, 30, 50, 75, 100, 125, 150, 200, 250, 300, 400, 500, 600 and 700 m) for the period 1955-2009.

In the current study, our focus is on the analysis of low-frequency climate oscillations separated from the signal associated with the global warming trend. So, estimating a global warming trend from the time series of ST _gm was necessary. One approach to estimate the global warming trend is to use physical models to simulate the temperature trend changes in response to anthropogenic radiative forcing (e.g., Schlesinger and Ramankutty, 1994; Andronova and Schlesinger, 2000). However, this requires information on the evolution of radiatively-active gases and aerosols of anthropogenic and biomass-burning origins (e.g., Lamarque et al., 2010). The complexity and uncertainty involved in such an approach prevented us from using it. Instead, we simply wanted to remove the centennial global mean trend from the data without attempting to remove the aerosol effects. So, we followed the approach of (Huang and Wu, 2008) and (Wu et al., 2011) to extract an intrinsically determined monotonic signal within the span of the available ST _gm datasets to identify it as a global warming trend. The method consists of empirical mode decomposition (EMD) (Huang et al., 1996, 1998, 1999) and Hilbert spectral analysis, and is an adaptive data analysis method. The method is designed specifically for nonlinear and non-stationary data analysis, which has been used extensively in geophysical research.

Since NCDC ST _gm uses ERSST and GISS ST _gm mainly uses HadISST, we applied Hilbert-Huang transform (HHT) to the two ST _gm datasets to extract their respective global warming trends, which in turn were used to detrend the two corresponding SST datasets. Similarly, the GISS ST _gm warming trend was used to detrend the ERA40 data, and the NCDC ST _gm warming trend was used to detrend the NCEP-NCAR and WOD datasets. Trend signals in each of the data fields were removed by subtracting at each grid the regressed signals corresponding to the global warming trend time series. The detrended datasets were then low-pass filtered using the Lancose filter with a 13-yr cutoff to remove ENSO and 11-yr solar forcing, and to retain multi-decadal variability. The low-pass filtered data were then analyzed by EOF methods to identify the dominant low-frequency oscillations.

First we show in Fig. 1 (top panel) the two sets of monthly mean ST _gm data and the corresponding warming trends during the period 1880-2009. Both data show a similar centennial warming trend of about 0.8 K during the past 130 years, which is attributed to greenhouse warming (IPCC, 2007). In addition to the warming trend, the data also show multi-decadal, decadal, and year-to-year variations. The spatial distribution of the warming trend over the oceans was obtained by regressing the monthly ERSST (HadISST) data at each grid against the centennial warming trend identified in NCDC (GISS) ST _gm, and the results are shown in Fig. 1 (bottom panel).

Figure 1.  The GISS (gray) and NCDC (green) monthly mean ST _gm and the corresponding warming trends during the period 1880-2009 (top panel). The spatial pattern of the warming trend in HadISST (bottom left panel) and ERSST.v3 (bottom right panel) obtained by regressing the monthly HadISST (ERSST.v3) data at each grid against the centennial warming trend identified in GISS (NCDC) ST _gm.

The two SST datasets show near basin-wide warming over all the oceans, with more significant warming than the centennial warming trend in ST _gm data over the tropical Indian Ocean and southern oceans (regions in Fig. 1, bottom panels, with regression coefficients >1), and less significant warming in the Pacific basin. The strong Indian ocean warming near 40^°-50^°S has been discussed previously in the literature (Levitus et al., 2005; Alory et al., 2007) as having a barotropic structure. The weaker surface warming over the Pacific than over other ocean basins is attributed to stronger mixing and/or air-sea exchanges due to ENSO and decadal to interdecadal oscillations in the Pacific, and the export of heat by the Indonesian Throughflow (e.g., Chen et al., 2008). Note that the trends in the equatorial eastern Pacific in HadISST are inconsistent with other observed quantities, as pointed by (Deser et al., 2010). However, the same study also concluded that the multi-decadal variability in HadISST is consistent with other SST datasets. South Greenland's cooling trends may be involved in the North Atlantic Oscillation trend, through the stronger wind speed and increasing heat loss at the sea surface (Deser et al., 2010). It may also be due to reduced poleward heat transport caused by the weakening of AMOC, by increasing freshwater input at high latitudes (Cubasch et al., 2001).

To better understand the ocean warming trend, we examine the spatial distribution of the warming trend in the upper-ocean temperature having regressed the monthly WOD data at each grid and level against the centennial warming trend identified in GISS ST _gm, as shown in Fig. 2. In the Atlantic, basin-wide warming trends are observed from the surface down to 700 m, except in the northwestern Atlantic where cooling trends are located north of 40^°N. The region with the strongest warming trend is located along the western basin and the zonal band near 40^°N. The above feature suggests that the surface warming trend in the north Atlantic is associated with increased heat transport by gyre circulation. In the Pacific basin, overall warming trends exist in the tropical surface layer (0-70 m), but cooling trends exist in a wavy zonal band near 40^°N (north of the Kuroshio extention zone) from the surface down to 400m, and in the tropical western region at 200 m, indicating a shoaling trend of the western Pacific thermocline. These features are consistent with a weakening trend in the equatorial Pacific trade winds (Vecchi et al., 2006). The shoaling trend of the western Pacific is also linked to the subsurface cooling trends in the tropical Indian Ocean shown in Fig. 2 (Alory et al., 2007).

Figure 2.  The spatial pattern of the warming trend in upper-ocean temperature obtained by regressing the WOD data for the period 1955-2009 at selected levels (75, 200, 400, and 700 m) against the centennial warming trend identified in GISS ST _gm.

Other than the above common features, the spatial patterns of the warming trend in Fig. 1 do show notable disagreements between the two datasets due to their differences in historical bias corrections, input data, and analysis procedures (Rayner et al., 2003; Smith and Reynolds, 2003, 2004). There is almost no cooling in the Pacific anywhere in ERSST.v3. The warming along the west coast of South and North America and over the Central and East Pacific region in ERSST.v3 is larger in HadISST. Furthermore, there are some differences in the warming details over the Indian and Atlantic oceans in the two datasets.

Next, we examine the low-frequency variability in the low-pass filtered SST data. The standard deviations of the filtered data (Fig. 3) show regions of large variability in the equatorial Pacific, North Pacific and North Atlantic with amplitudes in ERSST.v3 greater than those in HadISST. These regions of large standard deviation all coincide with the regions of small SST warming trend shown in Fig. 1, indicating the amplitude of the 20th century warming is significant and comparable to the amplitude of multi-decadal oscillations. This may serve as a justification for separating the warming trend from low-frequency variability in studying multi-decadal oscillations. Existing studies of interdecadal variability have revealed near global-scale features, so we choose the global oceans within 30^°S and 70^°N as the analysis domain of the SST datasets. The oceans south of 30^°S are not considered because of a lack of ship measurements there (e.g., Deser et al., 2010).

Figure 3.  The standard deviation in HadISST (left panel) and ERSST.v3 (right panel), detrended and low-pass filtered for the period 1880-2009.

Our EOF analysis of the two sets of low-pass filtered SST data show two dominant modes (Fig. 4). One of the dominant modes is represented by the first EOF mode in HadISST (29% of total variance) and the second EOF mode in ERSST (22% of total variance) shown in the two top panels. This mode is reminiscent of an "ENSO-like" pattern (Zhang et al., 1997) with warm SST anomalies in the equatorial central Pacific, tropical eastern Pacific, Indian Ocean, and cold SST anomalies in the North and South Pacific. The corresponding temporal variations in the two datasets show consistent features in the recent decades since 1945, with three regime shifts near 1945, the late 1970s, and 2000. However, the PCs from the dataset also show significant differences in the early century before 1945, especially during World War I (WWI) and World War II (WWII) between 1910 and 1945. To check the data quality issue, we performed separate EOF analyses of the same two sets of SST data but for a shorter period from 1950-2009. The PCs corresponding to the dominant mode from the two SST datasets (also shown in Fig. 4) are in closer agreement. The spatiotemporal variability of this leading mode is similar to the Interdecadal Pacific Oscillation (IPO) introduced by (Power et al., 1999) based on the work by (Folland et al., 1999) and several others, including the Pacific Decadal Oscillation (PDO), which was defined as the leading PC of extratropical North Pacific SST in (Mantua et al., 1997). In this study, this dominant mode is referred to as a PDO-like oscillation. Its temporal correlation with the PDO index (http://jisao.washington.edu/pdo/) and IPO index (http://www.iges.org/c20c/IPO_v2.doc) is 0.92 and 0.88, respectively, based on HadISST (0.79 and 0.89 based on ERSST).

Figure 4.  The spatial patterns and time series of low-frequency oscillations of the EOF analyses based on HadISST and ERSST.v3 for the period 1880-2009. The upper two panels show the PDO-like mode, which accounts for 29% and 22% of the total variance in HadISST and ERSST.v3, respectively. The lower two panels show the AMO-like mode, which explains 21% and 28% of the total variance in the two SST datasets. Note that the PCs from separate EOF analyses of the same data but for the period 1950-2009 are also plotted in the figures.

The other leading mode is represented by the second EOF mode in HadISST (21% of total variance) and the first EOF mode in ERSST (28% of total variance). Both datasets show a similar spatial pattern with large amplitudes in the North Atlantic and North Pacific that oscillates in a 60-70-yr cycle. These features resemble those of the Atlantic Multidecadal Oscillation (AMO), as discussed by (Enfield et al., 2001). The temporal correlation between the AMO index (http://www.esrl.noaa.gov/psd/data/timeseries/AMO/) and the corresponding PC in HadISST and ERSST is 0.94 and 0.72, respectively. Therefore, we will refer to this dominant mode as an AMO-like oscillation in the following discussion.

The two dominant modes combined account for about 50% of the total variance of the filtered SST in both datasets. In HadISST, the major variability associated with the PDO-like oscillation is confined to the Pacific, while the AMO-like oscillation evolves primarily in the Atlantic with weaker changes confined to the North Pacific. Some numerical studies have suggested that oscillations in the Atlantic can cause SST changes in the North Pacific (Zhang and Delworth, 2005, 2007; Wu et al., 2008). The major features of the PDO-like oscillations shown in the HadISST and ERSST data are similar, but the AMO-like oscillations in the two SST datasets show more significant differences. While both datasets show warming signals along the west coast of the North American continent associated with AMO (d'Orgeville and Peltier, 2007), extensive warming extends over the South Pacific in the ERSST data (Fig. 4, bottom panels). This appears to be inconsistent with some previous modeling results. In freshwater experiments (e.g., Zhang and Delworth, 2005), cold anomalies in the North Atlantic have been reported to be associated with a meridional shift in the Hadley cell and a warmer eastern Pacific. Such experimental results appear to be consistent with warmer SSTs in the North Atlantic associated with weaker cold SST anomalies in the eastern equatorial Pacific in the HadISST, unlike that shown in the ERSST data (Fig. 4).

In this section, we analyze the oceanic and atmospheric structure of low-frequency climate variability. To achieve this, we first applied combined EOF analysis to the upper ocean temperature field at 16 levels from the surface to 700 m provided in the WOD data (1955-2009). It was found that the two leading EOF modes account for about 38% and 22% of the total variance, respectively. The corresponding principal components (PCs) are shown in Fig. 5. They consistently show temporal variability similar to the PDO-like (PC1) and AMO-like (PC2) oscillations.

Figure 5.  Time series of the PDO-like (left panel) and AMO-like (right panel) oscillations obtained by EOF analysis of HadISST (black), ERSST.v3 (red) and combined EOF analyses of WOD (purple), NCEP (green), and ERA40 (blue).

Next, we analyze the available ERA-40 and NCEP-NCAR Reanalysis datasets. The combined EOF analysis was applied to the six basic variables (U, V, W, Z, T, Q) at eight pressure levels (1000, 925, 850, 700, 600, 500, 400, 300 hPa). The variables were regridded from the original 2.5^°× 2.5^° resolution to 5^°× 5^°, detrended and low-pass filtered. Considering the limited data period of the ERA-40 (1958-2001) and NCEP-NCAR (1958-2009) reanalyses, the low-pass filtering was performed with 8-yr cutoff periods instead of 13 yr as used for the SST analysis reported in the previous section. The two leading PCs identified by the combined EOF analysis of the two reanalysis datasets are also shown in Fig. 5. The first principal component obtained from the combined EOF analysis of ERA-40 data (24% variance) and the second principal component from the NCEP-NCAR data (16% variance) are consistent with the AMO-like oscillations identified by the SST analysis discussed in the previous section (Fig. 5, right panel). Likewise, the second principal component from the ERA-40 data (19% variance) and the first principal component from the NCEP-NCAR data (24% variance) are consistent with the PDO-like oscillations identified by the SST analysis discussed in the previous section (Fig. 5, left panel). If the combined EOF analysis is performed on the two datasets having been filtered in a similar way to the SST datasets with cutoff periods, the two dominant modes combined account for over 50% of the variance. This indicates the significance of the AMO-like and PDO-like oscillations in both SST and atmospheric variables.

Figure 9.  The spatial patterns of the horizontal wind (units: m s^-1) and height (units: gpm) fields at 850 and 300 hPa associated with the PDO-like oscillations obtained by combined EOF analysis of NCEP (left panels) and ERA40 (right panels).

The oceanic spatial patterns corresponding to the PDO-like oscillation are revealed in the eigenvectors of the combined EOF analysis of WOD data, as shown in Fig. 6. The primary feature is an ENSO-like east-west temperature contrast pattern in the tropical Pacific, with warm anomalies residing in the eastern basin within the upper 75-m layer and cold anomalies to the west in the upper 200-m layer. This ENSO-like structure is accompanied by cold anomalies in the North Atlantic and Pacific. Such features are similar to the simulated tropical response to a weakening or shutdown of the Atlantic overturning circulation (Zhang and Delworth, 2005).

Figure 6.  The PDO-like spatial pattern in upper-ocean temperature obtained by combined EOF analysis of the WOD data at selected levels below the sea surface (75, 200, 400, and 700 m).

For AMO-like oscillation, the corresponding eigenvectors obtained from WOD data are shown in Fig. 7. First, we note that the associated ocean temperature anomalies are confined within the upper oceans above 400 m in the hemisphere north of 10^°S. In the North Atlantic, cold anomalies exist along the western basin and the zonal band near 40^°N, revealing a southward shift of the Gulf Stream when the Atlantic thermohaline circulation intensifies, as suggested in many previous studies. In the Pacific, warm anomalies are confined to the Kuroshio extension (0-400 m) and the Oyashio current regions (above 75 m). The warm temperature in the Kuroshio-Oyashio regions is consistent with the features shown in (Zhang and Delworth, 2007) associated with AMO forcing. In the tropical Pacific, the oscillation is associated with warm anomalies in the eastern basin at 75 m and cold anomalies in the western basin at 200 m, indicating a linkage between AMO and east-west oscillation in the tropical Pacific.

Figure 7.  The AMO-like spatial pattern in upper-ocean temperature obtained by combined EOF analysis of the WOD data at selected levels below the sea surface (75, 200, 400, and 700 m).

Next, we show the atmospheric spatial patterns (eigenvectors) of combined EOF modes from the reanalyses data corresponding to the PDO-like oscillations. The temperature fields at 850 and 300 hPa from both the NCEP and ERA40 datasets are shown in Fig. 8. Similarly, the horizontal winds and height fields are shown in Fig. 9. In Fig. 8, we also show the 2-m air temperature fields regressed upon the first principle component (the PDO-like time series) from HadISST EOF analyses. These figures together show the PDO-like oscillations are characterized by warm temperature anomalies in the tropical lower troposphere, with the warmest anomalies located in the central and eastern Pacific where warm SST anomalies reside (Fig. 4). The corresponding circulation is characterized by low-level convergent flows toward the equatorial Pacific warm SST from the high pressure anomaly fields over the Maritime Continent/neighboring oceans and the Atlantic. The warm temperature anomalies in the lower troposphere in the tropics give rise to high pressure anomalies in the tropics at 300 hPa. Straddling the equatorial Pacific warm SST over the North and South Pacific, cold temperature anomaly, low pressure and cyclonic circulation centers exist near the dateline at 40^°S and 40^°N. These extratropical features appear at both 850 hPa and 300 hPa. The above atmospheric features resemble the dynamic and thermodynamic fields in response to equatorial Pacific SST forcing. The PDO-like pattern in the lower troposphere show hemispheric symmetric features in both the NCEP and ERA40 datasets (Figs. 8 and 9). However, the symmetric structures in the lower troposphere are not so evident at 300 hPa (Fig. 8) in the NCEP dataset, where warm temperature anomalies appear in the South Pacific but cold temperature anomalies appear in the ERA40 data.

Figure 8.  (Upper and middle panels) The spatial patterns of air temperature (units: K) at 300 and 850 hPa associated with the PDO-like oscillations obtained by combined EOF analysis of the two reanalysis datasets. (Lower panels) The spatial patterns of the surface 2-m air temperature field regressed upon the first principle component (the PDO-like time series) from HadISST EOF analyses. The above results obtained from NCEP and ERA40 are respectively shown in the left and right panels.

The spatial correlation coefficients between the corresponding EOF obtained from NCEP and ERA40 are shown in Table 1. The temperature, geopotential height, and horizontal wind fields in the two datasets are reasonably correlated, except temperature in the upper troposphere (300 hPa), where the two datasets show near zero correlation. The humidity and vertical velocity fields in the two datasets show poor correlations.

Corresponding to the AMO-like oscillations, the combined EOF modes in temperature at 850 and 300 hPa from both the NCEP and ERA40 datasets are shown in Fig. 10, and the horizontal winds and height fields are shown in Fig. 11. The associated AMO-like signal in the 2-m air temperature was obtained by regressing the near-surface temperature upon the second principle component (the AMO-like time series) from HadISST EOF analyses (Fig. 10). The atmospheric temperature fields in Fig. 10 show a hemispheric asymmetric distribution, with warm anomalies in the Northern Hemisphere and cold anomalies in the Southern Hemisphere near the surface, at 850 and 300 hPa, with significant anomalies all occurring at high latitudes. The temperature pattern in the lower troposphere (near-surface and at 850 hPa) is somewhat similar to the SST pattern. At 300 hPa, extensive warm anomalies exist over Eurasia and the North Pacific, and cold anomalies cover the southern oceans where the ERA40 data exhibit much stronger magnitude than in the NCEP data. The warm anomalies over Eurasia are accompanied by anticyclonic circulation, while the cold anomalies over the southern oceans are associated with cyclonic circulation. Another important feature is the low pressure center over the northern Atlantic in response to the warm SST there, which is attributed as the forcing for North Pacific change (high anomalies and anticyclonic circulation), as proposed by (Zhang and Delworth, 2007). The warm SST and low pressure system in the northern Atlantic is found to be associated with enhanced North Atlantic thermohaline overturning circulation (and colder SST in the eastern United States) (Polyakov et al., 2010). The dominant features associated with AMO-like oscillations in the troposphere show hemispheric asymmetric features, including northern warming and southern cooling through the troposphere (Figs. 10 and 11) in both reanalysis datasets.

Figure 10.  (Upper and middle panels) The spatial patterns of air temperature (units: K) at 300 and 850 hPa associated with the AMO-like oscillations obtained by combined EOF analysis of the two reanalysis datasets. (Lower panels) The spatial patterns of the surface 2-m air temperature field regressed upon the second principle component (the AMO-like time series) from HadISST EOF analyses. The above results obtained from NCEP and ERA40 are respectively shown in left and right panels.

Figure 11.  The spatial patterns of the horizontal wind (m s^-1) and height (gpm) fields at 850 and 300 hPa associated with the AMO-like oscillations obtained by combined EOF analysis of NCEP (left panels) and ERA40 (right panels).

Concerning the atmospheric structure of AMO-like oscillation, the two datasets show consistent features in their temperature fields, especially in the upper troposphere where the spatial correlations (Table 2) are most significant. The horizontal wind and height fields in the two datasets show reasonable correlations. Humidity and vertical velocity fields in the two datasets again show poor correlations. The atmospheric temperature distribution associated with the AMO is expected to cause regional differences in estimating low-frequency warming trends.In addition, the circulation changes in the atmosphere, together with changes in the ocean, must be considered when studying the corresponding mechanisms.

We performed an analysis of low-frequency climate trends in monthly SST from HadISST and the ERSST.v3. Both datasets cover the period 1880-2009. A centennial global warming trend was first identified by applying the HHT method to global SST data. The SST warming trend at each grid was then removed by linear regression against this centennial global warming trend. The detrended and low-pass filtered SST data were further analyzed by the EOF method. The dominant EOF modes in SST were then compared with another two analyses: an EOF analysis of the WOD ocean temperature data and a combined EOF analysis of six variables (U, V, W, Z, T, Q) at eight pressure levels (1000, 925, 850, 700, 600, 500, 400, 300 hPa) from the NCEP and ERA40 reanalysis datasets. The three independent analyses consistently showed two dominant climate oscillations. One is like the PDO/IPO with a bi-decadal period and near-55-yr period since 1945. The 55-yr period oscillation marks a sharp climate phase shift in the late 1970s and a reversed shift near to the year 2000. The other is like the AMO with a near 60-yr period. The oscillation in the period after WWII shows a change from warm to cold phase near the late 1960s and a reversed change near to 1990. The two oscillations can be traced in the ocean from the surface down to 400 m.

The PDO-like oscillation is characterized by ENSO-like warm ocean temperature anomalies in the tropical eastern Pacific and Indian Ocean in the upper 75-m layer, and cold ocean temperature anomalies in the tropical western, North and South Pacific in the upper 200-m layer.The warm tropical SST signals are associated with warm temperature anomalies in the tropical lower troposphere in the central and eastern Pacific, and low-level convergent flows from the high pressure anomaly fields over the Maritime Continent/neighboring oceans and the Atlantic toward the equatorial Pacific warm SST. Straddling the equatorial Pacific warm SST over the North and South Pacific, cold temperature anomaly, low pressure and cyclonic circulation centers exist near the dateline at 40^°S and 40^°N. The above hemispheric symmetric features in the atmosphere resemble the dynamic and thermodynamic fields in response to equatorial Pacific SST forcing.

The AMO-like oscillation features warm SST anomalies in North Atlantic and North Pacific, and associated warm ocean temperature anomalies in the upper 200 m north of 10^°S. Cold anomalies exist in the North Atlantic along the western basin and the zonal band near 40^°N, and in the tropical western Pacific at 200 m. Associated atmospheric oscillations feature a hemispheric asymmetric temperature distribution in the troposphere with extensive warm anomalies in Eurasia and the North Pacific, and cold anomalies cover the southern oceans. The temperature anomalies have larger magnitude in the upper troposphere with warm Northern Hemisphere (cold Southern Hemisphere) anomalies accompanied by high (low) pressure and anticyclonic (cyclonic) circulation, except in the northern Atlantic where warm SST and air temperature cause low pressure and cyclonic circulation.

Figure 12.  Linear trends of SST during the period 1980-2009 [units: K (30 yr)^-1] obtained at each grid from (upper panels) unfiltered HadISST and ERSST.v3 data, and (lower panels) detrended HadISST and ERSST.v3 data by removing the regressed signals against the corresponding global mean SST.

The Pacific and the Atlantic MDV in upper-ocean temperature appears to be mutually linked. Part of the observed features are in line with the argument that North Atlantic warm SST forces North Pacific change (high pressure anomalies and anticyclonic circulation), as proposed by (Zhang and Delworth, 2007). Other studies, such as (Latif, 2001) and (d'Orgeville and Peltier, 2007), suggest that the AMO and PDO of 60-yr cycles are signatures of the same oscillation. However, studies such as (Park and Latif, 2010) argue that the two climate oscillations are independent modes in their coupled climate model. Therefore, whether the two multi-decadal oscillations are physically related needs to be further investigated.

Figure 13.  The time series of global mean surface temperature (ST _gm) obtained from the GiSS dataset (thin black dashed line) and the warming trend (red solid line). Also shown are the ST _gm associated with PDO-like (purple solid line) and AMO-like (blue solid line) oscillations obtained by regressing the detrended ST _gm against the two corresponding PCs shown in Fig. 4. The sum of the three time series (orange solid line) accounts for the climate trend changes in ST_gm. The difference between ST _gm and the climate trend changes (thin green dotted line) denotes the high-frequency changes.

While the multi-decadal climate oscillations need to be investigated for understanding their nature, they contribute significantly to the spatial and temporal distributions in the climate trend changes. This was demonstrated in the spatial distribution of linear trend estimated from HadISST and ERSST.v3 for the period 1980-2009 (Fig. 12, top panels). This is to be compared with the bottom panels in the same figure, which show the linear trend map estimated from the same SST data for the same period except that the global mean trend has been removed from the data, as done in (Latif et al., 2006). The top panels of Fig. 12 show that the spatial pattern of SST in the Pacific resembles the PDO oscillation in negative phase and that in the Atlantic resembles the AMO oscillation in positive phase. This is consistent with the temporal change in the most recent 30 years of PDO from positive to negative phase, but an opposite change of AMO from negative to positive phase. Note further that during this period, the AMO acts to oppose the warming trend in the Atlantic western boundary current zone where strong centennial warming trends are found. The overall similarity between the top and bottom panels in Fig. 12, except the difference in the Indian Ocean, reveals that, in the most recent 30 years, the multi-decadal oscillations have contributed significantly to the warming trends in the Pacific and Atlantic, while the global warming trends have been most evident in the Indian Ocean.

In addition to the SST trend changes, our analysis further showed the contribution of multi-decadal oscillations to the trends in the tropospheric warming. While the thermodynamic response to CO₂ warming tends to produce a spatially uniform centennial warming, the transition of PDO-like oscillations from negative to positive phase contribute significant tropical warming and weaker extratropical cooling in the troposphere (Fig. 8). On the other hand, the transition of AMO-like oscillation from negative to positive phase gives rise to warming trends in the Northern Hemisphere and the opposite in the Southern Hemisphere, and the range of the Northern Hemisphere warming is greater than that of the Southern Hemisphere cooling (Fig. 10).

To summarize the effect of multi-decadal oscillation on global warming trends, we show in Fig. 13 the low-frequency variability of ST _gm that contains the centennial trend (red solid line) from Fig. 1 (HHT GISS), PDO-like (purple solid line) and AMO-like (blue solid line) ST _gm, and the sum of the three time series (orange solid line). The PDO-like/AMO-like ST _gm was obtained by regression of the detrended ST _gm against PDO-like/AMO-like index (Fig. 4, HadISST). In addition to the three low-frequency trend changes, the remaining high-frequency variability of ST _gm ( 0.2 K) results from ENSO (Chen et al., 2008). For the period 1975-2005, low-frequency ST _gm increased by 0.44 K (Wu et al., 2011), more than twice the centennial warming trend of about 0.22 K. The part of warming above 0.22 K is contributed by PDO-like and AMO-like oscillations, as shown in Fig. 13. This indicates that the two multi-decadal oscillations contribute about half of the ST _gm warming for the period 1975-2005, as indicated by (Qian et al., 2010) and (Wu et al., 2011). Furthermore, the ST _gm remains almost unchanged from 2005 to 2013 when the PDO-like oscillation turned to negative phase and the AMO-like oscillation passed its peak positive phase and headed toward negative phase. Moreover, (Kaufmann et al., 2011) found ST _gm did not rise in recent years with the combined effect of internal variability and anthropogenic factors.

Clearly understanding multi-decadal oscillations enables climate scientists to estimate global warming change signals with more certainty. Such knowledge is also crucially needed for developing decadal climate prediction systems.