1. Introduction
The concept of “climate mode” represents a useful tool for reducing the complexity of climate variability. A mode can be defined as a set of physical processes characterized by coherent large-scale variations and a quasiperiodic evolution in time. Identifying and understanding the mechanisms of specific climate cycles is of central importance in climate research. One of the climate modes with important socioeconomic impact is known as the Atlantic multidecadal oscillation (AMO: Kerr 2000). It can be defined as the first rotated EOF of the non-ENSO global SST field (Enfield and Mestas-Nuñez 1999; Mestas-Nunez and Enfield 1999). It has an ∼70-yr period and is global in scope. The AMO influences the climate over Europe and North America (Enfield et al. 2001; Sutton and Hodson 2005) and modulates the Atlantic tropical cyclone activity (Goldenberg et al. 2001). Owing to its multidecadal time scale, it can mask the anthropogenically induced variations (Hegerl et al. 1997), and therefore the understanding of mechanisms generating the AMO can increase the confidence in the detection of anthropogenic climate change.
The existence of a multidecadal mode was noted by Folland et al. (1986, 1991) as detected in global sea surface temperature and marine air temperature and was isolated by Schlesinger and Ramankutty (1994, 1995) in the global mean instrumental temperature record. This mode shows maximum amplitudes in the Atlantic basin (Mann and Park 1994) but also includes a positively correlated oscillation in the North Pacific, where multidecadal variations were also reported (Minobe 1997). The AMO was also identified in several proxy records (Stocker and Mysak 1992; Mann et al. 1995; Delworth and Mann 2000; Lohmann et al. 2004) and was reconstructed for the last centuries (Gray et al. 2004).
Spectral analyses of climatic time series (Delworth and Mann 2000) show a significant peak around 70 yr. The statistical significance of the peak and its sharpness support a deterministic origin of the cycle. Consistent with this, Schlesinger and Ramankutty (1994) proposed that the AMO is probably internally generated by deterministic processes. This internal origin hypothesis is supported by numerical experiments that suggest that an ∼70-yr cycle is not generated by external forcing (Andronova and Schlesinger 2000).
A complementary approach for investigating the multidecadal variability relies on numerical integrations of coupled atmosphere–ocean models. Delworth et al. (1993) describe an ∼50-yr multidecadal cycle as an oceanic mode excited by atmospheric stochastic forcing. Based on results from another numerical model integration, Timmermann et al. (1998) describe a Northern Hemisphere coupled air–sea mode with a 35-yr period. It involves interactions between the thermohaline circulation (THC) and the atmosphere in the North Atlantic and interactions between the ocean and the atmosphere in the North Pacific. The two ocean basins are coupled via an atmospheric teleconnection pattern. Ocean–atmosphere interactions in the Atlantic basin also play a major role in generating the 25–30-yr mode presented by Cheng et al. (2004). The memory of this cycle results from the mechanical spinup and thermal response time scales of the THC. Using a coupled atmosphere–ocean–land–sea ice model, Dai et al. (2005) describe a 24-yr period mode in which a dominant role is played by an advective mechanism and strong coupling between the THC and North Atlantic Oscillation (NAO). The time scale of the mode is determined by the ∼6-yr time lag between the midlatitude North Atlantic Ocean surface density and THC strength. All of these numerical experiments involve THC fluctuations, suggesting a key role played by this climate component in multidecadal variability (Latif et al. 2004; Knight et al. 2005).
The aim of this study is to investigate the mechanism of the AMO based on the instrumental and reconstructed data. The physical processes involved in multidecadal variability are inferred from the specific properties of the derived patterns, which provide constraints for the mechanism. The investigation is performed in a deterministic frame in which a memory and a negative feedback represent necessary elements for the generation of the oscillatory character of the mode. The negative feedback is responsible for the transition of the system from one extreme state to the other. However, if the feedback acts quasi-instantaneously, then it has only a damping effect and no oscillatory evolution results. Therefore, a memory is also necessary in order to generate an oscillation.
The data and methods are presented in section 2. The main components of the mechanism and their properties are described in section 3. Based on them, a dynamic picture of the AMO is presented in section 4. In our argumentation we will make use of instrumental and historical data and for some causal relationships we take advantage of previous modeling studies. The mechanism is synthesized in section 5 and discussed in section 6. The conclusions are drawn in section 7.
2. Data and methods
a. Data
Several datasets are used to investigate the properties and the mechanism of the AMO. SST on a 5° × 5° grid are obtained from the dataset compiled by Kaplan et al. (1998). These grids have been extended over the 1856–2002 period. Two sea level pressure (SLP) datasets are analyzed. One covers only the ocean’s surface on a 4° × 4° grid for the 1856–1991 period (Kaplan et al. 2000). The other set, compiled by Trenberth and Paolino (1980) on a 5° × 5° grid, extended northward from 15°N. These fields extend over the 1899–2002 period. Although the instrumental period covered by these datasets is not long in view of the investigated multidecadal variability, it was proven possible to isolate a 60–70-yr-period signal in observed data (Schlesinger and Ramankutty 1994, 1995).
Based on historical observations of multiyear sea ice off southwest Greenland, Schmith and Hansen (2003) compiled an annual record of the Fram Strait sea ice export (FSSIE) over the 1820–2000 period. The observations, called “storis,” are obtained from ship logbooks and ice charts. This reconstruction is used to investigate the role played by sea ice export from the Arctic Ocean on multidecadal variability.
b. Methods
Singular spectrum analysis (SSA) technique (Vautard et al. 1992) is applied on multicentennial climate reconstruction time series. The SSA method is designed to extract information from short and noisy time series by providing data-adaptive filters that help separate the time series into statistically independent components like trends, oscillatory signals, and noise (Allen and Smith 1997). Trends need not be linear and the oscillations can be modulated in amplitude and phase. We consider as “trend” any signal that has no more than a maximum/minimum over the analyzed period. Of course, in an extended time frame, the trend could be in fact an oscillatory signal, but usually this is of no importance for the results derived for the analyzed period. End effects are avoided by extending the initial time series, based on a fitted low-order autoregressive process.
The empirical orthogonal functions (Lorenz 1956) method is used to identify the eigenmodes of the SLP field. By construction, the EOF method is effective in separating modes with different characteristic spatial structures. To derive spatial patterns associated with the dominant principal component (PC), regression and composite maps were constructed based on several climatic fields. One may note that, unlike correlation, compositing is a nonlinear operation. However, as with correlation, it does not imply causality.
The canonical correlation analysis (CCA) method is applied to identify coupled ocean–atmosphere patterns (Preisendorfer 1988). It is a way of measuring the linear relationship between two multidimensional variables. It finds two bases, one for each variable, that are optimal with respect to correlations and, at the same time, it finds the corresponding correlations. For a detailed description of these methods, we refer to von Storch and Zwiers (1999).
Due to the limited time extension of the analyzed data relative to the multidecadal periodicity, the significance of the patterns is appreciated based on several criteria:
their physical relevance and consistency,
their robustness relative to different methods used to derive them, and
their similarity with patterns derived in modeling studies.
3. Static picture—AMO components
The mechanism that we propose involves physical processes in various locations over the Northern Hemisphere. Our presentation of the mechanism starts with the description of the main components and key physical processes that mediate the interactions between them. Based on the properties of these components and SST and SLP composite maps, the dynamic picture of the AMO is then constructed.
a. Thermohaline circulation
An AMO index is constructed based on the North Atlantic SST anomalies (Kaplan et al. 1998) in a similar way as in Enfield et al. (2001), Sutton and Hodson (2005), and Knight et al. (2005). In a preliminary step, annual values are calculated and the linear trend is removed for each grid point. The time series is obtained as an average over the region 0°–60°N, 75°–7.5°W. Finally, the AMO index is obtained by applying a 25-yr running mean filter to this time series to remove interannual and decadal variability (Fig. 1a). The index shows pronounced multidecadal variability with maxima around 1870 and 1950 and minima around 1915 and 1980. The amplitude of the AMO index over the 1856–2000 period is ∼0.4°C. This time component will be used to construct a dynamic picture for the AMO.
The associated pattern is derived by regressing the annual mean detrended SST field on the normalized (unit variance) AMO index (Fig. 1b). The main characteristic of the SST pattern is a quasi-monopolar structure in the North Atlantic. As indicated by several numerical experiments (Latif et al. 2004; Knight et al. 2005), such a quasi-monopolar SST structure is associated with THC variations, indicating that the THC plays an active role in the AMO mechanism.
b. Hemispheric wavenumber-1 SLP structure
To investigate a possible hemispheric extension of the AMO fingerprint in the atmosphere, an EOF analysis on detrended winter SLP anomalies is performed. The Trenberth and Paolino (1980) SLP dataset extends over the 1899–2002 period. The first EOF (32%) is identical with the Arctic Oscillation (Thompson and Wallace 2000; not shown here). EOF2 (12%) includes the western Atlantic pattern and the Aleutian center, while EOF3 (9%) includes two centers of negative anomalies over the North Pacific and Siberia, and two centers of positive values over the North Pole and the eastern North Atlantic. The fourth EOF (8%) is of specific importance in our study (Fig. 2). It includes two connected centers of negative anomalies over the North Pacific and Greenland and two other centers of positive anomalies over the North Atlantic and Siberia. A similar pattern was presented by Cavalieri (2002) in association with the Fram Strait sea ice export. The hemispheric wavenumber-1-like (hereafter HWN1) structure of this pattern is consistent with a zonally asymmetric oceanic forcing on the atmosphere. As will be shown below, the same structure is also obtained as a composite map. The associated time series (Fig. 2b) is dominated by a multidecadal time component (12%), which is identified through SSA using a 40-yr window.
Although EOF4 is not strictly separated from EOF3 according to North et al. (1982), the principal components associated with the first four EOFs have distinct properties. This is suggested by SSA performed on the first four time components (not shown). PC1 is dominated by interannual variability, PC2 by an ∼50-yr time scale, and PC3 by a centennial component. Only PC4 is dominated by an ∼70-yr time-scale component. Additional arguments for the physical relevance of this structure come from numerical experiments. A similar structure with that in Fig. 2b was presented as corresponding to the Atlantic multidecadal variability by Enfield and Mestas-Nuñez (1999, their Fig. 8). An SLP EOF with an analog pattern in the North Atlantic–European sector, resembling the east Atlantic pattern, was derived from numerical integrations with the coupled ECHAM5–OGCM model (Park and Latif 2005, their Fig. 5c). Its associated time component was significantly correlated with the North Atlantic THC index. An SLP pattern also including centers of opposite signs in the North Atlantic and the North Pacific was associated with a significant change of THC in a coupled ocean–atmosphere model (Manabe and Stouffer 1988, their Fig. 25). The projection of a large THC change on the 1000-hPa height anomalies [December–February (DJF)], as results from the Third Hadley Centre Coupled Ocean–Atmosphere General Circulation Model (HadCM3), has an almost identical structure with that of EOF4 (Vellinga and Wood 2002, their Fig. 5). Furthermore, recent numerical experiments show that the HWN1 structure is forced by the North Atlantic monopolar SST structure (Grosfeld et al. 2007, manuscript submitted to Tellus). Therefore, our results and several observational and numerical studies support the relevance of the HWN1 SLP pattern (EOF4) in the multidecadal variability and suggest a relation to THC variations.
c. Fram Strait sea ice export
A key feature of the HWN1 SLP pattern is its centers over Greenland, northern Canada, and Alaska on one hand and the opposite sign over Siberia and the North Atlantic on the other (Fig. 2a). The associated SLP gradient has high projection on Fram Strait. In these sectors sea ice dynamics is mainly driven by the large-scale wind anomalies in relation with SLP patterns (Hilmer et al. 1998; Rigor et al. 2002). Furthermore, the sea ice export from the Arctic is determined by wind-driven dynamics (Häkkinen and Geiger 2000).
To investigate a possible projection of multidecadal variability on the freshwater fluxes exported from the Arctic into the North Atlantic, we analyze a FSSIE reconstruction. The record has an annual resolution and extends over the 1820–2000 period (Schmith and Hansen 2003). This record shows pronounced interannual variability (Fig. 3a). Because our focus is on multidecadal variability, a 7-yr running mean filter was applied to the initial time series. The resulting record (Fig. 3a) is dominated by a quasiperiodic component with an approximately 110-yr period that explains 46% of the variance (not shown). It includes a maximum around 1890 and a minimum during the 1940s as obtained through SSA. This long-term component is subtracted from the record to obtain a residual time component (Fig. 3b). An SSA on this residual record using an 80-yr window is shown in Figs. 3c–e. The results of the analysis are not sensitive to variations of the window length. The first two eigenvalues (Fig. 3c) correspond to a pair of time EOFs (T-EOF) that are in quadrature (Fig. 3d), suggesting that they are associated with a robust quasiperiodic signal. The corresponding quasiperiodic time component (Fig. 3e) has an ∼70-yr time scale. It explains 29% of the variance in this residual time series and 17% of the variance in the initial record. As will be shown in section 4, the phase relations of the FSSIE and the other components of the AMO support an active role played by sea ice in multidecadal climate variability.
It was proposed that the sea ice export from the Arctic Ocean appears as an effective mechanism of transferring variability from the atmosphere into the ocean, participating in the North Atlantic climate fluctuations (Hilmer et al. 1998). The transport of sea ice within and out of the Arctic Ocean provides the most important direct connection between sea ice and the ocean. Therefore, the anomalous strong sea ice export from the Arctic would increase the freshwater fluxes into the North Atlantic. In line with a significant influence of sea ice variations on the THC (Hasumi and Suginohara 1995; Lohmann and Gerdes 1998; Gildor and Tziperman 2003), the Arctic sea ice export through Fram Strait can modulate the North Atlantic THC (Mauritzen and Häkkinen 1997; Häkkinen 1999). Because the thermohaline circulation is sensitive to freshwater intrusions, the resulting anomalous positive freshwater fluxes reaching the convection region contribute to a decrease of the meridional overturning due to increased buoyancy.
Modeling studies suggest that the THC can adjust to the forcing in time intervals of years to decades (Eden and Jung 2001) and that the basinwide North Atlantic cooling is associated with THC weakening (e.g., Lohmann 2003). Numerical integrations further show that the strength of the meridional overturning at 30°N correlates almost perfectly with a North Atlantic SST index, with the THC index leading the North Atlantic SSTs by several years (Latif et al. 2004). Therefore, sea ice export through Fram Strait can affect, with some lags, the meridional overturning and the SST in the Atlantic basin.
d. North Atlantic ocean–atmosphere interactions
To derive the North Atlantic SLP pattern associated with the AMO, a canonical correlation analysis is performed between the annual means of the SST and SLP fields in this sector. Their common period is 1856–1991. Prior to the analysis the linear trend is removed and a 25-yr running mean filter is applied. The first pair of SST and SLP obtained from CCA is shown in Fig. 4. By construction, the correlation between the time components associated with the SST and SLP structures (Fig. 4c) is high (0.98). The time components corresponding to the SST and SLP patterns (Fig. 4c) show pronounced multidecadal variability and are very similar to the AMO index shown in Fig. 1a. The SST pattern (Fig. 4a) explains 27% of variance and has a monopolar structure, which is specific to the AMO (Fig. 1a). The SLP structure is dominated by a low that covers most parts of the North Atlantic (Fig. 4b) and extends eastward over Europe, but also includes a positive center over southeastern Greenland. Although this pattern has some projection on the NAO, its axis is tilted northeastward and its negative center extends farther over Europe.
The two relatively homogeneous SST and SLP structures are expected because the flat SST pattern excludes strong horizontal gradients so that the surface baroclinicity is not changed by the SST anomaly pattern. Owing to the relatively short decorrelation time and relatively low spatial coherence of the atmosphere, it is unlikely that the atmosphere can generate such a wide and uniform SST pattern. Furthermore, the North Atlantic Ocean’s influence on the atmosphere is supported by observational (Kushnir 1994) and numerical studies (Kushnir and Held 1996; Pohlmann et al. 2005). A very similar relation between these SST and SLP structures, as expressions of multidecadal variability, is supported by numerical simulations (Park and Latif 2005). Furthermore, Grosfeld et al. (2007, manuscript submitted to Tellus) found an analog SLP response to Atlantic multidecadal SST forcing in a 20–ensemble member integration. Based on these considerations we interpret the SLP pattern as an atmospheric response to oceanic thermal forcing, which results from THC variations.
e. North Pacific ocean–atmosphere interactions
Previous studies reported multidecadal variability in the North Pacific (Minobe 1997; Wu et al. 2003; Frauenfeld et al. 2005). To derive the ocean–atmosphere coupled patterns with multidecadal time scales a CCA is performed for the North Pacific basin (0°–65°N, 120°–270°E). As in the preceding analyses, the linear trend is removed from the annual mean SST and SLP fields and a 25-yr running mean filter is applied. The second most coupled pair is of specific importance in our study. The SST pattern (Fig. 5a) explains 11% of variance. Its most prominent feature is the center of negative anomalies extending eastward from the northeast Asian coast. The associated SLP pattern is the Aleutian low, which has been found to be associated with multidecadal variations (Minobe 1999). The patterns of this second pair explain the largest part of the variance in both SST and SLP fields. The corresponding time components (Fig. 5c) are by construction highly correlated (0.98) and show pronounced multidecadal variability over the twentieth century.
A very similar pairing of multidecadal SST and SLP patterns was found in Mestas-Nuñez and Enfield [1999, their Figs. 5 (SST) and 8 (SLP: middle panel)]. Reversing the signs makes them correspond to Fig. 5 of this paper. This pair of patterns is consistent with the atmospheric forcing associated with fluctuations of position and strength of the Aleutian low. A deepened Aleutian low decreases the SST in the central North Pacific by advection of cold and dry air from the north by an increase of westerly winds and ocean–atmosphere turbulent heat fluxes, and by a strengthened equatorward advection of temperature by Ekman currents. In the eastern region, a deepened Aleutian low enhances poleward winds and leads to positive SST anomalies. The pair is also consistent with a barotropic atmospheric response to the SST anomaly forcing, as suggested by Peng and Robinson (2001). Based on these considerations we argue that ocean–atmosphere interactions in the North Pacific associated with these SST–SLP structures can act as a positive feedback that amplifies the perturbations. This is supported by previous studies regarding Pacific interdecadal and multidecadal variations.
A similar pair of coupled structures was termed “Interdecadal Pacific Signal” (IPS), and it was argued that including an atmospheric component allows the decomposition of the leading mode of the Pacific SST variability into its interdecadal and interannual patterns (Frauenfeld et al. 2005). Although this interdecadal signal is unrelated to interannual El Niño variability, it seems closely linked to tropical Pacific SSTs. Furthermore, it was argued that interdecadal variability in the Pacific Ocean and the Northern Hemisphere atmosphere, which is distinct from stochastic variability, results from interactions between ocean and atmosphere (Frauenfeld et al. 2005).
The SST pattern derived from CCA (Fig. 5a) is also known as the “North Pacific Multidecadal Mode” (NPM: Deser and Blackmon 1995; Nakamura et al. 1997; Zhang et al. 1997). It was proposed that the NPM originates from local atmospheric stochastic processes and coupled ocean–atmosphere interactions. Atmospheric stochastic forcing can generate a weak NPM-like pattern in both atmosphere and ocean with no preferred time scales. In contrast, coupled ocean–atmosphere feedback can substantially enhance the variability and generate a multidecadal mode in the North Pacific (Wu et al. 2003).
Such a positive feedback can involve the delayed response of the ocean to varying winds (Schneider et al. 2002). Based on a positive feedback in the North Pacific, it was suggested that, if there is a distinct time scale of variability in this sector, it should come from outside the North Pacific area (Seager et al. 2001).
4. Dynamical picture
We presented three climatic components/structures that are active elements in multidecadal climate variations: the THC, the HWN1 SLP pattern that extends over both Atlantic and Pacific sectors, and the sea ice export through Fram Strait. To construct the dynamic evolution of the multidecadal variations associated with the multidecadal mode, quasi-global SST and Northern Hemisphere SLP composite maps were constructed based on the AMO index (Fig. 1a). The annual SST and SLP fields from Kaplan et al. (1998) are detrended and a 25-yr running mean filter is applied.
At zero lag, the SST map includes positive SST anomalies in the North Atlantic, while the Pacific basin includes regions with both positive and negative anomalies (Fig. 6a). The highest values are disposed in the North Atlantic sector. This SST pattern for lag = 0 has been identified in previous studies (Enfield et al. 2001; Knight et al. 2005). The SLP pattern is dominated by the low disposed over the North Atlantic, which seems to extend also over eastern Asia. No coherent structure is observed in the North Pacific SLP field. Following the arguments presented in section 3d, we interpret this atmospheric structure (Fig. 6b) as a response to North Atlantic SSTs.
At 5-yr time lag, both SST and SLP structures are more pronounced in the North Pacific, while the North Atlantic patterns are almost unchanged (Figs. 6c,d). The area of positive SST anomalies extending in the region around 45°N, 160°E has a larger amplitude (Fig. 6c) and the pressure anomaly in the Aleutian low increased significantly (Fig. 6d). The SLP structure is very similar to the HWN1 SLP pattern (with reverse sign) shown in Fig. 2, including the oppositely signed SLP centered at 40°N in the Atlantic, forming a dipole with the Pacific. The association between the SST and SLP structures was underlined by the CCA for this region (Fig. 5), as parts of the IPS. They can be involved in the positive feedback in the North Pacific, which implies a memory of about 5 yr associated with oceanic adjustment (Schneider et al. 2002). Consistent with this, the SST and SLP anomalies continue to grow in the next 10 yr, as seen in the lag correlation maps for 10 and 15 yr (Figs. 6e–h). The signs of anomalies are not changed in the Atlantic basin at 10-yr lag (Fig. 6e) but negative SST anomalies start to emerge in this area at 15-yr lag (Fig. 6g). The SSTs’ decrease in this region is associated with a reduced low over this area, consistent with the forcing role played by the ocean.
It is important to notice that at lag 10 yr, the SLP gradient between the Atlantic and Pacific Oceans included in the HWN1 pattern has maximum values (Fig. 6f). The associated winds project strongly on Fram Strait and can enhance the sea ice and freshwater export from the Arctic into the North Atlantic Ocean. This causal relation is supported by the lag relation between the HWN1 time component (Fig. 2b) and the multidecadal component identified in the FSSIE reconstruction (Fig. 3e). The late 1920s maximum of the former (Fig. 2b) implies reduced anomalous wind from the Arctic through Fram Strait (Fig. 2a), consistent with the reduced sea ice export in the early 1930s (Fig. 3e). This causal relation is also valid for the early 1960s minimum of the HWN1 time component (Fig. 2b), which is followed by a maximum of sea ice export in the late 1960s (Fig. 3e). The atmosphere–sea ice forcing relation is supported also by Cavalieri (2002), who shows that the variability of the FSSIE export is closely related with the phase of zonal wave 1 in SLP. Furthermore, the phase of zonal wave 1 explains 60%–70% of the ice export variance in two dynamic–thermodynamic sea ice models (Cavalieri 2002). Therefore, the atmospheric wavenumber-1 structure appears to have a significant influence on the sea ice and freshwater export from the Arctic Ocean through Fram Strait. Consistent with the arguments presented in section 3c, the maxima/minima of AMO (Fig. 1a) lag the Fram Strait sea ice export record extremes (Fig. 3e) by about 10–15 yr. AMO index maxima around 1880 and 1950 (Fig. 1a) are preceded by sea ice export minima at approximately 1965 and 1935 (Fig. 3e). Similarly, the AMO index minima around 1915 and 1980 are preceded by maxima of sea ice export around 1900 and 1970. Furthermore, the HWN1 => FSSIE => THC causal chain as part of the multidecadal mode is accurately supported by a 2000-yr integration of a coupled model (Delworth et al. 1997).
The transition period related to the THC adjustment manifests differently in the two basins. At 20-yr lag, the Pacific structures are not modified (Figs. 7a,b), but the signs of the Atlantic SSTs start to change (Fig. 7a). Consequently, the atmospheric response in this sector decreases. At 25-yr lag, the Pacific SSTs have no well-defined structure, but the SLP pattern is still observed. In the North Atlantic negative anomalies start to dominate (Fig. 7c), and a high starts to develop southeast of Greenland (Fig. 7d). At lags 30 and 35 yr, the completion of the THC adjustment is indicated by the interhemispheric SST dipole in the Atlantic basin, which is the canonical SST signature for long-term THC changes (Crowley 1992; Stocker 1998). At this time, no coherent SST pattern is observed in the Pacific basin (Figs. 7e,g). An SLP high is developed in the North Atlantic over the negative SST anomalies. At this stage, the AMO is in its opposite phase, the negative feedback is closed, and half of the cycle is completed.
This dynamic picture, based on the composite maps, can also be followed based on the patterns and their associated time components identified in the previous analyses. For example, let us start with the positive SSTs that dominate the North Atlantic in the 1940s (Fig. 1a). This is instantaneously transmitted to the atmosphere above as a low in the SLP field (Figs. 4a–c). About one decade later, during the late 1950s and 1960s, the multidecadal signal has a maximum amplitude in the North Pacific (Figs. 5a–c, with reversed sign). The multidecadal signal is then reflected in the HWN1 SLP structure, which has a maximum in the 1960s (Figs. 2a,b, with reversed sign). In several years, in the late 1960s, the signal is visible in the FSSIE, which shows a maximum at this time, concurrent with the Great Salinity Anomaly (GSA: Figs. 3b,e). The associated freshwater fluxes can reduce the THC in about 10–20 yr, associated with a wave adjustment resulting in negative SST anomalies in the North Atlantic, which are observed in the late 1970s and early 1980s (Fig. 1a). At this time the AMO cycle is in its opposite phase.
5. Synthesis of the mechanism
The physical mechanism for the AMO is synthesized in Fig. 8. Conventionally, one starts with an enhanced THC that generates uniform positive SST anomalies in the North Atlantic. In this sector, the atmospheric response is represented by an SLP low that extends over the SSTs but also farther eastward over Eurasia. The multidecadal signal is also transferred in the Pacific basin via the Tropics. Further, it affects the North Pacific through atmospheric teleconnections, where it manifests as a weakened Aleutian low and associated positive SST anomalies extending eastward from the East Asian coast. The local positive feedback that involves oceanic adjustment amplifies these structures, which are reaching a maximum amplitude after 10–15 yr. At this time, the HWN1 SLP structure includes a maximum gradient over Fram Strait, increasing significantly the Arctic sea ice and freshwater export. Consequently, the meridional overturning in the ocean is reduced due to resulting freshwater fluxes in the North Atlantic (10–20 yr) and the cycle is turned into its opposite phase. Therefore, the negative feedback of the cycle results from ocean–atmosphere–sea ice interactions in the Atlantic, Pacific, and Arctic Oceans.
In view of this mechanism, the memory of the AMO has a heterogeneous nature, resulting from several different physical processes. The spinup/down time of the THC provides a part of the memory (∼15 yr). The North Atlantic SSTs lag the THC index by several years, which is supported by the model results of Latif et al. (2004). Another part of the memory (∼10 yr) is provided by the North Pacific basin through its ocean–atmosphere positive feedback that involves oceanic adjustment (Wu et al. 2003). Finally, several years are necessary for sea ice adjustment to wind stress forcing associated with the HWN1 SLP pattern. In this way, we find a cumulated memory of about 30–35 yr, which is half of the AMO periodicity.
6. Discussion
In the model integrations, which show pronounced multidecadal variability (Delworth et al. 1993, 1997; Timmermann et al. 1998; Cheng et al. 2004; Dai et al. 2005), the THC plays the key role. Because of its decadal-scale memory, even a stochastic forcing applied to the THC would result in a multidecadal periodic climatic evolution. In this view, a two-component system (e.g., forcing–delayed response) like the atmosphere–ocean one is enough to generate a multidecadal cycle, which does not rely on sea ice dynamics. Finally, the constraints based on observational and proxy data provide the validity criteria for the mechanisms identified in numerical simulation.
Our mechanism shares common features with that described by Delworth et al. (1993, 1997) and Timmermann et al. (1998). The present mechanism and that described in the last study include THC variations, ocean–atmosphere interactions, North Pacific manifestations, and hemispheric atmospheric variations. The main difference is represented by the physical processes generating the forcing for THC. In the numerical integration (Timmermann et al. 1998), ocean–atmosphere interactions in the North Atlantic are responsible for the freshwater fluxes, while in our mechanism these are related to the FSSIE forced by the HWN1 SLP structure. The multidecadal cycle with the closest period to the AMO time scale is that described in the Geophysical Fluid Dynamics Laboratory (GFDL) model (Delworth et al. 1993). A simple sea ice component included in the model, which allows sea ice to move with the ocean currents, generates physical links that support our mechanism. The causal chain HWN1 => FSSIE => THC from our mechanism is reproduced in the model simulation (Delworth et al. 1997).
The GSA and its suggested Arctic origin (Aagaard and Carmack 1989) appear as an integrated part of our mechanism. The persistent intensification of the northerly winds over the Greenland Sea, resulting in an increased injection of polar waters into the East Greenland Current, was considered as the main cause for the GSA (Dickson et al. 1975, 1988). This was accompanied by a shift in sea ice conditions, contributing to the polar freshwater excess during the GSA (Mysak et al. 1990; Hakkinen 1993). Consistent with this, in the frame of our mechanism, northerly winds are generated by the HWN1 SLP pattern, which enhances the FSSIE export during the 1960s (Fig. 3e). This would imply that the climate shift during the 1970s is linked to the GSA leading to an anomalous weak THC. This topic will be investigated in a forthcoming study.
One important aspect of our mechanism is the causal relation between the Atlantic and Pacific basins. One possibility is that the multidecadal signal is transferred from the Atlantic to the Pacific through the Tropics (Dima and Lohmann 2004) where the intertropical convergence zone acts as a zonal waveguide through which the phase of the decadal signal is propagating westward (White and Cayan 2000). Further, through atmospheric teleconnections it is transmitted from the tropical Pacific to the North Pacific (Seager et al. 2001). Here it is amplified by a local positive feedback resulting from ocean–atmosphere interactions, with a delay of several years (Peng and Robinson 2001; Wu et al. 2003). This causal chain is consistent with the spatiotemporal development of the climate shift around the 1970s. It is observed in the North Atlantic in the late 1960s and in the tropical Atlantic in 1974/75 (Dima et al. 2001). Further, in the North Pacific the shift is reported in 1976/77 and several studies suggest that it has a tropical origin (Nitta and Yamada 1989; Trenberth and Hurrell 1994), which is also supported by the symmetry of the associated SST pattern (Evans et al. 2001). Therefore, the shift followed the proposed causal chain, which starts in the North Atlantic and affects the North Pacific after several years. However, the mid/high-latitude atmosphere in the Northern Hemisphere can also be involved in transferring the multidecadal signal from the Atlantic to the Pacific basin.
The SST and SLP composite maps support the amplifier role played by the North Pacific basin. The anomalous cyclonic circulation developed in the North Pacific at 5-yr lag (Fig. 6d) does not decrease significantly in amplitude until uniform negative SST anomalies are developed in the North Atlantic (Fig. 7f). Furthermore, it stayed at quasi-constant values even during periods when North Atlantic SSTs were in a transition phase (Fig. 7a). This confirms that if there is a distinct time scale of variability in the North Pacific, then it must come from outside this region (Seager et al. 2001), and supports the effectiveness of the ocean–atmosphere positive feedback in this area.
7. Conclusions
The mechanism for the Atlantic multidecadal variability is investigated in a deterministic frame in which a memory and a negative feedback are necessary elements for the generation of a quasiperiodic mode. Putting together results from observational data with previous studies based on numerical experiments, we propose a mechanism responsible for the generation of the AMO. It involves the atmospheric dynamics of the HWN1 SLP pattern and sea ice/freshwater export from the Arctic to the North Atlantic ocean. The memory of the cycle is provided by the THC adjustment to freshwater fluxes and Atlantic SST response to meridional overturning, by oceanic adjustment in the North Pacific, and by sea ice response to wind stress. The negative feedback basically results from exchanges between the Arctic and Atlantic basin, while the Pacific basin acts as an amplifier of the signal.
The lead–lag relations between different components of the mechanism could be used to improve the climate predictability at multidecadal time scales in the North Atlantic realm. According to our mechanism, the minimum in the Fram Strait sea ice export around the year 2000 will result in increased THC in the years 2010–15, which would translate into warmer conditions over Europe and North America in the next decades.
Acknowledgments
We thank two anonymous reviewers for constructive and helpful comments and Dr. Norel Rimbu for fruitful discussions. The work has been supported by the Humboldt Foundation and the German Ministry of Education and Research through MARCOPOLI.
REFERENCES
Aagaard, K., and E. C. Carmack, 1989: The role of sea ice and fresh water in the Arctic circulation. J. Geophys. Res., 94 , 14485–14498.
Allen, M., and L. A. Smith, 1997: Optimal filtering in Singular Spectrum Analysis. Phys. Lett., 234 , 419–428.
Andronova, N. G., and M. E. Schlesinger, 2000: Causes of global temperature change during the 19th and 20th centuries. Geophys. Res. Lett., 27 , 2137–2140.
Cavalieri, D. J., 2002: A link between Fram Strait sea ice export and atmospheric planetary wave phase. Geophys. Res. Lett., 29 .1614, doi:10.1029/2002GL014684.
Cheng, W., R. Bleck, and C. Rooth, 2004: Multi-decadal thermohaline variability in an ocean-atmosphere general circulation model. Climate Dyn., 22 , 573–590.
Crowley, T. J., 1992: North Atlantic deep water cools the southern hemisphere. Paleoceanography, 7 , 489–497.
Dai, A., A. Hu, G. A. Meehl, W. M. Washington, and W. G. Strand, 2005: Atlantic thermohaline circulation in a coupled general circulation model: Unforced variations versus forced changes. J. Climate, 18 , 3270–3293.
Delworth, T. L., and M. E. Mann, 2000: Observed and simulated multidecadal variability in the Northern Hemisphere. Climate Dyn., 16 , 661–676.
Delworth, T. L., S. Manabe, and R. J. Stouffer, 1993: Interdecadal variability of the thermohaline circulation in a coupled ocean–atmosphere model. J. Climate, 6 , 1993–2011.
Delworth, T. L., S. Manabe, and R. J. Stouffer, 1997: Multidecadal climate variability in the Greenland Sea and surrounding regions: A coupled model simulation. Geophys. Res. Lett., 24 , 257–260.
Deser, C., and M. L. Blackmon, 1995: On the relationship between tropical and North Pacific sea surface temperature variations. J. Climate, 8 , 1677–1680.
Dickson, R. R., H. H. Lamb, S-A. Malmberg, and J. M. Colebrook, 1975: Climate reversal in the northern North Atlantic. Nature, 256 , 479–481.
Dickson, R. R., J. Meincke, S-A. Malmberg, and A. J. Lee, 1988: The “great salinity anomaly” in the northern Atlantic 1968–1982. Progress in Oceanography, 20 , Pergamon. 103–151.
Dima, M., and G. Lohmann, 2004: Fundamental and derived modes of climate variability. Concept and application to interannual variability. Tellus, 56A , 229–249.
Dima, M., N. Rimbu, S. Stefan, and I. Dima, 2001: Quasi-decadal variability in the Atlantic basin involving Tropics–midlatitudes and ocean–atmosphere interactions. J. Climate, 14 , 823–832.
Eden, C., and T. Jung, 2001: North Atlantic interdecadal variability: Oceanic response to the North Atlantic Oscillation (1865–1997). J. Climate, 14 , 676–691.
Enfield, D. B., and A. M. Mestas-Nuñez, 1999: Multiscale variabilities in global sea surface temperatures and their relationships with tropospheric climate patterns. J. Climate, 12 , 2719–2733.
Enfield, D. B., A. M. Mestas-Nunez, and P. J. Trimble, 2001: The Atlantic multidecadal oscillation and its relation to rainfall and river flows in the continental U.S. Geophys. Res. Lett., 28 , 2077–2080.
Evans, M. N., M. A. Cane, D. P. Schrag, A. Kaplan, B. K. Linsley, R. Villalba, and G. M. Wellington, 2001: Support for tropically driven Pacific decadal variability based on paleoproxy evidence. Geophys. Res. Lett., 28 , 3689–3692.
Folland, C. K., T. N. Palmer, and D. E. Parker, 1986: Sahel rainfall and worldwide sea temperatures, 1901–85. Nature, 320 , 602–607.
Folland, C. K., J. A. Owen, N. M. Ward, and A. W. Coleman, 1991: Prediction of seasonal rainfall in the Sahel region using empirical and dynamical methods. J. Forecasting, 10 , 21–56.
Frauenfeld, O. W., R. E. Davis, and M. E. Mann, 2005: A distinctly interdecadal signal of Pacific Ocean–atmosphere interaction. J. Climate, 18 , 1709–1718.
Gildor, H., and E. Tziperman, 2003: Sea-ice switches and abrupt climate change. Philos. Trans. Roy. Soc. London, B361 , 1935–1944.
Goldenberg, S. B., C. W. Landsea, A. M. Mestas-Nuñez, and W. M. Gray, 2001: The recent increase in Atlantic hurricane activity: Causes and implications. Science, 293 , 474–479.
Gray, S. T., L. J. Graumlich, J. L. Betancourt, and G. T. Pederson, 2004: A tree-ring based reconstruction of the Atlantic Multidecadal Oscillation since 1567 A.D. Geophys. Res. Lett., 31 .L12205, doi:10.1029/2004GL019932.
Hakkinen, S., 1993: An Arctic source for the great salinity anomaly: A simulation of the Arctic ice-ocean system for 1955–1975. J. Geophys. Res., 98 , 16397–16410.
Häkkinen, S., 1999: A simulation of thermohaline effects of a great salinity anomaly. J. Climate, 12 , 1781–1795.
Häkkinen, S., and C. A. Geiger, 2000: Simulated low-frequency modes of circulation in the Arctic Ocean. J. Geophys. Res., 105 , 6549–6564.
Hasumi, H., and N. Suginohara, 1995: Haline circulation induced by formation and melting of sea ice. J. Geophys. Res., 100 , 20613–20625.
Hegerl, G. C., K. Hasselmann, U. Cubasch, J. F. B. Mitchell, E. Roeckner, R. Voss, and J. Waskewitz, 1997: Multi-fingerprint detection and attribution analysis of greenhouse gas, greenhouse gas-plus-aerosol, and solar forced climate change. Climate Dyn., 13 , 613–634.
Hilmer, M., M. Harder, and P. Lemke, 1998: Sea ice transport: A highly variable link between Arctic and North Atlantic. Geophys. Res. Lett., 25 , 3359–3362.
Kaplan, A., M. A. Cane, Y. Kushnir, A. C. Clement, M. B. Blumenthal, and B. Rajagopalan, 1998: Analyses of global sea surface temperature 1856–1991. J. Geophys. Res., 103 , 27835–27860.
Kaplan, A., Y. Kushnir, and M. A. Cane, 2000: Reduced space optimal interpolation of historical marine sea level pressure: 1854–1992. J. Climate, 13 , 2987–3002.
Kerr, R. A., 2000: A North Atlantic pacemaker for the centuries. Science, 288 , 1984–1986.
Knight, J. R., R. J. Allan, C. K. Folland, M. Vellinga, and M. E. Mann, 2005: A signature of persistent natural thermohaline circulation cycles in observed climate. Geophys. Res. Lett., 32 .L20708, doi:10.1029/2005GL024233.
Kushnir, Y., 1994: Interdecadal variations in North Atlantic sea surface temperature and associated atmospheric conditions. J. Climate, 7 , 141–157.
Kushnir, Y., and I. M. Held, 1996: Equilibrium atmospheric response to North Atlantic SST anomalies. J. Climate, 9 , 1208–1220.
Latif, M., and Coauthors, 2004: Reconstructing, monitoring, and predicting multidecadal-scale changes in the North Atlantic thermohaline circulation with sea surface temperature. J. Climate, 17 , 1605–1614.
Lohmann, G., 2003: Atmospheric and oceanic freshwater transport during weak Atlantic overturning circulation. Tellus, 55A , 438–449.
Lohmann, G., and R. Gerdes, 1998: Sea ice effects on the sensitivity of the thermohaline circulation. J. Climate, 11 , 2789–2803.
Lohmann, G., N. Rimbu, and M. Dima, 2004: Climate signature of solar irradiance variations: Analysis of long-term instrumental and historical data. Int. J. Climatol., 24 , 1045–1056.
Lorenz, E. N., 1956: Empirical orthogonal functions and statistical weather prediction. Statistical Forecasting Project Scientific Rep. 1, Defense Doc. Center 110268, Massachusetts Institute of Technology, Cambridge, MA, 49 pp.
Manabe, S., and R. J. Stouffer, 1988: Two stable equilibria of a coupled ocean–atmosphere model. J. Climate, 1 , 841–866.
Mann, M. E., and J. Park, 1994: Global-scale modes of surface temperature variability on interannual to century timescales. J. Geophys. Res., 99 , 25819–25833.
Mann, M. E., J. Park, and R. S. Bradley, 1995: Global interdecadal and century-scale climate oscillations during the past five centuries. Nature, 378 , 266–268.
Mauritzen, C., and S. Häkkinen, 1997: Influence of sea ice on the thermohaline circulation in the Arctic-North Atlantic Ocean. Geophys. Res. Lett., 24 , 3257–3260.
Mestas-Nuñez, A. M., and D. B. Enfield, 1999: Rotated global modes of non-ENSO sea surface temperature variability. J. Climate, 12 , 2734–2746.
Minobe, S., 1997: A 50-70 year climatic oscillation over the North Pacific and North America. Geophys. Res. Lett., 24 , 683–686.
Minobe, S., 1999: Resonance in bidecadal and pentadecadal climate oscillations over the North Pacific: Role in climatic regime shifts. Geophys. Res. Lett., 26 , 855–858.
Mysak, L. A., D. K. Manak, and R. F. Marsden, 1990: Sea-ice anomalies observed in the Greenland and Labrador Seas during 1901–1984 and their relation to an interdecadal Arctic climate cycle. Climate Dyn., 5 , 111–133.
Nakamura, H., G. Lin, and T. Yamagata, 1997: Decadal climate variability in the North Pacific during recent decades. Bull. Amer. Meteor. Soc., 78 , 2215–2225.
Nitta, T., and S. Yamada, 1989: Recent warming of tropical sea surface temperature and its relationship to the Northern Hemisphere circulation. J. Meteor. Soc. Japan, 67 , 375–383.
North, G. R., T. L. Bell, R. F. Cahalan, and F. J. Moeng, 1982: Sampling erorrs in the estimation of empirical orthogonal functions. Mon. Wea. Rev., 110 , 699–706.
Park, W., and M. Latif, 2005: Ocean dynamics and the nature of air–sea interactions over the North Atlantic at decadal timescales. J. Climate, 18 , 982–995.
Peng, S., and W. A. Robinson, 2001: Relationship between atmospheric internal variability and the response to an extratropical SST anomaly. J. Climate, 14 , 2943–2959.
Pohlmann, H., F. Sienz, and M. Latif, 2006: Influence of the multidecadal Atlantic meridional overturning circulation variability on European climate. J. Climate, 19 , 6062–6068.
Preisendorfer, R. H., 1988: Principal Component Analysis in Meteorology and Oceanography. Elsevier, 425 pp.
Rigor, I. G., J. M. Wallace, and R. L. Colony, 2002: Response of sea ice to the Arctic Oscillation. J. Climate, 15 , 2648–2663.
Schlesinger, M. E., and N. Ramankutty, 1994: An oscillation in the global climate system of period 65–70 years. Nature, 367 , 723–726.
Schlesinger, M. E., and N. Ramankutty, 1995: Is the recently reported 65- to 70-year surface-temperature oscillation the result of climatic noise? J. Geophys. Res., 100 , 13767–13774.
Schmith, T., and C. Hansen, 2003: Fram Strait ice export during the nineteenth and twentieth centuries reconstructed from a multiyear sea ice index from southwestern Greenland. J. Climate, 16 , 2782–2791.
Schneider, N., A. J. Miller, and D. W. Pierce, 2002: Anatomy of North Pacific decadal variability. J. Climate, 15 , 586–605.
Seager, R., Y. Kushnir, N. Naik, M. A. Cane, and J. A. Miller, 2001: Wind-driven shifts in the latitude of the Kuroshio–Oyashio Extension and generation of SST anomalies on decadal timescales. J. Climate, 14 , 4249–4265.
Stocker, T. F., 1998: The seesaw effect. Science, 282 , 61–62.
Stocker, T. F., and L. A. Mysak, 1992: Climatic fluctuations on the century time scale: A review of high-resolution proxy data and possible mechanisms. Climatic Change, 20 , 227–250.
Sutton, R. T., and L. R. Hodson, 2005: Atlantic Ocean forcing of North American and European summer climate. Science, 309 , 115–118.
Thompson, D. W. J., and J. Wallace, 2000: Annular modes in the extratropical circulation. Part I: Month-to-month variability. J. Climate, 13 , 1000–1016.
Timmermann, A., M. Latif, R. Voss, and A. Grötzner, 1998: Northern Hemisphere interdecadal variability: A coupled air–sea mode. J. Climate, 11 , 1906–1931.
Trenberth, K. E., and D. A. Paolino Jr., 1980: The Northern Hemisphere sea-level pressure data set: Trends, errors and discontinuities. Mon. Wea. Rev., 108 , 855–872.
Trenberth, K. E., and J. W. Hurrell, 1994: Decadal atmosphere-ocean variations in the Pacific. Climate Dyn., 9 , 303–319.
Vautard, R., P. Yiou, and M. Ghil, 1992: Singular-spectrum analysis: A toolkit for short, noisy chaotic signals. Physica D, 58 , 95–126.
Vellinga, M., and R. A. Wood, 2002: Global climatic impacts of a collapse of the Atlantic thermohaline circulation. Climatic Change, 54 , 251–267.
von Storch, H. V., and F. W. Zwiers, 1999: Statistical Analysis in Climate Research. Cambridge University Press, 484 pp.
White, W. B., and D. R. Cayan, 2000: A global El-Niño Southern Oscillation wave in surface temperature and pressure and its interdecadal modulation from 1900 to 1997. J. Geophys. Res., 105 , 11223–11242.
Wu, L., Z. Liu, and R. Gallimore, 2003: Pacific decadal variability: The tropical Pacific mode and the North Pacific mode. J. Climate, 16 , 1101–1120.
Zhang, Y., J. M. Wallace, and D. S. Battisti, 1997: ENSO-like interdecadal variability: 1900–1993. J. Climate, 10 , 1004–1020.
(a) The AMO index derived based on North Atlantic SSTs (0°–60°N, 75°–7.5°W). A time series is obtained as an average of linearly detrended annual mean SST anomalies (Kaplan et al. 1998; see text for details). (b) The regression map of the detrended annual mean SST anomalies on the normalized AMO index.
Citation: Journal of Climate 20, 11; 10.1175/JCLI4174.1
(a) The fourth EOF of winter SLP anomalies from Trenberth and Paolino (1980) over the 1899–2002 period and (b) the associated time component (thin line) and its dominating component (thick line) identified through SSA.
Citation: Journal of Climate 20, 11; 10.1175/JCLI4174.1
Singular spectrum analysis of the reconstructed Fram Strait sea ice export (km3 yr−1) for the 1820–2000 period: (a) annual reconstruction (thin line) and its 7-yr running mean filtered version; (b) residual Fram Strait sea ice export (thin line) obtained after removing the long-term centennial component; (c) eigenvalue spectrum of the SSA performed on the residual record using a window of 80 yr; (d) time EOFs 1 and 2 resulting from SSA, which correspond to the first 2 points in the eigenvalue spectrum; and (e) reconstructed multidecadal component associated with the first two T-EOFs.
Citation: Journal of Climate 20, 11; 10.1175/JCLI4174.1
(a) SST and (b) SLP maps of the strongest coupled pair obtained through canonical correlation analysis for the North Atlantic sector (0°–70°N, 80°–20°W); prior to the analysis the linear trend is removed and a 25-yr running mean filter is applied. (c) Time components associated with SST (solid line) and SLP (dotted line) patterns in (a), (b).
Citation: Journal of Climate 20, 11; 10.1175/JCLI4174.1
As in Fig. 4, but for the second strongest coupled pair for the North Pacific sector (0°–65°N, 120°E–90°W).
Citation: Journal of Climate 20, 11; 10.1175/JCLI4174.1
SST (°C) and SLP (hPa) composite maps constructed based on the AMO index (Fig. 1a). The SST composite maps are calculated so that the SST field lags the AMO index by (a) 0, (c) 5, (e) 10, and (g) 15 yr; similarly, the SLP field lags the index by (b) 0, (d) 5, (f) 10, and (h) 15 yr.
Citation: Journal of Climate 20, 11; 10.1175/JCLI4174.1
As in Fig. 6, but for (a) 20-, (c) 25-, (e) 30-, and (g) 35-yr lag.
Citation: Journal of Climate 20, 11; 10.1175/JCLI4174.1
Schematic representation of the AMO mechanism. The thick arrows represent links that are associated with memory: THC adjustment to freshwater fluxes (horizontal line), Atlantic SST response to meridional overturning, oceanic adjustment in the North Pacific, and sea ice response to wind stress.
Citation: Journal of Climate 20, 11; 10.1175/JCLI4174.1