On the physics of the Atlantic Multidecadal Oscillation

The Atlantic Multidecadal Oscillation (AMO) is a pronounced signal of climate variability in the North Atlantic sea-surface temperature field. In this paper, we propose an explanation of the physical processes responsible for the timescale and the spatial pattern of the AMO. Our approach involves the analysis of solutions of a hierarchy of models. In the lowest member of the model hierarchy, which is an ocean-only model for flow in an idealized basin, the variability shows up as a multidecadal oscillatory mode which is able to destabilize the mean thermohaline circulation. In the highest member of the model hierarchy, which is the Geophysical Fluid Dynamics Laboratory R30 climate model, multidecadal variability is found as a dominant statistical mode of variability. The connection between both results is established by tracing the spatial and temporal expression of the multidecadal mode through the model hierarchy while monitoring changes in specific quantities (mechanistic indicators) associated with its physics. The proposed explanation of the properties of the AMO is eventually based on the changes in the spatial patterns of variability through the model hierarchy. Introduction The North Atlantic sea-surface temperature (SST) appears to have a distinct signal of multidecadal variability (Bjerknes 1964; Folland et al. 1984; Schlesinger and Ramankutty 1994). The difference of the SST pattern between the relatively warm years 1950–1964 and the relatively cold years 1970–1984 shows negative anomalies near Newfoundland and positive anomalies over the rest of the basin (Kushnir 1994). By subsequent analysis of longer and better quality SST and sea-level pressure (SLP) data, the pattern of multidecadal variability has been characterized more accurately (Moron et al. 1998; Tourre et al. 1999). Delworth and Mann (2000) extended the instrumental record with proxy data and demonstrated that there is a significant spectral peak in the 50to 70-year frequency band. The variability was named the Atlantic Multidecadal Oscillation (AMO) in Kerr (2000), and an AMO index was defined in Enfield et al. (2001) as the 10-year running mean of the detrendedAtlantic SSTanomalies north of the equator. Although detailed mechanisms are not clear in most cases, there appear to be climate variations which are well correlated with the AMO. Enfield et al. (2001) showed that there is a significant negative correlation with US continental rainfall, with less (more) rain over most of the central USA during a positive (negative) AMO index period. For example, the Mississippi outflow is about 5% less than average during a positive AMO phase. High positive correlations have been found between the AMO and Sahel rainfall and between the AMO and hurricane intensity in the Atlantic (Gray et al. 1997). During the positive AMO index period 1950–1964, there were 47 intense (class 3, 4, 5) hurricanes originating east of 60°W. In the same length (negative AMO index) period 1970– 1984, there were only 19. Multidecadal variability has also been identified in long-term observations of sea-ice concentration in the Arctic (Venegas and Mysak 2000). The periods where the sea-ice concentration is lower (higher) than average roughly coincide with periods of positive (negative) AMO index. Recently, Sutton and Hodson (2005) investigated the northern hemispheric Responsible Editor: Tal Ezer H. A. Dijkstra (*) . L. te Raa Institute for Marine and Atmospheric Research Utrecht, Utrecht University, 3766 HW Utrecht, The Netherlands e-mail: [email protected] M. Schmeits Royal Netherlands Meteorological Institute, De Bilt, The Netherlands J. Gerrits College of Oceanic and Atmospheric Sciences, Oregon State University, Corvallis, OR, USA climate impacts of the AMO and demonstrated that the difference in summer precipitation pattern between positive and negative AMO states shows a 5–15% increase over Western Europe. The difference pattern in atmospheric surface temperatures shows warm anomalies over the USA and over central Europe. North Atlantic multidecadal climate variability has also been found in coupled ocean–atmosphere models. Delworth et al. (1993) analyzed a 600-year simulation of the Geophysical Fluid Dynamics Laboratory (GFDL) R15 climate model. They found an average period of about 50 years in the strength of the thermohaline circulation (THC) as measured by a THC index, i.e., the maximum of the Atlantic’s meridional overturning streamfunction. The SST difference between highand low-THC-index states is in reasonable agreement with the pattern derived from observations (Kushnir 1994). Further analysis of the GFDL-R15 results (Delworth et al. 1993) showed that the multidecadal variability is associated with a coupling between the density anomalies in the sinking region (of the THC) and the strength of the Atlantic meridional overturning circulation. The GFDL-R15 model is not the only coupled general circulation model (GCM) in which Atlantic multidecadal variability has been found. In the Max Planck Institute Ocean Model 1 (MPI-OM1) (Latif et al. 2004), irregular variability is found with a mean period of about 50 years and an amplitude in SST about twice as large as in observations. The spatial pattern of SST anomalies (Fig. 2b in Latif et al. 2004) is similar to that in observations and in the GFDL-R15 model. Dong and Sutton (2005) found multidecadal variability in the HadCM3 model with an average period of about 24 years. Analysis of the lag correlation between different fields (SST, mixed layer depth, salinity anomalies, etc.) leads to the description of a similar mechanism as that in Delworth et al. (1993), although the timescale of variability is about half as that in the GFDL-R15 model. A nice set of sensitivity simulations with the GFDL-R15 model was presented by Delworth and Greatbatch (2000). They showed that when the ocean model is forced by the climatological seasonal cycle of surface fluxes, no multidecadal variability is found. When the ocean model is forced by annual-mean surface fluxes (allowing atmospheric “noise” in the surface fluxes), the variability is similar to that of the coupled model. The heat flux is shown to be the essential component of the surface fluxes causing the variability. It is concluded that the multidecadal variability can be attributed to “a damped mode in the ocean system, which is continuously excited by low-frequency atmospheric forcing.” Although this is an important result, it provides neither a mechanistic description for the physical processes setting the timescale of the AMO nor an explanation for its spatial pattern. The results in Delworth et al. (1993) and Delworth and Greatbatch (2000) suggest that the essential physics of the multidecadal variability can be understood from oceanonly models. Indeed, multidecadal oscillations have been found in many idealized sector models of the threedimensional ocean circulation, forced by only a surface heat flux (Greatbatch and Zhang 1995; Greatbatch and Petersen 1996; Chen and Ghil 1996). From these studies, it appears that a phase difference between changes in the meridional overturning circulation and the resulting temperature anomalies is essential to the variability. Moreover, critical thresholds also exist: for example, when the horizontal mixing of heat is too large, no multidecadal oscillations are found. By studying the sensitivity of the multidecadal oscillations to the representation of convective adjustment, the thermal and momentum boundary conditions, the coupling to the atmosphere, and the representation of mesoscale variability, Huck and coworkers (Colin de Verdière and Huck 1999; Huck et al. 1999) showed that the multidecadal oscillations exist under a wide range of conditions. Both Huck and Vallis (2001) and Te Raa and Dijkstra (2002) concluded that the threshold behavior is associated with a Hopf bifurcation, in which a multidecadal mode destabilizes the thermally driven steady flow. The physical mechanism of growth and propagation of this mode was described from the patterns near the Hopf bifurcation (Te Raa and Dijkstra 2002). The multidecadal mode also exists in a coupled ocean–atmosphere model, but it is damped over the volume in parameter space investigated (Te Raa and Dijkstra 2003b). Te Raa et al. (2004) studied finite amplitude transient flows displaying multidecadal variability in the singlehemispheric basin configuration with and without “realistic” continental boundaries. They showed that in situations with continental geometry, the variability still originates from the same physical processes as in the idealized model, although the patterns of variability become strongly deformed. The work in Te Raa and Dijkstra (2002) provides strong motivation for trying to connect the variability in the ocean-only models to that found in coupled GCMs such as the GFDL climate model (Delworth et al. 1993; Delworth and Greatbatch 2000). In this paper, we attempt to establish such a connection and propose an explanation for the timescale and spatial pattern of multidecadal variability in these models and in observations. The main idea pursued here is to trace the spatial and temporal characteristics of the multidecadal mode in Te Raa and Dijkstra (2002) through a model hierarchy while monitoring so-called mechanistic indicators to see if changes in the physics of the variability occur. First, we first analyze the multidecadal variability in the results of a 900-year simulation with the GFDL-R30 climate model. Next, we start with the results in Te Raa and Dijkstra (2002) and shortly recall the physical mechanism of the multidecadal mode as a necessary step to introduce the mechanistic indicators. Subsequently, the indicators are used to explain the physics of multidecadal variability in finite amplitude ocean flows. The connection between these results leads to the proposed explanation of the AMO which is discussed in the last section. Statistical modes in the GFDL-R30 model The GFDL-R30 model consists of general circulation models of both the atmosphere and ocean, with relatively simple formulations of land-surface and sea-ice processes (Delworth et al. 2002). The atmospheric portion of the coupled model solves the primitive equations on the sphere using a spectral technique with rhomboidal 30 truncation, which corresponds to a resolution of approximately 3.75° longitude by 2.25° latitude. There are 14 unevenly spaced levels in the vertical. The ocean component of the coupled model uses version 1.1 of the Modular Ocean Model (MOM). The resolution of this component is 1.875° longitude by 2.25° latitude, with 18 unevenly spaced levels in the vertical. The governing equations are solved numerically with the use of the Boussinesq, rigid-lid, and hydrostatic approximations; flux correction is applied to limit drift in the model. The model output used in this study [generously provided to us by Tom Delworth (GFDL, Princeton)] is from the most stable control run of the GFDL-R30 model as described in Delworth et al. (2002). Although the model was integrated for 900 years, we have only analyzed the last 500 years (years 401–900) because of substantial drift in the first 400 years of the simulation. Fields analyzed were the annual-mean potential temperature, potential density, salinity, and horizontal velocity at five vertical levels [level 1 (20 m), level 6 (301 m), level 9 (683 m), level 12 (1,456 m), and level 15 (2,798 m)] and the meridional overturning in the Atlantic. The meridional overturning streamfunction (Ψ) is computed from: