Spatial patterns of Atlantic Meridional Overturning Circulation variability

Variability of the Atlantic Meridional Overturning Circulation (AMOC) is usually represented by a simple index, typically the maximum meridional overturning streamfunction value at a certain latitude in the North Atlantic, or the principle components associated with stationary EOFs of the meridional overturning streamfunction field. However, natural variation of AMOC is known to exhibit different spatial patterns of behavior, suggesting the inadequacy of using a single index. There is an analogy with the behavior exhibited in the tropical Pacific ocean-atmosphere variability, where the Nino3.4 index was used to represent the state of the El Nino-Southern Oscillation, but actually combines two coupled ocean-atmosphere behaviors now known to be distinct - the so called ‘cold tongue’ and ‘central Pacific’ El Ninos. Following a method previously used to extract tropical ocean-atmosphere coupled modes of variability, we apply maximum covariance analysis (MCA) to understand the relationship between AMOC forcing - here upper-ocean temperature (T) and salinity (S) over the high-latitude North Atlantic – and AMOC response, in a long control simulation of the Community Earth System Model version 1 (CESM1). Two distinct MCA modes of significance are found, the first characterized by a basin-wide meridional overturning pattern that leads changes to T and S in the North Atlantic. The second is a forced response whereby T and S changes in the North Atlantic leads to a meridional overturning response localized over the North Atlantic. The ‘localized’ MOC response leads the ‘basinwide’ MOC response by a few years, suggesting that the former evolves to the latter response; indeed, this behavior is seen in the transient response of CESM1 to an NAO-like buoyancy forcing in the North Atlantic. A linear combination of the two modes accounts for most of a standard AMOC index, suggesting that the AMOC index captures a combination of these two behaviors. Application of the same method to a control simulation of the GFDL-ESM2M coupled model shows that the ‘basinwide’ pattern dominates, indicating the very different nature of AMOC variability in that model.