stratosphere. On the other hand, the very-low-frequency climate variations tend to be amplified in polar latitudes and have their greatest amplitude near the earth's surface.
The most unambiguous evidence for low-frequency climate variations with a period of several decades comes from surface temperature records over land. Figure 1 shows the historical temperature record averaged by latitude belts (Hansen and Lebedeff, 1987). Several features stand out in Figure 1. First, there is a clear upward trend in temperature over the past century, which is greatest in the higher latitudes of the Northern Hemisphere. Second, very-low-frequency
variations are superimposed on that trend. The most obvious features are pronounced minima at the beginning of the century and in the late 1960s and early 1970s. In between these two relatively cold events there was a period of relatively rapid warming in the 1920s. The most recent part of the record is not shown in Figure 1, but measurements indicate that Northern Hemisphere temperatures are rising rapidly in the 1980s, as they did in the 1920s (IPCC, 1990).
Bjerknes (1964) studied the pre-World War II climatic record, particularly the sea surface temperature archives of the British Meteorological Office. He was impressed by the relatively rapid rise of Northern Hemisphere temperature in the 1920s, which followed an anomalously cold period at the beginning of the century. In his analysis of the sea surface temperature and the surface atmospheric pressure fields at mid-latitudes he found a distinct difference between the fluctuations that had a time scale of seasons and years and the climatic fluctuations, which were on decadal time scales. Bjerknes concluded that the atmosphere played the dominant role in those fluctuations with a yearly time scale, while the ocean played the dominant role in the climatic fluctuations. He was aware of the heat-balance calculations carried out by Sverdrup (1957) and realized that the North Atlantic is much more important than the North Pacific in the poleward transport of heat in higher latitudes. He reasoned that fluctuations in the North Atlantic poleward heat transport could be the cause of decadal-scale Northern Hemisphere climate variations. Bjerknes' hypothesis received relatively little attention in 1964. Since so little was known about ocean circulation at the time, his ideas probably seemed to be difficult to check in any way. On the other hand, his explanation of the onset of the Southern Oscillation (Bjerknes, 1969) in terms of the Walker circulation was widely accepted within a few years after publication.
A few years previous to the publication of Bjerknes' study of the Atlantic, Stommel (1961) developed a simple model of the thermohaline circulation. He pointed out that air-sea interaction has a very different effect on the temperature field of the ocean than on the salinity field. The radiation balance of the atmosphere directly responds to the sea surface temperature field, while a similar feedback cannot exist for salinity. When this asymmetry was taken into account, Stommel demonstrated, his simple model of the thermohaline circulation could take on two stable solutions for the same external boundary conditions. In one solution the thermohaline solution was dominated by the thermal component of the density gradient, and in the other solution the salinity component of the density gradient was most important. In recent years it has been possible to test the concepts of Stommel's two-box thermohaline model in more complete two- and three-dimensional models (see Weaver and Hughes, 1992, for a review). Through these recent numerical