Land and marine temperatures may be combined in a number of ways. We describe below the Climatic Research Unit's combined data set, comprising the land temperatures discussed earlier and SST anomalies corrected for changes in measurement technique. The main change is from uninsulated-bucket to engine-intake measurements (and some insulated buckets) around the start of the 1940s. The SST correction scheme is discussed in detail by Jones and Wigley (1990) and Jones et al. (1991). The data, which are given as a set of 5° by 5° grid-box values, are anomalies from the 1950-1979 reference period. Where co-located land and marine grid data occur, the resulting value is the average of the land and marine components.
Annual and seasonal time series of hemispheric-mean temperatures, based on the combined data set, are shown in Figure 2. Detailed discussions of the features exhibited by the two sets of curves may be found elsewhere (see, for example, Folland et al., 1990, 1992; Jones and Briffa, 1992; Jones and Wigley, 1990). The differences in temperature between the second half of the record (1946-90) and the first half (1901-45) are listed in Table 2. In many respects the year-to-year and decadal-scale variations in Figure 2 are damped versions of those seen in Figure 1. There are, however, subtle differences. The greater nineteenth-century variability prominent in the land-only data is not as apparent in the land-plus-marine data. The cooling between 1940 and 1970 evident in the Northern Hemisphere land-only seasonal series is hardly apparent in the combined data. It is replaced by a period of little trend in 1940-1960, followed by a slight cooling from 1960 to the mid-1970s. Variability in the combined data set is now similar between the two hemispheres, whereas in the land-only data the Northern Hemisphere variability was clearly greater.
The intercorrelation between the two hemispheric combined series is remarkably high (for annual values between 1901 and 1990, r = 0.79). Some of this high correlation is due to coincident long-term trends, and the remainder is due to common high-frequency (year-to-year) variability. Some of the interannual correspondence derives from a response to the common forcing of the El Niño/Southern Oscillation (ENSO) phenomenon (Philander, 1983). Many of the years that are warm relative to the filtered curves in Figure 2 have been shown to be El Niño years or "warm event" years (Bradley et al., 1987b; van Loon and Shea, 1985). In contrast, many cold years, relative to the filtered curves, correspond to La Niña or "cold event" years (Philander, 1985). Using the Southern Oscillation Index (SOI) (Allan et al., 1991; Ropelewski and Jones, 1987), Jones (1989) has removed the ENSO influence from the hemispheric and global temperature series. The SOI explains about 25 to 30 percent of the high-frequency (less than 20 years) variations in hemispheric temperature anomalies