data. Over the Southern Hemisphere there is more agreement between the long-time-scale seasonal trends.
In both hemispheres, greater year-to-year variability is apparent during the nineteenth century. This increased variability is due to the sparser spatial coverage at that time. Although the estimates for individual years may be less reliable in the nineteenth than the twentieth century, the "frozen-grid" analyses undertaken by Jones et al. (1986a,c) indicate that the decadal-scale temperature fluctuations are reliably reproduced by the available data (for more discussion of this see Madden et al., 1993). The coolness of the 1880s compared with 1851 to 1880 in the Northern Hemisphere is, therefore, probably real. Thus, for the hemispheric temperature analyses of Hansen and Lebedeff (1987, 1988) and Vinnikov et al. (1990), who begin their analyses around 1880, warming rates measured over 1881 to 1990 are slightly greater than rates calculated over 1861 to 1990 (see also Folland et al. (1990, 1992) for more discussion of this point).
Since land represents only 29 percent of the area of the earth's surface, it is important to incorporate marine data into hemispheric averages if we want to get the "best possible" global series. Merchant and naval ships have taken weather observations and measured the temperature of the sea surface since the beginning of the nineteenth century. In the last 20 years, major international efforts have been made to transfer all of the climate data contained in ships' log books into computer data banks. One such compilation is the Comprehensive Ocean-Atmosphere Data Set (COADS) produced by NOAA workers at Boulder, Colorado (Woodruff et al., 1987). COADS contains about 80 million non-duplicated sea surface temperature (SST) observations. Another similar data set has been assembled by the U.K. Meteorological Office (Bottomley et al., 1990). Much of the data is common to both sets, but comparisons are under way to isolate the unique observations in each.
Unfortunately, as with land data, marine records are affected by inhomogeneities and errors. Correction schemes have been devised to adjust both marine air temperatures and sea surface temperatures for biases attributable to the method of measurement, time of day, etc. (see, e.g., Bottomley et al., 1990; Folland and Parker, 1990, 1991; Folland et al., 1990, 1992; Jones and Wigley, 1990; and Jones et al., 1991).