northern Urals set) interval rather than annual periods is considered.
The two summer temperature reconstructions for South America (Figure 8) are of widely different lengths. A comparison of the overlap period indicates that the Boninsegna series (see Table 1) exhibits a warming trend over the period of record. By contrast, the longer Villalba series exhibits some low-frequency variations but little if any trend over both the full data period and that of the Boninsegna period, beginning about the mid-fifteenth century.
Because of the longer record available, a variability index (described in the section on instrumental data above) was computed using the decadal averages, instead of the annual values. Our purpose here is to examine interdecadal temperature variability, with records spanning several centuries, noting that climate-sensitive tree-ring records are particularly useful for studying decadal-scale climatic fluctuations (Fritts, 1991). The changes in reconstructed climatic vari-
ability have been summarized on a century-by-century basis in Table 2. For comparison, we also show in Table 3 the mean values of interannual and interdecadal variability for the full-length record. The variability index was calculated by applying equation (1) to both the annual and decadal-average reconstructed series. We fail to see any consistent trends in interdecadal variability associated with these tree-ring temperature reconstructions (see indices 5-9, 12-13, and 16-18 in Table 2). Interdecadal variability is typically about half the interannual values. This implies that substantial low-frequency variance is present in the paleotemperature record, so that recent high values of reconstructed decadal-mean summer temperature may yet represent an oscillation within the range of natural variability.
The oxygen-isotope record has not been directly calibrated against temperature, as has been done for the tree-ring record. However, it is well established that the oxygen-isotope ratio (d18O) is dependent primarily on the temperature of formation of the precipitation, with increasingly negative d18O ratios associated with decreasing temperature. The problems lie in the interpretation of a local ice-core record, not only with respect to local temperature variations, but even in terms of the larger-scale temperature patterns. Several factors control the oxygen-isotope composition of the snow that falls on a given ice body, and temperature is only one of them. The sensitivity of these records to annual temperature variations has been discussed in a number of papers, including the source articles referred to in Table 1. Whether or not these oxygen-isotope records accurately represent a "local" temperature record, we consider them to be sensitive indicators of prevailing climatic conditions within a suitably broad source region. Since our stated purpose is to compare recent changes in a suite of climate-sensitive paleoindicators, we have included them in our comparisons.
Figures 9 and 10 illustrate the changes in decadal averages of the oxygen-isotope ratio of glacier ice cores for different locations in northern Canada and Greenland, and for two low-latitude, high-elevation glaciers (the Quelccaya ice cap in the Peruvian Andes and the Dunde ice cap in the Tibetan Plateau region of China (see Table 1 for sources)). The polar ice d18O record, which ends before 1970, shows relatively less warming (trend toward smaller negative values of d18O) than do the tropical ice cores. Furthermore, the warmest decades in the tropical record occur in the most recent time. The relation of this tropical d18O signal to air-temperature changes in the general location of these records may be partially evaluated with independent observations. It is known that significant melting took place at the Quelccaya site during the 1980s (Thompson, personal communication). Increases in tropospheric