correlations with surface temperature could lead to problems and should be done only if the proxy–temperature relationship has climatological justification.

• The observed discrepancy between some tree ring variables that are thought to be sensitive to temperature and the temperature changes observed in the late 20th century (Jacoby and D’Arrigo 1995, Briffa et al. 1998) reduces confidence that the correlation between these proxies and temperature has been consistent over time. Future work is needed to understand the cause of this “divergence,” which for now is considered unique to the 20th century and to areas north of 55°N (Cook et al. 2004).

• For tree ring chronologies, the process of removing biological trends from ring-width data potentially obscures information on long-term changes in climate.

• Temperature reconstructions for periods before about A.D. 1600 are based on proxies from a limited number of geographic regions, and some reconstructions are not robust with respect to the removal of proxy records from individual regions (see, e.g., Wahl and Ammann in press). Because the data are so limited, different large-scale reconstructions are sometimes based on the same datasets and thus cannot be considered as completely independent.

• Reconstructions of low-frequency variations in the temperature record that make use of proxies other than tree rings (Moberg et al. 2005b) are limited by the small number of available records, by dating uncertainties, and by the sensitivity of many proxies to hydrologic variables as well as to temperature. These data gaps highlight the need for continued coordinated efforts to collect proxy data over broad geographic regions.

Specifically concerning the reconstructed temperature variability over short time periods (year-to-decade scale), the committee identified the following as limitations that would benefit from further research:

• Large-scale surface temperature reconstructions demonstrate very limited statistical skill (e.g., using the CE statistic) for proxy sets before the 19th century (Rutherford et al. 2005, Wahl and Ammann in press). Published information, although limited, also suggests that these statistics are sensitive to the inclusion of small subsets of the data. Some of the more regionally focused reconstructions (D’Arrigo et al. 2006) have better demonstrated skill back to the 16th century or so, and possibly earlier. To improve the skill of reconstructions, more data need to be collected and possibly new assimilation methods developed.

• Accurately inferring the absolute values of temperature for single years and decades from proxies sensitive to variability at this timescale requires accurate reconstruction of the longer term mean.

Based on its deliberations and the materials presented in Chapters 1–11 and elsewhere, the committee draws the following overall conclusions regarding large-scale surface temperature reconstructions for the last 2,000 years:

• The instrumentally measured warming of about 0.6°C during the 20th century is also reflected in borehole temperature measurements, the retreat of glaciers, and other observational evidence, and can be simulated with climate models.