6.6.1.2 What Do Large-Scale Temperature Histories from Subsurface Temperature Measurements Show?

Hemispheric or global ground surface temperature (GST) histories reconstructed from measurements of subsurface temperatures in continental boreholes have been presented by several geothermal research groups (Huang et al., 2000; Harris and Chapman, 2001; Beltrami, 2002; Beltrami and Bourlon, 2004; Pollack and Smerdon, 2004); see Pollack and Huang (2000) for a review of this methodology. These borehole reconstructions have been derived using the contents of a publicly available database of borehole temperatures and climate reconstructions (Huang and Pollack, 1998) that in 2004 included 695 sites in the NH and 166 in the SH (Figure 6.11). Because the solid Earth acts as a low-pass filter on downward-propagating temperature signals, borehole reconstructions lack annual resolution; accordingly they typically portray only multi-decadal to centennial changes. These geothermal reconstructions provide independent estimates of surface temperature history with which to compare multi-proxy reconstructions. Figure 6.10b shows a reconstruction of average NH GST by Pollack and Smerdon (2004). This reconstruction, very similar to that presented by Huang et al. (2000), shows an overall warming of the ground surface of about 1.0°C over the past five centuries. The two standard error uncertainties for their series (not shown here) are 0.20°C (in 1500), 0.10°C (1800) and 0.04°C (1900). These are errors associated with various scales of areal weighting and consequent suppression of site-specific noise through aggregation (Pollack and Smerdon, 2004). The reconstruction is similar to the cooler multi-proxy reconstructions in the 16th and 17th centuries, but sits in the middle of the multi-proxy range in the 19th and early 20th centuries. A geospatial analysis by Mann et al. (2003; see correction by Rutherford and Mann, 2004) of the results of Huang et al. (2000) argued for significantly less overall warming, a conclusion contested by Pollack and Smerdon (2004) and Beltrami and Bourlon (2004). Geothermal reconstructions based on the publicly available database generally yield somewhat muted estimates of the 20th-century trend, because of a relatively sparse representation of borehole data north of 60°N. About half of the borehole sites at the time of measurement had not yet been exposed to the significant warming of the last two decades of the 20th century (Taylor et al., 2006; Majorowicz et al., 2004).

The assumption that the reconstructed GST history is a good representation of the surface air temperature (SAT) history has been examined with both observational data and model studies. Observations of SAT and GST display differences at daily and seasonal time scales, and indicate that the coupling of SAT and GST over a single year is complex (Sokratov and Barry, 2002; Stieglitz et al., 2003; Bartlett et al., 2004; Smerdon et al., 2006). The mean annual GST differs from the mean annual SAT in regions where there is snow cover and/or seasonal freezing and thawing (Gosnold et al., 1997; Smerdon et al., 2004; Taylor et al., 2006), as well as in regions without those effects (Smerdon et al., 2006). Observational time series of ground temperatures are not long enough to establish whether the mean annual differences are stable over long time scales. The long-term coupling between SAT and GST has been addressed by simulating both air and soil temperatures in global three-dimensional coupled climate models. Mann and Schmidt (2003), in a 50-year experiment using the GISS Model E, suggested that GST reconstructions may be biased by seasonal influences and snow cover variability, an interpretation contested by Chapman et al (2004). Thousand-year simulations by González-Rouco et al. (2003, 2006) using the ECHO-G model suggest that seasonal differences in coupling are of little significance over long time scales. They also indicate that deep soil temperature is a good proxy for the annual SAT on continents and that the spatial array of borehole locations is adequate to reconstruct the NH mean SAT. Neither of these climate models included time-varying vegetation cover.