Monday, November 28, 2005
A study by Mann et al (J. Clim. 18, 4097 (2005) indicates that the two most common statistical techniques used to reconstruct temperature data do not systematically underestimate temperature variability and accurately estimate temperature history. The question was significant because:
"Reconstructing a temperature record for the past from proxy data (e.g., tree rings, corals, and ice cores) is difficult because proxies are imperfect thermometers, and the noise that contaminates the temperature signal can introduce large uncertainties into any estimate. The two most common statistical techniques used to interpret these noisy data sets are the climate field reconstruction (CFR, well suited for spatial patterns) and composite-plus-scale (CPS, with a simpler statistical procedure) methods. Evaluating the fidelity of those approaches is difficult, however, because the direct observational temperature record is too short and too incomplete to allow them to be verified thoroughly."
So Mann used climate models to provide temperature outputs that were long and geographically complete and then tested the CFR and CPS methods, using a virtual climate record that is essentially perfect.