The IPCC Spaghetti-Diagram Reconstructions of Paleoclimate are Incoherent With Each Other
Bürger, G. 2010. Clustering climate reconstructions. Climate of the Past Discussions 6: 659-679.
Reconstructions of past climate thus have a central role to play in the global warming debate. In past IPCC reports, for example, the reconstructions in the spaghetti graph of past climate (by authors such as Mann, Bradley, Hughes, Jones, and Esper, see figure below) are claimed to be "remarkably consistent". Bürger, however, decided that this opinion needed to be more rigorously evaluated; and he took it upon himself to perform that evaluation.
IPCC spaghetti graph from Fig. 6.10 of Working Group 1.
Working with eight graphs from the IPCC and adding two more, he determined that the calibration process during the instrumental period would bias the degree of agreement, since the graphs were all fixed to largely agree during this period, so he only examined the period before 1850. In order to examine the shapes of the curves rather than arbitrary offsets, he rescaled them all to unit variance and centered them on zero, after which he computed the spectral coherence of each pair, and then -- from the similarity matrix -- conducted a clustering analysis.
The result was that five clusters were formed by the ten reconstructions, with three in the largest and one in the smallest cluster. Members within a cluster were similar at the 95% confidence level, based on standard tests. All of the clusters, however, were significantly incoherent with each other. In fact, they were incoherent not merely at some points, but at virtually all timescales of fluctuation, from decadal to centennial oscillations, which means that it is not meaningful to speak of somehow "averaging" the different reconstructions, whether by eye or numerically, because the incoherence will lead to a canceling out of the supposed climate signals in each, leaving merely a close-to-flat line.
This incoherence means that one cannot claim that the different temperature reconstructions are all "right" or "agree" in any sense of the word; and attempts to use these reconstructions for attribution studies or to calibrate climate models will give different results for any particular choice of reconstruction. In fact, the results of this study suggest that the reconstructions differ so much that there is no way to draw meaningful conclusions from them, nor can it be determined which one/ones is/are right.