On the statistical significance of correlations between synthetic mantle plumes and tomographic models

In a recent article, [Boschi, L., Becker, T.W., Stemberger, B., 2007. Mantle plumes: dynamic models and seismic images. Geochem. Geophys. Geosyst. 8, Q10006. doi:10.1029/2007GC001733] (BBS07) have reevaluated the degree to which slow seismic tomography anomalies correlate with the possible locations of plume-like mantle upwellings connected to surface hotspots. They showed that several, but not all, hotspots are likely to have a deep mantle origin. importantly, they found that when advection of plume conduits in mantle flow is considered, such correlations are significantly higher than when conduits are assumed to be vertical under hotspots. The validity of these statements depends, however, on the definition of statistical significance. BBS07 evaluated the significance of correlation through simple Student's t tests. Anderson (personal communication, July 2007) questioned this approach, given that the true information content of published tomography models is generally unknown, and proposed, instead, to evaluate the significance of correlation by comparing tomographic results with Monte Carlo simulations of randomly located plumes. Following this approach, we show here that the correlation found by BBS07 between advected plumes and slow anomalies in S-velocity tomography is less significant than previously stated, but still significant (at the 99.7% confidence level). We also find an indication that the seismic/geodynamic correlation observed by BBS07 does not only reflect the natural tendency of plumes to cluster in slow/hot regions of the mantle: although realistically advected, and thereby biased towards such regions, our random plumes correlate with slow tomographic anomalies significantly less than the plume models of BBS07. A less significant correlation with plume models characterizes P-velocity tomography; the correlation is, however, enhanced, if flow is computed from tomographic models with amplified heterogeneity, possibly accounting for the known resolution limits of global seismic data. In summary, the conclusions of BBS07 are confirmed: even at relatively long wavelengths, tomographic models are consistent with the presence of a number of tilted, whole-mantle plume-shaped slow anomalies, connected to surface hotspots. (C) 2008 Elsevier B.V. All rights reserved.