An alternative way to overcome the problem of uncertainty in the values of the component fields is to employ a processing method referred to as correlation analysis (Vibert-Charbonnel, 1996; Louvel and Galbrun, 2000). The principle employed by the correlation analysis method (Pozzi et al., 1993) is illustrated in Figure F6. Both Jr and Ji are dependant on the concentration of ferrimagnetic minerals in the sediment, and this concentration varies. Jr and Ji correlate in normal polarity intervals, and the linear regression line has a positive gradient; whereas, in reversed polarity intervals, Jr is inversely correlated with Ji, and the linear regression line has a negative gradient. The same applies to Bfr and Bfi on which the correlation analyses are actually performed. The linear regressions are applied to successive depth intervals ("windows") of various thickness (1.5-, 2.5-, 4.5-, 8-, and 13-m windows are presented here). Linear regressions with a correlation coefficient <0.5 are not plotted in the results (Figs. F7, F8). Prior to the correlation analyses, Bfr and Bfi are smoothed with an 11-sample (1.5 m) Hanning filter, so that the two measurements have comparable vertical resolution.
Some preconditions are necessary for this method to work well. First, the susceptibility of the formation must vary so that log features exist to correlate/anticorrelate. This is not usually a problem. Second, the remanent magnetization must be subparallel to the induced magnetization. This might not be the case in older strata whose position relative to the magnetic poles has changed as a result of plate motions. Third, it is assumed that the absolute ratio of the remanent to the induced magnetizations is constant. In fact, the remanent intensity depends on the type of sediment and the magnetic field intensity at the time of deposition as well as the ferrimagnetic mineral concentration (e.g., Tauxe, 1993, Williams et al., 1998). However, a change in this ratio will affect only the gradient of the linear regression, not its polarity. Fourth, since the peaks and troughs of the remanent and induced anomalies can be quite short (e.g., 1 m), the depth accuracy of the logs becomes important; if the peaks and troughs are not in phase, the correlation analysis will be disturbed. The SUMS takes measurements at a horizon some time before the NMRS (because it is higher up the tool string and the log is taken in the upward direction), and the ship's heave is never entirely removed from the tool motion resulting in (usually slight) depth offsets. Tool heave was especially significant in Hole 1095B (Fig. F2). Fifth, the magnetic induction logs from Hole 1095B contain spikes. We have removed these spikes and linearly interpolated values into the gaps left behind; however, the correlations from intervals where there were spikes cannot be confidently relied on. The fourth and fifth items are principally responsible for the inaccuracies in the results of the correlation analysis for Hole 1095B.