IODP Proceedings    Volume contents     Search



During Expedition 330, routine paleomagnetic and rock magnetic experiments were carried out aboard the JOIDES Resolution. Remanent magnetization was measured on archive core halves and on discrete cube samples taken from working core halves. All continuous archive halves were demagnetized in an alternating field (AF), whereas discrete samples were subjected to either stepwise AF or thermal demagnetization. Since the azimuthal orientations of core samples recovered by rotary drilling are not constrained, all magnetic data are reported relative to the sample core coordinate system. In this system, +x is perpendicular to the split-core surface and points into the working half (i.e., toward the double line), +z is downcore, and +y is orthogonal to x and z in a right-hand sense (Fig. F12).

Archive-half remanent magnetization data

Remanent magnetization of the archive halves, including any undisturbed sediment, was measured at 2 cm intervals using the automated pass-through direct-current superconducting quantum interference device (DC-SQUID) cryogenic rock magnetometer (2G Enterprises model 760R). An integrated in-line AF demagnetizer (2G Enterprises model 600) capable of applying peak fields up to 80 mT was used to progressively demagnetize the core. Demagnetization steps were typically (5), 10, (15), 20, (25), 30, 40, 50, (60), and (70) mT, where the fields in parentheses were not always performed.

With strongly magnetized materials, the maximum intensity that can be reliably measured (i.e., with no residual flux counts) is limited by the slew rate of the sensors. At a track velocity of 2 cm/s, it was possible to measure archive-half cores with a magnetization as high as ~10 A/m. Where even this slow measurement speed still resulted in residual flux counts, the data were nonetheless archived because they provide some indication of the character of the magnetization. Caution is therefore warranted in using archive-half data from strongly magnetized intervals. Although the baseline values measured just prior to and just after archive-half measurements were not saved in the database, the baseline drift, and thus the number of residual flux counts, can be determined indirectly from the archived directional and intensity data. We developed LabView software (WebTabularToMag) to reconstruct the baseline drift, allowing the residual flux counts to be logged while the data were converted for further processing.

A noise test was conducted on the magnetometer on a relatively calm day on station to assess the practical resolution of the instrument. This test was performed with a modified version of the Scripps Institution of Oceanography LabView code in order to allow more rapid communication with the magnetometer and therefore capture an unaliased record of wave noise. All three axes were sampled simultaneously every ~0.5 s, in contrast with the existing magnetometer software that sequentially accessed these data at a slower rate of ~3.5 s. The spectra from these magnetic noise measurements (Fig. F13) revealed a peak at the dominant period (8–12 s) of wave motion, corresponding to a moment of a ~1 × 10–9 Am2 and therefore a practical resolution of ~1 × 10–5 A/m for a core half.

The most recent compiled version of the LabView software (SRM_SAMPLE) is Build However, an uncompiled version of this program (BETA with speed control) was used throughout the expedition because this allowed the track speed to be adjusted, a critical parameter for hard rock expeditions. Two modifications to the program and the Galil motor system were made during the expedition. First, the speed at which the archive halves were moved when not measuring was increased to 20 cm/s. Second, the simultaneous sampling of the magnetometer axes described above was also incorporated into the magnetometer software midway through measurements on archive halves from Hole U1374A (a more rapid measurement routine was implemented on 14 January 2011 with Section 330-U1374A-31R-3). Together, these changes resulted in substantial time savings (on the order of 0.5 h per section with 6–8 demagnetization steps) and also allowed multiple measurements at each interval for weakly magnetized cores.

The response functions of the pick-up coils of the SQUID sensors have a full width of 7–8 cm at half height (Parker and Gee, 2002). Therefore, data collected within ~4 cm of piece boundaries (or voids) are significantly affected by edge effects. Consequently, all data points within 4.5 cm of piece boundaries (as documented in the curatorial record) were filtered out prior to further processing. It should be noted that edge effects may also occur in a contiguous core piece if substantial heterogeneity (in intensity or direction) is present in the piece. Such artifacts are more difficult to filter out, but calculating the average direction (using Fisher statistics) for each core piece provides a means of identifying these problems (see below).

Automated calculation of best-fit directions

Remanent magnetization directions were calculated for each 2 cm measurement using principal component analysis (PCA; Kirschvink, 1980). Manual selection of principal component directions was not practical given the substantial amount of data. Rather, we used an automated procedure to select the most linear segment of the demagnetization data. PCA was conducted for each permutation of four or more consecutive demagnetization steps (typically 6–8 steps). The following criteria were then used to calculate a single misfit value that could be minimized to select the best-fit direction and to assign a reliability index to this selection. First, the range of demagnetization steps must yield a fit with low scatter (e.g., Fig. F14A), as defined by the maximum angular deviation (MAD, as used in equations below) of the PCA fit. Second, the selected direction should trend toward the origin. Deviation from the origin (e.g., Fig. F14B, F14F) may indicate that an unresolved higher coercivity component is present or that a spurious magnetization was acquired during demagnetization. Deviation of the fit from the origin is calculated from the solid angle (α) between the unconstrained PCA fit and the fit anchored to the origin. Third, a particular fit is generally judged to be more reliable if a large percentage of the remanence (f) is used for the calculation (e.g., Fig. F14A, F14C). We use the vector difference sum (the cumulative sum of the magnetization vector removed at each demagnetization step) to characterize the percentage of the remanence, as this is monotonically decreasing. By selecting target values (αo, MADo, and fo) for each of these criteria, a single misfit value may be calculated for each trial PCA fit:

misfit = {(α/αo) + [(1 – f)/(1 – fo)] + (MAD/MADo)}.

Although minimizing the above misfit value finds a suitable best-fit direction in many cases, two additional constraints were introduced. Selecting the largest percentage of the remanence has the unfortunate side effect of preferentially excluding particularly high coercivity intervals (which in fact may provide the highest quality remanence data). We therefore allowed a best-fit direction to be anchored to the origin if the anchored fit represents a significant improvement (enforced through a penalty, arbitrarily set at 1.5, on anchored fits). The misfit for such anchored fits is given by

misfitA = {(CSD/αo) + [(1 – f)/(1 – fo)] + (MAD/MADo)} × 1.5,

where the circular standard deviation (CSD) of the remanence vectors used for the fit replaces α for the unconstrained fit. Finally, where a substantial low-coercivity component (e.g., from drilling) is present, the natural remanent magnetization step, and optionally the first demagnetization step, may be eliminated prior to finding the lowest misfit.

The best-fit PCA was calculated using target values of αo = 5°, MADo = 3°, and fo = 0.6, unless otherwise noted. The values of the misfit parameters provide a qualitative measure of reliability and have been used to filter out less reliable data. If all three parameters contributing to the misfit equal the target values, the misfit value will be equal to 3.0 for a free (not anchored) PCA fit. The minimum misfit distributions for the three data sets from Holes U1372A, U1373A, and U1374A are shown in Figure F15. The three cumulative distribution curves intersect at a misfit value of ~3.4, and for each data set ~40% of the data have misfit values lower than this value. We elected to retain only the most reliable 40% of the data for further analysis, which corresponds to a cutoff value of 3.40 for Hole U1372A and 3.42 for Holes U1373A and U1374A. The histograms in Figure F15 illustrate the distributions of MAD, α, and f values for the 40% of the data with misfit values below the cutoff value.

This automated selection routine generally provides a reasonable fit to the demagnetization data (Fig. F14). Low-coercivity overprints, such as drilling-induced remanent magnetization, are ignored whether they constitute a small or substantial percentage of the natural remanent magnetization (Fig. F14C, F14D, F14F, F14G). The program does not find a satisfactory straight line fit for samples exhibiting curvilinear behavior, as one would expect (Fig. F14B). Anchored PCA fits constitute 12% of the directions with misfits less than the cutoff value. Although some high-coercivity samples with apparently reliable directions are excluded (Fig. F14H), any inclination bias due to such anchored fits is minimized.

Note that the intensity reported for such PCA directions represents the length of the projection of the lowest- and highest-treatment vectors used in the PCA calculation onto the best-fit direction. Because the origin is not included in the PCA calculation and the remanence remaining after the highest treatment may be significant, the resulting characteristic remanent magnetization intensity values are systematically lower than those derived from the remanence at the lowest demagnetization step adopted for the PCA calculation.

Piece-average directions

The mean remanent magnetization of individual archive-half pieces >9 cm was calculated as an additional internal reliability check. All 2 cm individual PCA directions with sufficiently low misfits (see above) were averaged using Fisher statistics; the reported PCA intensity is the vector mean. Heterogeneous magnetization directions in a single piece can be identified from high values of the associated CSD. Figure F16 shows the data from a single piece and illustrates two aspects of the adopted piece-averaging approach that should be kept in mind. First, the piece average is calculated using the curated piece lengths assuming that all subpieces have the same relative orientation. Thus, any subpieces that have rotated relative to one another may adversely affect the mean direction (e.g., the break between subpieces at ~65 cm in Section 330-U1372A-9R-2 [Piece 1] in Fig. F16). Second, when boundaries separating flows or lithologic units with different directions or intensities occur within a piece, the average direction may be erroneous. Conversely, if they are sufficiently different, inclination variations in a piece may facilitate recognition of unit boundaries. Hence, piece averages with CSD > 20° were excluded from further shipboard analysis. However, it is likely that careful analysis of changes at subpiece boundaries would allow some of these rejected data to be retained in later shore-based analysis of the data.

Discrete sample data

All discrete samples taken from the working halves for shipboard magnetic analysis were 8 cm3 cubes. Although standard 2.5 cm diameter minicores are more commonly used, the cubic samples were preferred because they should have a more reproducible vertical reference (based on a double-bladed saw cut perpendicular to the core length) than the minicores (where the fiducial arrow on the split-core face must be transferred to the long axis of the sample). Discrete samples for shipboard analysis were selected on the basis of identified flow boundaries, the in situ confidence index (see “Igneous petrology and volcanology”), and results from archive-half stepwise AF demagnetization.

The remanent magnetization of discrete samples was measured exclusively with the JR-6A spinner magnetometer following tests of discrete sample measurements on the cryogenic magnetometer (described below). For samples measured on the spinner magnetometer, the automated sample holder was used, providing the most accurate discrete sample remanent magnetization directions and intensities. A set of four standard samples (previously measured in the shielded room at the Scripps Institution of Oceanography and archived as SIO 1, SIO 2, SIO 6, and SIO 8) was measured on the spinner magnetometer. The directions obtained were typically within 1°–2° of the known values, and intensities were accurate to within a few percent.

The current version of the software for discrete sample measurements on the cryogenic magnetometer (SRM_DISCRETE, Build allows for measurement in 24 separate orientations. The four SIO standards were measured in all orientations (Fig. F17). These experiments revealed considerable scatter among these 24 positions, ranging from ~10°–15° for strongly magnetized samples (~1–10 A/m) up to and exceeding 30° for weaker samples. The origin of this scatter is unknown. For this reason, all discrete samples were measured using the spinner magnetometer. The archive-half data measured on the cryogenic magnetometer agree with discrete sample data, and we have no reason to question their validity.

Discrete samples were subjected to stepwise AF demagnetization using a DTech AF demagnetizer (model D-2000) capable of peak fields as high as 200 mT. Typically 12–15 AF demagnetization steps were used, with the peak field values selected according to sample coercivity. The residual magnetic field at the demagnetizing position in this equipment was 5–50 nT, with the higher value apparently associated with vessel headings that are nearly east–west (the ambient field in this case is parallel to the mu-metal shield housing the instrument). Despite this low residual field, spurious magnetizations were nonetheless observed at high peak fields for some samples. At peak fields of >40 mT, samples were therefore typically demagnetized and measured twice (once after demagnetization along the sample +x, +y, and +z directions and again after demagnetization along the –x, –y, and –z directions) to identify and compensate for any spurious remanent magnetization acquisition caused by a bias field in the demagnetizing coil. An alternative method in which the last axis demagnetized was systematically varied was also used for some discrete sample demagnetizations.

Discrete samples were thermally demagnetized using a Schonstedt Thermal Specimen Demagnetizer (model TSD-1) capable of demagnetizing specimens up to 700°C. Each sample boat for thermal demagnetization included as many as nine samples, and sample orientations were varied at alternative steps to allow any interactions between adjacent samples to be identified. Samples were held at the desired temperature for 40 min and then cooled in a chamber with a residual magnetic field of <30 nT. Magnetic susceptibility was measured using a Bartington MS2F point magnetic susceptibility meter after every heating step to monitor thermal alteration of magnetic minerals during heating.

In addition to standard paleomagnetic measurements, the anisotropy of magnetic susceptibility was determined for all discrete samples using the KLY 4S Kappabridge with the software AMSSpin (Gee et al., 2008). The susceptibility tensor and associated eigenvectors and eigenvalues were calculated off-line following the method of Hext (1963). All bulk susceptibility values reported for discrete samples are based on a nominal 8 cm3 volume.

Inclination-only analysis

For azimuthally unoriented cores the simple arithmetic mean of inclination data will be biased to shallower values (e.g., Kono, 1980; McFadden and Reid, 1982; Arason and Levi, 2010). To compensate for this bias, we used the inclination-only statistics of Arason and Levi (2010) to calculate the overall mean inclination from discrete sample demagnetization data from each drill site, as well as the mean inclinations for each lithologic unit from the more abundant archive-half data. As one estimate, we simply calculate the inclination-only mean and associated α95 uncertainties (Arason and Levi, 2010) from all 2 cm PCA directions (with misfit below the cutoff value) from a single lithologic unit. Alternatively, the same inclination-only technique may be applied by averaging all Fisher piece-average inclinations from a single lithologic unit. These two approaches provide similar results, differing primarily in the uncertainty estimates where n is small: the α95 values for the piece-average data are necessarily larger than those calculated from the greater number of 2 cm measurements. The former error estimate is likely more realistic because the individual 2 cm data points are not all statistically independent.

Although the Arason and Levi technique is more robust than previous inclination-only methods (e.g., Kono, 1980; McFadden and Reid, 1982), this technique nonetheless fails to converge under certain circumstances. For example, if inclinations are steep and the scatter is substantial or if dual polarities (also with steep inclination) are present, no maximum likelihood estimate is possible. If initially no solution was found, we recalculated with only the dominant polarity data if <10% of the data were of opposite polarity (splitting the data set arbitrarily at an inclination of 0°). For unit average inclinations based on 2 cm archive-half data, means are reported only for units with more than one inclination measurement. In the case of piece-average directions, single inclination measurements were retained because each typically represents the mean direction from multiple samples.